Research Methods

Course overview

Objectives Identify types of studies performed Describe methods for analyzing study results Explain possible forms of bias or other limitations of each study type Content Research Designs Level 5 expert opinion case study Level 4 case series no comparison group cross sectional study Observational study survey cannot confirm exposure equals outcome Level 3 case control study Observational study with controls non diseased Cases and controls are from the same population Matching controls have similar characteristics to cases Level 3b retrospective cohort Observational exposed compared to non exposed Level 2 prospective cohort Observational exposed compared to non exposed follow patients to observe exposure and outcome Participants may leave the study Level 1 randomized control trial Experimental study participants assigned to study arm Treatment or experimental arm Control or standard arm Randomization equalizes measured and unmeasured baseline characteristics Blinding Unblinded single blinded double blinded triple blinded No investigators participants or researchers know treatment groups Analysis Intent to treat analyzed based on assignment even if treatment not received As treated analyzed based on treatment received may interfere with randomization Per protocol excludes non compliant participants Limitations Reduced external validity difficult to generalize due to strict inclusion criteria Difficulty maintaining and gaining compliance Time consuming and expensive Ethical challenges Level 1 reviews Narrative review random selection of studies Systematic review uses predefined criteria and includes all applicable studies Meta analysis combines applicable studies for comprehensive analysis Limitations Publication bias studies with significant findings more likely to be published Funnel plot issues Heterogeneity Bias Selection bias participants do not represent the population of interest Enrollment bias Loss to follow up Information bias Recall bias inaccurate memory of past events Reporting bias intentional misreporting Misclassification incorrect participant categorization Confounding unaccounted variable affects outcome Statistics Parameter describes a population Statistic describes a sample Goal of research determine whether outcomes differ and if results are applicable Hypothesis Testing Null hypothesis H0 no difference in population Alternative hypothesis Ha a difference exists Statistical analysis used to reject the null hypothesis P values and Confidence Intervals Confidence interval level of certainty that the estimate reflects the true effect P value probability of obtaining results if no association exists typically 0.05 Types of Data Numerical Continuous values can take any number Discrete values limited to defined numbers Categorical Ordinal data with natural order such as underweight normal overweight Nominal data without natural order such as gender Types of Analysis Statistical tests normally distributed data T test Paired T test ANOVA Chi squared test Non parametric tests not normally distributed data Mann Whitney U test Wilcoxon Signed Rank test Kruskal Wallis Fisher Exact test small samples T test and Mann Whitney U compare two independent groups Paired T test and Wilcoxon Signed Rank compare two related groups ANOVA and Kruskal Wallis compare three or more groups Chi square and Fisher Exact assess association between two categorical variables Regression Models predict outcomes based on variables Logistic regression predicts binary outcomes Linear regression predicts continuous outcomes Cox proportional hazards predicts time to event risk Survival analysis evaluates time to event Measuring Disease Frequency Count number of cases at a specific time Incidence proportion of at risk population over time Measures of Association Association does not imply causation Relative Risk incidence in exposed compared to unexposed