Net Reclassification Improvement and IDI
Net reclassification improvement (NRI) and integrated discrimination improvement (IDI) measure how much adding predictors moves predicted risks in the right direction for people with and without the event.
순재분류개선도(NRI)와 IDI는 예측변수를 더했을 때 사건군의 위험은 올라가고 비사건군의 위험은 내려가는 방향으로 얼마나 옮겨졌는지 측정합니다.
Does adding a new biomarker to my risk model improve how patients are classified?
위험예측모형에 새 바이오마커를 추가하면 환자 분류가 나아지는지 어떻게 보나요?
NRI adds the net proportion of events whose risk moves up and the net proportion of non-events whose risk moves down, either for any change or across risk categories. IDI is the change in the difference between mean predicted risk in events and non-events, reported alongside the change in AUC.
When to use Net Reclassification Improvement and IDI
- Comparing a baseline and an extended risk model on the same binary outcome.
- Clinically meaningful risk categories exist for categorical NRI.
- Two sets of predicted risks are already available from external models.
When to use something else
- You need a test of whether the new predictor adds information to nested models; use a likelihood-ratio test because NRI and IDI tests can inflate type I error.
- The outcome is time-to-event with censoring; use survival-specific NRI.
Assumptions and requirements
- Binary outcome observed for everyone.
- Both risks are calibrated reasonably well.
- For categorical NRI, thresholds are set before seeing the results.
What the analysis reports
- Continuous or categorical NRI with event and non-event components
- IDI, relative IDI and discrimination slopes
- AUC of both models and delta AUC
- Pencina z tests and bootstrap percentile CIs
How to report it
Adding [predictors] gave a net reclassification improvement of [value], 95% CI [[value], [value]] (events [value], non-events [value]) and an IDI of [value] (p = [value]); the AUC changed from [value] to [value].
Related methods
- ROC Curve and AUC
- Decision Curve Analysis (DCA)
- Binary Logistic Regression
Run Net Reclassification Improvement and IDI on Quorant
- Data you need
- a table (CSV or Excel) with an outcome variable and predictor (feature) columns
- Price
- 600 credits for one analysis up to 10,000 rows (complexity level 4); larger or wider data costs more. Optional methodologist review: 2400 credits.
- Checking
- Checked by property tests in every deployment
- Before you pay
- A free pre-analysis diagnosis checks your data and gives an exact quote; nothing is deducted until you approve it.