Clinical Research: Basic Statistics and Appraising the Literature
Epidemiology and Biostatistics Epidemiology: Study design and interpretation Biostatistics: Methods for analysis
Importance of Understanding Basic Statistics in Medicine Research –Design Studies –Plan Analyses –Data Interpretation Clinical Medicine –Understanding the Literature –Evidence-based practice
Learning the Language Sampling Variable types –Determine analysis method(s) Continuous Categorical (nominal, ordinal) Independent vs. Correlated Data Parametric vs. Non-parametric
Sampling: Is the study group representative? CAD case:Control Study n=328/group Non-diabetic Middle-aged Italian Men Colomba F et al. ATVB 2005; 25: 1032
Sampling: Is the study group representative? Dallas Heart Study Probability-based sample Over-sampling Minorities
Statistical Testing: Principles Question: Is blood pressure associated with stroke? Study 1Study 2 Stroke No Stroke Average= 136 mm/Hg 132 mm/Hg Average= 136 mm/Hg 132 mm/Hg
Statistical Testing: Principles Question: Is blood pressure associated with stroke? Study 1Study 2 Stroke No Stroke 132 mm/Hg Average= 136 mm/Hg Average= 136 mm/Hg
Statistical Testing Observed effect (what we see) – Expected (under null) Variability of the data Test Statistic = Use test statistic to generate a p-value
Learning the Language Sampling Variable types –Determine analysis method(s) Continuous Categorical (nominal, ordinal) Independent vs. Correlated Data Parametric vs. Non-parametric
Categorical Data Data where the results are in categories of some qualitative trait (yes/no) –Can be nominal or ordinal
Nominal v. Ordinal Nominal data (no order to the categories) –Smoking status (smoker, non- smoker) –Hair color (blonde, red, black) –Race (black, white, hispanic, other) Ordinal data (order to categories) –Med school year (1 st, 2 nd, 3 rd, 4 th ) –Heart failure class (NYHA 1, 2, 3, or 4)
Continuous Data Data that are quantitative and measured (can perform arithmetic on) (can be divided into smaller values) –Blood pressure –Age –Cholesterol levels
Variable Types: Ordinal, Numerical and Categorical Svensson AM, et al. Eur Heart J 2005; 26: 1255
Learning the Language Sampling Variable types –Determine anlaysis method(s) Continuous Categorical (nominal, ordinal) Independent vs. Correlated Data Parametric vs. Non-parametric
Data from Independent Samples Park L et al. Nat Med 4: g IP day g IP day g IP day g IP day -1 Diabetic ApoE null mice Control ApoE null mice Control ApoE null mice
Baseline 24 Hours Control GIK Data from Repeated Measures: Correlated Data Addo T, et al. Am J Cardiol 2004; 94: 1288
Learning the Language Sampling Variable types –Determine anlaysis method(s) Continuous Categorical (nominal, ordinal) Independent vs. Correlated Data Parametric vs. Non-parametric
Parametric (Gaussian) Distribution
Skewed Data
Statistical Tests: What Type of Data? NominalOrdinalParametricNon-Para ContinousCorrelatedPaired t-test Wilcoxon Sign Rank Independt-testWilcoxon Rank Sum CategoricalCorrelatedMcNemar Test IndependFisher’s Exact Chi-square trend test
Power and Sample Size
Power: What is it Power = (1- ): –The probability of rejecting the null hypothesis when it is false –English: the probability of detecting a true association between an exposure and an outcome when there is one
Sample Size and Power: The assumptions Sample size: –To determine sample size, enter three parameters: Power : (80 or 90%) Effect size –Control value and variance, or event rate –dependent on parameter of interest –best to have pilot data Significance level ( ) : (0.05) –1-tailed or 2-tailed testing (Confounders) –Non-compliance, Cross-overs (Drop Ins/Outs), Lost to follow up
Standards for Effect Size Small –20% Medium – 50% Large – 80% –only rough guidelines Small, medium and large are subject dependent
Adequacy of Sample: Size Matters Total # of eventsSample Size if risk 10% Power for 25% RRR Adequacy of size 0-50(under 500)<10%Utterly inadequate (1000)10-30%Probably inadequate (3000)30-70%Possibly adequate, possibly not (6000)70-90%Probably adequate Over 650(10,000)>90%Definitely Adequate
Effect of trial size on results: 24 trials of -blockade vs. Placebo Total deaths Mean Sample Size p<0.5 against Trend against Trend favorable p<0.5 favorable 0-50(255) (861) (2925) N/A---- Over 650N/A---- TOTAL(866)06153
Ways to Reduce Required Sample Size Higher Event Rate –High risk populations –Composite Endpoints Larger Effect Size Lower power Larger –1-tailed or 2 Change analysis type –Time dependent
Sample size planning How much money do you have? How much time to you have? How many patients/subjects can you expect to reasonably get? “What sample size and study design can I afford?”
The words to use to describe this The study was designed to have >80% power to detect an effect size of >20% with a 2-tailed significance level of 0.05, with a planned sample size of 400 participants in each group.
Suggested Reading Reference texts –Dawson-Saunders B, Trapp RG. Basic and Clinical Biostatistics, Appleton and Lange, Norwalk, CT, 2 nd Edition, –Sackett DL. Clinical Epidemiology: a basic science for clinical medicine. Little Brown, Boston, MA, 2 nd Edition, Selected papers: –Bias Sackett DL. Bias in analytic research. J Chron Dis 1979; 32:51-63 –Power Moher D, Dulberg CS, Wells GA. Statistical power, sample size, and their reporting in randomized controlled trials. JAMA 1994; 272: –Subgroup analyses Assmann SF, Pocock SJ, Enos LE, Kasten LE. Subgroup analysis and other (mis)use of baseline data in clinical trials. Lancet 2000; 355: Yusuf S, Wittes J, Probstfield J, Tyroler HA. Analysis and interpretation of treatment effects in subgroups of patients in randomized clinical trials. JAMA 1991; 266: