Drug Information Observational Studies & Epi Measures

OBSERVATIONAL STUDIES & EPIDEMIOLOGIC MEASURES KRISTIN M. JANZEN, PHARMD, BCPS CLINICAL ASSOCIATE PROFESSOR | PHARMACY PRACTICE OBSERVATIONAL STUDIES STUDY DESIGN OBJECTIVES 1. Demonstrate how a 2x2 table is used to represent the relationship between disease/outcome and exposure according to di...

OBSERVATIONAL STUDIES & EPIDEMIOLOGIC MEASURES KRISTIN M. JANZEN, PHARMD, BCPS CLINICAL ASSOCIATE PROFESSOR | PHARMACY PRACTICE OBSERVATIONAL STUDIES STUDY DESIGN OBJECTIVES 1. Demonstrate how a 2x2 table is used to represent the relationship between disease/outcome and exposure according to different observational study designs 2. Critique each observational study design according to the following: ¡ Purpose of each study design ¡ Advantages and disadvantages/ limitations of each study design ¡ Components of each study design and how each is conducted ¡ Measure of association(s) commonly used with each design 3. Describe the basic concepts of sources of errors for observational studies in terms of random errors and systematic errors 4. Determine what types of biases are commonly associated with each observational study design 5. Illustrate how confounding variables compete with the hypothesized risk factors as an explanation for the observed response 6. At an introductory level, explain how biases and confounding may be controlled or minimized STUDY DESIGN Descriptive: document and communicate experiences; begin search for explanations Explanatory: examine etiology, cause, efficacy, using the strategy of comparisons USE OF THE 2X2 TABLE Outcome # with exposure # without exposure + Outcome - Outcome  Classify and compare Begin Begins in the present and looks to the past for exposure OBSERVATIONAL Exposed + Cases Compares case vs. controls for exposure in the past STUDIES: People with Exposed - disease/ outcome CASE-CONTROL Exposed + DESIGN Controls People without disease/ outcome (RETROSPECTIVE, CASE Exposed - HISTORY, CASE-REFERENT, TROHOC) Past Present cohort spelled backwards! + Outcome - Outcome  ¡ Purpose: OBSERVATIONAL ¡ Retrospectively identify potential risk factors of diseases or outcomes STUDIES: ¡ Not designed to determine effect CASE-CONTROL ¡ Establish association, not causality DESIGN ¡ Advantages: (RETROSPECTIVE, CASE ¡ Suited for initial, explanatory studies HISTORY, CASE-REFERENT, ¡ Study rare diseases or one that occurs infrequently TROHOC) ¡ Study a disease that occurs many years after exposure ¡ Discover possible disease etiology/ risk factors ¡ Relatively inexpensive ¡ Can be completed in shorter time frame OBSERVATIONAL ¡ Selection of cases: STUDIES: a. Subjects with a particular characteristic or outcome of interest CASE-CONTROL b. (e.g., disease) Case definition: broad criteria vs. rigorous criteria DESIGN ¡ ¡ standard diagnostic criteria severity of the disease or outcome (RETROSPECTIVE, CASE ¡ objective vs. subjective HISTORY, CASE-REFERENT, c. Sources of cases TROHOC) ¡ Do the cases represent the larger population of those who have the disease/ outcome? d. Incident or prevalent cases ¡ Incidence: development of new outcomes ¡ Prevalence: existing cases at the time of measure  ¡ Selection of controls: OBSERVATIONAL ¡ Subjects who have not experienced the characteristic or outcome STUDIES: ¡ Ideally, controls would have to have the same characteristics as the cases, except for the exposure of interest CASE-CONTROL ¡ Sources of control DESIGN 1. Population-based? (voter registration, Medicaid/Medicare) 2. Institution-based? (one hospital of patients) (RETROSPECTIVE, CASE 3. Do the controls represent the larger population? –most important! HISTORY, CASE-REFERENT, ¡ Use of multiple control groups: Controls of the same type (i.e. only the ratio is different) TROHOC) 1. ¡ For every 1 case: 4 controls ¡ Has to do with statistics 2. Controls of different types ¡ Are the controls from one group or multiple groups? ¡ An inappropriate control group can bias results  ¡ Matching: OBSERVATIONAL ¡ A method used to assure that cases and controls are similar STUDIES: ¡ Control for variables that are known to be related to the outcome and may confuse results CASE-CONTROL ¡ Reduces competing explanations for the outcome in question ¡ Most commonly matched variables? DESIGN ¡ Age, sex, race, socioeconomic status (RETROSPECTIVE, CASE ¡ Tests are run to see how similar case and control groups are to match them ¡ Data collected about past exposures: HISTORY, CASE-REFERENT, ¡ Sources of data collection: TROHOC) ¡ Interviews, health records ¡ Relies on patient recall, accuracy of medical records ¡ Difficult to validate data ¡ Measure of association – odds ratio (OR): ¡ Will see this when trying to determine if there is an association ¡ What are the odds of being exposed? STUDY DESIGN  Measure outcome/ compare OBSERVATIONAL Begin STUDIES: Disease/ outcome + PROSPECTIVE Exposed + FOLLOW-UP Disease/ outcome - Free of disease DESIGN Sample from a defined (COHORT STUDY, population Disease/ PROSPECTIVE STUDY Exposed - outcome + INCIDENCE STUDY, LONGITUDINAL STUDY) Exclude subjects already with Disease/ outcome - outcome Present Future Prospective – takes place in present and follows group forward (could be 10-20 years) No intervention! Measures incidence (new cases) OBSERVATIONAL § Advantages STUDIES: § Strongest observational study design – permits direct PROSPECTIVE determination of risk FOLLOW-UP § § Prospective data collection Exposure known at the beginning of study DESIGN § Study determines outcome occurrence (COHORT STUDY, § Allow examination of multiple outcomes PROSPECTIVE STUDY § Administratively easier and cheaper than clinical trials INCIDENCE STUDY, LONGITUDINAL STUDY) § But long ones are still expensive § Disadvantages § Difficult to address rare events § Results frequently not available for a long time § Time-consuming and expensive § Ethical concerns OBSERVATIONAL § Sources of cohorts: STUDIES: § Population-based PROSPECTIVE § Exposure-based Look for people from a certain place where they were likely FOLLOW-UP § exposed (i.e. factory) DESIGN § Measure of association: (COHORT STUDY, § Calculate incidence rate PROSPECTIVE STUDY INCIDENCE STUDY, § Relative risk LONGITUDINAL STUDY) FOLLOW-UP STUDIES: OTHER DESIGNS ¡ Nested Case-Control Study ¡ Case-control study within a cohort study ¡ Advantages o Efficient to conduct o Comparability of cases and controls is high because they come from a common source population o Quality and comparability of data are usually good STUDY DESIGN Descriptive: document and communicate experiences; begin search for explanations Explanatory: examine etiology, cause, efficacy, using the strategy of comparisons  Begin Measure/ Classify & Compare Snapshot of one point in OBSERVATIONAL Exposure or risk factor + time (could be on day or a Free of disease/ couple months) STUDIES: Exposure or Sample from a outcome risk factor - Prevalence study (existing defined population cases) CROSS- Exposure or Have disease/ risk factor + outcome SECTIONAL Exposure or risk factor - DESIGN Present (PREVALENCE STUDY)  ¡ Description: OBSERVATIONAL ¡ Exposure and disease measures are obtained at the level of the individual STUDIES: ¡ Single period of observation – exposure and disease histories are CROSS- ¡ collected simultaneously (1 day, 2 months, etc) Provides a quantitative estimate of the magnitude of a problem SECTIONAL ¡ Descriptive in nature DESIGN ¡ Findings serve as a source of hypotheses for future detailed etiologic studies (PREVALENCE STUDY) ¡ Use of Cross-Sectional Studies: ¡ Evaluate a new test or the new application of an old one ¡ Evaluate the predictive capability of clinical features ¡ Identify etiological agents or causative factors ¡ Determine the prevalence of a problem OBSERVATIONAL ¡ Advantages: STUDIES: ¡ Suited for initial, explanatory studies ¡ Study rare conditions/diseases CROSS- ¡ Inexpensive and quick to conduct SECTIONAL ¡ Less involvement/effort for patients then in a cohort or controlled clinical trial DESIGN ¡ May be used to identify cases and controls for a case-control study (PREVALENCE STUDY) ¡ Measurement: ¡ Prevalence (existing cases) TYPES OF BIAS  Berkson’s bias (Berkson’s fallacy, admission rate bias) CONCERNS AND Admission rates for comparison groups differ, resulting in a distorted association between exposure and Disease in hospital-based studies study design: case-control BIASES: Healthy worker bias Do working, healthy individuals have healthier habits? SELECTION Study design: cohort Volunteer bias BIAS Is there something different about those willing to volunteer? Study design: cross-sectional; cohort Migration bias – change in habits i.e. If someone in a smoking study quit smoking Non-random or biased selection Study design: cohort Non-response bias of participants Non-respondents may exhibit exposures or outcomes that differ from responders, resulting in over or under estimation of odds or risk loss to follow-up: cohort study design: case-control; cross-sectional; cohort Neyman bias (prevalence-incidence bias) Timing of exposure identification causes some cases to be missed (quick recovery or high mortality) study design: case-control; cohort Unmasking bias An innocent exposure causes a sign or symptom that precipitates search for a disease, but does not itself cause the disease study design: case-control; cohort CONCERNS AND Recall bias (most common) BIASES: Occurs when cases are more likely to recall exposure than controls Study design: case-control; cross-sectional; cohort INFORMATION Prevarication (lying ) bias BIAS Occurs when a respondent has ulterior motives for answering a question and thus may underestimate or exaggerate exposures undesirable Study design: case-control; cross-sectional; cohort Incomparable medical records Collection of Study design: any study design requiring use of medical records inaccurate Interviewer/abstractor bias information -- Knowledge of a subject’s disease/outcome status may influence the intensity of a search for exposure Study design: case-control; cross-sectional; cohort misclassification or Surveillance bias measurement Following a high-risk group more closely Study design: cohort CONCERNS AND Post hoc significance bias BIASES: Occurs when decisions regarding level of significance are selected a posteriori DATA ANALYSIS Authors conduct analysis of outcomes they didn’t originally state BIAS Data dredging bias Occurs when data are analyzed for all possible associations without prior hypotheses Misinterpretation of Outside of the study objectives and primary outcomes Significance bias data Occurs when statistical significance is confused with clinical significance p