Skip to content
Omnikalc

Net reclassification improvement (NRI) calculator

Enter how many patients with and without the event moved up or down a risk category when a marker was added to get the NRI for events, for non-events, the total NRI, its z score, p-value and confidence interval.

Updated on

Educational tool, not medical advice: this page does not diagnose and does not replace a doctor, nurse or pharmacist. It has not been reviewed by a healthcare professional. Editorial and medical policy

Your inputs

Value out of range

Value out of range

Value out of range

Value out of range

Value out of range

Value out of range

Results

Net reclassification improvement (NRI)

0.0875

z score

2.248

Two-sided p-value

0.0246

NRI for events
0.0500
NRI for non-events
0.0375
Standard error
0.0389
Lower limit of the 95 % interval
0.0112
Upper limit of the 95 % interval
0.1638
Share of all patients reclassified
22.0 %

The net reclassification improvement (NRI) calculator gives the NRI for events, the NRI for non-events and the total NRI when a marker is added to a risk model, with the standard error, z score, p-value and 95 percent interval.

How the calculation works

  • Enter the number of patients with the event and how many moved up or down a risk category.
  • Enter the number of patients without the event and how many moved up or down.
  • Read the NRI, its components and the significance.

Example

Among 200 events, 30 moved up and 20 moved down, an NRI of 0.0500. Among 800 non-events, 70 moved up and 100 moved down, an NRI of 0.0375. The total NRI is 0.0875, with a standard error of 0.0389, a z score of 2.248, a p-value of 0.0246 and an interval from 0.0112 to 0.1638. 22.0 % of all patients were reclassified.

Why it matters

The calculator helps clinical researchers, epidemiologists and students who evaluate whether a new biomarker, score or imaging test improves a risk prediction model, and who need to report the reclassification table in a paper or a thesis.

Good to know

Count the moves from the cross table of the old and the new risk categories, separately for events and non-events. Use categories with clinical meaning, such as low, intermediate and high risk, and avoid choosing the cut points after seeing the data. Report the NRI with its components, the reclassification tables, a measure of discrimination such as the AUC and the calibration of both models. A significant NRI is not enough to justify a new marker without clinical benefit.

The confusion matrix calculator summarizes classification, the McNemar test calculator compares paired classifiers, the Matthews correlation coefficient calculator scores a classifier and the Brier score decomposition calculator checks calibration.

Frequently asked questions

What is the net reclassification improvement?

The NRI measures how much a new risk model or marker improves the classification of patients into risk categories compared with an old one. Events moving up and non-events moving down are correct reclassifications, and the opposite moves are errors. NRI = (up − down) among events / events + (down − up) among non-events / non-events.

How is the result read?

The NRI ranges from −2 to +2, as it is the sum of two proportions from −1 to +1. A positive NRI means that the new model classifies better overall. It is useful to look at the two components, because an improvement in events can be offset by a loss in non-events. A value of 0.10 to 0.20 is often seen as a meaningful gain.

How is the p-value found?

Under the null of no improvement, the NRI divided by its standard error follows approximately a standard normal distribution. The standard error is √((up + down)events / n_events² + (up + down)nonevents / n_nonevents²). The calculator gives the two-sided p-value and a 95 percent interval.

What are the cautions?

The NRI depends on the risk categories chosen and on how many patients are near a cut point. It can look significant when the new model is poorly calibrated, and the test is criticized for inflated type I errors. Pair the NRI with a discrimination measure such as the AUC and a calibration check before drawing a conclusion.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.