Community Dent Oral Epidemiol 2012; 40: 3745 All rights reserved

2011 John Wiley & Sons A/S

Development of the Dundee Caries Risk Assessment Model (DCRAM) risk model development using a novel application of CHAID analysis
MacRitchie HMB, Longbottom C, Robertson M, Nugent Z, Chan K, Radford JR, Pitts NB. Development of the Dundee Caries Risk Assessment Model (DCRAM) risk model development using a novel application of CHAID analysis. Community Dent Oral Epidemiol 2012; 40: 3745. 2011 John Wiley & Sons A S Abstract Objectives: To use a novel statistical analysis in the development of caries risk assessment models for preschool children for use in a particular community setting. Methods: Data were collected longitudinally on a cohort of approximately 1500 children born in one calendar year in the city of Dundee, Scotland. A dental examination and oral microbiological saliva sample, together with parental and health visitor questionnaires, were completed for each child at ages 1, 2, 3 and 4 years. The 1-year data were analysed using chi-squared automated interaction detector analysis (CHAID) to produce a set of caries risk assessment models for predicting caries in 4-year-olds. Results: Four risk models were developed using CHAID analysis for caries at 4 years of age using risk assessment data collected at age 1. These models included two any caries risk models (n = 697, dmft >0) and two high cariesrisk models (n = 784, dmft 3) depending on the use of the d1 (enamel and dentine) or d3 (dentine only) level of caries detection. The most appropriate model developed for use was shown to be the CHAID high cariesrisk model at the d3 level of detection (d3mft 3). This had a sensitivity of 65% and specicity of 69%. Conclusions: An appropriate risk assessment model for use in a particular community setting predicting caries at age 4 years from data collected at age 1 year was developed. This has been termed the Dundee Caries Risk Assessment Model.

Heather M. B. MacRitchie1, Christopher Longbottom2, Margaret Robertson3, Zoann Nugent2, Karen Chan2, John R. Radford4 and Nigel B. Pitts2
1 Kings Cross Health and Community Care Centre, Dundee, UK, 2Dental Health Services and Research Unit, Dundee, UK, 3NHS Tayside, Dundee, UK, 4University of Dundee Dental Hospital and School, Dundee, UK

Key words: community; dental caries; preschool children; risk assessment model Dr. Heather M. B. MacRitchie, BDS, BMSc, PhD, MSND RCS, Dental Department, Kings Cross Health and Community Care Centre, Hospital Street, Dundee DD3 8EA, UK Tel.: +44 (0)1382 596990 Fax: +44 (0)1382 596995 e-mail: heather.macritchie@nhs.net Submitted 23 July 2008; accepted 1 July 2011

Scottish epidemiological surveys have shown that the prevalence of dentinal decay in Scotland (as measured by d3mft) is much higher than in England and Wales (1). They also emphasize the skew, or polarization, of decay within the population, a trend evident in many populations (2). Most recently, in Scotland, all established decay in 5year-old children was observed in 38% of the population, and half of the severe decay was found in just 3% (1). These high cariesrisk children are often the most difcult to access and are the least
doi: 10.1111/j.1600-0528.2011.00630.x

likely to be registered with a dentist, especially in the preschool years when dental habits are being established (3). Therefore, if caries prevalence in young children is to be reduced, some method is required to enable any healthcare personnel in contact with very young children to quickly assess caries risk status prior to the development of overt manifestation of the disease and direct children to dental services for preventive care. A simple but scientically robust caries prediction model for use in the community by healthcare teams could

37

MacRitchie et al.

facilitate such focussed prevention in a cost-effective fashion, without reducing the community wide benets of mass preventive methods (such as water uoridation) a so-called twin-track approach. Although opinions vary as to the merits of the risk targeting approach (48), Burt (9) concluded that a combination of some population-based preventive measure and some geographical targeting would seem to be the best way to proceed to reduce the burden of caries in our children. As noted by Zero et al. (10), risk assessment must be considered a necessary component in the clinical decision-making process. The development of caries prediction models to date has involved complex statistical techniques with resultant complex risk models (1122). The aim of this study was to develop a user-friendly caries prediction model for caries at age 4 years from data collected at age 1 year, within a particular community setting, using a novel application of chi-squared automated interaction detector analysis (CHAID) analysis.

Method
The methods employed in this study were part of a 4-year longitudinal study and have been described elsewhere in detail (6, 23, 24). In summary, a total of 56 variables were investigated as potential caries risk markers, and these data were collected annually over a 4-year period by the following means:

Microbiological saliva sampling. The tongue-loop method of saliva sampling (26) was carried out by the childs health visitor (HV), and mutans streptococci, lactobacilli and yeasts were cultured and analysed as described previously (23, 24). Health Visitor Questionnaire (HVQ). This annual questionnaire was completed by the childs HV, as an addition to their routine child health monitoring programme, at the same time as saliva sampling or at a later date. HVs are qualied nurses working within the community with a wide remit that includes child health monitoring. All 57 HVs employed in Dundee at that time participated in the study. The HVQ provided data such as height, weight and head circumference; immunization status; ethnic origin, illnesses, medication, weaning, use of comforter, vitamin supplementation, feeding problems, family history, parental employment, parental health, parental smoking and housing status. One of the initial questions asked the HVs to give their opinion as to whether the child was at caries risk (Yes No). This was a subjective assessment (hunch) and involved no specic training or calibration. The Deprivation Category (DEPCAT) score (27), a measure of deprivation, was also obtained from information provided by the questionnaire. Parental Questionnaire. This annual questionnaire was given to the parent guardian of the study child by the HV at the time of saliva sampling. This provided data on breast bottle feeding, meals, drinks, snacks, toothbrushing, uoride supplementation as well as other sociodemographic variables.

Subjects
The target cohort consisted of all children born and resident in Dundee, Scotland, in one complete calendar year (n = 1890). These children were followed longitudinally for 4 years.

Statistical analysis
Risk model development. An any-risk model involved using a dmft score of 1 or more. To aid development of any-risk and high-risk models, the distribution of disease within the population was examined and the nearest point to which the relatively smallest proportion of the population had the relatively largest proportion of disease was identied. For the purposes of this study, a d1mft value of 3 for each child was classied as high risk because those with a d1mft of 3 or more, which represented 32% of the population with 89% of the disease at 4 years, were the most appropriate value for detection of the minority of children with the majority of the disease. At the d3mft threshold of detection, a value of 3 was selected as the cutoff point for high caries risk because 19% of the population had 87% of the disease at 4 years when using the d3mft value.

Data collection
Dental examination. This used a combination of direct vision and illumination by a pen-light, with the child in the supine position at age 1 year (and upright position for the remaining years). Examination was carried out within 1 month of the childs birthday wherever possible. Caries detection was at the d1 caries into both enamel and dentine threshold. All lesions were recorded according to the criteria developed for the Dundee selective threshold methods for caries detection detection (25). All children were examined by a calibrated, single examiner (HBM).

38

Development of the Dundee Caries Risk Assessment Model (DCRAM)

These data were analysed using CHAID analysis to provide any-risk and high-risk prediction models for caries in 4-year-olds using data collected at age 1 year. Chi-squared automatic interaction detector analysis. This novel method of analysis comprised a treebased classication system derived by Kass (28), shown to improve upon traditional approaches (29) and enhanced by statistical innovations (SPSS PC + CHAID). This was a software package designed to exibly generate prediction trees from nonparametric data and has been fully described elsewhere (6). All 56 factors from the fully completed child data forms were entered into the analysis to obtain statistically signicant predictors (284 child data forms). The factors that were not statistically signicant were then excluded and the CHAID analysis applied to the data from 506 child data forms for which complete data were available for all these statistically signicant factors. This resulted in a new set of predictors for this larger pool of data. The process was then repeated exclusion of nonpredictive factors and application of CHAID analysis to the 784 child data forms (high risk) and 697 data forms (any risk) for which complete data were available for that new set of statistically signicant predictors. This analysis created branching groups until the predetermined critical value of probability and or cell size was reached. The predetermined critical value was the prevalence of disease in the data to which the CHAID analysis was applied, i.e. in the primary cell. The predetermined minimum cell size was 10. Ends of the branches were labelled as high or low risk depending on the prevalence of disease within that data relative to that in the primary cell a higher relative prevalence being designated with a high-risk label. Sensitivities and specicities were calculated by generation of two-by-two tables of predicted versus actual disease. This analysis was carried out for d1mft >0, d3mft >0, d1mft 3 and d3mft 3.

Table 1. Number of eligible children consented to participate in year 1 and year 4 of the study Number of eligible children Year 1 Year 4 1890 1888 Number of children consented 1683 (89%) 1681 (89%)

Results
Participation rate
The overall participation rate was 89% of those children born and resident in Dundee in a 1-year cohort (n = 1890) (Table 1). A total of 1681 children remained in the study after 4 years. (There was no statistical difference in the sociodemographic prole between the consented and nonconsented groups, conrming the representative nature of the sample.) Tables 1 and 2 show the number of children consented to participate, the number of dental examinations carried out and the caries prevalence at ages 1 and 4 years at d1 and d3 levels. The caries prevalence at age 4 years for d1mft 3 was 32% and d3mft 3 was 19%. Complete data sets were not obtained from all participating subjects.

Risk model development

Chi-squared automatic interaction detector analysis. The complete CHAID analyses have been provided elsewhere (6). The denitive risk models for each detection threshold are shown in Fig. 1a-d accompanied by the sensitivity and specicity values for each model, as well as the number of subjects with the data available for the particular factors in that specic risk model. The predictive factors for each of the four models are listed below: d1mft at age 4 years >0 any cariesrisk model (Fig. 1a) The predictive factors were HV opinion of risk, DEPCAT, parental smoking, breastfeeding and use of a dummy. d3mft at age 4 years >0 any cariesrisk model (Fig. 1b)

Table 2. The number of dental examinations (of eligible cohort) and the caries prevalence at the d1 and d3 thresholds of caries diagnosis Number and percentage of dental examinations Year 1 Year 4 1419 (84%) 1365 (81%) Number and percentage of children with decay at d1 39 (3%) 674 (49%) Number and percentage of children with d1mft 3 442 (32%) Number and percentage of children with decay at d3 6 (0.4%) 449 (33%) Number and percentage of children with d3mft 3 262 (19%)

39

MacRitchie et al.
(a)
All children at age 1-year

(b)

All children at age 1-year

HV opinion of risk No Yes

HV opinion of risk Yes No

DEPCAT 1:2 3:4:5:6

Parental Smoking Both One None

*
Both/one

Parental Smoking None

Ok

Breast fed No Yes

Use of dummy Yes No

Food or drinks at night

Ok

Yes

No

*
*
(c)

Ok

Ok

*
*
(d)

Ok

At risk of decay (d1mft > 0)

At risk of decay (d3mft > 0)

All children at age 1-year

All children at age 1-year

Type of housing

Type of housing
Council Private

Council

Private

HV opinion of risk Yes No

Use of a feeder cup

Yes

No

Use of vitamins Yes No

Ok

Ok

*
*
Ok
1

* At high risk of decay (d mft 3)

At high risk of decay (d3mft 3)

Fig. 1. (a) Diagrammatic chi-squared automated interaction detector analysis (CHAID) risk assessment model for prediction at age 1-year of d1mft >0 at age 4-years (Se = 67%, Sp = 57%) (n = 697). (b) Diagrammatic CHAID risk assessment model for prediction at age 1-year of d3mft >0 at age 4-years (Se = 53%, Sp = 77%) (n = 697). (c) Diagrammatic CHAID risk assessment model for prediction at age 1-year of d1mft 3 at age 4-years (Se = 69%, Sp = 60%) (n = 784). (d) Diagrammatic CHAID risk assessment model for prediction at age 1-year of d3mft 3 at age 4-years (Se = 65%, Sp = 69%) (n = 784).

The predictive factors were HV opinion of risk, parental smoking and food or drink at night. d1mft at age 4 years 3 high cariesrisk model (Fig. 1c) The predictive factors were type of housing and use of a feeder cup. d3mft at age 4 years 3 high cariesrisk model (Fig. 1d) The predictive factors were type of housing, HV opinion of risk and use of vitamins.

Discussion
There has been some improvement in the dental health of Scottish 5-year-olds in recent years, but a consistent proportion in the more deprived strata of society continues to develop high levels of disease (1). These high levels of disease have a great cost impact on the government, and although theoretical cost-benet analyses of highly specic scenarios indicate that further research is required

40

Development of the Dundee Caries Risk Assessment Model (DCRAM)

in this eld (30), it has been suggested that potential savings could be made with early identication of caries risk preschool children (31, 32). Petersen (33) highlighted the importance of risk assessment in estimating the potential for prevention, and more recently, in a consensus paper, Young (7) emphasized the need for risk assessment and also the importance of doing this in the early years of life. For populations with polarization of decay, such as Scotland, a combination of some population-based preventive measures and geographical targeting (i.e. areas of social deprivation) would seem the best way to proceed (2, 6, 8, 34). Batchelor and Sheiham (4) concluded that the main emphasis should be centred on a population approach but the data used to justify this conclusion were cross-sectional and from much older children than in the present study. More recently, those authors stated that the ndings from their study challenged the basis for adoption of a highrisk strategy (5). In their paper, data were collected from 7-year-old children, who were followed for 4 years. However, data from Scottish epidemiological studies show that by 5-year old, half of this population has experienced obvious dentinal decay (1). A major window of opportunity has, therefore, been lost in terms of prevention, by age 7 years. A recent study by Thitasomakul et al. (35) concluded that an extremely high caries-affected rate was found even before the age of 18 months, and Messer (36) and Wendt et al. (37) have also emphasized the need for identication of risk children, ideally, before the start of the caries attack. Data collected from the longitudinal study described in the current paper showed that 3%, 12% and 27% of 1, 2 and 3-year-olds, respectively, had d1 caries (6) and that 49% (at d1) of 4-year-olds had dental caries (Table 2). Therefore, any risk assessment system must be employed at a very early stage, ideally at, or even before, birth. Such a system would necessitate involvement of those healthcare personnel closely involved with children (across geographical areas) at very young ages or with expectant mothers. This study, therefore, aimed to develop simple community-based, but scientically robust, caries risk models for the identication of caries risk preschool children using a novel application of CHAID analysis. As noted by Reich et al. (38), in diagnosing caries risk, no single test can simultaneously measure host resistance, microbial pathogens and cariogenicity of the diet. A systematic review by Harris et al. (39) concluded that many studies looking for

predictors of caries in young children have not used the optimum study design, which is a longitudinal study. They also noted that the evidence pointed most consistently to a young child being most likely to develop caries if they acquire Streptococcus mutans at a young age. Many recent studies have again found factors associated with caries, but cross-sectional data have been employed (4050). Previous caries risk assessment studies using multiple factors in the production of caries risk models for children have shown varied results. Comprehensive reviews of these studies have been documented elsewhere (6, 21, 39, 51). Although a vast number of related reports have appeared in the literature, there have been few large-scale (n > 200), multidisciplinary, population-based, longitudinal studies of caries risk assessment in preschool children, especially in a demographic area similar to Scotland. To summarize, some authors have linked distinct factors with the assessment of caries risk, for example, immigrant status (16), race (15), mothers nationality (52), mutans streptococci (15, 5355), sociodemographic variables (56, 57), clinical variables (53, 58 61) and immune system factors (62). Several statistical methods have been employed to create risk or prediction models, for example, regression analysis (11, 14, 52, 54, 58, 6365), multivariate survival analysis (66), Markov models (22), Dentoprog method (61) and classication and regression trees (67). The predictive capability of risk models in the literature has varied. Schroeder (68) showed that with none of the predictors was it possible to nd a screening level that combined high sensitivity with high specicity. Stamm (14) noted that their models were below the current suggested minimum for caries prediction (69) but the sum of the sensitivities and specicities exceeded 1.40. In a review, Zero (10) assessed the validity of multifactorial test models for caries prediction and noted that using data from the best level of evidence (those studies rated as good), previous caries experience was the best predictor for caries in primary teeth. This was followed by parents education and socioeconomic status. However, of the 15 articles analysed, only three studies were rated as good and seven were poor the authors emphasized the need for further research to identify and validate caries risk assessment strategies that can be applied in dental practice. The predictive capabilities of the models developed from the data collected during the 4-year longitudinal caries risk assessment study described

41

MacRitchie et al.

in this paper are shown in Fig. 1a-d. This shows the four possible risk models developed, which have been termed the Dundee Caries Risk Assessment Models (DCRAMs). There were 284 forms (from 1419 subjects), which had completed questions for all 56 variables whilst this appears to represent only a 20% completion rate, the vast majority of questions were in fact completed for the large majority of subjects. The missing data ranged from 0% to 26% for the nine questions factors that were included in the nal models. Why different specic questions were not answered by different individuals was not investigated, but could have been the result of one or more of several factors, including simple omission error, sensitivity to the question subject, questionnaire fatigue or even literacy limitations, amongst others. A statistical analysis was carried out to determine whether there were any biases, in the four models, because of missing data in the subjects data forms this analysis indicated that there were no statistically signicant differences (at the 1% level) between the missing data and nonmissing data groups except for the DEPCAT and breastfeeding factors in the d1-any cariesrisk model (Fig. 1a). Further investigation of this model is being carried out using a multiple imputing technique for missing data, and until the results of this analysis are determined, the utility validity of this d1-any-risk model requires conrmation. All models had sensitivity (Se) and specicity (Sp) values less than the suggested target minimum of 80% (69, 70). As noted by Zero (10), at an individual level, it might be desirable, both from an ethical and economical perspective, to increase the sensitivity of the predictive test to reduce the number of false negatives. By increasing the sensitivity of the risk assessment test, the number of false positives would increase, but if the dentist uses a preventive management strategy, this would not result in any harm to the patient. It must be noted, however, that the caries prevalence for 4-year-old children is not the peak caries prevalence for the deciduous dentition in Scotland and many of those children apparently caries-free at 4 years will go on to develop caries at 5 years (and beyond) and some of those with a dmft <3 will develop more caries after 4 years of age; hence, the sensitivity of the models will increase. Indeed, the survey of 5-year-olds carried out the year following this study, which would have included this cohort, showed that 57% of 5-year-olds had obvious caries (d3mft >0). Thus, the specicity of the prediction model is also likely to increase

considerably as the true-negative gure decreases, and it would be of great benet to develop future caries risk models based on the population at the peak level of disease in the deciduous (and permanent) dentitions. The numbers predicted to be high risk by the CHAID models, i.e. the prevalence values, for both d1 and d3 are almost identical (within a few percentage points) to the true prevalence gures, suggesting that the numbers are acceptable and manageable. The fact that the caries prevalence is still rising and has not peaked by 4 years of age suggests that the prediction model will rise in accuracy as the children age and does not under- or overestimate the absolute numbers in terms of need for prevention. In terms of practical use (in the community) of the risk models developed in this study, the predictive factors involved, combined with the predictive power of the model, need to be considered. Three of the models incorporated the HVs opinion (or hunch) in relation to the caries risk. This was a simple yes no subjective assessment by the HVs but has been shown by the CHAID analysis to be one of the best predictors. This may be related to the HVs ability to assimilate healthrelated information about individual childrens families. The fact that HVs visit all well babies in their homes provides a unique opportunity to use the model in a community setting. As they have access to all infants across socioeconomic backgrounds, the high-risk children can be identied individually, even within geographical areas unlikely to have high dmf levels, and directed towards appropriate preventive dental services. A follow-up study in a different geographical and sociodemographic area of the UK, on a larger sample size, produced almost identical predictive models (and their values). These results validated the DCRAMs and demonstrated its generalizability in the UK (71). Another simple predictive factor was the type of housing in which the child lived (a marker of deprivation in Scotland), and this information can be easily obtained. The DCRAM is, therefore, relatively easy to apply, because it only requires between two and ve (depending on the particular DCRAM to be used) simple questions (predictors), structured in a decision tree format to identify caries risk individuals; hence, this model is suitable for a community-based setting involving HVs. This ease of use, combined with a sensitivity of 65% and specicity of 69% of the risk assessment model for prediction at age 1 year of d3mft 3 at age 4 years (Fig. 1d), suggests that this targeting

42

Development of the Dundee Caries Risk Assessment Model (DCRAM)

model could be used, in conjunction with population strategies, for future implementation of preventive care. Outside of the UK, to assess the utility of the DCRAM, it would be necessary to: (i) identify healthcare workers with similar responsibilities to the UK HVs (which may limit its use outwith the UK) and (ii) test the applicability of the various predictive factors within the model. The Scottish government, in response to the ndings from this work, including the further evaluation in another population (71) and along with evidence-based guidelines for caries management (72), is in the process of introducing preventive interventions based on this targeted approach as part of the Childsmile Programme in Scotland (http://www.child-smile.org). The NIH Consensus Development Conference Statement (73) noted, in relation to assessment of caries risk, the need for comprehensive, longitudinal, multifactorial studies of implicated risk factors to obtain rm support for their association with caries incidence, as well as the requirement for multiple indicators, combined on an appropriate scale and accounting for possible interactions the novel approach reported here is a start to addressing those stated gaps in the evidence base in relation to the specic population studied. In conclusion, a risk model, the DCRAM, has been developed for potential use in a particular community setting, by healthcare personnel, which could identify high cariesrisk children at age 1 year. These children could then be appropriately directed to health and dental services for preventive care.

Acknowledgements
This study was funded by the Chief Scientist Ofce of the Scottish Government Health Department (formally Scottish Executive). Thanks to all the Health Visitors, parents and children in the study and to Val Wilson for her technical support. The views expressed in this paper are those of the authors and are not necessarily those of any other body or organization.

References
1. Merrett MCW, Goold S, Jones CM, Levin KA, McCall DR, Macpherson LMD et al. National Dental Inspection Programme of Scotland: Report of the 2006 Survey of P1 Children. Scottish Dental Epidemiological Co-ordinating Committee 2006 ISBN 978-09550957-2-6.

2. Spencer AJ. Skewed distributions new outcome measures. Community Dent Oral Epidemiol 1997;25:529. 3. Jones CM. Capitation, registration and social deprivation in England; and inverse dental care law. Br Dent J 2001;190:2036. 4. Batchelor PA, Sheiham A. The limitations of a highrisk approach for the prevention of dental caries. Community Dent Oral Epidemiol 2002;30:30212. 5. Batchelor PA, Sheiham A. The distribution of burden of dental caries in schoolchildren: a critique of the high-risk caries prevention strategy for populations. BMC Oral Health 2006;6:3. 6. Ballantyne-MacRitchie HM. A partnership between dentists and health visitors to identify high caries risk Scottish pre-school children. PhD Thesis, University of Dundee, 2000. 7. Young DA, Featherstone JDB, Roth JR, Anderson M, Autio-Gold J, Christensen GJ et al. Caries management by risk assessment: implementation guidelines. CDA J 2007;35:799805. 8. Hausen H. Caries prediction. In: Fejerskov O, Kidd E, editors. Dental caries. The disease and its clinical management, 2nd edn. Copenhagen: Blackwell Munksgaard Ltd, 2008; 52751. 9. Burt BA. Concepts of risk in dental public health. Community Dent Oral Epidemiol 2005;33:2407. 10. Zero D, Fontana M, Lennon AM. Clinical applications and outcomes of using indicators of risk in caries management. J Dent Educ 2001;65:112632. 11. Disney JA, Graves RC, Stamm JW, Bohannan HM, Abernathy JR, Zack DD. The University of North Carolina Caries Risk Assessment Study: further developments in caries risk prediction. Community Dent Oral Epidemiol 1992;20:6475. 12. Stamm JW, Stewart PW, Bohannan HM, Disney JA, Graves RC, Abernathy JR. Conference prevention revisited London 1990 risk assessment for oral diseases. Adv Dent Res 1991;5:417. 13. Koch GG, Beck JD. Statistical methodologies useful for the analysis of data from risk assessment studies. J Public Health Dent 1992;3:14667. 14. Stamm JW, Disney JA, Beck JD, Weintraub JA, Stewart PW. The University of North Carolina Caries Risk Assessment Study: nal results and some alternative modeling approaches. In: Bowen WH, Tabak LA, editors. Cariology for the nineties. Rochester, New York: University of Rochester, 1993; 20934. 15. Reisine S, Litt M, Tinanoff N. A biopsychosocial model to predict caries in pre-school children. Paediatr Dent 1994;16:4138. 16. Grindefjord M, Dahllof G, Nilsson B, Modeer T. Prediction of dental caries development in 1-year old children. Caries Res 1995;29:3438. 17. Grindefjord M, Dahllof G, Nilsson B, Modeer T. Stepwise prediction of dental caries in children up to 3.5 years of age. Caries Res 1996;30:25666. 18. Helfenstein U, Steiner M, Menghini G. The use of generalized additive models (GAM) in dentistry. Community Dent Health 1997;14:2216. 19. Hausen H. Caries prediction state of the art. Community Dent Oral Epidemiol 1997;25:8796. 20. Al Ghanim NA, Adenubi JO, Wyne AA, Khan NB. Caries prediction model in pre-school children in

43

MacRitchie et al. Riyadh, Saudia Arabia. Int J Paediatr Dent 1998;8:11522. Powell LV. Caries prediction: a review of the literature. Community Dent Oral Epidemiol 1998;26:361 71. Kopycka-Kedzierawski DT, Billings RJ. Application of nonhomogenous Markov models for analyzing longitudinal caries risk. Community Dent Oral Epidemiol 2006;34:1239. Radford JR, Ballantyne HM, Nugent ZJ, Beighton D, Robertson M, Longbottom C et al. Caries associated microorganisms in infants from different socio-economic backgrounds in Scotland. J Dent 2000;28:307 12. Radford JR, Ballantyne HM, Nugent ZJ, Beighton D, Robertson M, Longbottom C et al. Does social deprivation in 1, 2, 3 and 4-year-old Scottish infants inuence the frequency of isolation of caries associated microorganisms. J Dent 2001;29:32532. Fyffe HE, Deery C, Nugent ZJ, Nuttall NM, Pitts NB. Effect of diagnostic threshold on the validity and reliability of epidemiological caries diagnosis using the Dundee Selectable Threshold Method for caries diagnosis. Community Dent Oral Epidemiol 1997;28:4251. Beighton D. A simplied procedure for estimating the level of Streptococcus mutans in the mouth. Br Dent J 1986;160:32930. Carstairs V, Morris R. Deprivation and health in Scotland. Aberdeen: Aberdeen University Press; 1991. Kass GV. An exploratory technique for investigating large quantities of categorical data. Appl Stat 1980;29:11927. Magidson J. Improved statistical techniques for response modeling. Journal of Direct Marketing 1988;2:618. Grifn SO, Grifn PM, Gooch BF, Barker LK. Comparing the costs of three sealant delivery strategies. J Dent Res 2002;81:6415. Zavras AI, Edelstein BL, Vamvakidis A. Health care savings from microbiological caries risk screening of toddlers: a cost estimation model. J Public Health Dent 2000;60:1828. Axelsson P. The effect of a needs related caries preventive program in children and young adults results after 20 years. BMC Oral Heath 2006;6(Suppl 1):S7. Petersen PE. Sociobehavioural risk factors in dental caries international perspectives. Community Dent Oral Epidemiol 2005;33:2749. Pitts NB. Inequalities in childrens caries experience: the nature and size of the UK problem. Community Dent Health 1998;15(Suppl):296300. Thitasomakul S, Thearmontree A, Piwat S, Chankanka O, Pithpornchaiyakul W, Teanpaisan R et al. A longitudinal study of early childhood caries in 9- to 18-month old Thai infants. Community Dent Oral Epidemiol 2006;34:42936. Messer LB. Assessing caries risk in children. Aust Dent J 2000;45:106. Wendt LK, Carlsson E, Hallonsten AL, Birkhed D. Early dental caries risk assessment and prevention in pre-school children: evaluation of a new strategy for dental care in a eld study. Acta Odontol Scand 2001;59:2616. 38. Reich E, Lussi A, Newbrun E. Caries risk assessment. Int Dent J 1999;49:1526. 39. Harris R, Nicoll AD, Adair PM, Pine CM. Risk factors for dental caries in children: a systematic review of the literature. Community Dent Health 2004;21(1 Suppl):7185. 40. Hallett KB, ORourke PK. Early childhood caries and infant feeding practice. Community Dent Health 2002;19:23742. 41. Southward LH, Robertson A, Wells-Parker E, Eklund NP, Silberman SL, Crall JJ et al. Oral health status of Mississippi Delta 3- to 5-year olds in child care: an exploratory study of dental health status and risk factors for dental disease and treatment needs. J Public Health Dent 2006;66:1317. 42. Rupesh S, Nayak UA. Genetic sensitivity to the bitter taste of 6-n propythiouracil: a new risk determinant for dental caries in children. J Indian Soc Pedod Prev Dent 2006;24:638. 43. Slade GD, Sanders AE, Bill CJ. Risk factors for dental caries in the ve-year-old South Australian population. Aust Dent J 2006;51:1309. 44. Tsai AI, Chen C-Y, Li L-A, Hsiang C-L, Hsu K-H. Risk indicators for early childhood caries in Taiwan. Community Dent Oral Epidemiol 2006;34:43745. 45. Du M, Luo Y, Zeng X, Alkhatib N, Bedi R. Caries in preschool children and its risk factors in 2 provinces in China. Quintessence Int 2007;38:14351. 46. Folayan M, Sowole A, Kola-Jebutu A. Risk factors for caries in children from South-Western Nigeria. J Clin Pediatr Dent 2007;32:1715. 47. Iida H, Auinger P, Billings RJ, Weitzman MD. Association between infant breastfeeding and early childhood caries in the United States. Pediatrics 2007;120:94452. 48. Kopycka-Kedzierawski DT, Auinger P, Billings RJ, Weitzman M. Caries status and overweight in 2- to 18-year-old US children: ndings from national surveys. Community Dent Oral Epidemiol 2007;36:15767. 49. Ohlund I, Holgerson PL, Backman B, Lind T, Hernell T, Johansson I. Diet intake and caries prevalence in four-year-old children living in a low-prevalence country. Caries Res 2007;41:2633. 50. Qin M, Li JL, Zhang S, Ma W. Risk factors for severe early childhood caries in children younger than 4-years old in Beijing, China. Pediatr Dent 2008;30: 1227. 51. Pitts NB. Risk assessment and caries prediction. J Dent Educ 1998;62:76270. 52. Martens L, Vanobbergen J, Williams S, Aps J, De Maeseneer J. Determinants of early childhood caries in a group of inner-city children. Quintessence Int 2006;37:52736. 53. Clarke P, Fraer-Lee NJ, Shimono T. Identifying risk factors for predicting caries in school-aged children using dental health information collected at preschool age. J Dent Child 2001;68:3023. 54. Pienhakkinen K, Jokela J, Alanen P. Assessment of caries risk in preschool children. Caries Res 2004;38:15262. 55. Nishimura M, Oda T, Naoyuke K, Matsumara S, Shimono T. Using a caries activity test to predict caries risk in early childhood. J Am Dent Assoc 2008;139:6371.

21. 22.

23.

24.

25.

26. 27. 28. 29. 30. 31.

32.

33. 34. 35.

36. 37.

44

Development of the Dundee Caries Risk Assessment Model (DCRAM) 56. Marshall TA, Eichenberger-Gilmore JM, Broftt BA, Warren JJ, Levy SM. Dental caries and childhood obesity: roles of diet and socioeconomic status. Community Dent Oral Epidemiol 2007;35:44958. 57. Jeppesen BA, Foldspang A. Can the development of new dental caries in Danish schoolchildren be predicted from surveillance data in the school dental service. Community Dent Oral Epidemiol 2006;34:20512. 58. Pereira da Silva Tagliaferro E, Pereira AC, Meneghim MC, Ambrosano GMB. Assessment of dental caries predictors in a seven-year longitudinal study. J Public Health Dent 2006;66:16979. 59. Kanellis MJ. Caries risk assessment and prevention: strategies for head start, early head start and WIC. J Public Health Dent 2000;3:2107. 60. Li Y, Wang W. Predicting caries in the permanent teeth from caries in primary teeth: an eight year cohort study. J Dent Res 2002;81:5616. 61. van Palenstein Helderman WH, Mulder J, van Hof MA, Truin GJ. Validation of a Swiss method of caries prediction in Dutch children. Community Dent Oral Epidemiol 2001;29:3415. 62. Bagherian A, Nematollahi H, Afshari JT, Moheghi N. Comparison of allele frequency for HLA-DR and HLA-DQ between patients with ECC and caries-free children. J Indian Soc Pedod Prevent Dent 2008;26:1821. 63. Sakuma S, Kakamura M, Miyazaki H. Predictors of dental caries development in 1.5-year-old high-risk children in the Japanese public health service. J Public Health Dent 2007;7:149. 64. Yonezu T, Yotsuya K, Yakushiji M. Longitudinal study of prolonged breast- or bottle-feeding on dental caries in Japanese children. Bull Tokyo Dent Coll 2006;47:15760. Tiberia MJ, Milnes AR, Feigal RJ, Morley KR, Richardson DS, Croft WG et al. Risk factors for early childhood caries in Canadian preschool children seeking care. Pediatr Dent 2007;29:2018. Ollila P, Larmas M. A seven-year survival analysis of caries onset in primary second molars and permanent rst molars in different caries risk groups determined at age two years. Acta Odontol Scand 2007;65:2935. Stewart PW, Stamm JW. Classication tree prediction models for dental caries from clinical, microbiological and interview data. J Dent Res 1991;70:123951. Schroeder U, Widenheim J, Peyron M, Hagg E. Prediction of caries in 1.5-year old children. Swedish Dent J 1994;18:95104. Kingman A. Statistical issues in risk models for caries. In Bader JD, editor. Risk assessment in dentistry. Chapel Hill, NC: University of North Carolina Dental Ecology, 1990; 193200. Hausen H, Karkkainen S, Seppa L. Application of the high-risk strategy to control dental caries. Community Dent Oral Epidemiol 2000;28:2634. Anderson CE. Primary care based targeted caries prevention for pre-school children. PhD Thesis, University of Dundee, 2007. Prevention and management of dental decay in the pre-school child. SIGN 83: Nov 2005; Scottish Intercollegiate Guidelines Network; Edinburgh. Diagnosis and management of dental caries throughout life, National Institute of Health Consensus Development Conference Statement, March 2628, 2001. [No authors] J Dent Educ 2001;65:11628.

65.

66.

67. 68. 69.

70. 71. 72. 73.

45

This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material.