SPSC Screening Test cheat sheet

Screening converts a continuous or multi-item signal into a pass/fail decision against a pre-set threshold; the threshold is fixed before testing, never after seeing scores.

Sensitivity = TP / (TP + FN): probability a truly positive case is flagged; a highly sensitive screen yields few false negatives and is preferred when missing a case is costly.

Specificity = TN / (TN + FP): probability a truly negative case passes as negative; low specificity produces false positives and confirmatory burden.

Positive predictive value = TP / (TP + FP) and negative predictive value = TN / (TN + FN); both depend on prevalence, so PPV falls as prevalence falls even at constant sensitivity and specificity.

Likelihood ratio positive = sensitivity / (1 - specificity); LR+ above 10 gives strong evidence for the condition, LR+ near 1 changes nothing.

Likelihood ratio negative = (1 - sensitivity) / specificity; LR- below 0.1 substantially argues against the condition.

Receiver-operating characteristic plots sensitivity against (1 - specificity) across thresholds; the area under the curve equals the probability a random positive outranks a random negative.

Raising the cut-off generally increases specificity and lowers sensitivity; lowering it does the reverse. Report the threshold with every screening statistic.

Number needed to screen = 1 / (absolute risk reduction from screening); it expresses effort per outcome prevented and rises sharply as event prevalence falls.

Reliability concerns repeatability (coefficient of variation, intraclass correlation, kappa); validity concerns agreement with a reference standard. A test can be reliable but invalid.

Cohen's kappa = (observed agreement - chance agreement) / (1 - chance agreement); kappa below 0.20 is slight, 0.41-0.60 moderate, above 0.80 almost perfect.

Lead-time bias: earlier detection lengthens survival time from diagnosis without delaying death; compare mortality, not survival from diagnosis.

Length bias: slow-progressing disease stays detectable longer, so screening preferentially samples indolent cases and inflates apparent benefit.

Selection bias appears when volunteers for screening differ in risk from non-participants; randomization or adjustment is required before attributing benefit.

Verification (work-up) bias: patients with positive screens receive more thorough reference testing than negative patients, so false negatives are undercounted.

Spectrum bias: sensitivity measured in severe, hospital-referred cases overstates performance in the mild, asymptomatic population actually screened.

Reference standard must be applied blinded and independently; an imperfect gold standard biases sensitivity and specificity toward each other.

Overdiagnosis is detection of disease that would never have caused symptoms; it is a harm of screening, not a benefit, and is invisible in survival statistics.

Randomized controlled trial with mortality or incidence as the endpoint is the reference design; observational comparisons of screened versus unscreened groups are confounded by self-selection.

Intention-to-screen analysis preserves randomization; per-protocol analysis of attenders only reintroduces healthy-adherer bias.

Absolute measures (risk difference, number needed to screen) convey benefit better than relative risk reduction, which can look large when baseline risk is tiny.

Wilson-Jungner criteria: important health problem, accepted treatment, recognizable latent stage, suitable test, agreed policy on whom to treat, cost-effective, continuous case-finding.

A screening test should be acceptable, safe, inexpensive, and reproducible; a positive result must lead to a defined diagnostic pathway with available capacity.

Informed consent for screening must disclose the possibility of false positives, false negatives, and overdiagnosis, not only the intended benefit.

Cost per quality-adjusted life-year and the false-positive rate per true case found are standard program metrics; a low PPV implies many work-ups per case detected.

Harms to count: false-positive anxiety and procedures, radiation or perforation from confirmatory tests, overdiagnosis, and opportunity cost to the health system.

Regulatory framing: a screening test used to guide treatment decisions is often held to a higher evidence standard than one used for risk awareness alone.