DETAILED COURSE OUTLINE Epidemiologic Methods I & II Epidem 201AB Winter & Spring 2002 Hal Morgenstern, Ph.D. Department of Epidemiology UCLA School of Public Health
Page 1 I. THE NATURE OF EPIDEMIOLOGIC RESEARCH A. What is Epidemiology? 1. Definition: Disease and health; types of populations a. study population (sample) b. source population (sampling frame) c. base population (at risk) d. target population 2. Aims of epidemiologic research: Describe, explain, predict, and control 3. Natural history of disease: Understanding vs. intervention 4. Types of epidemiologic research a. laboratory research b. epidemic (outbreak) investigation c. population (or survey) research B. Basic Methodologic Features 1. Empirical/quantitative, probabilistic approach 2. Use of comparisons; statistical associations 3. Quantitative methods: Measurement, estimation, and testing C. Types of Research Strategies 1. Experiments: Random assignment of subjects to treatment groups 2. Quasi experiments: Assignment of subjects to treatment groups without randomization 3. Observational studies: No assignment of subjects to treatment groups II. CAUSATION AND ETIOLOGIC RESEARCH A. Models of Causation 1. Deterministic/counterfactual model: Causal/preventive types; necessary and sufficient causes 2. Koch's postulates
Page 2 3. Rothman's model of component and sufficient causes 4. Probabilistic model: Risk and prognostic factors; dose-response relations B. Logic and Inference 1. Deduction and induction 2. Refutation of hypotheses (Popper) and strong inference 3. The scientific method in epidemiology C. Hypotheses 1. Formulating hypotheses 2. Operationalizing and testing hypotheses 3. Proving vs. disproving hypotheses III. OBSERVATIONAL STUDY DESIGN A. Options in Design 1. Methods of subject selection a. restriction b. random sampling c. stratification 2. Methods of observation a. type of population: cross-sectional vs. longitudinal; fixed cohort vs. dynamic (open) population b. type of outcome measure: prevalence vs. incidence c. timing: prospective vs. retrospective data d. unit of analysis: individual vs. group e. method of data collection: primary vs. secondary B. Types of Study Designs 1. Basic designs a. cohort study: incident cases b. cross-sectional study: prevalent cases c. case-control study: incident cases d. case-control study: prevalent cases
Page 3 2. Hybrid designs a. population-based case-control study b. case-crossover study c. two-stage study d. follow-up prevalence study e. selection prevalence study f. cross-sectional incidence study g. repeated survey h. survey follow-up study i. repeated follow-up study j. intervention follow-up study 3. Incomplete designs a. proportional studies: mortality and morbidity b. ecologic studies: exploratory, multiple-group, trend, and mixed studies c. space-time cluster studies d. family cluster studies: familial aggregation, twin, and pedigree studies C. Criteria for Comparing Study Designs 1. Type of information sought 2. Quality of the information 3. Cost of acquiring the information IV. QUANTITATIVE MEASURES A. Basic Concepts 1. Proportions, rates, and ratios 2. Types of epidemiologic measures: frequency, association/effect, and impact measures B. Measures of Disease Frequency and Estimation 1. Incidence measures: New cases a. risk (cumulative measure) vs. rate (instantaneous measure) b. risk estimation: cumulative incidence or incidence proportion (average risk) c. rate estimation: incidence density (weighted average rate) d. other methods of risk estimation: density, life-table, and product-limit methods e. choosing among incidence measures and estimators
Page 4 2. Prevalence measures: Existing cases a. point and period prevalence b. relation with incidence under steady-state conditions 3. Mortality measures: Deaths a. disease-specific mortality, case fatality, and total mortality b. relation with incidence under steady-state conditions 4. Cohort analysis a. age, period, and cohort effects b. the identifiability problem C. Measures of Association/Effect and Estimation 1. Causal parameters for a population: Counterfactual model a. causal risk ratio b. causal risk difference 2. Measures of effect: Reflect causal parameters assuming comparability of exposed and unexposed populations a. risk ratio b. risk difference 3. Ratio measures of association a. risk, rate, and prevalence ratios b. odds ratios and their interpretations 4. Difference measures of association a. risk, rate, and prevalence differences b. relation with ratio measures 5. Correlation and model coefficients a. product-moment correlation coefficient b. linear-regression coefficient (slope) c. difference in means (continuous outcome) d. phi coefficient (product-moment correlation coefficient for two binary variables) D. Impact Measures 1. Attributable number and fractions (causal risk factors) 2. Prevented number and fractions (protective risk factors) 3. Preventable number and fraction (protective risk factors)
Page 5 V. CRUDE ANALYSIS A. Elements of Analysis 1. Estimation: Ratio measures a. point estimates and variance estimates b. interval estimates--confidence intervals 2. Tests of significance and P values: Mantel-Haenszel tests a. case-noncase vs. person-time data b. tests for no association (binary exposures) vs. tests for trend (ordinal exposures) B. Applications to Specific Study Designs 1. Cohort study a. cumulative analysis: fixed cohort (case-noncase data) b. density analysis (person-time data) 2. Cross-sectional study 3. Case-control study a. density sampling of controls b. cumulative sampling of controls 4. Population-based case-control study a. density sampling of controls b. cumulative sampling of controls c. case-base sampling of controls 5. Proportional mortality study 6. Multiple-group ecologic study VI. ERRORS OF MEASUREMENT A. Basic Concepts 1. Measurement and classification (categorical variables) 2. Concrete vs. abstract factors 3. Quality of measurement: Error/misclassification 4. Sources of measurement error: Observers, classification system, subjects,
Page 6 instrumentation, and data processing 5. Reliability vs. validity B. Reliability of Measurement 1. Types of reliability a. intraobserver or temporal b. interobserver agreement c. internal consistency 2. Quantifying reliability a. intraobserver: problems with test-retest correlation coefficient b. interobserver: marginal heterogeneity and kappa coefficient c. internal consistency: coefficient alpha C. Validity of Measurement 1. Types of validity a. content and face b. criterion c. construct 2. Quantifying criterion validity of disease classification a. sensitivity, specificity, and predictive value b. applications to screening tests and programs c. application of two tests: parallel vs. series testing d. application to clinical diagnosis e. application to estimation of disease frequency 3. Dealing with measurement error: In the planning stage, data collection, and analysis VII. ERRORS OF ESTIMATION A. Accuracy of Estimation 1. Lack of random error: Precision 2. Lack of Nonrandom error (bias): Internal validity B. Precision 1. Estimation precision vs. test power
Page 7 2. Sources of precision a. sample size b. statistical efficiency C. Internal Validity 1. Types of bias; expressing the direction of bias 2. Selection bias a. general definition and analytic structure b. types of selection bias: inappropriate reference group; self-selection bias and healthy-worker effects, selective loss to follow-up, selective survival, Berkson's bias, detection bias, diagnostic and reporting bias, inclusion/exclusion bias, temporal ambiguity and "protopathic" bias c. dealing with selection bias 3. Information bias a. misclassification bias: nondifferential vs. differential misclassification of exposure and/or disease status; independent vs. correlated errors between variables b. correcting for misclassification c. ecologic bias/fallacy d. other problems of categorization 4. Confounding a. confounding as the lack of comparability between exposure groups b. confounders and their properties c. identification of confounders: prior knowledge, observed associations, and study design d. causal confounders, proxy confounders, and selection confounders e. confounders vs. intermediate variables f. methods of controlling for confounding: design features and analytic methods VIII. STRATIFIED ANALYSIS A. Basic Concepts 1. Stratification on covariates; data layout 2. Adjusted estimates as weighted averages 3. Standardized measures vs. common measures B. Standardization
Page 8 1. Frequency measures 2. Association/effect measures a. risk and rate ratios b. odds ratios 3. Direct vs. indirect methods of adjustment (misleading designations) 4. Impact measures: attributable and prevented fractions C. Common Ratio Effect Measures 1. Woolf's method of estimation 2. Mantel-Haenszel method of estimation D. Interval Estimation of Effect (ratio measures) 1. Standardized measures 2. Common measures E. Statistical Testing: Mantel-Haenszel Tests 1. Case-noncase data: Binary exposure; ordinal exposure (test for trend) 2. Person-time data: Binary exposure; ordinal exposure (test for trend) F. Applications to Specific Study Designs 1. Cohort study: Cumulative analysis--risk ratios 2. Cohort study: Density analysis--rate ratios 3. Case-control study: Odds ratios G. Problems with Stratified Analysis 1. Heterogeneity of effect across strata 2. Choice and comparability of the standard population (with standardization) 3. Residual confounding within strata 4. Loss of precision with sparse data; possible gain in precision in cohort studies
Page 9 IX. MATCHING AND MATCHED ANALYSIS A. Basic Concepts and Principles 1. Matching in cohort vs. case-control studies 2. Types of matching: individual (category or caliper) vs. frequency; fixed- vs. variable-ratio 3. Major statistical advantage of matching: Gain in statistical efficiency 4. Matched analysis as a form of stratification B. Matched Analysis: Fixed Cohort Studies 1. Pairwise matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing 2. Fixed-ratio matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing 3. Variable-ratio matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing C. Matched Analysis: Density Cohort Studies D. Matched Analysis: Case-control Studies 1. Pairwise matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing 2. Fixed-ratio matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing 3. Variable-ratio matching: Mantel-Haenszel procedures a. effect estimation b. statistical testing 4. Controlling for nonmatched variables E. Criteria for Matching and Selection of Matching Variables
Page 10 1. Validity: Matching on intermediates 2. Statistical efficiency: Gain vs. overmatching; overmatching in case-control vs. cohort studies 3. Cost efficiency X. INTERACTION AND EFFECT MODIFICATION A. Basic Concepts 1. Effect modification--in the source/base population: Heterogeneity of effect; assessment of effect modification is model dependent (depends on the choice of effect measure) 2. Statistical interaction (or interaction effect)--in the study population: Heterogeneity of the estimated effect; to assess effect modification B. Assessment of Effect Modification 1. Testing for homogeneity of effect (no effect modification) a. Woolf's test for homogeneity across strata b. Testing for homogeneity, controlling for confounders 2. Estimating interaction effects of two exposures a. joint effect and components effects b. additive vs. multiplicative effects 3. Effect modification (and interaction) vs. confounding 4. Use of mathematical models; interpretation of product terms a. linear model: estimation of risk/prevalence differences b. logistic model: estimation of odds ratios c. proportional hazards model: estimation of rate ratios C. Biological (Causal) Interaction: Interdependence of Effects at the Individual Level 1. Biological interaction and Rothman's sufficient causes model 2. Synergistic, antagonistic, and competing effects 3. Problems of assessment in a population 4. Complete independence of effects and additivity of risks
Page 11 D. Public-health Assessment of Interaction Effects: Causal Risk Factors 1. Impact estimation in the total population 2. Impact estimation in the exposed population 3. Walker's index of net synergy XI. CAUSAL INFERENCE A. Basic Concepts 1. Accuracy vs. generalizability of findings 2. Assessment of generalizability a. extrapolating vs. integrating findings b. relation with effect modification c. confusion with selection bias B. Criteria for Making Causal Inferences: Hill's criteria and their limitations 1. Strength of the association 2. Dose-response gradient 3. Lack of temporal ambiguity 4. Consistency of the findings 5. Biological plausibility of the hypothesis 6. Coherence of the evidence 7. Specificity of the association (not useful) C. Multiple Comparisons 1. The problem of testing multiple associations in the same dataset using conventional statistical methods 2. Methods of correcting significance tests and confidence limits for multiple comparisons 3. Arguments against significance-testing corrections; and alternative methods for dealing with multiple comparisons
Page 12 D. Studying Multifactorial Etiologies 1. Simple causal diagrams: Planning for the measurement, analysis, and interpretation of single covariates as confounders, intermediates, and/or modifiers 2. Complex causal diagrams; conceptual frameworks 3. Modeling the exposure effect to deal with confounders, modifiers, and intermediate variables; covariate selection XII. PLANNING AN EPIDEMIOLOGIC STUDY A. The Research Protocol 1. Objectives of the protocol a. rationale for the study b. appropriateness of the methods c. feasibility (resources and time) d. competence of the investigators 2. Parts of the protocol and preparation guidelines a. introduction: brief overview and purpose of the study b. background: literature synthesis and rationale for the proposed study c. specific objectives and hypotheses d. methods: overall design, subject selection, data sources and variables, statistical methods, and sample-size justification c. administration: schedule, budget, and human-investigation issues d. discussion: summary, expected problems/limitations and solutions, research significance and implications B. Sample-size Determination 1. Relation between sample size and power 2. Specification of relevant parameters a. type I and type II errors (alpha and beta) b. the minimum magnitude of effect that is regarded as important to detect or that is expected c. the expected frequency of disease or exposure d. the sampling ratio of comparison to index subjects 3. Applications to basic designs a. fixed cohort studies: risk-ratio estimation
Page 13 b. density cohort studies: rate-ratio estimation c. case-control studies: odds-ratio estimation 4. Testing for interactions