Observational Studies & Epidemiologic Measures

Observational Study Designs

  • 2x2 Table: Represents the relationship between disease/outcome and exposure.
    • Rows: Exposure Status (Yes/No).
    • Columns: Disease Status (Yes/No).
    • AA = Exposed with Disease; BB = Exposed No Disease; CC = Unexposed with Disease; DD = Unexposed No Disease.
Case-Control Design (Retrospective)
  • Purpose: Retrospectively identify potential risk factors; establish association, not causality.
  • Advantages: Good for initial studies, rare diseases, long latency, inexpensive, quicker.
  • Disadvantages: Cannot determine effect, prone to selection and recall bias.
  • Components: Cases (with disease) vs. Controls (without disease).
  • Measure of Association: Odds Ratio (OR).
  • Matching: Ensures cases and controls are similar on known variables to reduce confounding.
Prospective Follow-Up Design (Cohort Study)
  • Purpose: Begin in present, follow groups (exposed/unexposed) forward to measure outcome occurrence; measures incidence.
  • Advantages: Strongest observational design, permits direct risk determination, prospective data, examines multiple outcomes.
  • Disadvantages: Difficult for rare events, long-term, expensive, ethical concerns.
  • Measure of Association: Incidence Rate, Relative Risk (RR).
Cross-Sectional Design (Prevalence Study)
  • Purpose: Snapshot at one point in time; exposure and disease measured simultaneously; provides magnitude of a problem.
  • Advantages: Suited for initial studies, inexpensive, quick, less patient effort.
  • Measure of Association: Prevalence.

Sources of Errors in Observational Studies

  • Random Error: Fluctuations due to sampling variability; cannot be controlled.
  • Systematic Error (Bias): Deviation of results from truth.
    • Minimization: Clear case definitions, blinding, standardized training and data collection.
    • Confounding: Distortion of exposure-effect estimate by an extraneous factor.
      • Criteria for Confounder: Risk factor for disease, associated with exposure, not intermediate in causal path.
      • Control: Restriction, Matching, Statistical adjustment in data analysis.
Types of Bias
  1. Selection Bias: Non-random participant selection (e.g., Berkson's, healthy worker, non-response, Neyman).
  2. Information Bias: Inaccurate data collection (e.g., recall, prevarication, interviewer, surveillance).
  3. Data Analysis Bias: Misinterpretation of data (e.g., post hoc significance, data dredging, significance, correlation).

Epidemiologic Measures

  • Epidemiology: Study of distribution and determinants of health and disease in populations.
    • Components: Describe health status, explain etiology, predict occurrence, control distribution.
Prevalence vs. Incidence
FeaturePrevalenceIncidence
NumeratorAll existing casesNew cases during a period
DenominatorAll people examined (cases + non-cases)All susceptible people at period start
TimeOne point in timeDuration of the period
Study DesignCross-sectionalCohort
RelationshipextPrevalencehickapproxextIncidenceimesextDurationext{Prevalence} hickapprox ext{Incidence} imes ext{Duration}
Prevalence (P)
  • Definition: Number of existing cases of a condition in a population at a designated time.
  • Point Prevalence: (extNumberofexistingcases)/(extTotalpopulationatapointintime)( ext{Number of existing cases}) / ( ext{Total population at a point in time})
  • Period Prevalence: (extNumberofexistingcasesduringtimeperiod)/(extAveragepopulationduringtimeperiod)( ext{Number of existing cases during time period}) / ( ext{Average population during time period})
Incidence Rate (IR)
  • Definition: Rate of development of new disease cases in a group over a certain time period.
  • extIR=(extNumberofnewcases)/(extPopulationatriskoveratimeperiod)imesextMultiplierext{IR} = ( ext{Number of new cases}) / ( ext{Population at risk over a time period}) imes ext{Multiplier}
Relative Risk (RR)
  • Measure: Association for cohort studies; "how many times more/less likely are exposed to get the disease?"
  • Formula: RR=(A/(A+B))/(C/(C+D))RR = (A / (A+B)) / (C / (C+D)). A,B,C,D from 2x2 table.
  • Interpretation:
    • RR=1RR = 1: No difference in risk.
    • RR > 1 : Increased risk in exposed.
    • RR < 1 : Decreased risk in exposed.
Odds Ratio (OR)
  • Measure: Association for case-control studies; "how much more/less likely cases are to be exposed than controls?"
  • Formula: OR=(AimesD)/(BimesC)OR = (A imes D) / (B imes C). A,B,C,D from 2x2 table.
  • Interpretation:
    • OR=1OR = 1: Equal odds of exposure.
    • OR > 1 : Increased odds of exposure for cases.
    • OR < 1 : Decreased odds of exposure for cases.
  • Caution: OR approximates risk; direct risk from prospective studies only.
Confidence Intervals (CI)
  • Purpose: Statistical measure for RR and OR to determine statistical significance.
  • 95% CI: Contains the "true" population estimate 9595% of the time.
  • Significance: For RR/OR, a CI that does not overlap 11 implies statistical significance.