Study Designs and the Evidence Pyramid

🟢 Lite — Quick Review (1h–1d)

Rapid summary for last-minute revision before your exam.

Evidence-Based Medicine (EBM) integrates the best available research evidence with clinical expertise and patient values to guide primary-care decisions. The hierarchy of evidence ranks systematic reviews and meta-analyses of RCTs at the top, descending through individual RCTs, cohort studies, case-control studies, cross-sectional surveys, and case reports. A clinical question is structured using the PICO framework (Population, Intervention, Comparison, Outcome). Diagnostic test performance is read off a 2×2 table: Sensitivity = TP/(TP+FN) rules out disease when negative (SnNout), while Specificity = TN/(TN+FP) rules it in when positive (SpPin). NNT = 1/ARR quantifies treatment benefit. Two high-yield exam pointers: (1) cohort studies yield Relative Risk, case-control studies yield Odds Ratio; (2) a non-significant p-value means absence of evidence, not evidence of absence — always check the confidence interval.

🟡 Standard — Regular Study (2d–2mo)

Standard content for students with a few days to months.

Study Designs and the Evidence Pyramid

Each design answers a different question and carries a distinct risk of bias. RCTs randomise participants to intervention vs control, balancing confounders and supporting causal inference. Cohort studies follow exposed and unexposed groups forward in time to calculate incidence and Relative Risk (RR = incidence_exposed / incidence_unexposed). Case-control studies start with diseased vs non-diseased subjects and look backward for exposure, producing an Odds Ratio (OR = ad/bc); they are efficient for rare diseases but cannot directly give incidence. Cross-sectional studies measure exposure and disease simultaneously at one time point, yielding prevalence rather than incidence.

Diagnostic Test Appraisal (2×2 Table)

Test result	Disease +	Disease −
Positive	TP	FP
Negative	FN	TN

Sensitivity = TP/(TP+FN) — proportion of diseased correctly identified.
Specificity = TN/(TN+FP) — proportion of non-diseased correctly identified.
PPV = TP/(TP+FP) and NPV = TN/(TN+FN) depend on disease prevalence in the tested population.

Measures of Treatment Effect

The Absolute Risk Reduction (ARR) = control event rate − experimental event rate. NNT = 1/ARR tells the clinician how many patients must be treated for one additional patient to benefit. Relative Risk Reduction (RRR) looks larger than ARR and is often misreported in promotional material.

Biostatistics Essentials

The p-value is the probability of observing the data (or more extreme) assuming the null hypothesis is true; p < 0.05 is the conventional threshold but is not a measure of effect size. A 95% confidence interval (CI) provides a range of plausible values for the true effect — narrow intervals reflect precision, and intervals crossing 1.0 (for RR/OR) or 0 (for mean differences) indicate non-significance. Intention-to-treat analysis preserves randomisation by analysing participants in their assigned group regardless of crossover or dropout.

Critical Appraisal Steps

Validity — appropriate design, randomisation, blinding, complete follow-up.
Importance — magnitude and precision of the effect.
Applicability — match between study population and the Saudi primary-care patient.

🔴 Extended — Deep Study (3mo+)

Comprehensive coverage for students on a longer study timeline.

Bias, Confounding, and Causation

Selection bias distorts who enters the study (e.g. healthy-worker effect in occupational cohorts). Information/recall bias distorts how exposures are measured — particularly problematic in case-control designs relying on memory. Confounding occurs when a third variable influences both exposure and outcome (e.g. smoking confounding the alcohol–cancer relationship); it is controlled by randomisation, restriction, matching, stratification, or multivariate regression. Publication bias selectively favours positive results, inflating effect estimates in meta-analyses — assessed by funnel plots and Egger’s test.

Screening Concepts

Screening tests require higher sensitivity than confirmatory tests because the goal is to miss no cases. Lead-time bias falsely extends apparent survival by earlier diagnosis without altering mortality. Length-time bias over-represents slowly progressive cases in prevalence-based screening studies. Wilson and Jungner criteria guide whether screening is justified in a Saudi MoH primary-care context (e.g. diabetes, breast cancer).

Worked Example

A screening mammography study of 10,000 women: 200 have breast cancer (180 test positive, 20 false negative); among the 9,800 without cancer, 980 false positive. Sensitivity = 180/200 = 90%; Specificity = 8,820/9,800 = 90%; PPV = 180/(180+980) = 15.5%; NPV = 8,820/8,840 = 99.8%. Even with strong sensitivity and specificity, PPV is modest because prevalence is only 2%.

Common Mistakes

Reporting RRR instead of ARR inflates perceived benefit; always calculate NNT.
Applying OR as if it were RR in cohort studies overestimates risk when outcomes are common (>10%).
Treating p > 0.05 as “no effect” ignores study power and CI width.

Practice Prompts

A cohort study of 1,000 statin users and 1,000 non-users records 30 vs 60 MIs over 5 years. Calculate RR, ARR, RRR, and NNT.
From the 2×2 data above (180 TP, 20 FN, 980 FP, 8,820 TN), explain why PPV would rise if the same test were applied to a high-risk clinic where prevalence is 20%.

Content adapted based on your selected roadmap duration. Switch tiers using the selector above.

Study Designs and the Evidence Pyramid