PICO Format Population + Intervention + Comparator + Outcome
2. Study Designs Randomised Control Trial Therapy / intervention Systematic Review Randomised Control Trial Cohort Study Case-control study Harm / Aetiology (Randomised control trial) Cohort study Case control study Diagnosis Cross-sectional analytic Prognosis Cohort Study Frequency of occurrence Descriptive (survey)
Cohort Study Case Control Study
Clinical Question Evidence Sources of bias Time RCT Therapy/ Intervention Harm/aetiology (Unethical) Strongest Evidence No Allocation concealment Loss to follow up No analysis by intention to treat Non-blind assessment of outcome Begins with study factor-exposure Randomised Cohort Therapy/ Intervention Harm/aetiology Prognosis Observational Loss to follow up Non blind assessment of outcome factor Confounding Begins with study- factor exposure Non-Randomised Case- Control Therapy/ Intervention Harm/Aetiology Observational Recall Bias Confounding Low response rates Non-blind assessment of outcome factor Controls not chosen from same population as cases Begins with outcome factor Cross- Sectional analytic Diagnosis Best study to assess diagnostic accuracy Non-independent blind comparison with reference standard Verification bias Inappropriate spectrum of disease Compares a diagnostic test to a gold standard Therapy/ Intervention Harm/Aetiology Association or hypothesis generating Selection of participants non-random (volunteers) Confounding Low response rate Small sample size Study and outcome factors measured at one point in time 3. Causality
Bias in Studies Cohort Studies Case Control Studies Cross-sectional studies Loss of follow up Non-blind outcome assessment Confounding Poorly-defined study base Low response rates Recall bias Confounding Incidence-prevalence bias Low response rat Recall bias Confounding
Confounders = prognostic factors that are unevenly distributed between groups of study subjects Randomisation removes confounders it equally distributes factors in control and exposure groups Need a large study group to ensure even distribution of confounders
Controls aim is to choose controls who are representative of the population from which the cases arise. Case and controls should come from the same population (study base) Eg hospital based case control studies are not good not representative of the general population + badly defined
Assembled study subjects Subjects exposure measured by investigators Exposed group Non-exposed group Measure outcomes of interest in both groups and compare Choose cases Choose controls (Outcome known at beginning) Measure past exposures in both cases and controls and compare Assembled study subjects Randomize subjects Intervention group Control group Measure outcomes of interest in both groups and compare Causal Criteria Strength of association Temporality Dose-response Biological plausibility Consistency across studies Reversibility Analogy
4. Frequency and Prognosis
Prevalence = proportion of a population with an attribute of interest at a specified time Prevalence = Total no. people with disease at a given time Total population at given time
Measure with descriptive study Population of interest random sample measure health characteristic of interest Survey Sampling o Random (probability) sampling electoral roll, random digit dialling o Non-probability sampling quota sampling, snow-balling, volunteers
Incidence = New occurrences in a population over time Incidence = No. of person experiencing a new even during a time period No. of persons at risk at beginning of time period
Measure with descriptive study population of interests representative sample o Loss to follow up (include in analysis) good response rate >80% o Measure health characteristic of interest
Prognostic Factors affect outcome once you have disease
Relative Risk <1 protective > 1 harmful Eg RR = 0.77 means the study factor is protective against disease 23% less likely to have disease RR/OR = 2.3 means you are over twice more likely to get disease
Confidence Intervals If the same study were repeated 100 times, then the results of 95 of these 100 studies would like somewhere between the 95% confidence interval You can be 95% confident that the result is within the 95% CI Narrow confidence interval more confident of the result If the CI crosses the null (=1) then the result is not statistically significant
P Values = probability of obtaining the study result if there really is no effect Probability the study result is due to chance
Steps for Appraising Articles on Causality 1. Are the results valid? Were there clearly identified comparison groups? Were outcomes and exposures measured in the same way in the two groups Was the follow up sufficiently long and complete Is the temporal relationship correct Is there a dose-response gradient? 2. What are the results? How strong is the association between exposure and outcome How precise is the estimate of risk 3. Will the results help me is caring for my patients Are the results applicable to my practice What is the magnitude of the risk Should I attempt to stop the exposure
Prognosis Articles (Cohort Studies) 1. Are the results valid? Representative sample of patients at a particular point in course of disease? Was follow up sufficiently long and complete? Were objective and unbiased outcome criteria used? Was their adjustment for important prognostic factors? 2. What are the results? How large is the likelihood of outcome events in a specified time? How precise are the estimates of likelihood? 3. Can I apply study to my patients? 5. Intervention Studies
Randomisation makes the two groups on average the same, so that any change or difference in outcome can be attributable to the intervention
Analysis by intention to treat preserves randomisation analyse patient outcomes based on into which group they were randomised regardless of whether they actually receive the planned intervention
Event rate = number of people who experienced the outcome/total number of people Experimental (EER) and Control event rates (CER)
Relative Risk (RR) = Event rate in treatment group / Event rate in control group or EER/CER
Relative Risk reduction (RRR) = 1 RR (expressed as a percentage)
Absolute risk reduction (ARR) = Event rate in control event rate in treatment group
Number needed to treat (NNT) = 1/ARR Number of people that need to be treated to prevent one adverse outcome
Analysing Intervention Studies (RCT) 1. Are the results valid? Was the assignment of patients to treatments randomised? Were the groups similar at the start of the trial? Were all patients who entered the trial properly accounted for and attributed at its conclusion? o Was follow-up sufficiently long and complete? o Were patients analysed in the groups to which they were randomised? (analysis by intention to treat) Were patients, study staff and outcome assessors blind to treatment? Aside from the experimental intervention, were the groups treated equally? 2. What were the results? How large was the treatment effect? How precise was the estimate of treatment effect? (CI) 3. Are the results applicable to my patients? 6. Diagnostic Questions
Study type = cross sectional analytical study
Pre-test probability = prevalence in the general population
Post-test probability = chance of patient having disease after the test (depends on pre-test probability)
Sensitivity = TP / TP + FN Proportion of people with the disease who test +ve for disease High sensitivity test is good at ruling OUT probability of disease (if ve, do not have disease)
Specificity = TN / FP + TN Proportion of people without disease who test ve for disease High Specificity test rules IN disease (if +ve, have disease)
Predictive values tell you the probability of a test being correct (prevalence dependent)
Positive Predictive Value = TP / TP + FP
Negative Predictive Value = TN / FN + TN
Likelihood ratio LR = probability of test result in diseased patients Probability of result in non-disease patients LR can be combined with pre-test probability post-test probability
+ve LR = Sensitivity / 1- specificity Want LR+ >1 the higher the better
-ve LR = 1 sensitivity / specificity Want LR- <1 the lower the better
Analysing Diagnostic Studies 1. Is the study valid? Was there an independent blind comparison to the gold standard? Were the test and reference standard measured independently? Was the choice of patients who were assessed by the reference standard independent of the test result? (avoids verification bias) Was there an appropriate spectrum of disease? Is the test reproducible? 2. What are the results? Likelihood ratios 3. Is the study applicable to my patient?