16.4 Clinical Epidemiology, Biostatistics & Study Design

1. Clinical Epidemiology & Biostatistics for Board Exams

A. Diagnostic Test Performance & 2x2 Contingency Matrix

• Sensitivity (True Positive Rate): The probability that a test is positive among individuals who have the disease. SnNOut: A test with high Sensitivity, when Negative, rules Out the disease (minimal false negatives). Ideal for screening tests (e.g., ELISA for HIV, D-dimer for PE, ANA for SLE).
• Specificity (True Negative Rate): The probability that a test is negative among individuals who do NOT have the disease. SpPIn: A test with high Specificity, when Positive, rules In the disease (minimal false positives). Ideal for confirmatory tests (e.g., Western blot/differentiation immunoassay for HIV, biopsy for cancer).
• Positive Predictive Value (PPV): The probability that a patient with a positive test actually has the disease. PPV depends heavily on disease prevalence: As disease prevalence increases, PPV increases and NPV decreases.
• Negative Predictive Value (NPV): The probability that a patient with a negative test does not have the disease. As disease prevalence decreases, NPV increases and PPV decreases.

B. Likelihood Ratios (Prevalence-Independent Diagnostic Power)

Likelihood ratios quantify the odds of a given test result in diseased versus non-diseased individuals and are independent of disease prevalence.

$\text{Positive Likelihood Ratio (LR+)} = \frac{\text{Sensitivity}}{1 - \text{Specificity}} = \frac{\text{True Positive Rate}}{\text{False Positive Rate}}$

$\text{Negative Likelihood Ratio (LR-)} = \frac{1 - \text{Sensitivity}}{\text{Specificity}} = \frac{\text{False Negative Rate}}{\text{True Negative Rate}}$

• Interpretation of Likelihood Ratios:
  • LR+ > 10: Generates a large, conclusive increase in post-test disease probability (rules in disease).
  • LR+ 5 to 10: Moderate increase in post-test probability.
  • LR 1.0: No diagnostic value (post-test probability equals pre-test probability).
  • LR- 0.1 to 0.2: Moderate decrease in post-test probability.
  • LR- < 0.1: Generates a large, conclusive decrease in post-test disease probability (rules out disease).

C. Measures of Association & Clinical Treatment Effects

2. Epidemiological Study Designs & Methodological Biases

A. Hierarchy of Study Designs

1. Randomized Controlled Trial (RCT): Gold standard for demonstrating causality. Participants are randomly allocated to experimental or control arms. Minimizes selection bias and baseline confounding.
   • Intention-to-Treat (ITT) Analysis: Analyzes all randomized patients in their assigned treatment groups, regardless of protocol non-adherence, crossover, or study dropout. Preserves randomization balance and avoids bias.
   • Per-Protocol (As-Treated) Analysis: Analyzes only subjects who strictly adhered to the assigned protocol. Introduces selection bias and confounding.
2. Cohort Study: Observational design where subjects are identified by exposure status and followed prospectively (or retrospectively) over time to measure disease incidence and Relative Risk (RR). Prone to Loss to Follow-up / Attrition Bias.
3. Case-Control Study: Observational design where subjects are selected based on outcome status (Cases with disease vs. Controls without disease) and evaluated retrospectively for prior exposures. Measures Odds Ratio (OR). Prone to Recall Bias.
4. Cross-Sectional Study: Observational snapshot measuring exposure and disease status simultaneously at a single point in time. Measures disease Prevalence. Cannot establish temporal sequence or causality.

B. High-Yield Methodological Biases

• Selection Biases:
  • Berkson Bias: Selection bias occurring when hospitalized patients are used as control subjects (hospitalized patients have higher rates of comorbidities and risk exposures than the general population).
  • Healthy Worker Effect: Bias where working individuals exhibit lower mortality rates than the general population, because severely ill/disabled persons are excluded from the workforce.
• Information & Measurement Biases:
  • Recall Bias: Patients with adverse clinical outcomes recall past exposures more vividly (or inaccurately) than healthy controls (classic in retrospective case-control studies).
  • Observer / Pygmalion Effect: Investigator knowledge of group assignment influences measurement of outcomes (prevented by double-blinding).
  • Hawthorne Effect: Study subjects alter their behavior simply because they know they are being observed.
• Confounding: A distortion that occurs when an extraneous variable is associated with both the exposure and the outcome, without being an intermediate step in the causal pathway.
  • Methods to Control Confounding in Design Phase: Randomization, Restriction (limiting enrollment criteria), and Matching.
  • Methods to Control Confounding in Analysis Phase: Stratification and Multivariable Regression Analysis.
• Screening Biases:
  • Lead-Time Bias: Early detection of disease creates the false statistical illusion of prolonged survival, even though the natural history and actual time of death are entirely unchanged.
  • Length-Time Bias: Screening preferentially identifies slow-growing, indolent, less aggressive cases with longer preclinical phases and inherently better prognoses, while missing rapidly fatal, aggressive cases that present symptomatically between screening intervals.