Statistical & Clinical Analysis
Net Reclassification Improvement (NRI) & Integrated Discrimination Improvement (IDI)
Evaluate Michael Pencina’s biomarker incremental value metrics: additive Event + Non-Event NRI and probability-slope separation gain IDI = (IS_new - IS_old).
Why a clinically decisive biomarker like coronary calcium or hs-Troponin barely nudges ΔAUC by 0.01—yet moves patients across statin thresholds!
Michael J. Pencina et al. (2008, Statistics in Medicine — Framingham Heart Study) showed that adding a strong independent risk marker to an established clinical model often yields a tiny Δc-statistic while substantially reclassifying patients across actionable treatment cutoffs (e.g., <5%, 5–20%, >20% 10-year CVD risk)! Categorical Net Reclassification Improvement separately credits upward movement among true Events and downward movement among true Non-Events: NRI = (Up_ev - Down_ev) / N_ev + (Down_ne - Up_ne) / N_ne!
Simultaneously, threshold-free Integrated Discrimination Improvement IDI = (p̄_new,ev - p̄_old,ev) - (p̄_new,ne - p̄_old,ne) measures the net widening of Yates’s discrimination slope! With N_ev = 200 cases (54 up, 22 down -> NRI_e = +0.1600) and N_ne = 800 controls (168 down, 96 up -> NRI_ne = +0.0900), overall NRI = +0.2500 (25.0%) and IDI = (0.310 - 0.250) - (0.080 - 0.095) = +0.0750!
| Biomarker Reclassification Component | Formula Definition | Cohort Result (200 Events / 800 Controls) | Clinical Interpretation |
|---|---|---|---|
| Event NRI_e = (54 - 22) / 200 | Net cases moved to higher tier | +0.1600 (+16.0%) | More future MI cases treated |
| Non-Event NRI_ne = (168 - 96) / 800 | Net controls moved to lower tier | +0.0900 (+9.0%) | Fewer healthy controls overtreated |
| Total Additive NRI & IDI | NRI_e + NRI_ne ; ΔSlope | NRI = +0.2500 / IDI = +0.0750 | Strong net reclassification gain |
Pencina Overall Net Reclassification Improvement (NRI = NRI_events + NRI_nonevents)
Sums net upward reclassification in events (B3 - B4)/B2 and net downward reclassification in non-events (B6 - B7)/B5.
=IF(OR(B2 <= 0, B5 <= 0, B3 + B4 > B2, B6 + B7 > B5), "Invalid Reclassification Counts", ROUND((B3 - B4) / B2 + (B6 - B7) / B5, 4))Integrated Discrimination Improvement IDI = (p̄_new,ev - p̄_old,ev) - (p̄_new,ne - p̄_old,ne)
Computes the increase in discrimination slope between the updated biomarker model (B8, B10) and baseline clinical model (B9, B11).
=IF(OR(B2 <= 0, B5 <= 0), "Invalid Reclassification Counts", ROUND((B8 - B9) - (B10 - B11), 4))Try the example
Use the original values or change them to test your understanding.