Biostatistics
Biostatistics & Study Design — Complete
Biostatistics

Biostatistics & Study Design — Complete

Study designs, RR vs OR, sensitivity/specificity, biases, statistical tests.

Select any text to highlight it or make a flashcard.

◆Study designs (strongest to weakest)

  • •Meta-analysis > Systematic review > RCT > Cohort > Case-control > Cross-sectional > Case series/report
  • •RCT: gold standard for causation; randomization eliminates confounding
  • •Cohort: prospective; exposure → outcome; calculates incidence + RR
  • •Retrospective cohort: uses past records (both exposure and outcome already occurred)
  • •Case-control: starts with disease, looks back for exposure; OR; good for rare diseases; recall bias risk
  • •Cross-sectional: snapshot; measures prevalence; no temporal relationship
  • •Ecological: population-level data; ecological fallacy = main bias
  • •Twin concordance: monozygotic vs dizygotic for heritability

◆Clinical trial phases

  • •Phase I: small group of HEALTHY volunteers — SAFETY + pharmacokinetics
  • •Phase II: patients with disease — efficacy + side effects + dosing
  • •Phase III: large RCT — confirms efficacy, compares to standard or placebo, FDA approval
  • •Phase IV: post-marketing surveillance — long-term/rare side effects

◆Risk measures

  • •Relative Risk (RR) = Incidence in exposed / Incidence in unexposed (cohort/RCT)
  • •Odds Ratio (OR) = ad/bc (case-control)
  • •Relative Risk Reduction (RRR) = 1 − (Rt/Rc)
  • •Absolute Risk Reduction (ARR) = Rc − Rt
  • •Number Needed to Treat (NNT) = 1 / ARR (round UP)
  • •Number Needed to Harm (NNH) = 1 / ARI
  • •Attributable Risk = Ie − Iu

◆Diagnostic test characteristics

  • •Sensitivity = TP / (TP + FN) — rules OUT (SnNout); intrinsic test property
  • •Specificity = TN / (TN + FP) — rules IN (SpPin); intrinsic test property
  • •PPV = TP / (TP + FP) — depends on prevalence
  • •NPV = TN / (TN + FN) — depends on prevalence
  • •LR+ = Sn / (1 − Sp) — large value (>10) rules IN
  • •LR− = (1 − Sn) / Sp — small value (<0.1) rules OUT
  • •Likelihood ratios: independent of prevalence

◆Cutoff effects (lower threshold)

  • •Lower cutoff: ↑ Sn, ↑ NPV, ↓ Sp, ↓ PPV — more false positives
  • •Higher cutoff: ↑ Sp, ↑ PPV, ↓ Sn, ↓ NPV — more false negatives
  • •Screening test: lower cutoff (want high sensitivity)
  • •Confirmatory test: raise cutoff (want high specificity)

◆Biases

  • •Selection bias: Berkson (hospital controls), healthy worker, attrition (loss to follow-up)
  • •Recall bias: case-control inaccurate memory
  • •Observer bias: researcher expectation; prevent with double-blinding
  • •Hawthorne effect: behavior change from being observed
  • •Confounding: 3rd variable linked to both; address with randomization, matching, stratification, regression
  • •Effect modification: NOT a bias — true difference in effect across subgroups
  • •Lead-time bias: earlier diagnosis without survival benefit
  • •Length-time bias: screening preferentially detects indolent disease
  • •ITT (intention-to-treat) analysis: addresses attrition bias

◆Statistical tests by data type

  • •2 means, normal distribution → t-test (paired or independent)
  • •≥3 means → ANOVA
  • •Categorical × categorical → chi-square (or Fisher exact if cells <5)
  • •Paired categorical → McNemar
  • •Continuous × continuous (linear) → Pearson correlation
  • •Non-normal small samples → Mann-Whitney (2 groups), Kruskal-Wallis (≥3)
  • •Survival data → Kaplan-Meier curves; log-rank test for comparison

◆Hypothesis testing

  • •Null hypothesis: NO effect/difference
  • •Type I error (α, false positive): rejecting a TRUE null; usually 0.05
  • •Type II error (β, false negative): failing to reject a FALSE null
  • •Power = 1 − β (probability of correctly rejecting false null)
  • •Increase power: ↑ sample size, ↑ effect size, ↓ variability, ↑ α
  • •P-value: probability of observing data if null is true; p < α → reject null
  • •Confidence interval: if CI includes null value (0 for differences, 1 for ratios) = NOT significant

◆Distributions & central tendency

  • •Normal distribution: 68% ±1 SD, 95% ±2 SD, 99.7% ±3 SD
  • •Right-skewed: mean > median > mode
  • •Left-skewed: mean < median < mode
  • •Outliers pull MEAN; median is robust
  • •Standard error (SE) = SD / √n

High-yield pearls

  • ◆Cohort → RR; Case-control → OR
  • ◆NNT = 1/ARR (round UP)
  • ◆Sn/Sp don't depend on prevalence; PPV/NPV do
  • ◆Power = 1 − β; increase n to increase power
  • ◆Lead-time bias: 🕒 earlier clock; Length-time bias: 🐢 turtle (slow) cancers
Quick check

5-question quiz on this note

Test yourself before moving on. ~1 min.

Done reading?
Track your progress by marking this complete.
Next in Biostatistics