Chapter 10, CBS

Chapter 10.
COMPUTER-ASSISTED OPTIMIZATION OF PHARMACEUTICAL FORMULATIONS AND PROCESSES

Bhupinder Singh, R. K. Gupta and Naveen Ahuja
INTRODUCTION Pharmaceutical product and process design problems are normally characterized by multiple objectives (Fonner et al., 1970; Banker & Anderson, 1987). In an attempt to achieve such objectives, a pharmaceutical scientist has to fulfill various control limits for a formulation. Some characteristics for ascertaining the control limits, common for all dosage forms include unit cost, physico-chemical stability and physiological availability of the active ingredient. Apart from meeting these common traits, a particular dosage form must also satisfy certain "individual" quality performance characteristics. In case of tablets, for instance, hardness, friability, disintegration test, dissolution rate, etc., would be most appropriate to control. As most of the objectives of a formulation are often differing, accepting a suitable compromise between one or more properties (e.g., dissolution rate at the expense of hardness) usually becomes unavoidable. Thus the primary aim of the formulator is to find a suitable compromise under the given set of restrictions rather than designing the best formulation (Fonner et al., 1970; Shekh et al., 1980; Banker & Anderson, 1987; Podczeck, 1996). Since decades, drug formulations are being developed by trial and error. The previous experience, knowledge and wisdom of the formulator have been the key factors in formulating new dosage forms or modifying the existing ones. At times, when the developer is intuitive, skilled and "fortunate", such nonsystematic approach may yield surprisingly successful outcomes. Invariably however, when skill, wisdom or luck is not in his favour, it leads to squandering remarkable volume of time, energy and resources (Lewis, 2002). Though a new product may be developed, yet it may retain any defects or problems inherent in the old product. The modification of the formulation or the process is carried out by studying the influence of composition and process variables on dosage form characteristics, changing one separate/single factor at a time (COST), while keeping others as constant. Using this 'COST' approach, the solution of a specific problematic property can be achieved somehow, but attainment of the true optimum composition or process is never guaranteed (Tye, 2004). This may be ascribed to the presence of interactions, i.e., the influence of one or more factors on others. The final product may be satisfactory but mostly sub-optimal, as a better formulation might still exist for the studied conditions. Therefore, the conventional 'COST' approach of drug formulation
273
PDF created with FinePrint pdfFactory trial version http://www.fineprint.com
274 Pharmaceutical Product Development
development suffers from several pitfalls (Fonner et al., 1970; Schwartz et al., 1973; Belloto Jr. et al., 1985; Stetsko, 1986; Lewis et al., 1999; Lewis, 2002; Myers, 2003; Singh & Ahuja, 2004). These drug product inconsistencies are generally due to inadequate knowledge of causal factor and response relationship(s). The said approach is quite: time consuming,
energy utilizing, uneconomical, unpredictable, unsuitable to plug errors, ill-suited to reveal interactions, and yielding only workable solutions.
Computer-based systematic design and optimization techniques, on the other hand, have widely been practiced to alleviate such inconsistencies (Irvin & Notari, 1991; Singh et al., 2005a; Tye, 2004). Such techniques are usually referred to as 'computer-aided dosage form design' (CADD). Their implementation invariably encompasses the statistical design of experiments (DoE), generation of mathematical equations and graphic outcomes, thus depicting a complete picture of variation of the response(s) as a function of the factor(s) (Doornbos & Haan, 1995; Schwartz & Connor, 1996; Lewis, 2002). Optimization techniques possess much greater benefits, as they surmount several pitfalls inherent to the traditional approaches (Lewis et al., 1999; Tye, 2004). The meritorious features that such techniques offer include: best solution in the presence of competing objectives,
fewer experiments needed to achieve an optimum formulation, significant saving of time, effort, materials and cost, easier problem tracing and rectification, possibility of estimating interactions, simulation of the product or process performance using model equation(s), and comprehension of process to assist in formulation development and subsequent scale-up.
Thus, the trial and error COSTapproach requires many experiments for little gain in information about the system under investigation. In contrast, systematic optimization methodology offers an organized approach that connects experiments in a rational manner, giving more precise information from fewer experiments. (Tye, 2004). Hence of late, DoE optimization techniques have become a regular practice globally in the design and development of an assortment of dosage forms. However, implementation of such rational approaches usually involves a great deal of mathematical and statistical complexities. Manual calculation of such optimization data being quite cumbersome, often calls for the indispensable help of an apt computer interface (Banker & Anderson, 1987). With the advent of the pertinent computer software coupled with the powerful hardware, the erstwhile arduous task has grossly been simplified and streamlined. The computational hiccups involved during optimization of pharmaceutical products have greatly been reduced by the availability of comprehensive and user-interactive software (Lewis et al., 1999; Singh, 2003). Conduct of systematic DoE studies using computers usually obviates the requirement of an in-depth knowledge of statistical and mathematical precepts. Nevertheless, the comprehension of varied concepts underlying these methodologies is certainly a must for the successful conduct of optimization studies. The information on such rational techniques, however, lies scattered in different books and journals, and complete description on variegated vistas of optimization is not available from a single textual source. The current chapter is an attempt to acquaint the reader with the fundamental principles and precepts of systematic optimization methodologies, and to present a concise and lucid account on the use of its methodologies in the computer-assisted design and development of wide-ranging pharmaceutical formulations and processes.
Chapter 10.
Computer-Assisted Optimization of Pharmaceutical Formulations and Processes
275
1.2
OPTIMZATION: BASIC CONCEPTS AND TERMINOLOGY
The word, optimize simply means to make as perfect, effective or functional as possible (Schwartz & Connor, 1996). The term optimized has been used in the past to suggest that a product has been improved to accomplish the objectives of a development scientist (Singh & Ahuja, 2004). However, today the term implies that computers and statistics have been utilized to achieve the objective(s). With respect to drug formulations or pharmaceutical processes, optimization is a phenomenon of finding "the best" possible composition or operating conditions (Lewis, 2002). Accordingly, optimization has been defined as the implementation of systematic approaches to achieve the best combination of product and/or process characteristics under a given set of conditions (Tye, 2004). 1.2.1 Variables Design and development of drug formulation or pharmaceutical process usually involve several variables (Lewis, 2002). The input variables, which are directly under the control of the product development scientist, are known as independent variables, e.g., compression force, excipient amount, mixing time, etc. Such variables can either be quantitative or qualitative. Quantitative variables are those that can take numeric values (e.g., amount of disintegrant, suspending agent, temperature, time, etc.) and are continuous. Instances of qualitative variables, on the other hand, include the type of emulgent, solubilizer or tabletting machine. Their influence can be evaluated by assigning dummy values to them. The independent variables, which influence the formulation characteristics or output of the process, are labeled as factors. The values assigned to the factors are termed as levels, e.g., 30 and 50are the levels for the factor, temperature. The restrictions placed on the factor levels are known as constraints (Bolton, 1990; Schwartz & Connor, 1996). The characteristics of the finished drug product or the in-process material are known as dependent variables, e.g., drug release profile, friability, size of tablet granules, disintegration time, etc. (Box et al., 1960; Bolton, 1990; Doornbos & Haan, 1995; Lewis et al., 1999; Montgomery, 2001). Popularly termed as response variables, these are the measured properties of the system to estimate the outcome of the experiment. Usually these are the direct function(s) of any change(s) in the independent variables. Accordingly, a drug formulation (product) with respect to optimization techniques can be considered as a system, whose output (Y) is influenced by a set of controllable (X) and uncontrollable (U) input variables via a transfer function (T). Fig. 10. 1 depicts the same graphically (Cochran & Cox, 1992; Doornbos & Haan, 1995). The nomenclature of T depends upon the predictability of the output as an effect of change of the input variables. If the output is totally unpredictable from the previous studies, T is termed as black box. The term, white box is used for a system with absolutely true predictability, while the term, gray box is used for moderate
U
Fig. 10. 1. System with controllable input variables (X), uncontrollable input variables (U), transfer function (T) and output variables (Y).
predictability. Using optimization methods, the attempt of the formulator is to attain a white box or nearly white box status from the erstwhile black or gray box status observed in the traditional studies (Lewis et al., 1999; Montgomery, 2001). The more is the number of variables in a given system, the more complicated becomes the job of optimization. Nevertheless, regardless of the number of variables, there exists a distinct relationship between a given response and the independent variables (Schwartz & Connor, 1996). 1.2.2 Effect, Interaction and Confounding The magnitude of the change in response caused by varying the factor level(s) is termed as an effect. The main effect is the effect of a factor averaged over all the levels of other factors (Bolton, 1990; Cochran & Cox, 1992). However, an interaction is said to occur, when there is "lack of additivity of factor effects". This implies that the effect is not directly proportional to the change in the factor levels (Bolton, 1990). In other words, the influence of a factor on the response is nonlinear (Doornbos & Haan, 1995; Lewis et al., 1999). Also, an interaction may be said to take place when the effect of two or more factors is dependent on each other, e.g., effect of factor A depends on the level given to the factor B (Montgomery, 2001; Stack, 2003; Tye, 2004). The measured property of the interacting variables not only depends on their fundamental levels, but also on the degree of interaction between them. Fig. 10. 2 illustrates the concept of interaction graphically. The term orthogonality is used, if the estimated effects are due to the main factor of interest and are independent of interactions (Box et al., 1960; Bolton, 1990). Conversely, lack of orthogonality (or independence) is termed as confounding or aliasing (Cochran & Cox, 1992). When an effect is confounded (or aliased), one cannot assess how much of the observed effect is due to the factor under consideration. The effect is influenced by other factors in a manner that cannot easily be explored. The measure of the degree of confounding is known as resolution (Tye, 2004). Confounding is a bias that must be controlled by suitable selection of the design and data analysis. Interaction, on the other hand, is an inherent quality of the data, which must be explored. Confounding must be assessed qualitatively; while interaction may be tested more quantitatively (Stack, 2003). 1.2.3 Code transformation The process of denoting a natural variable into a dimensionless coded variable Xi such that the central value of experimental domain is zero is known as coding or normalization (Bolton, 1990; Schwartz & Connor, 1996; Lewis et al., 1999; Montgomery, 2001). Various salient features of the transformation include:
NO INTERACTION D I S S O L U T I O N Low DRUG D I S S O L U T I O N Low DRUG
INTERACTION Low Polymer level
Low polymer level
High Polymer level
High Polymer level High
High
Fig. 10. 2. Diagrammatic depiction of interaction. The unparallel lines in the figure 2(b) describe the phenomenon of interaction between drug and polymer levels affecting drug dissolution. Linear ( ); nonlinear lines (...).
Chapter 10.
277
depiction of effects and interaction using signs (+) or (-), allocation of equal significance to each axis, easier calculation of the coefficients, easier calculation of the coefficient variances, easier depiction of the response surfaces, and orthogonality of the effects.
Generally, the various levels of a factor are designated as -1, 0 and +1, representing the lowest, intermediate (central) and the highest factor levels investigated, respectively. For instance, if starch, a disintegrating agent, is studied as a factor in the range of 5 to 10% (w/w), then codes -1 and +1 signify 5% and 10% concentrations, respectively. The code 0 would represent the central point at the mean of the two extremes, i.e., 7.5% w/w. 1.2.4 Factor Space The dimensional space defined by the coded variables is known as factor space (Lewis et al., 1999). Fig. 10. 3 illustrates the factor space for two factors on a bidimensional (2-D) plane during a typical tablet compression process. The part of the factor space that is investigated experimentally for optimization is the experimental domain (Doornbos & Haan, 1995; Lewis et al., 1999). Also known as the region of interest, it is enclosed by the upper and lower levels of the variables. The factor space covers the entire figure area and extends even beyond it, whereas the design space of the experimental domain is the square enclosed by X1 = 1, X2 = 1. 1.2.5 Experimental Design Conduct of an experiment and subsequent interpretation of its experimental outcome are the twin essential features of the general scientific methodology (Cochran & Cox, 1992; Lewis, 2002). This can be accomplished only if the experiments are carried out in a systematic way and the inferences are drawn accordingly. An
Span 80 (% w/v)
Fig. 10. 3. Quantitative factors and the factor space. The axes for the natural variables, Ethyl cellulose:Drug and Span 80 are labelled as U1 and U2 and those of the corresponding coded variables as X 1 and X2.
experimental design is the statistical strategy for organizing the experiments in such a manner that the required information is obtained as efficiently and precisely as possible (Kettaneh-Wold, 1991; Cochran & Cox, 1992). Runs or trials are the experiments conducted as per the selected experimental design (Bolton, 1990; Doornbos & Haan, 1995). Such DoE trials are arranged in the design space in such a way that reliable and consistent information is achievable with minimum experimentation. The layout of the experimental runs in a matrix form, as per the experimental design, is known as design matrix (Lewis et al., 1999). The choice of the design depends upon the proposed model, shape of the domain and the objective of the study. Primarily, the experimental (or statistical) designs are based on the principles of randomization (the manner of allocations of treatments to the experimental units), replication (the number of units employed for each treatment) and error control or local control (grouping of specific type of experiments to increase the precision) (Das & Giri, 1994; Montgomery, 2001). 1.2.6 Response Surfaces Conduct of DoE trials, as per the chosen statistical design, yields a series of data on response variables explored. Such data can be suitably modeled to generate mathematical relationship between the independent variables and the dependent variable. Graphical depiction of the mathematical relationship is known as response surface (Lewis et al., 1999; Myers, 2003). A response surface plot is a 3-D graphical representation of a response plotted between two independent variables and one response variable. The use of 3-D response surface plots allows understanding of the behaviour of the system by demonstrating the contribution of the independent variables. The geometric illustration of a response, obtained by plotting one independent variable versus another, while holding the magnitude of response level and other variables as constant, is known as a contour plot (Singh & Ahuja, 2004). Such contour plots represent the 2-D slices of 3-D response surfaces. The resulting curves are called contour lines. Fig. 10. 4 depicts the response surface and contour lines for the response variable of percent drug entrapment in liposomal vesicles of nimesulide (Singh et al., 2005b). For complete response depiction amongst 'n' independent variables, a total of nC2 number of response surfaces and contour plots may be required. In other words, 1, 3, 6 or 10 number of 3-D and 2-D plots are needed to provide depiction of each response for 2, 3, 4 or 5 number of variables, respectively.
5 0 -6 5
95
6 5 -8 0
8 0 -9 5
80
65
1 -1 0 PL 1 -1 0 CHOL
50
(a)
(b)
Fig. 10. 4. (a) A typical response surface plotted between a response variable, percent drug entrapment and two factors, cholesterol (CHOL) and phospholipid (PL) in case of vesicular systems; (b) the corresponding contour lines.
Chapter 10.
279
1.3 MATHEMATICAL MODELS Mathematical model, simply referred to as the model, is an algebraic expression defining the dependence of a response variable on the independent variable(s). Mathematical models can either be empirical or theoretical (Doornbos & Haan, 1995; Lewis, 2002). An empirical model provides a way to describe the factor-response relationship. It is most frequently, but not invariably, a set of polynomial equations of a given order (Box & Draper, 1987; Myers & Montgomery, 2002). Most commonly used linear models are shown in equations 1-3:
E( y ) = 0 + 1X1 + 2 X 2 E( y ) = 0 + 1X 1 + 2 X 2 + 12 X 1X 2
2 + 22 X 2 E(y ) = 0 + 1X1 + 2 X 2 + 12 X 1X 2 + 11X 1 2
(1) (2) (3)
where, E(y) represents the measured response, X i, the value of the factors, and 0, i, ii, and ij are the constants representing the intercept, coefficients of first-order terms, coefficients of second-order quadratic terms and coefficients of second-order interaction terms, respectively. Equations 1 and 2 are linear in variables, representing a flat surface and a twisted plane in 3-D space, respectively. Equation 3 represents a linear second-order model that describes a twisted plane with curvature, arising from the quadratic terms. A theoretical model or mechanistic model may also exist or be proposed. It is most often a nonlinear model, where transformation to a linear function is usually not possible. However, theoretical relationships are rarely employed in pharmaceutical product development. 1.4 FACTOR STUDIES Systematic screening and factor influence studies are usually carried out as a prelude to DoE optimization (Lewis, 2002). These are often sequential stages in the development process. Screening methods are used to identify important and critical effects (Murphy, 2003). Factor studies aim at quantitative determination of the effects as a result of a change in the potentially critical formulation or process parameter(s). Such factor studies usually involve statistical experimental designs, and the results so obtained provide useful leads for further response optimization studies. 1.4.1 Screening of Influential Factors As the term suggests, "screening" is analogous to separating "rice" form "rice husk", where "rice" is a group of factors with significant influence as response, and "husk" is a group of the rest of the noninfluential factors. A product development scientist normally has numerous possible input variables to be investigated for their impact on the response variables. During initial stages of optimization, such input variables are explored for their influence on the outcome of the finished product to see if they are factors (Lewis et al., 1999; Myers, 2003). The process, called as screening of influential variables, is a paramount step. An input variable, identified as a factor increases the chance of success, while an input variable that is not a factor has no consequence. Further, an input variable falsely identified as a factor unduly increases the effort and cost, while an unrecognized factor leads to wrong picture and a true optimum may be missed (Lewis, 2002). The entire exercise aims at selecting the active factors and excluding the redundant variables, but not at obtaining complete and exact numerical data on the system properties. Such reduction in the number of factors becomes necessary before the pharmaceutical scientist invests the human, financial and industrial resources in more elaborate studies. This phase may be omitted if the process is known well enough from the analogous studies. Even after elimination of the noninfluential variables, the number of factors may still be too large to optimize in terms of available resources (time, manpower, equipment, etc.). Generally, more influential variables are optimized, keeping the less influential ones as constant at their best levels. The number of experiments is kept as small as possible to limit the volume of work carried out during the initial stages (Singh et al., 2005a). The experimental designs employed for the purpose are commonly termed as
screening designs (Murphy, 2003; Myers, 2003). Usually these designs are first-order and low-resolution designs. 1.4.2 Factor Influence Study Having screened the influential variables, a more comprehensive study is subsequently undertaken to quantify the effect of factors, and to determine the interactions, if any (Bolton, 1990; Lewis et al., 1999; Montgomery, 2001). Herein, the studied experimental domain is less extensive, as quite fewer active factors are studied. The models used for this study are neither predictive nor capable of generating a response surface. The number of levels is usually limited to two (i.e, at the extremes). However, sufficient experimentation is carried out to allow for the detection of interactions amongst factors. The experiments conducted at this step may often be reused during optimization or response modeling phase by augmenting with additional design points. Central points (i.e., at the intermediate level), if added at this stage, are not included in the calculation of model equations (Doornbos & Haan, 1995; Lewis, 2002). Nevertheless, they may prove to be useful in identifying the curvature in the response, in allowing the reuse of the experiments at various stages; and if replicated, in validating the reproducibility of the experimental study. 1.5 OPTIMIZATION METHODOLOGIES Broadly, DoE optimization methodologies can be categorized into two classes, i.e., simultaneous optimization, where the experimentation is completed before the optimization takes place and sequential optimization, where experimentation continues sequentially as the optimization study proceeds (Doornbos & Haan, 1995; Schwartz & Connor, 1996). The whole optimization endeavour is attempted in several steps, commencing from the screening of influential factors, factor influence studies, and applying one or more of the various techniques to reach an optimum (Lewis et al., 1999; Singh & Ahuja, 2004). 1.5.1 Simultaneous Optimization Methodology Generally termed as response surface methodology (RSM), simultaneous optimization approach is a modeldependent technique (Doornbos & Haan, 1995). The key elements in its implementation encompass, the experimental designs, mathematical models and the graphic outcomes. One or more selected experimental response(s) is (are) recorded for a set of experiments, carried out in a systematic way, to predict an optimum and the interaction effects. This is followed by the determination of the mathematical model for each response in the zone of interest, i.e., the experimental domain. Rather than estimating the effects of each variable directly, RSM involves fitting the coefficients into the model equation of a particular response variable and mapping the response, i.e., studying the response over whole of the experimental domain in the form of a surface (Lewis et al., 1999; Myers & Montgomery, 2002; Myers, 2003). Principally, RSM is a group of statistical techniques for empirical model building and model exploitation (Box & Draper, 1987; Myers, 2003). By careful design and analysis of experiments, it seeks to relate a response to a number of predictors affecting it by generating a response surface. A response surface is an area of space defined within upper and lower limits of the independent variables depicting the relationship of these variables to the measured response. 1.5.1.1 Experimental designs The designs used for simultaneous methods are frequently referred to as response surface designs. Various experimental designs frequently involved in the execution of RSM can broadly be classified as: A Factorial design and modifications B Central Composite design and modifications C Mixture designs D D-optimal designs
Chapter 10.
281
A. Factorial design and modifications Factorial designs (FDs; full or fractional) are the most frequently used response surface designs. These are generally based upon first-degree mathematical models (Bolton, 1990; Myers & Montgomery, 2002; Li, 2003). Full FDs involve studying the effect of all the factors (n) at various levels (x), including the interactions amongst them, with the total number of experiments as x n. The simplest FD involves study of two factors at two levels, with each level coded suitably. FDs are said to be symmetric, if each factor has same number of levels, and asymmetric, if the number of levels differs for each factor (Lewis et al., 1999). Besides RSM, the design is also used for screening of influential variables and factor influence studies. Fig. 10. 5 represents a 22 and 23 FD pictorially, where each point represents an individual experiment. The mathematical model associated with the design consists of the main effects of each variable plus all the possible interaction effects, i.e., interactions between the two variables, and in fact, between as many factors as are there in the model. The mathematical model generally postulated for FDs is given as Equation 4.
Y = 0 + 1X1... + 12 X1X 2 ... + 123 X1X 2 X 3 ... +
(4)
where, i, ij and represent the coefficients of the variables and the interaction terms, and the random experimental error, respectively. The effects (coefficients) in the model are estimated usually by multiple linear regression analysis (MLRA). The topic is discussed in greater detail later under section 1.5.1.2, 'Model selection'. Their statistical significance is determined and then a simplified model can be written. In a full FD, as the number of factors or factor levels increases, the number of required experiments exceeds the manageable levels. Moreover with a large number of factors, it is plausible that the highest-order interactions have no significant effect. In such cases, the number of experiments can be reduced in a systematic way, with the resulting design called as fractional factorial designs (FFD). An FFD is a finite fraction (1/x r) of a complete or full FD, where r is the degree of fractionation and x n-r is the total number of experiments required (Doornbos & Haan, 1995; Lewis et al., 1999; Li, 2003). However, by reducing the number of experiments, the ability to distinguish some of the factor effects is partly sacrificed, i.e., the effects can no longer be uniquely estimated. The degree of fractionation should not be large because this leads to confounding of factor effects not only with the interactions but also with other factor effects. Table 10. 1 illustrates the layout of the experiments as per an FD. Lines 1-4 of columns 1 and 2 show a 2 2 design for two factors, lines 1-8 of columns 1-3 a 2 3 design for three factors and lines 1-16 of columns 1-4 a 2 4 design for four factors. Lines 1-16 of columns 1-5 describe a 2 5-1 FFD with degree of fractionation, r, as 1 (Menon et al., 1996).
Fig. 10. 5. (a) 22 full factorial design (b) 23 full factorial design.
Table 10. 1. Experimental layout as per full and fractional factorial designs for two to five factors
Experiment run 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 X1 -1 +1 -1 +1 -1 +1 -1 +1 -1 +1 -1 +1 -1 +1 -1 +1 X2 -1 -1 +1 +1 -1 -1 +1 +1 -1 -1 +1 +1 -1 -1 +1 +1 X3 -1 -1 -1 -1 +1 +1 +1 +1 -1 -1 -1 -1 +1 +1 +1 +1 X4 -1 -1 -1 -1 -1 -1 -1 -1 +1 +1 +1 +1 +1 +1 +1 +1 X5 +1 -1 -1 +1 -1 +1 +1 -1 -1 +1 +1 -1 +1 -1 -1 +1
Table 10. 2. A Plackett-Burman design for 8 experiments

Experiment run 1 2 3 4 5 6 7 8 X1 +1 -1 -1 +1 -1 +1 +1 -1 X2 +1 +1 -1 -1 +1 -1 +1 -1 X3 +1 +1 +1 -1 -1 +1 -1 -1 X4 -1 +1 +1 +1 -1 -1 +1 -1 X5 +1 -1 +1 +1 +1 -1 -1 -1 X6 -1 +1 -1 +1 +1 +1 -1 -1 X7 -1 -1 +1 -1 +1 +1 +1 -1
Plackett-Burman Design (PBD) is a special two-level FFD used generally for screening of (K = N-1) factors, where N is a multiple of 4 (Plackett & Burman, 1946). Also known as Hadamard design or symmetrically reduced 2k-r FD, the design is easily constructed. Table 10. 2 presents the PBD layout for 8 experiments. Star designs can be used to provide a simple way to fit a quadratic model (Cochran & Cox, 1992; Doornbos & Haan, 1995). These designs alleviate the problem encountered with FFDs, which do not allow detection of curvature unless more than two levels of a factor are chosen. The number of experiments required in a star design is given by 2n + 1. A central experimental point is located, from which other factor combinations are generated by moving the same positive and negative distance (= step size). For two factors, the star design is simply a 22 FD rotated over 45with an additional center point. The design is invariably orthogonal and rotatable (Lewis, 2002). In general, the first-order experimental designs must enable estimation of the first-order effects, preferably free from interference by the interactions between factors and other variables. These designs should also allow the testing for the goodness of fit of the proposed model (Myers & Montgomery, 2002; Lewis, 2002). Even if they are able to determine the existence of the curvature of the response surface, they should normally be used only in the absence of curvature of the response surface.
Chapter 10.
283
B Central composite design and its modifications Also known as Box-Wilson design, it is the most often used design for second-order models (Box & Wilson, 1951). Central composite design (CCD) is comprised of the combination of a two-level factorial points (2n), axial or star points (2n) and a central point. Thus the total number of factor combinations in a CCD is given by 2n +2n + 1. The axial points for a two-factor problem include, ( , 0) and (0, ), where is the distance of the axial points from the center. A two factor CCD is identical to a 3 2 FD with square experimental domain ( is 1), as shown in Fig. 10. 6 (a). On the other hand, when a is 2 = 1.414, the experimental domain is spherical in shape, as shown in Fig. 10. 6 (b). A face centered cube design (FCCD) results when the same positive and negative distance is taken from the center in a CCD (Doornbos & Haan, 1995). A rotatable CCD (RCCD) is identical to FCCD except that the points defined for the star design are changed to [ (2 n )1/4 , 0] and those generated by the FD remain unchanged. In this way, the design generates information equally well in all the directions, e.g., the variance of the estimated response is same at all the points on a sphere centered at the origin (Schwartz & Connor, 1996; Lewis et al., 1999). The second-order polynomial for two factors, generally used for the composite designs, is given as Equation 5.
2 + 22 X2 Y = 0 + 1X1 + 2 X2 + 12 X1X2 + 11X1 2 +
(5)
Box-Behnken Design (BBD) is a specially made design that requires only 3 levels (-1, 0, 1) (Box & Behnken, 1960). It overcomes the inherent pitfalls of CCD, where each factor has to be studied at 5 levels (except for 2 factors with = 1, where the number of levels per factor is 3), thus the number of experiments increases with rise in the number of factors (Myers, 2003). A BBD is an economical alternative to CCD. Also called as orthogonal balanced incomplete block design, these are available for 3 to 10 factors. Because the design involves study at three levels, the quadratic model is considered to be most appropriate. Table 10. 3 lists a documented instance of a BBD design layout for preparing sustained release pellets, employing 15 experiments with 3 factors at three levels each (Kramar et al., 2003). Doehlert design is an equiradial design where experimental points are uniformly distributed on the surface of a hypersphere, thus providing a good basis for interpolation (Doehlert, 1970; Doornbos & Haan, 1995).
Fig. 10. 6.
(a) Central composite design (square domain) with = 1. (b) Central composite design (spherical domain) with = 1.414.
Table 10. 3. Design layout as per Box-Behnken design.*

Run 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Factors -1 X1: Plasticizer concentration (%) X2: Polymer ratio X3: Quantity of coating dispersion (g) 10 2/1 300 Variable factors X1 30 30 10 10 30 30 10 10 20 20 20 20 20 20 20 X2 6/1 2/1 6/1 2/1 4/1 4/1 4/1 4/1 6/1 6/1 2/1 2/1 4/1 4/1 4/1 Levels 0 20 4/1 500 1 30 6/1 700 Y1: Cumulative percent drug release after 3 h Y2: Cumulative percent drug release after 4 h Y3: Cumulative percent drug release after 6 h X3 500 500 500 500 700 300 700 300 700 300 700 300 500 500 500 Y1 20.0 0.8 33.0 1.0 42.4 0.9 66.1 1.3 15.4 1.1 53.9 1.4 32.9 0.8 82.4 2.0 10.8 1.0 47.4 1.1 22.1 1.2 75.3 0.9 26.5 1.0 24.0 1.5 25.0 1.2 Response variables Y2 27.5 1.1 45.4 1.1 58.7 1.5 85.6 1.2 21.1 1.6 71.7 1.8 46.5 1.3 91.0 2.0 14.8 1.1 62.7 1.3 30.4 1.2 87.1 2.0 36.4 1.5 32.9 2.0 34.9 1.5 Y3 38.0 1.2 65.2 1.1 80.5 0.9 94.1 2.0 29.5 1.1 85.1 1.0 68.5 1.2 93.8 2.0 20.3 1.3 80.3 1.5 42.8 1.7 94.0 1.9 50.8 1.3 47.0 2.0 49.3 2.0
Response variables
* data taken from Kramar et al., 2003.
Hence, such designs are also known as uniform shell designs. The total number of experiments is given as n2+n+1. For two factors, the design is geometrically shaped in the form of a regular hexagon with a center point, thus requiring a total of 7 experiments. This design has the advantage that the experimental domain can be shifted in any direction by adding experiments on one side of the domain and eliminating them at the other. The design is highly recommended for pharmaceutical product development (Lewis, 1999). C. Mixture designs In FDs and the CCDs, all the factors under consideration can simultaneously be varied and evaluated at all the levels. This may not be possible under many situations. Particularly, in pharmaceutical formulations with multiple excipients, the characteristics of the finished product usually depend not so much on the quantity of each substance present but on their proportions. Here, the sum total of the proportions of all excipients is unity and none of the fractions can be negative. Therefore, the levels of the various components can be varied with the restriction that the sum total should not exceed one. Mixture designs are highly recommended in such cases (Cornell, 1990; Lewis et al., 1999; Lewis, 2002; Singh & Ahuja, 2004). In a two-component mixture, only one factor level can be independently varied, while in a three-component mixture only two factor levels, and so on. The remaining factor level is chosen to complete the sum to one. Hence, they have often been described as experimental designs for the formulation optimisation (Cornell, 1990; Schwartz & Connor, 1996). For process optimisation, however, the designs like FDs and CCDs are preferred. There are several types of mixture designs, the most popular being the simplex designs. A simplex is the simplest possible n-sided figure in a (n-1) dimensional space. It is represented as a straight line for two components, as a 2-D triangle for three components, as a 3-D tetrahedron for four components and so on.
Chapter 10.
285
Fig. 10. 7. Simplex mixture designs a) linear model; b) quadratic model; c) special cubic model.
Scheff s designs, also at times referred as simplex mixture designs (SMD), can either be centroid or lattice designs (Scheff , 1958; Doornbos & Haan, 1995). Both of these are identical for first and second-order models, but differ from third-order onwards. The design points are uniformly distributed over the factor space and form the lattice. The design point layout for three factors using various models is shown in Fig. 10. 7, where each point refers to an individual experiment. Scheff s polynomial equations are used for estimating the effects. General mathematical models for 3 components are given as under:
Linear : Quadratic :
Y = 1X1 + 2 X 2 + 3 X 3 Y = 1X1 + 2 X 2 + 3 X 3 + 12 X1X 2 + 13 X1X 3 + 23 X 2 X 3
(6) ...(7) (8)
Special cubic model: Y = 1X1 + 2 X 2 + 3 X 3 + 12 X1X 2 + 13 X1X 3 + 23 X 2 X 3 + 123 X1X 2 X 3
The mathematical model of mixture designs does not have the intercept in its equations. As a consequence, these Scheff models are not calculated by linear regression. Special regression algorithms are required (Doornbos & Haan, 1995). Table 10. 4 shows the design matrix for a simplex lattice design generated for optimization of dissolution enhancement of an insoluble drug (prednisone) with the physical mixtures of superdisintegrants (Ferrari et al., 1996). Extreme vertices design, another type of a mixture design, is used when there are restrictions on the levels of the factors (Doornbos & Haan, 1995; Lewis et al., 1999). For instance, in a study involving a tablet
Table 10. 4. Design layout for simplex lattice design
Formulation X1 1 2 3 4 5 6 7 8 9 10 X1: X2: X3: 1 0 0 0.5 0.5 0 0.33 0.667 0.167 0.167 X2 0 1 0 0.5 0 0.5 0.33 0.167 0.667 0.167 X3 0 0 1 0 0.5 0.5 0.33 0.167 0.167 0.667 Percent drug dissolved in 10 min 15.2 2.8 23.1 55.3 59.5 20.6 82.4 44.7 45.5 71.6
Croscarmellose Sodium Dicalcium Phosphate Dihydrate Anhydrous -Lactose
* data taken from Ferrari et al., 1996.
formulation for direct compression, use of more than 2% of lubricant or more than 30% of disintegrant is meaningless. Usually, there are restrictions on both the lower and upper limits of the factors. In such designs, the observations are made at the corners of the bounded design space, at the middle of the edges, and at the center of the design space, which can be evaluated only by regression. D D-optimal designs If the experimental domain is of a definite shape, e.g., cubic or spherical, the standard experimental designs are normally used. However, in case the domain is irregular in shape, D-optimal designs can be used (Lewis et al., 1999). These are non-classical experimental designs based on the D-optimum criterion, and on the principle of minimization of variance and covariance of parameters (de Aguiar et al., 1995; Doornbos & Haan, 1995). The optimal design method requires that a correct model is postulated, the variable space defined and the number of design points fixed in such a way that will determine the model coefficients with maximum possible efficiency. One of the ways of obtaining such a design is by the use of exchange algorithms using computers (Chariot et al., 1988; Lewis et al., 1999). These designs can be continuous, i.e., more design points can be added to it subsequently, and the experimentation can be carried out in stages. D-optimal designs are also used for screening of factors. Depending upon the problem, these designs can also be used along with factorial, central composite and mixture designs. Table 10. 5 gives a comparative account of important experimental designs employed for RSM, listing their advantages and disadvantages. 1.5.1.2 Model selection
"All models are wrong. But some are useful." This assertion of Box & Draper (1987) characterizes the situation that a formulation scientist faces while optimizing a system. Accordingly, the success of optimization study depends substantially upon the judicious selection of the model. In general, a model has to be proposed before the start of the DoE optimization study (Myers & Montgomery, 2002). Model selection depends upon the type of the variables to be investigated and the type of the study to be made, i.e., factor screening, description of the system, or prediction of the optima or feasible regions. The choice also depends on the a priori knowledge of the experimenter about possible interactions and quadratic effects (Doornbos & Haan, 1995). If the model chosen is too simple, higher-order interactions and effects may be missed because the relevant terms are not part of the model. If the model selected is too complicated, over fitting of the data may occur. The effect is a larger variance in the predictions, and reliability of the predicted optimum would be too low. The models mostly employed to describe the response are first, second and very occasionally, third order polynomials. A first-order model is initially postulated. If a simple model is found to be inadequate for describing the phenomenon, the higher order models are followed. After hypothesizing the model, a series of computations are performed subsequently to calculate the coefficients of polynomials and their statistical significance to enable the estimation of the effects and interactions. A Calculation of the coefficients of polynomial equations Regression is the most widely used method for quantitative factors (Bolton, 1990; Myers, 1990). It cannot be used for qualitative factors, because interpolation between discrete (dummy) factor values is meaningless. In ordinary least-squares regression (OLS), a linear model, as shown in Equation 9, is fitted to the experimental data, i.e., in estimating the values of in such a way that the sum of squared differences between predicted and observed responses is minimized.
2 E( y ) = 0 + 1X1 or E(y ) = 0 + 1X1 + 11X1
... (9)
Multiple nonlinear regression analysis (MLRA) can be performed for more factors, X i, interactions, XiXj, and higher order terms, as depicted in Equation 10.
Chapter 10.
287
Table 10. 5. Popular experimental designs for response surface optimization with merits and limitations (Doornbos & Haan, 1995; Schwartz & Connor, 1996; Lewis et al., 1999; Montgomery, 2001; Singh & Ahuja, 2004)
Des i g n Factorial Merits Efficient in estimating main effects and interactions Maximum usage of data Used for screening of factors, factor influence studies Suitable for large number of factors or factor levels Limitations Reflection of curvature not possible in a 2 level design Large number of experiments required Prediction outside the region is not advisable Effects cannot be uniquely estimated, as are confounded with interaction terms. Difficult to construct
Fractional factorial
Plackett-Burman
Suitable for very large number of factors, where Fixed designs in which runs are predetermined even FFDs require a large number of experiments and are limited to 16 experiments Effects confounded as suitable for two levels only Study keeps a central point, hence, suitable for second-order effects Allows the work to proceed in stages, i.e., if linear design does not adequately fit the data, suitable number of experiments can be added to run a CCD and determine the quadratic effects. Combines the advantages of FDs and star designs Requires fewer experiments Suitable for formulations in which a constraint is imposed on the combination of factor levels Does not reveal interactions Difficult to practice with fractional values of
Star Central composite
Simplex lattice
As the numbers of coefficients in the model are exactly equal to the number of design points, it is not possible to estimate residual error. Even replication allows only the estimation of experimental error Interactions and quadratic effects are not estimated Calculations can only be performed by regression
Extreme-vertices
Suitable for formulations in which a constraint is imposed on levels of the factors and/or on the combination of factor levels
D-Optimal
Can be employed even if experimental domain is Involves a relatively complex model irregular in shape
E( y ) = 0 + 1X1 + 2 X 2 + 12 X1X 2 + ...
(10)
MNLRA may also be performed in certain situations, wherein the factor-response relationship is nonlinear. Regression analysis can only be performed on the coded data or the original values after one or several models have been postulated, the choice being based on some expectation of the response surface. B Estimation of the significance of coefficients and model Significance of coefficients can be estimated using ANOVA followed by Student's t-test (Box et al., 1960; Bolton, 1990). ANOVA computation can be performed using Yates algorithm to find the significance of each coefficient. It is always advisable to retain only significant coefficients in the final model equation. This ANOVA helps in determining the significance of the model as well as of the lack of fit. The values of Pearsonian coefficient of determination (r 2) and that adjusted for degrees of freedom (r 2adj) of the polynomial equation are also compared. The value of r 2 is the proportion of variance explained by the regression according to the model, and is the ratio of the explained sum of squares to that of the total sum of squares. The closer the value of r2 to unity, the better is the fit and better apparently is the model (Myers, 1990; Lewis et al., 1999).
However, there are limitations to its use in MLRA, especially in comparing the models with different number of coefficients fitted to the same data set. A saturated model will inevitably give a perfect fit, and a model with almost as many coefficients as data is likely to yield a higher value for r2. In such cases, r2adj is preferred, which corrects the r2 value for the number of degrees of freedom (Bolton, 1990; Myers, 1990). The value of r2adj is calculated using equivalent mean squares in place of sum of squares, and has value usually less than r2. Finally, all these parameters are assessed to help in choosing the most appropriate model for a particular response. The final polynomial equation is subsequently used to calculate the magnitudes of effects and interactions. C Model diagnostic plots One or more of the model diagnostic plots can be plotted to investigate the goodness of fit of the proposed model: Actual vs predicted: A graph is plotted between the actual and the predicted response values (Montgomery, 2001; Singh & Agarwal, 2002). It helps in detecting a value, or group of values, that are not easily predicted by the model. Ideally, such plots passing through origin should be highly linear, i.e., with r2 values close to unity. These plots are simple to construct and comprehend. They reveal the most pragmatic information of prognosis, i.e., whether the experimentally observed values of responses are analogous with those predicted using optimization methodology. Fig 8 (a) illustrates the same. Residuals vs predicted: Residuals (or error) is the magnitudinal difference between the observed and the predicted response(s). Studentized residuals is the residuals converted to their standard deviation units (Bolton, 1990; Singh & Ahuja, 2002). The residuals (or studentized residuals) are plotted versus the predicted values of the response parameters. It tests the assumption of constant variance. The plot should have a random and uniform scatter with points close to zero axis and a constant range of residuals across the graph (Fig.8 (b)). Distinct patterns like expanding variance (megaphone pattern) in the plots are indicative of the need for a suitable data transformation (like logarithmic, exponential, square root, inverse, etc.). Residuals vs run: This is a plot of the residuals versus order of the experimental run (Montgomery, 2001). It checks for lurking variables that may have influenced the response during the experiment. The plot should show a random and uniform scatter as in Fig. 10. 8(c). Trends indicate a time-related variable lurking in the background. Residuals vs factor: This is a plot of the residuals versus any selected factor (Myers & Montgomery, 2002). It checks whether the variance not accounted for by the model is different for different levels of a factor. Ideally, the plot should exhibit a random scatter. Pronounced curvature may indicate a systematic contribution of the independent factor that is not accounted for by the model. Normal probability plot: The plot indicates whether the residuals follow a normal probability distribution, in which case the points will follow a straight line when plotted on a probit scale (Fig 8(d)). Definite patterns like an "S-shaped" curve, suggest that transformation of the response data may provide a better analysis (Lewis, 2002). Outlier T: This is a measure of how many standard deviations the actual value deviates from the value predicted after deleting the point in question. Many a times, this is referred to as an "externally studentized residual", since the individual case is not used in computing the estimate of variance (Montgomery, 2001). Outliers should be investigated to find out if a special cause can be assigned to them. If a cause is found, then it may be acceptable to analyze the data without that point. If no special cause is identified, then the point probably should remain in the data set. The graphical plots provide a better perspective on whether a case (or two) grossly deviates from the others or not. Fig. 10. 8(e) depicts the same with one distinct outlier. Cook's distance: It provides measures of the influence, potential or actual, of the individual runs (Montgomery, 2001). This is a measure of the effect that each point has on the model. A point that has a very high distance value relative to the other points may be an outlier, as shown in Fig. 10. 8 (f).
Chapter 10.
289
Leverage: This is a measure of degree of influence of each point on the model fit (Montgomery, 2001). If a point has a leverage of 1, then the model must go through that point (Fig. 10. 8 (g)). Verily, such a point controls the model. The point with leverage near one, should be reduced by adding or replicating points. Box-Cox plot for power transforms: The Box-Cox plot is a tool to help in determining the most appropriate power transformation for application to response data (Lewis et al., 1999; Montgomery, 2001). Most data transformations can be described by the power function, = fn( ), where is the standard deviation, is the mean and is the power. If the standard deviation associated with an observation is proportional to the mean raised to the power, then transforming the observation by the (1 - ) (or ) power gives a scale satisfying the equal variance requirement of the statistical model.
L o g 1 0 (t 9 0 % ) 1 .7 1
P re d i c te d vs . A c tu a l
3 .0
L o g 1 0 (t 9 0 % )
R e s i d u a ls v s . R u n
S tu d e n ti z e d R e s i d u a l s
3 .0 0
S tu d e n ti ze d R e s i d u a l s
1 .5
1 .4 9
1 .5 0
P r e d i c te d
1 .2 6
0 .0
0 .0 0
1 .0 4
-1 . 5
-1 . 5 0
0 .8 1 0 .8 1 1 .0 4 1 .2 6 1 .4 9 1 .7 1
-3 . 0 0 .8 1 1 .0 4 1 .2 6 1 .4 9 1 .7 1
-3 . 0 0 1 2 3 4 5 6 7 8 9
L o g 1 0 (t 9 0 % )
a.
N o r m a l % P r o b a b i l i ty
99
A ct tuo af l R e s i d u a ls N o rm a l P lo t9 0 %
3 .5
b.
P re du i ctli te e dr O
L o g 1 0 (t 9 0 % ) 4 .3 8
c.
C o o k 's D is ta nc e
R un N um ber
C o o k 's D i s ta n c e
95 90
0 .1
3 .2 8
50 30 20 10 5 1
O u tl i e r T
80 70
-3 . 3
Outlier T
2 .1 9
-6 . 7
1 .0 9
-1 0 . 1
-0 . 0 1 3 -0 . 0 0 8 -0 . 0 0 3 0 .0 0 2 0 .0 0 7
0 .0 0
d.
L o g 1 0 (t 9 0 % ) 1 .0
L e ve ra Re s ig de u a vs l . R un
L o g 1 0 (t9 0 % ) Lam bda C u rre n t = 0 B e st = 0 .2 L o w C .I . = 0 . 0 4 H i g h C . I . = 0 .3 6
e.
f.
Run Num ber

7 .1 3
R un N um ber
0 .8
L n (R e s i d u a l S S )
L e ve r a g e
0 .7
3 .6 2
0 .5
R e c o m m e n d t ra n sfo rm : Log (L a m b d a = 0 )
0 .1 0
0 .3
0 .2
-3 . 4 1
0 .0
g.
h.
-6 . 9 2 -3 -2 -1 0 1 2 3
R un N um ber
Lam bda
Fig 10.8. Various types of diagnostic plots for selecting suitable model(s). a) predicted vs. actual; b) Studentized residuals vs. predicted; c) Studentized residuals vs. run; d) normal probability plots; e) outlier T plot; f) Cook's distance plot; g) leverage plot; h) Box-Cox plot.
Fig 8 (h) shows a typical Box-Cox plot plotted between Ln(Residuals) and . Here, the value of near 0 suggests no power transformation. 1.5.1.3. Search for an optimum Optimization of one response, or the simultaneous optimization of multiple responses can be accomplished either graphically or numerically. A Graphical optimization (Response surface analysis) Graphical optimization displays the area of feasible response values in the factor space. For this, the graphical optimization criterion is set (Schwartz & Connors, 1996; Lewis et al., 1999; Myers, 2003). Selection of optima in graphical methods is not based upon minimization or maximization of any function. Hence, graphical methods require only computability but not continuity or differentiability of the function(s) as in the classical techniques. The experimenter has to make a choice, 'trading off' one objective for other(s), according to acceptability, i.e., the relative importance of the objectives considered. The success in locating an optimum lies in the sagacious interpretation and/or comparison of the resulting plots, leading to attainment of the best compromise . One or more of the following techniques may be employed for the purpose: 1. Search methods These methods are employed for choosing the upper and lower limits of the responses of interest (Schwartz & Connors, 1996). In these search methods, the response surfaces, as defined by the appropriate equations, are searched to find the combination of independent variables yielding the optimum. Two major steps are used viz. feasibility search and grid search. Together, these techniques are also referred to as brute force method (Bolton, 1990; Doornbos & Haan, 1995). The feasibility search method is used to locate a set of response constraints that are just at the limit of possibility. One selects several values for the responses of interest and a search of the response surface is made to determine whether a solution is feasible. The feasibility search method yields the possibilities satisfying the constraints. Subsequently, the exhaustive grid search is applied, wherein the experimental range is divided into a grid of specific size, and searched methodically. Grid search method can provide a list of possible formulations and the corresponding response values. 2. Overlay plots The response surfaces or contour plots are superimposed over each other to search for the best compromise visually. Minimum and maximum boundaries are set for acceptable objective values. The region is highlighted where all the responses are acceptable. Within this area, an optimum is located, trading off the different responses. The use of overlay diagrams is limited only to three or four response variables (Doornbos & Haan, 1995; Lewis, 2002). Fig. 10. 9 depicts an instance of overlay plots used for locating optimum formulation with response values of release till 18 h, (Rel18h) between 80-85% and bioadhesive strength (F) between 24-28 g (Singh et al., 2003a). 3. Pareto optimality charts In order to find the most optimum factor combinations satisfying various objectives of a formulation, a pareto optimality approach may also be used (Doornbos & Haan, 1995). In this method, a graph is plotted between the predicted values of the objectives and the variables. These are also called multiple criterion decision making plots. The space occupied by the resulting cloud of points is called the feasible criterion space. Special subsets of the points (forming a shell partly around the cloud) are the pareto-optimal (PO) points. A PO point is a point in the feasible criterion space, when there exists no other point in that space which yields an improvement in one criterion without causing degradation in the other. Graphical analysis is usually preferred in case of single response. However, in case of multiple responses, it is usually advisable to conduct numerical or mathematical optimization first to uncover a feasible region.
Chapter 10.

1.0
291
O v erlay P lot X = A : hpm ck15m Y = B : nacm c
F: 28
0.5
R el18h: 80 F: 24
So d . CM C
0.0
R el18h: 85
-0.5
-1.0 -1.0
-0.5
0.0
0.5
1.0
H PM C Fig 10.9. A contour overlay plot, plotted between two excipients X1 : HPMC and X2 : Sod. CMC shows the region between the two set criterions, i.e., Release till 18 h, Rel18h should be between 80 - 85% and bioadhesive strength, F should be between 24 to 28 g.
B. Mathematical optimization methods 1. Desirability functions This technique involves a way of overcoming the difficulty of multiple, sometimes opposing responses (Derringer & Suich, 1980). Each response is associated with its own partial desirability function. If the value of the response is optimum, its desirability equals 1, and if it is totally unacceptable, its value is zero. Thus the desirability for each response can be calculated at a given point in the experimental domain. An overall desirability function can be calculated by multiplying all of the r partial functions together and taking its r th root. The optimum is the point with the highest value for the desirability. The contour plots of desirability surface around the optimum should be studied along with the contour plots of the other responses, as described in overlay plots. 2. Objective functions These methods are used to seek an optimum formulation by solving the equation (objective function) either for a maximum or a minimum in the presence of equality and/or inequality constraints (Benkerrour et al., 1984; Das & Giri, 1994; Schwartz & Connor, 1996). Objective function may be expressed as Equation 11, and the inequality and equality constraints as Equation 12 and 13.
Y = f( X1X 2 )
(11) (12) (13)
i.e., inequality constraint G(X ) = f1(X1, X 2 ) 0 i.e., equality constraint

H( X ) = f2 ( X1, X 2 ) = 0
If the objective function is expressed as a function of a single variable, i.e., Y = f(X), calculus based mathematical approach is applied to find the maximum or minimum of a function. First derivative of the function can be taken and by setting it equal to zero, the value of X can be solved to obtain the maximum or
minimum. When the relationship for the response Y (objective function) is given as a function of two or more independent variables, as in Equation 11 for X 1 and X2, the problem is slightly more involved. Mathematically, appropriate manipulations with partial derivatives of the function can locate the necessary pair of X values for the optimum. This type of optimization is known as classical optimization and is applicable only to unconstrained problems. Particularly, these techniques find limited use in the optimization of pharmaceutical dosage forms, where the problems generally are the constrained ones (Fonner et al., 1970; Schwartz & Connor, 1996). 3. Sequential unconstrained minimization technique (SUMT) The above-mentioned technique can also be used for solving the objective function for a maximum or a minimum (Takayama et al., 1985). In this method, the constrained optimization problem is transformed to an unconstrained one by adding a penalty function, with the resulting function called as transformed unconstrained objective function. However, as different starting points may lead to different optimum solutions, application of a suitable random number technique like Monte Carlo approach can be used. 4. Lagrangian method The method can be used for optimization of functions expressed in Equations 11-13 using a series of steps viz. determining objective functions and constraints, changing the inequality constraint to equality constraint by introducing a slack variable (q) for each inequality constraint (Schwartz & Connor, 1996). Several equations are combined into a Lagrange function (F) with one Lagrange multiplier (l) for each constraint. Lagrange function is then partially differentiated for each variable and a set of simultaneous equations are solved by setting derivatives equal to zero. 1.5.2 Sequential Optimization Methodology Despite the numerous meritorious visages of simultaneous approaches, there are situations where there is hardly any a priori knowledge about the effects of variables (Schwartz et al., 1973; Doornbos & Haan, 1995; Araujo & Brereton, 1996). Such situations call for the application of the sequential methods. In sequential approach, optimization is attempted in a step-wise fashion. Experimentation is started at an arbitrary point in the experimental domain and responses are evaluated. Subsequent experiments are designed based upon the results of these studies, according to an algorithm that directs newer experiments towards the optimum. Whether the chosen optimum is a maximum or a minimum, the general term used for this approach is "hill climbing" (Doornbos & Haan, 1995; Lewis et al., 1999). An important aspect of sequential designs is to know when the goal has been accomplished. There are many different 'stopping criteria' to choose from. Nonetheless, sometimes the best method involves the experimenter's skill in judging the true optimum, which generally is a local maximum and minimum. There are two main model-based methods for extrapolating outside the domain, steepest ascent or steepest descent (first-order model) and optimum path (second-order). In addition, there is another model-independent sequential-simplex method. The inherent advantages of these methods are: no need of planning all the experiments simultaneously,
a priori knowledge of the response surface not essential, and
interactive. However, various disadvantages encompass: number of experiments to reach an optimum can not be predicted, optimum found may not be the global optimum, robustness is not known, unsuitable for multiple objective problems,
Chapter 10.
293
attainment of optimum is judged only by the expert developmental scientist, mathematical model and complete response surface is not generated, yields unreliable results when multiple optima exist, and applicable only when response surface is continuous.
1.5.2.1. Steepest ascent (descent) methods These methods are direct optimization methods for first-order designs (Lewis et al., 1999; Myers, 2003). They are good choice when the optimum is outside the domain and is to be arrived at rapidly. These approaches are an amalgamation of model-independent and model-dependent methods. The direction of the steepest increase of the response in terms of coded variables is determined, and then experiments are carried out along this line (Lewis, 2002). This is followed by measurement of the response and is continued until an optimum is reached. 1.5.2.2. Optimum path method This method is just analogous to steepest ascent method, where the optimum is also searched outside the experimental domain by extrapolation. Such situations arise when choosing a very extensive experimental domain is difficult or the possible experimental domain is not known at the beginning of the study. However, this method is used for searching the optimum by extrapolation from a second-order design along a curved path. 1.5.2.3. Sequential simplex techniques The technique consists of first generating data from n + 1 experiments, where n is the number of independent variables or factors (Shekh et al., 1980; Bolton, 1990; Araujo & Brereton, 1996). Based on n + 1 responses and predetermined rules, one result is eliminated and a new experiment is performed. A decision is made as a result of experimentation, eventually terminating the study at an optimal response. Fig. 10. 10 illustrates various steps involved under the approach using an arbitrary example. A simplex is constructed by selecting three combinations (A, B and C) of two variables (X1 and X2). Three experiments are carried out and evaluated, and the worst response illustrated as point A is identified. The next experiment is conducted for a combination moving away from point A. This is achieved by reflecting the triangle ABC around BC axis. The experiment at point D is performed, and the response is compared with the response at point A, B and C. The next move depends on the relative values of the four responses: If the response at point D is greater than the responses at A, B and C, the next experimental point is E.
If the response at point D is greater than the response at B but smaller than the response at C, the triangle BCD is reflected about CD axis and the next experimental point is F. If the response at point D is lower than the responses at B and C, but greater than at point A, the next experimental point is G. If the response at point D is lower than the responses at A, B and C, the next experimental point is H.
1.5.2.4. Evolutionary operations (EVOP) It is a popular technique in several industrial processes (Schwartz & Connor, 1996; Lewis et al., 1999). The underlying basis for this approach is that the production procedure (formulation and process) is allowed to evolve to the optimum by careful planning and constant repetition. The process is run in such a way that it produces a product that meets all the specifications and at the same time, generates information on product improvement. Generally, these involve factorial and simplex designs requiring a large number of experiments. In a typical industrial process, this extensive experimentation is usually not a problem, since the process will be run repeatedly over and over again. However, in the most complex situations involving product development, it is not so because there is often insufficient freedom in the formula or process to allow necessary experimentation. In pharmaceutical product development setup, however, more efficient methods are desirable.
Fig. 10. 10. Schematic diagram illustrating the stages of optimization using a simplex method.
1.5.3 Artificial Neural Networks Of late, the application of artificial neural networks (ANNs) in the field of pharmaceutical development and optimization of dosage forms has become a blown out topic of discussion in the pharmaceutical literature (Takayama et al., 1999; Takayama et al., 2003). The ANNs are model-independent computational paradigms that can simulate the neurological processing ability of the human brain. The neural networks, consisting of inter-connected adaptive processing units, so-called neurons, are able to discern complex and latent patterns in the information presented to them. ANN is a computer-based learning system that can be applied to quantify a nonlinear relationship between causal factors and pharmaceutical responses by means of iterative training of data obtained from a designed experiment (Achanta et al., 1995; Bourquin et al., 1997). The results obtained from implementation of an experimental design are used as input information for learning. Once trained, the neurons of an ANN may be used to forecast outputs from new sets of input conditions (Peck et al., 1989; Achanta et al., 1995; Zupancic Boic et al., 1997; Bourquin et al., 1997). A typical ANN must have one input layer and one output layer, and may contain one or more hidden layers as depicted in Fig.11. The information is passed from input layer to the output layer through hidden layer(s) by the network connections or synapses. Modeling starts with a random set of synaptic weights and proceeds in iterations. During each iteration, connection weights are adapted via selected modeling. The basis of such modeling technique is to minimize the error, i.e., the difference between the momentary network signal and the aimed signal based on the experimental results. When the minimal " error" is obtained, learning is completed and connection weights become the memory units. After this, the test set of values can be applied on a learned ANN to evaluate it. Subsequently, it can be used for output prediction on the basis of the new input values. The modeling is invariably done via a suitable computer software. The prediction ability (PA) or reliability of an ANN output depends heavily on the training data (So & Karplus, 1996). Two problems that tend to diminish PA are overfitting (i.e., few data points per network connection) and overtraining (long network training period). Thus, ANN does not work well with many variables and few formulations. Further, the results from ANN cannot be treated statistically and no definitive reasons can be given for the same. In an attempt to improve PA and to reduce training efforts, genetic neural networks (GNN), and generalized regression networks (GRN) have been used with fruition, respectively. While the former employs a combination of genetic algorithms with ANN, the latter utilizes the modelization of the function more or less directly from the training data. Since ANNs require a great deal of iterative
Chapter 10.
295
Fig. 10. 11. Schematic diagram illustrating various parts of an Artificial Neural Network. X 1 - X3 represent the input factors; Y as the response variable connected to the input layer via various nodes of hidden layer (H 1-H9). W11 and W93 represent the connections between the corresponding input factors and the nodes of the hidden layer while W 1y, W5y and W9y denote the connections between the corresponding respective hidden nodes and output layer, Y.
computations, the use of versatile computer software dedicated for the purpose becomes almost obligatory for their execution (Bourquin et al., 1997). 1.5.4 Choosing an Optimization Methodology In case of single response, graphical analysis is opted for (Lewis et al., 1999). However, in case of multiple response variables, certain responses can oppose one another. Accordingly, changes in a factor that improve one response may have a negative effect on another. Since it is not usually possible to obtain the best values for all the responses, optimization principally embarks upon finding experimental conditions where different responses are most satisfactory, over all. Nevertheless, there is a certain degree of subjectivity in weighing up their relative importance. 1.6 COMPUTER USE IN OPTIMIZATION
Development of the principles behind optimization, now known as DoE, dates back to the 1920s with its
Table 10. 6. Suitability of various optimization methods under variegated situations
Optimization method GRAPHICAL ANALYSIS DESIRABILITY FUNCTION STEEPEST ASCENT OPTIMUM PATH SEQUENTIAL SIMPLEX EVOLUTIONARY OPERATIONS Model situations for use Mathematical model of any order, Normally no more than 4 factors, Preferably in single response Mathematical model of any order, Number of factors between 2 and 6, Multiple responses First-order model, Optimum outside the domain, Single response Second-order model, Optimum outside the domain, Single response No mathematical model, Direct optimization, Single or multiple responses, Industrial situation, Little variation possible
discovery by British statistician, Ronald Fisher. Optimization, however, lay virtually dormant due to the complex and tedious hand calculations it required. Software that automates the designed-experiment optimization studies was invented in the early days of mainframe computers (Potter, 1994). Mainframes, requiring programming skills beyond most statisticians' scope, chugged through complicated DoE equations. Nevertheless, it wasn't until those room-sized computers became desktop PCs, that affordable DoE software for the non-statisticians first appeared. Now a days, computer use is considered almost indispensable in the design and optimization methods, as a great deal of intricate statistical and mathematical calculations are involved (Doornbos & Haan, 1995; Podczeck, 1996; Tye, 2004). Particularly, the ANN optimization is based totally upon the computer interface, tailor-made for the purpose (Bourquin et al., 1997). The computer software have been used almost at every step during the entire optimization cycle ranging from selection of design, screening of factors, use of response surface designs, generation of the design matrix, plotting of 3-D response surfaces and 2-D contour plots, application of optimum search methods, interpretation of the results, and finally the validation of the methodology (Potter, 1994). Verily, many software packages lead the user through the data analysis even without a mathematical model or statistical equations in sight. Use of pertinent software can make the DoE optimization task a lot easier, faster, more elegant and economical (Singh, 1997; Singh, 2003; Tye, 2004). Specifically, the erstwhile impossible task of generating varied kinds of 3-D response surfaces manually is accomplished with phenomenal ease using appropriate software (Bolton, 1987; Potter, 1994). 1.6.1 Choice of Computer Software Package: Many commercial software packages are also available, which are either dedicated to a set of experimental designs or are of a more general statistical nature with modules for select experimental design(s). The dedicated computer software is frequently better as the user pays only for the DoE capabilities (Potter, 1994). In contrast, the more powerful, comprehensive and expensive statistical packages like SPSS, SAS, BBN, BMDP, MINITAB, etc. are geared up for larger enterprises offering diverse facilities for statistical computing, support for networking and client-server communication, and portability with a variety of computer hardware (Potter, 1994; Singh, 2003, Singh et al., 2005a). When selecting a DoE software, it is important to look for not only a statistical engine that is fast and accurate but also the following: A simple graphic user interface (GUI) that's intuitive and easy-to-use.
A well-written manual with tutorials to get you off to a quick start. A wide selection of designs for screening and optimizing processes or product formulations. A spreadsheet flexible enough for data entry as well as dealing with missing data and changed factor levels. Graphic tools displaying the rotatable 3-D response surfaces, 2-D contour plots, interaction plots and the plots revealing model diagnostics Software that randomizes the order of experimental runs. Randomization is crucial because it ensures that "noisy" factors will spread randomly across all control factors. Design evaluation tools that will reveal aliases and other potential pitfalls. After-sales technical support, online help and training offered by manufacturing vendors
Table 10. 7 lists some commonly used computer software for optimization along with their salient features. Today, these off-the-shelf software packages commonly sell for divergent prices, varying widely from $99 to $2500, depending upon the features provided with these software. The actual number of computer systems, however, is much more as the field is still rapidly growing. 1.7 PLAN TO IMPLEMENT OPTIMIZATION: AN OVERVIEW
The overall approach for conduct of computer-assisted optimization studies in the development of drug
Chapter 10.
Co mp u t e r As s i s t e d Op t i mi z a t i o no fPh a r ma c e u t i c a lFo r mu l a t i o n sa n d Pr o c e s s e s
Table 10. 7. Important computer software for optimization and their salient features
297
Software Design Expert
Salient features Powerful, comprehensive and popular package used for optimizing pharmaceutical formulations and processes; allows screening and study of influential variables for FD, FFD, BBD, CCD, PBD and mixture designs; provides 3D plots that can be rotated to visualize the response surfaces and 2D contour maps; numerical and graphical optimization Powerful DoE software for automated data analysis, graphic and help features, MS-Excel compatibility, includes almost all designs of RSM DoE software for automated data analysis of various designs of RSM, graphic and help features Powerful DoE software for automated data analysis, includes graphic and help features MS-Excel compatible DoE software for automated data analysis using Taguchi, FD, FFD and PBD. The relatively inexpensive software, DoE KISS is, however, applicable only to single response variable. Excel compatible optimization software with facilities for various experimental designs and Taguchi design. DoE software with features for executing various experimental designs Used for designing and analyzing optimization experiments Mathematical optimization program to search for the maximum or minimum of a function with or without constraints Used for designing the optimization experiments ANN-based software based on GRN technique Suitable for FDs and CCDs, has features for numerical optimization and graphic outputs Suitable for response surface modeling and evaluation of fitting of model Comprehensive statistical software with facilities for implementing experimental designs Only for mixture designs; only program that supports multicriterion decision making by Pareto- optimality, upto six objectives and has various statistical functions Supports designs for screening, D-optimal, Taguchi and user defined designs, also options are available for pareto optimality charts Optimization software for systematic DoE and response surface methodology studies with state-of-art mathematical search techniques Generates the experimental design, fits a mathematical equations to the data and graphically depicts response surfaces Aids in selection of an experimental design, has modules for numerical optimization and graphic outcomes Aids in the optimization of formulation using various FDs, and other designs through development of polynomials and grid search; includes computer-aided-education module for optimization
Source www.statease.com
MINITAB
www.minitab.com
JMP CARD DoE PRO XL & DoE KISS
www.jmp.com www.s-matrix.com www.sigmazone.com
MATREX Cornerstone ECHIP GRG2 DoE PC IV STATISTICA NEMROD@ MODDE SPSS Omega
http://www.rsd-associates.com/ matrex.htm www.brooks.com www.echip.com www.fp.mcs.anl.gov/otc/Guide/ SoftwareGuide/Blurbs?grg2.html http://www.adeptscience.co.uk/as/ products/qands/qasi/doepciv/ www.statsoftinc.som www.umt.ciw.uni-karlsruhe.de/ 22713 www.umetrics.com www.spss.com www.winomega.com
DoE WISDOM
www.launsby.com
COMPACT
www-fp.mcs.anl.gov/otc/ guide/ SoftwareGuide/Blurbs/ compact.html www.optimasoftware.co.uk www.amazon.com www.puchd.ac.in/uips/bhoop.html
OPTIMA XSTAT FACTOP
Software iSIGHT SOLVER Multisimplex AB
Salient features General DoE software with features for implementation of Taguchi, CCDs and FDs. Optimization software for linear and nonlinear problems with state-of-art mathematical programs Aids in optimization based on simplex and D-optimal designs
Source www.engenious.com/ release1_11isightenhance.html www.solver.com www.multisimplex.com
ANN: Artificial neural network; DoE: Design of experiments; FD: Factorial design; FFD: Fractional factorial design; CCD: Central composite design; GRN: Genetic regression network; PBD: Plackett-Burman design; RSM: Response surface methodology; @ New efficient methodology for research using optimal designs
product systems can be described by a DoE optimization strategy. Although there is no infallible plan, yet its choice depends on the diverse characteristics of the problem at hand, the required quality of remedy and the quantum of experimental effort to gain the information (Lewis, 2002; Myers, 2003; Singh, 2003; Singh & Ahuja, 2004). The salient steps involved in an optimization plan encompass: 1. Defining the objective: The optimization objective, i.e., the property of interest is clearly defined (e.g., drug release from a compressed tablet). Selection of the response variables should be made with dexterity. Selected response variables should be such that they provide maximum information with minimal experimental effort and time. 2. Choice of appropriate computer interface: Since the use of computers is nearly obligatory for implementing an optimization plan, the choice of apposite software is vital. The computer package selected for the purpose should ideally encompass the facilities of executing several experimental designs for screening as well as response surface optimization, generating design matrices and response surfaces, and conducting statistical analysis and graphics for model diagnostic analysis. 3. Screening of the factors and selection of their levels: The independent variables to be investigated should be quantifiable. These are first screened for the identification of the possible factors. Once selected as factors, these variables are investigated for their effect on the dependent variable. The study is usually carried out at two levels to determine the main effect and the possible interactions. Then, the levels of each variable are established either from the prior experience or from the pilot studies. Factor level selection should be judicious enough to include the optima in the region of interest. 4. Selecting the optimization methodology: Based on the choice of independent variables and the type of response expected, a suitable statistical method is selected. A choice has to be exercised between the various simultaneous or sequential methodologies available. If no information is available on the type of response, a quadratic model should be preferred. 5. Generating the design matrix: The experimental runs are conducted as per the chosen experimental design and the data are collected. The number of experimental trials required is dictated by the selected design. To measure the inherent variability, a sufficient number of replicates should also be determined. 6. Conducting the experimental studies: The pharmaceutical dosage forms are formulated corresponding to the requisite number of experiments as per the selected design model and evaluated for the desired response(s). 7. Finding the optimum: The experimental data are used for generation of a mathematical model and an optimum formula is determined using search methods and/or mathematical optimization methods. This is usually facilitated with the help of suitable computer software. 8. Validating the results: The predicted optimal formulations are formulated, responses evaluated and inferences drawn. The results are implemented in the process/product development cycle or subsequent planning. The flow chart in Fig 12 provides a bird's eye view on the myriad steps involved in implementing an optimization plan.
Chapter 10.
299
PROBLEM ANALYSIS - DEFINITION OF AIMS
SELECTION OF EXCIPIENTS
SELECTION OF PROCESS
IDENTIFICATION OF INDEPENDENT VARIABLES
SCREENING FOR INFLUENTIAL VARIABLES
IDENTIFICATION OF FACTOR LEVELS
ESTIMATION OF REPRODUCIBILITY OF RESPONSE
R E A T T E M P T I N G
PRELIMINARY CHOICE OF MATHEMATICAL MODEL
CHOICE OF EXPERIMENTAL SET-UP
SEQUENTIAL APPROACH
SIMULTANEOUS APPROACH
FORMULATING AND EVALUATING DRUG DELIVERY SYSTEMS
STATISTICAL ANALYSIS
MODEL SELECTION
STOP CRITERION REACHED
RESPONSE SURFACES
SEARCH FOR AN OPTIMUM RANGES VERIFICATION BY NEW EXPERIMENTS
IMPLEMENTATION OF RESULTS IN PRODUCT / PROCESS DEVELOPMENT
UNSUCCESSFUL
SUCCESSFUL
PRODUCTION CYCLE
Fig. 10. 12. Overall plan for computer-aided optimization of drug formulations
1.8 OPTIMIZATION OF VARIOUS DRUG PRODUCTS AND PHARMACEUTICAL PROCESSES: A LITERATURE INSIGHT 1.8.1 Historical Background Optimization employing various experimental designs has been used for a long time like, the FDs since 1926, the screening designs since 1946, the CCDs since 1951 and mixture designs since 1958. (Fisher, 1935; Plackett & Burman, 1946; Box & Wilson, 1951, Scheff , 1958). The use of DoE optimization techniques, however, permeated into the field of pharmaceutical product development around four decades ago. The first literature report on the rational use of optimization appeared in 1967, when a tablet of sodium salicylate was optimized using an FD (Marlowe & Shangraw, 1967). Since then, these systematic approaches have been put into practice in routine drug product/process development at steady pace. A product development scientist has to handle a heterogeneous group of formulations, including uncoated or coated tablets, controlled release tablets, effervescent tablets, soluble tablets, dispersible tablets, microspheres, granules, pellets, etc. amongst the solid dosage forms; gels, ointments, creams and suppositories amongst semisolids; solutions, emulsions, suspensions, lotions, inhalations, etc., amongst the liquid dosage forms; and aerosols, patches, films, plasters amongst the others. The characteristics of these dosage forms vary markedly from immediate release (fast disintegration, fast dissolution) to (very) slow release matrix tablets. Extensive literature search carried out in pharmaceutical journals and texts till date, reveals that the DoE optimization techniques have been employed for almost all of these dosage forms, ranging from the simple conventional ones to that of the most intricate novel drug delivery systems. The current literature reports unequivocally point out a significant shift of the focus of the formulator from optimization of the conventional formulations to that of the modern drug delivery devices. Since 1990, there has been a sudden spurt in the number of published work on the use of rational optimization in drug product development. This may largely be attributed to the realization of stellar benefits of CADD techniques coupled with the ready accessibility of effectual and cost-effectual computational tools in the armamentarium of a formulator. The number continues to swell further, rather at higher pace. Already with over 400 publications reported till date on diverse type of dosage forms, optimization is becoming a common place in drug industry in the developed world. With the growing popularity of such rational methods, it is anticipated that drug formulation optimization would soon become a global phenomenon in routine product development. In the developing countries like India, however, the switch-over from the traditional to the modern approach is relatively a recent phenomenon and the number of literature reports on optimization of drug formulations are picking up lately. Amongst conventional dosage forms, tablets have been most widely investigated for the purpose, whereas, amongst various drug delivery systems, microparticulates and CR matrices have majorly been studied, followed by macroparticulates, fast release (FR) dosage forms, transdermal drug delivery systems (TDDS) and vesicular systems. Figure 13 gives the pictorial depiction on the proportion of various kinds of dosage forms formulated using DoE approach. A succinct account of the literature search on diverse type of drug formulations and processes is being presented here in with select literature instances of each type. 1.8.2 Tablet Formulations Tablet is the most popular dosage form and its formulation is still a tedious and complicated process (Banker & Anderson, 1987). Although optimization has been carried out on tablet since 1967, yet most studies are reported in the last two decades. Amongst the conventional tablet dosage forms, the targeted responses for optimization include, crushing strength, weight uniformity, disintegration time, physico-chemical stability, thickness, content uniformity and in vitro dissolution profile. Various tablet constituents have been most frequently investigated as the independent variables. The process variables that have been optimized include,
Chapter 10.
301
Transdermals Tablets Macroparticulates Vesicular systems Capsules Topicals Microparticulates Liquids
Fast release dosage forms
Controlled release matrices
Fig. 10. 13. Pie chart showing the proportion of various dosage forms formulated using optimization approach.
Table 10. 8. Select optimization reports on tablet formulations

Formulation optimization Sr. No. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. Drug Chlorthalidone Artemisinin Vitamin C Pyridoxal HCl Theophylline Aspirin & Paracetamol Aspirin Oxazepam Paracetamol Phenylpropanolamine HCl Sodium salicylate Ticlopidine HCl Acetaminophen Des i g n RSM FD BBD FD RCCD FD FD SMD FFD FD FD RSM D-optimal MLRA FD CCD IV 2 3 3 6 2 3 4 3 4 2 1 3 2 5 5 RV 4 2 4 2 1 1 3 2 2 5 4 2 1 2 3 8 Reference Ceschel et al., 1999 Ngo et al., 1997 Chang, 1994 Durig & Fassihi, 1993 Dawoodbhai et al., 1991 Mahmud & Po, 1991 Devay et al., 1988 Van Kamp et al., 1987 Sanderson et al., 1984 Fonner et al., 1970 Marlowe & Shangraw, 1967 Gabrielsson et al., 2003 Sabir et al., 2001 Chowhan et al., 1982 Malinowski & Smith, 1974 Schwartz et al., 1973
Process optimization
HCl: hydrochloride; IV: Input variable; RV: Response variable; FD: Factorial design; FFD: Fractional factorial design; CCD: Central composite design; FCCD: Face centered composite design; RCCD: Rotatable central composite design; RSM: Response surface methodology; MLRA: Multiple linear regression analysis; SMD: Simplex mixture design; BBD: Box Behnken design.
the compaction pressure and lubrication time. The number of investigated independent variables generally ranged between 2 to 6, while the optimized responses between 1 to 4, the broader range being 1 to 10. Table 10. 8 indicates some literature reports on the use of optimization techniques for conventional tablets. 1.8.3 Pre-Tablet Granulation Optimization techniques have also been successfully applied since 1975 to the process of pre-tabletting granulation, using wet or dry method, to produce optimum granules for compression (Malinowski & Smith, 1975). The responses taken into consideration for optimization have primarily been the physical properties of granules, i.e., geometric mean diameter, particle size distribution, bulk and tap densities, friability, flow rate, compactibility index, recompressibility of roller compacted dry granules, etc. The main independent variables for wet granulation pertain to the composition of the binder solution, i.e., percentage of binder, total amount, amount of alcohol in a hydroalcoholic binder solution, etc. The process variables, which usually have been investigated, include, impeller speed, kneading time, time for granulation, etc. In the process of dry granulation using roll compactor, the roll speed, horizontal feed speed and vertical feed speed have been the key input variables. The number of independent variables has ranged between 2 and 8, and the response variables between 1 and 8. For the purpose of optimization of granulation process, most studies have been carried out excluding the drug. The FD and FFD have been the most popular designs, followed by the CCD and others. Table 10. 9 enlists some literature instances on the use of optimization techniques in pre-tablet granulation.
Table 10. 9. Select optimization reports on pre-tabletting granulation process
Sr. No. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Des i g n FFD FCCD FD FFD SMD FD CCD D-Optimal RSM FD IV 4 4 8 4 3 2 2 5 3 5 RV 4 2 2 4 3 5 2 7 6 4 Reference Gao et al., 2002 Rambali et al., 2001 Voinovich et al., 1999 Miyamoto et al., 1997 Vojnovic et al., 1994 Wehrl et al., 1993 Shirakura et al., 1991 Chariot et al., 1988 Benkerrour et al., 1984 Malinowski & Smith, 1975
1.8.4 Tablet Coating Tablet coating operations are considered to be critical and intricate processes involving stringent control of variables (Banker & Anderson, 1987). Optimization of variegated coating processes has been carried out since 1977 (Dincer & Ozdurmus, 1977), which includes, aqueous-, non-aqueous- and enteric-film coating. Important independent variables studied comprise, film former concentration, plasticizer concentration, viscosity of the solution, spraying time, length of drying interval, pan speed, bed temperature, etc. In some instances (Hutchings et al. 1994; AlKhatib & Sakr, 2003), curing time and curing temperature have also been evaluated. The responses optimized for all type of coating processes are principally the defects in the physical appearance of coating viz. mottling, picking, peeling, cracking, pitting and pin-hole formation. Literature indicates that the number of independent variables has ranged between 3 to 6, and the response variables between 1 and 6. Factorial designs (full and fractional) appear to be more suitable for this purpose. Table 10. 10 summarizes a few literature reports on the optimized coating operations.
Chapter 10.
303
Table 10. 10. Optimization instances on tablet coating process

Sr. No. 1. 2. 3. 4. 5. 6. 7. 8. Typ e Aqueous film coating Film coating Film coating Film coating Granule coating Film coating Enteric film coating Enteric coating Design IV FD ANN CCD EVD BBD SMD FD FFD 2 6 2 3 3 6 2 4 RV 1 2 1 1 2 3 1 2 Reference AlKhatib & Sakr, 2003 Plumb et al., 2003 Hutchings et al., 1994 Li & Peck, 1991 Turkoglu & Sakr, 1992 Thoennes & McCurdy, 1989 Devay et al., 1982 Dincer & Ozdurmus, 1977
ANN: artificial neural network; EVD: Extreme vertices design.
1.8.5 Capsule Dosage Forms Relatively limited reports are available on the optimization of capsule dosage form. The independent variables, which have been considered for optimizing capsules encompass, quantities of the ingredients, capsule size, operating rate of semi-automatic filling machine, etc. for responses like, powder blend homogeneity, flow, weight, dissolution rate, etc. Table 10. 11 presents the list of a few reports on the use of optimization techniques on capsule formulations.
Table 10. 11. Select optimization reports on capsules
Sr. No. 1. 2. 3. 4. Drug(s) -Methyl dopa Tetracycline Oxytetracycline Nitrofurazone Nitrofurantoin Ethinamate Des i g n Simplex FFD FFD FD IV 7 4 3 3 RV 1 2 4 1 Reference de Saavedra & Caudra, 2001 Iskandarani et al., 1993 McGurk et al., 1991 Newton & Razzo 1974
5.
FD
Newton et al., 1971
1.8.6 Liquid Formulations There are several reports on optimization of diverse liquid formulations viz. emulsions, solutions, suspensions, lotions, etc. Amongst the ophthalmic and parenteral formulations, solubility, chemical stability and viscosity have been optimized by varying the formulation composition. Table 10. 12 enlists some literature reports on the use of optimization techniques in the development of liquid formulations. Emulsions, microemulsions, solutions, lotions or suspensions have been optimized for responses as turbidity, cloud point, physical stability, preservative efficacy etc., primarily by altering the levels of the ingredients. The FD, CCD or SMD have been utilized as the experimental designs, with the number of independent variables ranging between 2 to 8 and responses between 1 and 5. 1.8.7 Nonoral Semisolids The gel, ointment and suppository formulations have been optimized for percutaneous absorption, spreadability, chemical stability, appearance, skin irritation, etc., by changing the formulation composition using RSM. The independent variables range between 2 and 5 and the response variable between 1 and 5.
Table 10. 12. Optimization reports on liquid dosage forms

Formulation optimization Sr. No. Type 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Syrup Emulsion Lotion Emulsion Suspension Suspension Oral solution Ophthalmic formulation Parenteral nutrition Solution Emulsion Drug Acetaminophen Oxybenzone, octylmethoxycinnamate Erythromycin Glucose Erythromycin ethylsuccinate Rifampicin Lamivudine Enalkrien Nutrient mixtures Diazepam Des i g n RSM SMD CCD FD FD FD CCD SMD FD PBD SMD BBD IV 4 3 3 3 3 4 5 4 3 5 3 6 RV 2 3 1 1 4 5 2 1 1 1 1 4 Reference Worakul et al., 2002 Marti-Mestres et al., 2000 Brisaert et al., 2000 Bjerregaard et al., 1999 Elkheshen et al., 1997 Elkheshen et al., 1996 Nguyen et al., 1995 Pattarino et al., 1993 Patel et al., 1990 Ozil & Rochat, 1988 Belloto Jr. et al., 1985 Lemaitre-Aghazarian et al.,2004
Microemulsion Retinol
The studied drugs have usually been the NSAIDs. Amongst the nonoral semisolids, bioadhesive gels and semisolids have been optimized for the bioadhesion strength of the formulation and release profile of drug candidate. The input variables vary from 2 to 3 and the response variables vary from 1 to 4. Mainly, FD has been employed for optimization with isolated studies using mixture design. Table 10. 13 lists some of the literature citations on the use of optimization techniques in nonoral semisolids. 1.8.8 Oral Controlled Release Formulations Application of statistical DoE techniques in optimizing the controlled release (CR) dosage forms has relatively been a recent phenomenon, though the first report appeared in 1985 (Harris et al., 1985). The studied matrix dosage forms encompass, the inert, hydrophilic/hydrocolloid, silicone elastomer and the wax matrices. The common independent variable for all of these have been the quantities of the ingredients and the optimized response was invariably the in vitro dissolution profile. Other responses include, the disintegration time, bioavailability, etc. The optimization methodology of choice has been the FD and the FFD with some reports on use of CCD, BBD and others. The independent variables have ranged between 2 to 7 and the response variables between 1 and 6. A detailed updated literature review on optimization of CR systems has been presented by us elsewhere (Singh & Ahuja, 2004). Limited number of reports on systematic optimization of osmotic, hydrodynamically balanced (floating) and bioadhesive drug delivery systems are also available in literature. For osmotic delivery systems, apart from polymer content, the cure time and cure temperature have also been considered as important independent variables (Appel et al., 1992; Wu et al., 2000). The response variables that have been studied include various parameters of the release behaviour and coat burst strength. FDs, BBD and ANN have been used in these systems. Floating systems have been optimized mainly for the buoyant properties of the dosage form and release profile of the drug using FDs and CCDs. Besides drug release profile, the bioadhesive dosage forms have invariably been optimized for the bioadhesive strength. Table 10. 14 gives a concise account on the use of optimization on various oral CR formulations. Few reports on the use of optimization in CR solid dispersions are also available, wherein the dispersions have been prepared with release rate retarding polymers/carriers.
Chapter 10.
305
Table 10. 13. Select optimization reports on semisolids formulations

Sr. No. Type 1. 2. 3. 4. 5. 6. 7. 8. Cataplasm Dental semisolid system Topical cream Vaginal suppository Topical gel Ocular nanoparticulate system Nasal gel Topical ointment Drug Ibuprofen Tetracycline Methotrexate Progesterone AGN-190168 Propranolol Indomethacin Des i g n FCCD FD FD RSM RSM FD SMD CCD IV 3 2 3 2 2 3 2 5 RV 2 4 2 2 2 2 1 5 Reference Agarwal et al., 1999 Jones et al., 1996 Hwang et al., 1995 Iwata et al., 1995 Keyhani-Morison et al., 1995 Das et al., 1995 Chu et al., 1991 Takayama et al., 1990
1.8.9 Microparticulate Drug Delivery Systems Though the first report appeared in 1982 (El-Banna & Effimova, 1982), yet the major work on optimization of microparticulate drug delivery systems viz. microspheres, microcapsules, nanoparticles and liposomes have been reported in the last decade. The optimized formulation variables have been the emulsifier concentration, pH, drug/polymer ratio, composition of the internal phase of the emulsion, etc. The process variables studied have been the nature of mixing, baffled/non-baffled container, homogenization pressure, stirring rate and time, duration of cross linking process, needle gauge, injection rate, etc. The response variables of interest for optimization have been the characteristics of the particulate system in the resulting formulation, in vitro release, entrapment efficiency, percent yield, fraction of loose surface crystals, etc. The optimization methodologies include, the FD, FFD, CCD, BBD and other designs like SMD. The number of independent variables has ranged between 2 to 7 and the response variables between 1 and 5. Table 10. 15 encompasses the selected examples of the various optimized microparticulate drug delivery systems. 1.8.10 Macroparticulate Dosage Forms The macroparticulates including pellets, beads and microgranules prepared usually by the process of extrusion spheronization have also been optimized for yield, mean particle size, size distribution, granulate hardness, granule friability, porosity, flow rate, etc. Many of such formulations include, oral CR macroparticles. Also, the influence of extruder speed, spheronizer speed, spheronizer time, load, screen size, etc. has been studied in macroparticulates. The experimental designs used have been the FD, FFD, BBD and SMD with the number of independent variables ranging between 3 to 6 and the response variable ranging between 1 to 6. Table 10. 16 construes a few literature reports on macroparticulate optimization. 1.8.11 Fast Release Dosage Forms Some reports are also available on optimization of diverse kinds of immediate release oral drug delivery systems. The optimized systems comprise dispersible tablets, solid dispersions, coevaporates, inclusion complexes, phospholipid micelles and self-emulsified drug delivery systems. The number of independent variables broadly ranged between 2 to 6 and the response variables between 1 to 6. Amongst these, the independent variables have been the quantity of ingredients and the solvent used for the preparation. Nonetheless, the dissolution profile and chemical stability were the prime response variables. Table 10. 17 comprises the literature instances of the optimized solid dispersions. 1.8.12 Transdermal Drug Delivery Systems Sizeable number of optimization studies on TDDS have been reported in the literature. The iontophoretic type of transdermal (TD) drug delivery systems have been optimized for penetration rate using independent variables as pH of the buffer, ionic strength in the solution, current amplitude, frequency of pulsed current,
Table 10. 14. Select optimization reports on oral controlled release formulations
Natural and semisynthetic polymer matrices Sr. No. Drug(s) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. Diltiazem HCl Ketorolac Verapamil HCl Diclofenac sodium Theophylline Nafronyl oxalate Theophylline Sodium sulfathiazole Aspirin Didanosine Aspirin Ibuprofen Des i g n FD FD CCD SCD CCD SMD MLRA FFD ANN D-Optimal ANN FD IV 2 3 2 4 3 3 2 5 2 3 2 3 5 4 3 3 4 RV 3 3 3 1 2 4 1 4 1 3 3 3 1 2 3 1 2 Reference Singh et al., 2004a Vatsaraj et al., 2002 Singh et al., 2002 Gohel et al., 1998 Matsumara et al., 1994 Waaler et al., 1992 Joly & Brossard, 1987 Harris et al., 1985 Ibric et al., 2003 Sanchez-Lafuente et al., 2002 Ibric et al., 2002 Khan et al., 1995 Li & Tu, 1991 Dangprasirt & Ritthidej, 1997 Wu et al., 2000 Sastry et al., 1997 Appel et al., 1992
Synthetic matrices
DextromethorphanHBr FFD Diclofenac sodium Salbutamol sulphate Atenolol Potassium chloride CCD ANN BBD FD
Solid dispersions Osmotic drug delivery systems
Gastroretentive dosage forms Floating dosage forms 18. 19. 20. 21. 22. 23. 24. 25. Calcium Captopril Calcium Sotalol HCl Furosemide Atenolol Diltiazem HCl Sotalol HCl HBr: Hydrobromide FD FD CCD CCD FD FD FD CCD 2 3 3 2 2 3 2 2 4 3 2 4 4 3 2 Li et al., 2003 Khattab et al., 2001 Li et al., 2001 Chueh et al., 1995 Menon et al., 1994 Singh et al., 2003a Singh & Ahuja, 2002 Chueh et al., 1995
Bioadhesive dosage forms
etc. The matrix type TDDS have been optimized for the concentration of solvents by studying the influence of drying conditions, taking drying time and temperature as the controllable independent variables. Also, TDDS have been optimized for in situ penetration rate and lag time using type and concentration of penetration enhancer, and vehicle composition as the factors. The methodologies used have been the FD and CCD. Table 10. 18 indicates the succinct list of various TDDS that have been optimized. 1.8.13 Inhalations Optimization techniques have also been exploited in inhalations and metered dose inhalers using formulation constituents as independent variables and physico-chemical stability, release profile, etc. as the response
Chapter 10.
307
Table 10. 15. Optimization reports on Microparticulates

Microspheres and Microcapsules Formulation optimization Sr. No. Drug(s) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. Nanoparticles Formulation optimization 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. Camptothecin Cyclosporin A Indomethacin Metipranolol Nimesulide Riboflavin Sodium ascorbate Tetracaine Taguchi FD FD FD CCD CCD & FD CCD CCD PBD FFD FD 4 3 3 2 2 3 5 2 4 2 3 3 1 1 4 3 3 4 4 2 3 4 Yang and Zhu, 2002 Calvo et al., 1996 Wehrl et al., 1995 Losa et al., 1993 Mller et al., 1996 Julienne et al., 1992 McLeod et al., 1988 Singh et al., 2005b Loukas, 2001 Loukas, 1998 Foldvari et al., 1993 Theophylline Felodipine 5-Fluorouracil Diltazem HCl Insulin Mitoxantrone Oxantrazole Cyclophosphamide Sulfamethoxazole Sodium chloride Des i g n FD CCD FD FD FD CCD CCD FD FD FD FD FFD FFD IV 2 3 2 2 2 4 5 2 3 2 3 5 6 RV 6 1 3 3 3 2 3 3 2 3 1 1 1 Reference Bayomi, 2003 Ko et al., 2003 Parikh et al., 2003 Singh & Agarwal, 2002 Hu et al., 2000 Luftensteiner & Vierstein, 1998 Hassan et al., 1992 Vural et al., 1991 Devay & Racz, 1984 Narayan et al., 2001 Bibby et al., 1998 Zeng et al., 1994 El-Banna & Effimova, 1982
Liposomal systems
variables. The reports are listed in Table 10. 19. 1.8.14 Miscellaneous Formulations Apart from the above formulations, optimization techniques have also been exploited in various other kind of formulations like films, lyophilized injectables, nonpareils, plaster, etc. Table 10. 20 enlists the reports on use of optimization in such cases. CONCLUSIONS Computer-assisted optimization using experimental designs, beyond doubt, has immense potential in development of diverse pharmaceutical products or processes. The industrial payoffs of the approach are
Table 10. 16. Select optimization reports on macroparticulates

Formulation optimization Sr. No. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Type Beads CR Beads CR Beads CR Pellets CR Pellets CR Granules Granules Pellets Spheroid Pellets Drug Ibuprofen Diltiazem HCl Verapamil HCl Bumetanide Propranolol HCl Indomethacin Des i g n FD RSM ANN RSM BBD FFD & SMD MLRA FD RSM FD IV 2 4 3 2 3 3 4 3 2 3 RV 4 2 1 2 4 5 2 5 2 6 Reference Maheshwari et al., 2003 El-Kamel et al., 2003 Vaithiyalingam & Khan, 2002 Hamed & Sakr, 2001 Bodea & Leucuta, 1998 Takayama & Nagai, 1989 Rambali et al., 2003 Paterakis et al., 2002 Heng et al., 1996 Lalla & Bhat, 1992
Table 10. 17. Select optimization reports on fast release dosage forms
Sr. No. 1. 2. 3. 4. 5. 6. 7. 8. Drug(s) Captopril Flurbiprofen Ibuprofen Roxithromycin Prochlorperazine maleate Prednisone Indomethacin Griseofulvin Des i g n BBD FD RCCD CCD FD SMD SMD CCD IV 3 2 3 3 2 3 3 2 RV 5 3 4 3 1 1 3 2 Reference Lee et al., 2003 Singh et al., 2003b Schiermeier & Schmidt, 2002 Weon et al., 2000 Nagarsenker & Garad, 1998 Ferrari et al., 1996 Takayama et al., 1985 Takai et al., 1984
Table 10. 18. Select optimization reports on transdermal drug delivery systems
Sr. No. 1. 2. 3. 4. 5. 6. 7. Type TD gel TDDS TD gel TDDS TD gel Iontophoretic TDDS TDDS Drug Tenoxicam Verapamil HCl Ketoprofen Terbutaline sulphate Ketoprofen Thyrotropin releasing hormone Nitrendipine Des i g n FD FD ANN FD ANN FD & CCD FD IV 2 3 3 3 2 4 3 RV 4 6 2 1 3 1 1 Reference Singh et al., 2004b Devi et al., 2003 Wu et al., 2001 Manna et al., 2000 Takayama et al., 1999 Huang et al., 1996 Giannakou et al., 1995
Table 10. 19. Optimization reports on inhalations

Sr. No. 1. 2. Type Inhalation Metered dose inhaler Drug Insulin Des i g n FD FCCD IV 2 2 RV 4 1 Reference St hl et al., 2002 Fedina et al., 1998
Chapter 10.
309
Table 10. 20. Optimization reports on miscellaneous formulations

Sr. No. 1. 2. 3. 4. 5. Type Aerosol dry powder Nonpareil seeds Films Lyophilized injectable Plaster Drug Flurbiprofen Ketoprofen Des i g n CCD FD RSM FFD IV 2 2 3 RV 2 1 4 Reference Bosquillon et al., 2001 Zhang et al., 1999 Nagarsenker & Hegde, 1999 Kagkadis et al., 1998 Akitoshi et al., 1985
phenomenal, as it offers product development solutions with fairly small involvement of men, materials, machination and money. A pharmaceutical scientist should earnestly consider the use of optimization studies particularly when finding the correct compromise is not simple and straightforward. Besides helping the formulation scientist in selecting a true optimum for the objectives, the techniques tend to reveal the degree of improvement in the formulation characteristics, both graphically and mathematically, as a function of the change in any ingredient or system parameter. Once the product is explained systematically using DoE models, one can determine how much a response variable gains or suffers with any alteration in the inputs proposed later. Such systematic experimental studies can also prove to be resourceful in the product and/or process validation and subsequent scale-up operations. Notwithstanding the outstanding benefits of DoE optimization, the experimenter should certainly not consider it as a panacea to all the product development problems, as there is no substitute to the pharmaceutical experience or brain power. A wise scientist can even choose the influential variables through his vast experience and observation, bypassing the rigours of screening and factor influence studies. The optimized formulation has to be one amongst the experimental domain investigated, as extrapolation outside the variable limits is neither advisable nor useful. As per the famous computer adage, "Garbage-In-Garbage-Out (GIGO)", the degree of predictability depends chiefly upon the accuracy and enormity of the input data. Incorrect choice of experimental design during computer-aided optimization can diminish the reliability of prognosis, while inappropriate choice of experimental range may either miss the optimum or require much more number of experiments to find the same. In any case, the variables, input and response, should both be quantifiable, and the relationship between them should be continuous over the range studied. The literature search indicates undeniably that the practice of formulation optimization has steadily increased over the past few years. Nonetheless, it is far from being adopted as a standard practice; many more efforts need to be made and many papers signifying the value of this method need to be published before this can happen. The broad selection of examples cited here has been used to highlight the growing utility of computer-based optimization in drug formulation research. Still, with short and rapid development cycles becoming the endeavour of pharmaceutical houses, DoE optimization would further gain increased acceptance as a priceless developmental tool. What is essentially needed for realization of this quality breakthrough is the workable basic knowledge of computers and statistics, coupled with persistence, patience, perseverance and passion. The day is not very far when these computer-based rational approaches of formulation development would be reaped to the fullest advantage by putting them into the routine practice in drug industry and research. ACKNOWLEDGEMENTS Research grants received from Council of Scientific and Industrial Research, New Delhi for funding the research projects on optimization of drug delivery devices is gratefully acknowledged. The authors appreciate the help rendered by Dr. (Ms.) Monika Bakshi, Mr. Madella Ram Babu, Mr. Rajesha B.C., Ms. Sukhwinder Kaur Chakkal and Ms. Ashu Rani in the extensive survey of optimization literature.
REFERENCES
Achanta, A. S.; Kowalski, J. G. and Rhodes, C. T. (1995) "Artificial neural networks: Implications for pharmaceutical sciences." Drug Dev. Ind. Pharm. 21, 119-155. Agarwal, V.; Singh, S. K.; Reddy, I. K.; Durrani, M. J. and Khan, M. A. (1999) "Cataplasm-based controlled drug delivery: Development and optimization of a novel formulation." Drug Dev. Ind. Pharm. 25, 659-665. Akitoshi, Y.; Takayama, K.; Machida, Y. and Nagai, T. (1985) "Computer optimization of the formulation of acrylic plaster." Chem. Pharm. Bull. 33, 4536-4543. AlKhatib, H. S. and Sakr, A. (2003) "Optimization of methacrylic acid ester copolymers blends as controlled release coatings using response surface methodology." Pharm. Dev. Technol. 8, 87-96. Appel, L. E.; Clair, J. H. and Zentner, G. M. (1992) "Formulation and optimization of a modified microporous cellulose acetate latex coating for osmotic pumps." Pharm. Res. 9, 1664-1667. Araujo, P. W. and Brereton, R. G. (1996) "Experimental design II Optimization." Trends in Anal. Chem. 15, 63-70. Banker, G. S. and Anderson, N. R. (1987) "Tablets." In: Lachman, L.; Lieberman, H. A. and Kanig, J. L. (Eds.) The theory and practice of industrial pharmacy. 3rd edn. Varghese Publishing House, Bombay, pp 293-345. Bayomi, M. A. (2003) "Influence of polymers weight ratio and pH of polymers solution on the characteristics of chitosancarboxymethyl cellulose microspheres containing theophylline." Boll. Chim. Farm. 142, 336-342. Belloto Jr, R. J.; Dean, A. M.; Moustafa, M. A.; Molokhia, A. M.; Gouda, M. W. and Sokoloski, T. D. (1985) "Statistical techniques applied to solubility predictions and pharmaceutical formulations: An approach to problem solving using mixture response surface methodology." Int. J. Pharm. 23, 195-207. Benkerrour, L.; Duchene, D.; Puisieux, F. and Maceario, J. (1984) "Granule and tablet formulae study by principal component analysis." Int. J. Pharm. 19, 27-34. Bibby, D. C.; Davies, N. M. and Tucker, I. G. (1998) "Preparation and characterization of beta-cyclodextrin and poly(acrylic acid) microspheres." J. Microencapsul. 15, 629-637. Bjerregaard, S.; Soderberg, I.; Vermehren, C. and Frokjaer, S. (1999) "Formulation and evaluation of release and swelling mechanism of a water-in-oil emulsion using factorial design." Int. J. Pharm. 193, 1-11. Bodea, A. and Leucuta, S. E. (1998) "Optimization of propranolol hydrochloride sustained release pellets using box-behnken design and desirability function." Drug Dev. Ind. Pharm. 24, 145-155. Bolton, S. (1987) "Statistical applications in the pharmaceutical sciences." In: Lachman, L.; Lieberman, H. A. and Kanig, J. L. (Eds.) The theory and practice of industrial pharmacy. 3rd edn. Varghese Publishing House, Bombay, pp 243-289. Bolton, S. (1990) "Pharmaceutical statistics: Practical and clinical applications." 2nd edn. Marcel Dekker, Inc., New York. Bosquillon, C.; Lombry, C.; Pr at, V. and Vanbever, R. (2001) "Influence of formulation excipients and physical characteristics of inhalation dry powders on their aerosolization performance." J. Control. Release 70, 329-339. Bourquin, J.; Schmidli, H.; van Hoogevest, P. and Leuenberger, H. (1997) "Application of artificial neural netrworks (ANN) in the development of solid dosage forms." Pharm. Dev. Tech. 2, 111-121. Box, G. E. P. and Wilson, K. B. (1951) "On the experimental attainment of optimum conditions." J. Royal Stat. Soc. Ser. B 13, 1-45. Box, G. E. P. and Behnken, D. W. (1960) "Some new three-level designs for the study of quantitative variables." Technometrics 2, 455-475. Box, G. E. P.; Connor, L. R.; Cousins, W. R.; Davies, O. L.; Himsworth, F. R.; Sillitto, G. P. and (Eds). (1960) "The design and analysis of industrial experiments." 2nd edn. Oliver and Boyd, London. Box, G. E. P. and Draper, N. R. (1987) "Empirical model-building and response surfaces." 1st edn. John Wiley & Sons, New York. Brisaert, M.; Gabriels, M. and Plaizier-Vercammen, J. (2000) "Investigation on the chemical stability of an erythromycintretinoin lotion by the use of an optimization system." Int. J. Pharm. 197, 153-160. Calvo, P.; Snchez, A.; Mart nez, J.; Lpez, M. I.; Calonge, M.; Pastor, J. C. and Alonso, M. J. (1996) "Polyester nanocapsules as new topical ocular delivery systems for cyclosporin A." Pharm. Res. 13, 311-315.
Chapter 10.
311
Ceschel, G. C.; Maffei, P. and Badiello, R. (1999) "Optimization of a tablet containing chlorthalidone." Drug Dev. Ind. Pharm. 25, 1167-1176. Chang, K. T. (1994) "Application of experimental design to optimize vitamin C 90 tabletting performance." J. Kor. Pharm. Sci. 24, S67-S77. Chariot, M.; Lewis, G. A.; Mathieu, D.; Phan Tan Luu, R. and Stevens, H. N. E. (1988) "Experimental design for pharmaceutical process characterisation and optimization using an exchange algorithm." Drug Dev. Ind. Pharm. 14, 2535-2556. Chowhan, Z. T.; Yang, I. C.; Amaro, A. A. and Chi, L. (1982) "Effect of moisture and crushing strength on tablet friability and in vitro dissolution." J. Pharm. Sci. 71, 1371-1375. Chu, J. S.; Amidon, G. L.; Weiner, N. D. and Goldberg, A. H. (1991) "Mixture experimental design in the development of a mucoadhesive gel formulation." Pharm. Res. 8, 1401-1407. Chueh, H. R.; Zia, H. and Rhodes, C. T. (1995) "Optimization of sotalol floating and bioadhesive extended release tablet formulations." Drug Dev. Ind. Pharm. 21, 1725-1747. Cochran, W. C. and Cox, G. M. (1992) "Experimental design." 2nd edn. John Wiley & Sons, Inc., New York. Cornell, J. A. (1990) "Experiments with mixtures: Designs, models, and the analysis of mixture data." 2nd edn. John Wiley & Sons, Inc., New York. Dangprasirt, P. and Ritthidej, G. C. (1997) "Development of diclofenac sodium controlled release solid dispersion tablet using optimization strategy." Drug Dev. Ind. Pharm. 23, 843-848. Das, M. N. and Giri, N. C. (1994) "Design and analysis of experiments." 2nd edn. Wiley Eastern Limited, New Age International Limited, New Delhi. Das, S. K.; Tucker, I. G.; Hill, D. J. and Ganguly, N. (1995) "Evaluation of poly(isobutylcyanoacrylate) nanoparticles for mucoadhesive ocular drug delivery. Part 1. Effect of formulation variables on physicochemical characteristics of nanoparticles." Pharm. Res. 12, 534-540. Dawoodbhai, S.; Suryanarayan, E. R. and Rhodes, C. T. (1991) "Optimization of tablet formulations containing talc." Drug Dev. Ind. Pharm. 17, 1343-1371. de Aguiar, P. F.; Bourguignon, B.; Khots, M. S.; Massart, D. L. and Phan Tan Luu, R. (1995) "D-optimal designs." Chemom. Intell. Lab. Syst. 30, 199-210. de Saavedra, M. S. and Cuadra, I. I. S. (2001) "Application of a mixed optimization strategy in the design of a pharmaceutical solid formulation at laboratory scale." Drug Dev. Ind. Pharm. 27, 675 - 685. Derringer, G. and Suich, R. (1980) "Simultaneous optimization of several response variables." J. Qual. Tech. 12, 214-219. Devay, A.; Kovacs, B.; Uderszky, J. and Domotor, Z. (1982) "Multifactorial design optimization of film coating technique." Pharm. Ind. 44, 830-838. Devay, A. and Racz, I. (1984) "Examination of parameters determining particle size of sulfamethoxazole microcapsules prepared by a new melt dispersion method." Pharm. Ind. 46, 101-103. Devay, A.; Uderszky, J. and Racz, I. (1988) "Formulation of effervescent tablets containing acetylsalicylic acid. Part 1. Investigation of granulation parameters." Acta Pharm. Hung. 58, 97-100. Devi, K. V.; Saisivam, S.; Maria, G. R. and Deepti, P. U. (2003) "Design and evaluation of matrix diffusion controlled transdermal patches of verapamil hydrochloride." Drug Dev. Ind. Pharm. 29, 495-503. Dincer, S. and Ozdurmus, S. (1977) "Mathematical model for enteric film coating of tablets." J. Pharm. Sci. 66, 1070-1073. Doehlert, D. H. (1970) "Uniform shell designs." Appl. Stat. 19, 231-239. Doornbos, C. A. and Haan, P. D. (1995) "Optimization techniques in formulation and processing." In: Swarbrick J. and Boylan, J. C. (Eds.) Encyclopedia of pharmaceutical technology. Marcel Dekker, New York, pp 77-160. Durig, T. and Fassihi, A. R. (1993) "Identification of stabilizing and destabilizing effects of excipient-drug interactions in solid dosage form design." Int. J. Pharm. 97, 161-170. El Banna, H. M. and Efimova, L. S. (1982) "Construction and use of factorial design in fluidized bed microencapsulation." Pharm. Ind. 44, 641-645. El-Kamel, A. H.; Al-Gohary, O. M. and Hosny, E. A. (2003) "Alginate-diltiazem hydrochloride beads: Optimization of formulation factors, in vitro and in vivo availability." J. Microencapsul. 20, 211-225.
Elkheshen, S. A.; Badawy, S. S. and Badawi, A. A. (1996) "Optimization of a reconstitutable suspension of rifampicin using 24 factorial design." Drug Dev. Ind. Pharm. 22, 623-630. Elkheshen, S. A.; Abd El-Gawad, N. A. and Badawi, A. A. (1997) "Factorial design and computer optimization of a reconstitutable suspension of erythromycin ethylsuccinate." Pharm. Ind. 59, 439-443. Fedina, L. T.; Zelko, R.; Fedina, L. I.; Szabados Z and Szanto, M. (1998) "Comparative evaluation of surfactant and hydrophobizing agent concentration in relation to the optimal particle size of metered-dose inhalers." Drug Dev. Ind. Pharm. 24, 853-856. Ferrari, F.; Bertoni, M.; Bonferoni, C. M.; Rossi, S. and Caramella, C. (1996) "Dissolution enhancement of an insoluble drug by physical mixture with a superdisintegrant: Optimization with a simplex lattice design." Pharm. Dev. Tech. 1, 159-164. Fisher, R. A. (1935) "The design of experiments." 1st edn. Oliver and Boyd, Edinburgh. Foldvari, M.; Gesztes, A.; Mezei, M.; Cardinal, L. and Behl, M. (1993) "Topical liposomal local anesthetics: Design, optimization and evaluation of formulations." Drug Dev. Ind. Pharm. 19, 2499-2517. Fonner, D. E.; Buck, J. R. and Banker, G. S. (1970) "Mathematical optimization techniques in drug product design and process analysis." J. Pharm. Sci. 59, 1587-1596. Gabrielsson, J.; Lindberg, N. O.; Palsson, M.; Nicklasson, F.; Sjostrom, M. and Lundstedt, T. (2003) "Multivariate methods in the development of a new tablet formulation." Drug Dev. Ind. Pharm. 29, 1053-1075. Gao, J. Z. H.; Jain, A.; Motheram, R.; Gray, D. B. and Hussain, M. A. (2002) "Fluid bed granulation of a poorly water soluble, low density, micronized drug: Comparison with high shear granulation." Int. J. Pharm. 237, 1-14. Giannakou, S. A.; Dallas, P. P.; Rekkas, D. M. and Choulis, N. H. (1995) "Development and in vitro evaluation of nitrendipine transdermal formulations using experimental design techniques." Int. J. Pharm. 125, 7-15. Gohel, M. C.; Jani, G. K.; Patel, N. K. and Gondaliya, D. P. (1998) "Optimization of hydrophilic matrix tablet formulation of diclofenac sodium using a mixture design." Pharm. Pharmacol. Commun. 4, 433-438. Hamed, E. and Sakr, A. (2001) "Application of multiple response optimization technique to extended release formulations design." J. Control. Release 73, 329-338. Harris, M. R.; Schwartz, J. B. and McGinity, J. W. (1985) "Optimization of a slow release tablet formulation containing sodium sulphathiazole and a montmorrilonite clay." Drug Dev. Ind. Pharm. 11, 1089-1110. Hassan, E. E.; Parish, R. C. and Gallo, J. M. (1992) "Optimized formulation of magnetic chitosan microspheres containing the anticancer agent, oxantrazole." Pharm. Res. 9, 390-397. Heng, P. W. S.; Wan, L. C. S. and Tan, Y. T. F. (1996) "Optimization of spheroid production by centrifugal rotary processing." Int. J. Pharm. 143, 107-112. Hu, Y. Q.; Guo, J. X.; Wang, L. J.; Tan, R. and Zhen, L. Y. (2000) "Preparation and evaluation of insulin-loaded polylactide microspheres using factorial design." Drug Dev. Ind. Pharm. 26, 1309-1313. Huang, Y. Y.; Wu, S. M. and Wang, C. Y. (1996) "Response surface method: A novel strategy to optimize iontophoretic transdermal delivery of thyrotrophin releasing hormone." Pharm. Res. 13, 547-552. Hutchings, D.; Kuzmak, B. and Sakr, A. (1994) "Processing considerations for an EC latex coating systems: Influence of curing time and temperature." Pharm. Res. 11, 1474-1478. Hwang, G. C.; Lin, A. Y.; Chen, W. and Sharpe, R. J. (1995) "Development and optimization of a methotrexate topical formulation." Drug Dev. Ind. Pharm. 21, 1941-1952. Ibric, S.; Jovanovic, M.; Djuric, Z.; Parojcic, J. and Solomun, L. (2002) "The application of generalized regression neural network in the modeling and optimization of aspirin extended release tablets with eudragit RS PO as matrix substance." J. Control. Release 82, 213-222. Ibric, S.; Jovanovic, M.; Djuric, Z.; Parojcic, J.; Petrovic, S.; Solomun, L. and Stupar, B. (2003) "Artificial neural networks in the modeling and optimization of aspirin extended release tablets with eudragit L 100 as matrix substance." AAPS PharmSciTech. 4, 62-70. Irvin, J. R. and Notari, R. E. (1991) "Computer-aided dosage form design. III. Feasibility assessment for an oral prolongedrelease phenytoin product." Pharm. Res. 8, 232-237. Iskandarani, B.; Clair, J. H.; Shiromani, P. K. and Dempski, R. E. (1993) "Simultaneous optimization of capsule and tablet
Chapter 10.
313
formulation using response surface methodology." Drug Dev. Ind. Pharm. 19, 2089-2101. Iwata, M.; Takayama, K.; Machida, Y.; Shirotake, S. and Nagai, T. (1995) "Formulation optimization of progesterone suppository using mixtures of witepsol and ethylene vinyl acetate copolymer as a base." Jap. J. Hosp. Pharm. 21, 269273. Joly, F. and Brossard, C. (1987) "Development of a hydrophilic matrix for theophylline. Part 1. Physical and mechanical properties of materials and optimization of formulation." STP Pharma 3, 556-568. Jones, D. S.; Woolfson, A. D.; Djokic, J. and Coulter, W. A. (1996) "Development and mechanical characterization of bioadhesive semi-solid, polymeric systems containing tetracycline for the treatment of periodontal diseases." Pharm. Res. 13, 1734-1738. Julienne, M. C.; Alonso, M. J.; Gomez Amoza, J. L. and Benoit, J. P. (1992) "Preparation of poly(d,l-lactide/glycolide) nanoparticles of controlled particle size distribution: Application of experimental designs." Drug Dev. Ind. Pharm. 18, 1063-1077. Kagkadis, K. A.; Rekkas, D. M.; Dallas, P. P. and Choulis, N. H. (1998) "A freeze-dried injectable form of flurbiprofen: Development and optimization using response surface methodology." Int. J. Pharm. 161, 87-94. Kettaneh-Wold, N. (1991) "Use of experimental design in the pharmaceutical industry." J. Pharm. Biomed. Anal. 9, 605-610. Keyhani-Morrison, E.; Nadkarni, S. and Sakr, A. (1995) "Optimization of gel formulations of a novel retinoid based on experimental design." Pharm. Ind. 57, 333-337. Khan, M. A.; Karnachi, A. A.; Singh, S. K.; Sastry, S. V. and Bolton, S. (1995) "Controlled release coprecipitates: Formulation considerations." J. Control. Release 37, 131-141. Khattab, I. S.; Abdel-rahman, S.; Khidr, S. and Abdel-Hakiem, O. (2001) "Development and optimization of sustained release captopril tablets and capsules." Egyptian J. Biotech. 10, 432-445. Ko, J. A.; Park, H. J.; Park, Y. S.; Hwang, S. J. and Park, J. B. (2003) "Chitosan microparticle preparation for controlled drug release by response surface methodology." J. Microencapsul. 20, 791-797. Kramar, A.; Turk, S. and Vrecer, F. (2003) "Statistical optimisation of diclofenac sustained release pellets coated with polymethacrylic films." Int. J. Pharm. 256, 43-52. Lalla, J. and Bhat, S. (1992) "Non-pareil seed: 23! Factorial approach to optimization of coating for pellet preparation: Coating pan vs dish pelletizer." Indian Drugs 29, 527-536. Lee, K. J.; Kang, A.; Delfino, J. J.; West, T. G.; Chetty, D.; Monkhouse, D. C. and Yoo, J. (2003) "Evaluation of critical formulation factors in the development of a rapidly dispersing captopril oral dosage form." Drug Dev. Ind. Pharm. 29, 967-979. Lemaitre-Aghazarian, V.; Piccerelle, P.; Reynier, J. P.; Joachim, J.; Phan-Tan-Luu, R. and Sergent, M. (2004) "Texture optimization of water-in-oil emulsions." Pharm. Dev. Technol. 9, 125-134. Lewis, G. A.; Mathieu, D. and Phan-Tan-Luu, R. (1999) "Pharmaceutical experimental design." 1st edn. Marcel Dekker, New York; Singh, B.; Ahuja, N. (2000) Book Review on "Pharmaceutical Experimental Design" Int. J. Pharm. 195, 247-248. Lewis, G. A. (2002) "Optimization methods." In: Swarbrick, J. and Boylan, J. C. (Eds.) Encyclopedia of pharmaceutical technology. 2 edn. Marcel Dekker, Inc., New York, pp 1922 - 1937. Li, J. (2003) "Factorial designs." In: Chow, S.-C. (Ed.) Encyclopedia of biopharmaceutical statistics. Marcel Dekker Inc., New York, pp 364 - 378. Li, L. C. and Peck, G. E. (1991) " Water based silicone elastomer controlled release tablet film coating V approach." Drug Dev. Ind. Pharm. 17, 27- 38. A statistical
Li, L. C. and Tu, Y. H. (1991) "In vitro drug release from matrix tablets containing a silicone elastomer latex." Drug Dev. Ind. Pharm. 17, 2197-2214. Li, S.; Lin, S.; Chien, Y. W.; Daggy, B. P. and Mirchandani, H. L. (2001) "Statistical optimization of gastric floating system for oral controlled delivery of calcium." AAPS PharmSciTech 2, 1-12. Li, S.; Lin, S.; Daggy, B. P.; Mirchandani, H. L. and Chien, Y. W. (2003) "Effect of HPMC and carbopol on the release and floating properties of gastric floating drug delivery system using factorial design." Int. J. Pharm. 253, 13-22. Losa, C.; Marchal Heussler, L.; Orallo, F.; Jato, J. L. V. and Alonso, M. J. (1993) "Design of new formulations for topical
ocular administration: Polymeric nanocapsules containing metipranolol." Pharm. Res. 10, 80-87. Loukas, Y. L. (1998) "Computer-based expert system designs and analyzes a 2(k-p) fractional factorial design for the formulation optimization of novel multicomponent liposomes." J. Pharm. Biomed. Anal. 17, 133-140. Loukas, Y. L. (2001) "A Plackett-Burman screening design directs the efficient formulation of multicomponent drv liposomes." J. Pharm. Biomed.l Anal. 26, 255-263. Luftensteiner, C. P. and Viernstein, H. (1998) "Statistical experimental design based studies on placebo and mitoxantroneloaded albumin microspheres." Int. J. Pharm. 171, 87-99. Maheshwari, M.; Ketkar, A. R.; Chauhan, B.; Patil, V. B. and Paradkar, A. R. (2003) "Preparation and characterization of ibuprofen-cetyl alcohol beads by melt solidification technique: Effect of variables." Int. J. Pharm. 261, 57-67. Mahmud, A. and Po, A. L. W. (1991) "Investigation of the effect of additives on the dissolution rates of aspirin and paracetamol using a factorial design." Drug Dev. Ind. Pharm. 17, 709-724. Malinowski, H. J. and Smith, W. E. (1974) "Effects of spheronization process variables on selected tablet properties." J. Pharm. Sci. 63, 285-288. Malinowski, H. J. and Smith, W. E. (1975) "Use of factorial design to evaluate granulations prepared by spheronization." J. Pharm. Sci. 64, 1688-1692. Manna, A.; Ghosh, I.; Sen, N.; Thakur, R. S.; Ghosh, L. K. and Gupta, B. K. (2000) "Statistical optimization of transdermal drug delivery system of terbutaline sulfate by factorial analysis." Boll. Chim. Farm. 139, 34-40. Marlowe, E. and Shangraw, R. F. (1967) "Dissolution of sodium salicylate from tablet matrices prepared by wet granulation and direct compression." J. Pharm. Sci. 56, 498-504. Marti-Mestres, G.; Nielloud, F.; Fortune, R.; Fernandez, C. and Maillols, H. (2000) "Effect of the addition of oxybenzone or octyl methoxycinnamate on particle size of submicron emulsions." Drug Dev. Ind. Pharm. 26, 349-356. Matsumara, M.; Nakagami, H.; Yamao, T.; Takayama, K. and Nagai, T. (1994) "Computer optimization for the formulation of controlled release theophylline tablet made of micronized low substituted hydroxypropylcellulose and methylcellulose." Chem. Pharm. Bull. 42, 1902-1908. McGurk, J. G.; Lendrem, D. W. and Potter, C. J. (1991) "Use of statistical experimental design in laboratory scale formulation optimization and progression to plant scale." Drug Dev. Ind. Pharm. 17, 2341-2358. McLeod, A. D.; Lam, F. C.; Gupta, P. K. and Hung, C. T. (1988) "Optimized synthesis of polyglutaraldehyde nanoparticles using central composite design." J. Pharm. Sci. 77, 704-710. Menon, A.; Ritschel, W. A. and Sakr, A. (1994) "Development and evaluation of a monolithic floating dosage form for furosemide." J. Pharm. Sci. 83, 239-245. Menon, A.; Dhodi, N.; Mandella, W. and Chakrabarti, S. (1996) "Identifying fluid-bed parameters affecting product variability." Int. J. Pharm. 140, 207-218. Miyamoto, Y.; Ogawa, S.; Miyajima, M.; Matsui, M.; Sato, H.; Takayama, K. and Nagai, T. (1997) "An application of the computer optimization technique to wet granulation process involving explosive growth of particles." Int. J. Pharm. 149, 25-36. Montgomery, D. C. (2001) "Design and analysis of experiments." 5th edn. John Wiley & Sons, Inc., New York. Mller, B. G.; Leuenberger, H. and Kissel, T. (1996) "Albumin nanospheres as carriers for passive drug targeting: An optimized manufacturing technique." Pharm. Res. 13, 32-37. Murphy, J. R. (2003) "Screening design." In: Chow, S.-C. (Ed.) Encyclopedia of biopharmaceutical statistics. Marcel Dekker Inc., New York, pp 920. Myers, R. H. (1990) "Classical and modern regression with applications." PWS-KENT Publishing, Boston. Myers, R. H. and Montgomery, D. C. (2002) " Response surface methodology: Process and product optimization using designed experiments." 2 edn. John Wiley & Sons, New York. Myers, W. R. (2003) "Response surface methodology." In: Chow, S.-C. (Ed.) Encyclopedia of biopharmaceutical statistics. Marcel Dekker Inc., New York, pp 858 - 869. Nagarsenker, M. S. and Garad, S. D. (1998) "Physical characterization and optimization of dissolution parameters of prochlorperazine maleate coevaporates." Int. J. Pharm. 160, 251-255.
Chapter 10.
315
Nagarsenker, M. S. and Hegde, D. D. (1999) "Optimization of the mechanical properties and water-vapor transmission properties of free films of hydroxypropylmethylcellulose." Drug Dev. Ind. Pharm. 25, 95-98. Narayan, P.; Marchant, D. and Wheatley, M. A. (2001) "Optimization of spray drying by factorial design for production of hollow microspheres for ultrasound imaging." J. Biomed. Mater. Res. 56, 333-341. Newton, J. M.; Rowley, G. and Tornblom, J. F. (1971) "Further studies on the effect of additives on the release of drug from hard gelatin capsules." J. Pharm. Pharmacol. 23, 156S-160S. Newton, J. M. and Razzo, F. N. (1974) "Influence of additives on the release of drugs from HG capsules." J. Pharm. Pharmacol. 26, 30P. Ngo, T. H.; Vertommen, J. and Kinget, R. (1997) "Formulation of artemisinin tablets." Int. J. Pharm. 146, 271-274. Nguyen, N. A.; Wells, M. L. and Cooper, D. C. (1995) "Identification of factors affecting preservative efficacy and chemical stability of lamivudine oral solution through statistical experimental design." Drug Dev. Ind. Pharm. 21, 1671-1682. Ozil, P. and Rochat, M. H. (1988) "Experimental design, an efficient tool for studying the stability of parenteral nutrition." Int. J. Pharm. 42, 11-14. Parikh, R. H.; Parikh, J. R.; Dubey, R. R.; Soni, H. N. and Kapadia, K. N. (2003) "Poly(d,l-lactide-co-glycolide) microspheres containing 5-fluorouracil: Optimization of process parameters." AAPS PharmSciTech. 4, 103-110. Patel, J. P.; Marsh, K.; Carr, L. and Nequist, G. (1990) "Factorial designs in ophthalmic formulation development of enalkiren." Int. J. Pharm. 65, 195-200. Paterakis, P. G.; Korakianiti, E. S.; Dallas, P. P. and Rekkas, D. M. (2002) "Evaluation and simultaneous optimization of some pellets characteristics using a 3(3) factorial design and the desirability function." Int. J. Pharm. 248, 51-60. Pattarino, F.; Marengo, E.; Gasco, M. R. and Carpignano, R. (1993) "Experimental design and partial least squares in the study of complex mixtures: Microemulsions as drug carriers." Int J. Pharm. 91, 157-165. Peck, G. E.; Baley, G. J.; McCurdy, V. E. and Banker, G. S. (1989) "Tablet formulation and design." In: Lieberman, H.; Lachman, L. and Schwartz, J. (Eds.) Pharmaceutical dosage forms tablets. 2nd Edition, Revised and Expanded edn. Marcel Dekker, Inc., New York and Basel, pp 83-86. Plackett, R. L. and Burman, J. P. (1946) "The design of optimum multifactorial experiments." Biometrica 33, 305-325. Plumb, A. P.; Rowe, R. C.; York, P. and Doherty, C. (2003) "Effect of varying optimization parameters on optimization by guided evolutionary simulated annealing (GESA) using a tablet film coat as an example formulation." Eur. J. Pharm. Sci. 18, 259-266. Podczeck, F. (1996) "The development and optimization of tablet formulations using mathematical methods." In: Banker, G. S. and Rhodes, C. T. (Eds.) Modern pharmaceutics. Marcel Dekker, Inc., New York, pp 561-593. Potter, C. D. (1994) "Experiment Design Software: Better Data, Less Work." The Scientist 8, 18-20. Rambali, B.; Baert, L. and Massart, D. L. (2001) "Using experimental design to optimize the process parameters in fluidized bed granulation on a semi-full scale." Int. J. Pharm. 220, 149-160. Rambali, B.; Van Aelst, S.; Baert, L. and Massart, D. L. (2003) "Using deepest regression method for optimization of fluidized bed granulation on semi-full scale." Int. J. Pharm. 258, 85-94. Sabir, A.; Evans, B. and Jain, S. (2001) "Formulation and process optimization to eliminate picking from market image tablets." Int. J. Pharm. 215, 123-135. Sanchez-Lafuente, C.; Furlanetto, S.; Fernandez-Arevalo, M.; Alvarez-Fuentes, J.; Rabasco, A. M.; Faucci, M. T.; Pinzauti, S. and Mura, P. (2002) "Didanosine extended-release matrix tablets: Optimization of formulation variables using statistical experimental design." Int. J. Pharm. 237, 107-118. Sanderson, I. M.; Kennerley, J. W. and Parr, G. D. (1984) "Evaluation of the relative importance of formulation and process variables using factorial design." J. Pharm. Pharmacol. 36, 789-795. Sastry, S. V.; Reddy, I. K. and Khan, M. A. (1997) "Atenolol gastrointestinal therapeutic system: Optimization of formulation variables using response surface methodology." J. Control. Release 45, 121-130. Scheff , H. (1958) "Experiments with mixtures." J. Royal Stat. Soc. Ser. B 20, 344-360. Schiermeier, S. and Schmidt, P. C. (2002) "Fast dispersible ibuprofen tablets." Eur. J. Pharm. Sci. 15, 295-305. Schwartz, J.; Flamholz, J. and Press, R. (1973) "Computer optimization of pharmaceutical formulations I: General procedure."
J. Pharm. Sci. 62, 1165-1170. Schwartz, J. B. and Connor, R. E. (1996) "Optimization techniques in pharmaceutical formulation and processing." In: Banker, G. S. and Rhodes, C. T. (Eds.) Modern pharmaceutics. 3rd edn. Marcel Dekker Inc., New York, pp 727-752. Shekh, E.; Ghani, M. and Jones, R. E. (1980) "Simplex search in optimization of capsule formulation." J. Pharm. Sci. 69, 1135-1142. Shirakura, O.; Yamada, M.; Hashimoto, M.; Ishimaru, S.; Takayama, K. and Nagai, T. (1991) "Particle size design using computer optimization technique." Drug Dev. Ind. Pharm. 17, 471-483. Singh, B. (1997) "Computer-aided education in pharmaceutical sciences." Indian J. Pharm. Educ. 31, 93-102. Singh, B.; Kumar, R. and Ahuja N. (2005a) "Optimizing drug delivery systems using systematic "design of experiments". Part I: Fundamental aspects." Crit. Rev. Ther. Drug Carrier Syst. 22, 27-105. Singh, B.; Mehta, G.; Kumar, R.; Bhatia, A.; Ahuja, N. and Katare, O. P. (2005b) "Design, development and optimization of nimesulide-loaded liposomal systems for topical application." Curr. Drug Deliv. 2, 143-153. Singh, B.; Dahiya, M.; Saharan, V. and Ahuja N. (2005c) "Optimizing drug delivery systems using systematic "design of experiments". Part II: Retrospect and prospects." Crit. Rev. Ther. Drug Carrier Syst. 22, 215-292. Singh, B. and Agarwal, R. (2002) "Design development and optimization of controlled release microcapsules of diltiazem hydrochloride." Indian J. Pharm. Sci. 64, 378-385. Singh, B. and Ahuja, N. (2002) "Development of controlled-release buccoadhesive hydrophilic matrices of diltiazem hydrochloride: Optimization of bioadhesion, dissolution, and diffusion parameters." Drug Dev. Ind. Pharm. 28, 433-444. Singh, B.; Buhary, M. S.; Ahuja, N. and Chaturvedi, S. C. (2002) Formulation optimization and characterization of hydrophilic matrices of verapamil hydrochloride using central composite design (CCD). Proceedings of 54th Indian Pharm. Congress, Pune, A 54. Singh, B. (2003) Invited lecture on "Computer-aided design and optimization of diverse drug delivery systems." FAMU College of Pharmacy, Florida A & M University, Tallahassee, USA, 17 October, 2003. Singh, B.; Chakkal, S. K. and Ahuja, N. (2003a) Computer-aided design, development and optimization of controlled release mucoadhesive formulations of atenolol. Proceedings of National Seminar on "Pharmaceutics in the Light of Drug Delivery Challenges". Panjab University, Chandigarh, P 18. Singh, B.; Singla, Y. P.; Dhiman, A. and Ahuja, N. (2003b) Formulation, optimization and characterization of flurbiprofen solid dispersions using biodegradable polymers. Proceedings of National Seminar on "Pharmaceutics in the Light of Drug Delivery Challenges". Panjab University, Chandigarh P 1. Singh, B. and Ahuja, N. (2004) "Response surface optimization of drug delivery system." In: Jain, N. K. (Ed.) Progress in controlled and novel drug delivery systems. 1st edn. CBS Publishers & Distributors, New Delhi, pp 470-509. Singh, B.; Gupta, B. and Rani, A. (2004a) Design and development of oral controlled release hydrophilic matrices: Computeraided optimization of drug release characteristics. Proceedings of Indian Science Congress, January 3-7, Panjab University, Chandigarh, A 16. Singh, B.; Singh, S. K. and Goel, P. (2004b) Formulation, optimization and evaluation of transdermal hydrogels of tenoxicam. Proceedings of Indian Science Congress, January 3-7, Panjab University, Chandigarh, A 4. So, S.-S. and Karplus, M. J. (1996) "Evolutionary optimization in quantitative structure activity relationship: An application of genetic neural networks." J. Med. Chem. 39, 1521-1530. Stack , C. B. (2003) "Confounding and interaction." In: Chow, S.-C. (Ed.) Encyclopedia of biopharmaceutical statistics. Marcel Dekker Inc., New York, pp 235 - 240. Sthl, K.; Claessonb, M.; Lilliehornc, P.; Lind nd, H. and B ckstrmd, K. (2002) "The effect of process variables on the degradation and physical properties of spray dried insulin intended for inhalation." Int. J. Pharm. 233, 227-237. Stetsko, G. (1986) "Statistical experimental design and its application to pharmaceutical development problem." Drug Dev. Ind. Pharm. 12, 1109-1123. Takai, T.; Takayama, K.; Nambu, N. and Nagai, T. (1984) "Optimum formulation of griseofulvin/hydroxypropylcellulose solid dispersions with desirable dissolution properties." Chem. Pharm. Bull. 32, 1942-1947. Takayama, K.; Imaizumi, H.; Nambu, N. and Nagai, T. (1985) "Mathematical optimization of formulation of indomethacin/
Chapter 10.
317
polyvinylpyrrolidone /methyl cellulose solid dispersions by the sequential unconstrained minimization technique." Chem. Pharm. Bull. 33, 292-300. Takayama, K. and Nagai, T. (1989) "Novel computer optimization methodology for pharmaceutical formulations investigated by using sustained release granules of indomethacin." Chem. Pharm. Bull. 37, 160-167. Takayama, K.; Okabe, H.; Obata, Y. and Nagai, T. (1990) "Formulation design of indomethacin gel ointment containing d limonene using computer optimization methodology." Int. J. Pharm. 61, 225-234. Takayama, K.; Takahara, J.; Fujikawa, M.; Ichikawa, H. and Nagai, T. (1999) "Formula optimization based on artificial neural networks in transdermal drug delivery." J. Control. Release 62, 161-170. Takayama, K.; Fujikawa, M.; Obata, Y. and Morishita, M. (2003) "Neural network based optimization of drug formulations." Adv. Drug Deliv. Rev. 55, 1217-1231. Thoennes, C. J. and McCurdy, V. E. (1989) "Evaluation of a rapidly disintegrating moisture resistant lacquer film coating." Drug Dev. Ind. Pharm. 15, 165-185. Turkoglu, M. and Sakr, A. (1992) "Mathematical modeling and optimization of a rotary fluidized bed coating process." Int. J. Pharm. 88, 75-87. Tye, H. (2004) "Application of statistical 'design of experiments' methods in drug discovery." Drug Discovery Today 9, 485491. Vaithiyalingam, S. and Khan, M. A. (2002) "Optimization and characterization of controlled release multi-particulate beads formulated with a customized cellulose acetate butyrate dispersion." Int. J. Pharm. 234, 179-193. Van Kamp, H. V.; Bolhuis, G. K. and Lerk, C. F. (1987) "Optimization of a formulation for direct compression using a simplex lattice design." Pharm.Weekbl. Sci. Edn. 9, 265-271. Vatsaraj, N.; Zia, H. and Needham, T. (2002) "Formulation and optimization of a sustained-release tablet of ketorolac tromethamine." Drug Deliv. 9, 153-159. Voinovich, D.; Campisi, B.; Moneghini, M. ; Vincenzi, C. and Phan-Tan-Luu, R. (1999) "Screening of high shear mixer melt granulation process variables using an asymmetrical factorial design." Int. J. Pharm. 190, 73-81. Vojnovic, D.; Moneghini, M. and Rubessa, F. (1994) "Optimization of granulates in a high shear mixer by mixture design." Drug Dev. Ind. Pharm. 20, 1035-1047. Vural, I.; Kas, H. S.; Oner, L. and Hincal, A. A. (1991) "Cyclophosphamide loaded albumin microspheres. Part 4. Application of factorial design." S.T.P. Pharm. Sci. 1, 318-320. Waaler, P. J.; Graffner, C. and Muller, B. W. (1992) "Optimization of a matrix tablet formulation using a mixture design." Acta Pharm. Nordica. 4, 9-16. Wehrl , P.; Nobelis, P.; Cuin , A. and Stamm, A. (1993) "Response surface methodology: An interesting statistical tool for process optimization and validation: Example of wet granulation in a high shear mixer." Drug Dev. Ind. Pharm. 19, 16371653. Wehrl , P.; Magenheim, B. and Benita, S. (1995) "Influence of process parameters on the PLA nanoparticle size distribution, evaluated by means of factorial design." Eur. J. Pharm. Biopharm. 41, 19-26. Weon, K. Y.; Lee, K. T. and Seo, S. H. (2000) "Optimization study on the formulation of roxithromycin dispersible tablet using experimental design." Arch. Pharm. Res. 23, 507-512. Worakul, N.; Wongpoowarak, W. and Boonme, P. (2002) "Optimization in development of acetaminophen syrup formulation." Drug Dev. Ind. Pharm. 28, 345 - 351. Wu, P.-C.; Obata, Y.; Fujikawa, M.; Li, C. J.; Higashiyama, K. and Takayama, K. (2001) "Simultaneous optimization based on artificial neural networks in ketoprofen hydrogel formula containing o-ethyl-3-butylcyclohexanol as percutaneous absorption enhancer." J. Pharm. Sci. 90, 1004-1014. Wu, T.; Pan, W.; Chen, J. and Zhang, R. (2000) "Formulation optimization technique based on artificial neural network in salbutamol sulfate osmotic pump tablets." Drug Dev. Ind. Pharm. 26, 211-215. Yang, S. C. and Zhu, J. B. (2002) "Preparation and characterization of camptothecin solid lipid nanoparticles." Drug Dev. Ind. Pharm 28, 265 - 274. Zeng, X. M.; Martin, G. P. and Marriott, C. (1994) "Albumin microspheres as a means of drug delivery to the lung: Analysis
of the effects of process variables on particle sizes using factorial design methodology." Int. J. Pharm. 107, 205-210. Zhang, T.; Chan, L. and Zhang, W. (1999) "Computer-aided optimization of non-pareil seeds using central composite design." Zhongyuo Yiyao Gongye Zazhi 30, 444-449. Zupancic Boic, D.; Vrecer, F. and Kozjek, F. (1997) "Optimization of diclofenac sodium dissolution from sustained release formulations using an artificial neural network." Eur. J. Pharm. Sci. 5, 163-169.

Chapter 10, CBS

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Chapter 10, CBS

Hochgeladen von

Copyright:

Verfügbare Formate

Chapter 10.

COMPUTER-ASSISTED OPTIMIZATION OF PHARMACEUTICAL FORMULATIONS AND PROCESSES

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

274 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

OPTIMZATION: BASIC CONCEPTS AND TERMINOLOGY

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

276 Pharmaceutical Product Development

INTERACTION Low Polymer level

Low polymer level

High Polymer level

High Polymer level High

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

278 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

(1) (2) (3)

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

280 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

282 Pharmaceutical Product Development

Table 10. 2. A Plackett-Burman design for 8 experiments

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

284 Pharmaceutical Product Development

Table 10. 3. Design layout as per Box-Behnken design.*

* data taken from Kramar et al., 2003.

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

Y = 1X1 + 2 X 2 + 3 X 3 Y = 1X1 + 2 X 2 + 3 X 3 + 12 X1X 2 + 13 X1X 3 + 23 X 2 X 3

(6) ...(7) (8)

Special cubic model: Y = 1X1 + 2 X 2 + 3 X 3 + 12 X1X 2 + 13 X1X 3 + 23 X 2 X 3 + 123 X1X 2 X 3

Croscarmellose Sodium Dicalcium Phosphate Dihydrate Anhydrous -Lactose

* data taken from Ferrari et al., 1996.

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

286 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

Star Central composite

E( y ) = 0 + 1X1 + 2 X 2 + 12 X1X 2 + ...

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

288 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

Run Num ber

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

290 Pharmaceutical Product Development

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

O v erlay P lot X = A : hpm ck15m Y = B : nacm c

(11) (12) (13)

i.e., inequality constraint G(X ) = f1(X1, X 2 ) 0 i.e., equality constraint

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

292 Pharmaceutical Product Development

a priori knowledge of the response surface not essential, and

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com

Computer-Assisted Optimization of Pharmaceutical Formulations and Processes

PDF created with FinePrint pdfFactory trial version http://www.fineprint.com