Transcript lect1
Applied Epidemiologic Analysis Patricia Cohen, Ph.D. Henian Chen, M.D., Ph. D. Teaching Assistants Julie Kranick Chelsea Morroni Applied Epidemiologic Analysis Fall 2002 Sylvia Taylor Judith Weissman Outline Lecture 1 1) Overview of unique and common aspects of epidemiological analysis in the larger scientific context 2) Review of the changing epidemiological research climate 3) Review of measures of disease frequency Applied Epidemiologic Analysis Fall 2002 Learning Objectives • To understand how study design and statistical analyses are linked components in sound epidemiologic investigations • To appreciate the changing nature of disease definitions and the consequent impact on design and analysis • To review alternative measures of disease frequency Applied Epidemiologic Analysis Fall 2002 Some Themes to Review in R&G Modern Epidemiology • Discussion of the meaning of causality • New understanding of the case-control study as a special kind of cohort study • Views on statistical hypothesis testing Applied Epidemiologic Analysis Fall 2002 Considerations Determining Study Design •Classic issues – Disease frequency – Exposure measures • Increasing importance – Participation rate problems – Concern about biased reports and recall – Difficulty isolating risk factors – Appreciation of multiple contributing factors – Appreciation of potential contingent effects – Wide availability of sophisticated statistical analytic techniques Applied Epidemiologic Analysis Fall 2002 Why Review Epidemiological Study Designs? • Measures of disease frequency depend on study designs • Measures of exposure effect depend on study designs • Appropriate statistical considerations depend on study designs including sampling issues Applied Epidemiologic Analysis Fall 2002 Cohort Designs: • Classical: sampling two or more disease-free groups differing on exposure to a potential disease cause (e.G. Snow study of cholera: “sampling” of two residential areas differing on water source) • Current: occupational or residentially-based cohorts • Extended: sampling of a single cohort heterogeneous on exposure, e.g. birth or age cohorts, samples from selected occupational groups or residential areas Applied Epidemiologic Analysis Fall 2002 Cohort Designs (Cont.) • May be prospective, retrospective, or crosssectional (x-sec may be considered a distinct category especially suited for “fixed” exposures) • Prospective: usually current or recent risks assessed at original sampling and disease assessed at follow-up • Retrospective: often risks assessed by means of prior records: may also be assessed by retrospective report of participants Applied Epidemiologic Analysis Fall 2002 Case-control Designs: • Classic method for assessing risk for rare diseases • Ideally samples incident cases, but may sample prevalent cases (note sampling bias associated with disease duration) • Notes R & G discussion of case-control designs as a “weighted sample” kind of cohort design Applied Epidemiologic Analysis Fall 2002 Study Designs We Will Not Focus On: • Ecological studies • Proportional mortality studies • Infectious disease studies Applied Epidemiologic Analysis Fall 2002 Nature of the Disease • Major influence on study design, the disease measure, and the statistical analysis • Classically, especially infectious, brief onset period and clear, often time-limited disease • Increasingly, chronic and insidious, gradual onset or progression – e.g many mental illnesses, artherosclerosis, obesity, diseases of old age – often also a certain arbitrariness in the “cut-point’ defining disease Applied Epidemiologic Analysis Fall 2002 Frequency of Both Disease and Studied Exposures • Increasing interest in high frequency disorders (like many of the chronic ones discussed) • Increasing interest in high frequency and graded “exposures”: diet, environmental risks) • Increasing felt need to incorporate genetic issues Applied Epidemiologic Analysis Fall 2002 Other Influences on Selected Study Designs • Measures of disease frequency – Incidence measures: what data are needed/how obtained – Prevalence measures: limitations – Scaled measures of prevalence Applied Epidemiologic Analysis Fall 2002 A review of some recent studies We recently surveyed the types of studies and analyses reported in three consecutive issues of American Journal of Epidemioloy. For your first homework assignment you will be asked to do a similar task, although a bit more thorough. Applied Epidemiologic Analysis Fall 2002 Survey results 27 articles produced the following results: Design type Outcome variable type Statistical analyses 12 Cohort 17 Dichotomous 14 Logistic regression 4 Cross sectional 10 Scaled 4 Survival analysis 8 Case-control 9 Multiple regression 1 Clinical trial 1 Conditional logistic regression 1 Poisson regression 2 ANOVA 1 Multilevel regression 1 Repeated measure analysis Applied Epidemiologic Analysis Fall 2002 Design and outcome determinants The selection of both the outcome measure (incidence; prevalence) and the study/analytic design depends on: The nature of the morbidity measure The nature of the exposure(s) of interest. Applied Epidemiologic Analysis Fall 2002 Issues relating to exposures • Is there a clear distinction between exposed and unexposed? • Is the exposure another disease? • Is the exposure to be found in one area/occupation/ but not in another otherwise comparable population? • Is the exposure a prevention or treatment program? Applied Epidemiologic Analysis Fall 2002 Other current exposure measures: • Biological markers • Environmental measures • Scaled measures of behavior (e.g. dietary intake, exercise) Applied Epidemiologic Analysis Fall 2002 Incidence measures: When are they feasible? Point exposures: Industrial accidents (e.g. radiation exposure) Natural disasters (e.g. hurricane Unnatural disasters (e.g. 9/11) Infection exposure Note: must be known and considerations of sequellae experiences complicate the question of timing of effective exposures. Applied Epidemiologic Analysis Fall 2002 Period or duration of exposures prior to onset Occupational exposures Exposures related to own behavior (e.g. smoking, diet) Age may be used as a proxy for duration of chronic exposures (e.g. as in mortality studies) Note: Considerations of level of exposure often complicate these analyses, although in a case-control design these may be treated separately by stratification. Applied Epidemiologic Analysis Fall 2002 Prevalence measures: When are they used? • Insidious onset • Fluctuating or recurring course • Indistinct or somewhat arbitrary “cut-off” points to define Applied Epidemiologic Analysis Fall 2002 Measures of Disease Frequency Incidence Time: times at which new disease occurs among population members Incidence Rate: occurrence of new disease per unit of person-time Incidence Proportion: proportion of people who develop new disease during a specified period of time Prevalence: proportion of people who have a disease at a specific time Applied Epidemiologic Analysis Fall 2002 Induction Time and Latent period Induction time Causal action of a component cause Latent period Disease Occurrence Disease Detection Induction time only in relation to a specific cause Latent period - characteristic of disease, can be influenced by screening Applied Epidemiologic Analysis Fall 2002 Definitions of terms in equations for measures of disease N = number of person in sample (or population) at a given time or at a given interval t = a time unit, t = the disease-free (pre-onset) time observed for each person in sample A = number of disease onsets R = risk of disease onset in a specified time interval S = survival proportion I = incidence rate Open population -- population size is balanced by equal entrances and exits, at least within particular age, sex or other strata Closed sample -- censoring needs to be taken into account Applied Epidemiologic Analysis Fall 2002 Incidence Rate Numerator: count onsets among people whose event occurred while contributing to denominator Denominator: sum of time periods (e.g. months, years) prior to onset across all persons at risk, defined as those without disease Need unit of time to interpret numerical value Applied Epidemiologic Analysis Fall 2002 Incidence and survival proportions Proportion of closed population with an onset in a given (constant) period Measure of average risk (risk applies to individuals) Range from 0 to 1, dimensionless S = Survival proportion = 1-R Applied Epidemiologic Analysis Fall 2002 Relationship among measures Incidence odds = R R S 1 R R S 1 , R S If R is small, But S < 1, so R R S If R is small, population declines only slightly N A N A At R It N Nt Applied Epidemiologic Analysis Fall 2002 Relationship among measures (cont) For a given interval k Ak Rk Nk - population is closed - no competing risks - number of events at each time k is small Ak Ik N k tk S k 1 Rk 1 I k tk S sk 1 I k tk Applied Epidemiologic Analysis Fall 2002 S exp I k tk R 1 S 1 exp I k tk • Open populations: – If open population, estimate incidence rate, use exponential formula to get estimate of incidence proportion – plausible if open population subset of closed population and loss to followup unrelated to risk • Competing risks: – population size not constant over subinterval Applied Epidemiologic Analysis Fall 2002 Prevalence Under stationarity assumptions D=duration of disease 1 I N P t P t D inflow = outflow Prevalence proportion - dimensionless - range 0 to 1 - useful for health planning - certain health outcomes P I D NP If P < .1 then P I D N More generally, Applied Epidemiologic Analysis Fall 2002 P ID N 1 I D Learning Objectives • To understand how study design and statistical analyses are linked components in sound epidemiologic investigations • To appreciate the changing nature of disease definitions and the consequent impact on design and analysis • To review alternative measures of disease frequency Applied Epidemiologic Analysis Fall 2002