Skip to main content

Table 2 Performance of risk prediction models for incident type 2 diabetes among 65,684 participants in the risk prediction population

From: Predictive value of circulating NMR metabolic biomarkers for type 2 diabetes risk in the UK Biobank study

Performance metric

Concise modela

Concise modela plus metabolic biomarkersb

Full modelc

Full modelc plus metabolic biomarkersb

C-statistic (CI)d

0.802 (0.791, 0.812)

0.830 (0.822, 0.841)

0.829 (0.819, 0.838)

0.837 (0.831, 0.848)

Metrics of relative performance

 χ2 e,f

453 (p < 0.0001)

177 (p < 0.0001)

 %increase χ2

17

6

 Absolute IDI f,g

1.5 (1.0, 1.9)

0.7 (0.4, 1.1)

 Relative IDI (%) (CI) f,g

15.0 (10.5, 20.4)

6.3 (4.1, 9.8)

 Continuous NRI (CI)f,h

 

  Events

0.15 (0.12, 0.20)

0.10 (0.06, 0.14)

  Non-events

0.28 (0.26, 0.31)

0.12 (0.09, 0.14)

  Overall

0.44 (0.38, 0.49)

0.22 (0.17, 0.28)

  1. DF degrees of freedom, IDI integrated discrimination improvement, NRI net reclassification improvement; T2D type 2 diabetes
  2. aConcise model: age, sex, parental history of diabetes, body mass index and HbA1c
  3. bMetabolic biomarkers comprise the first 11 metabolic biomarker principal components
  4. cFull model: concise model plus waist circumference, blood pressure, triglycerides and HDL cholesterol
  5. dThe c-statistic measures the ability of a model to rank participants from low to high risk. Given two randomly selected individuals, one who develops T2D and one who does not, the c-statistic is the probability that the model will give a higher predicted risk for the individual who develops T2D. An uninformative model will have a c-statistic of 0.5 and a model that discriminates perfectly will have a c-statistic of 1.0
  6. e11 DF
  7. fBias-corrected estimates and confidence intervals were derived using 200 bootstrap samples
  8. gThe IDI quantifies the difference between two models in their ability to predict risk. It is calculated as the difference between the two models in the mean predicted T2D risk among those who did develop T2D minus the mean predicted risk of T2D in those who did not develop T2D (i.e. it is the difference between two differences). When metabolic biomarkers were added to the concise model, the separation in the mean predicted T2D risk between those who did develop T2D, compared with those who did not develop T2D, increased in relative terms by 15.0%. Positive IDI values indicate improved T2D risk classification following the addition of metabolic biomarkers to the risk prediction model
  9. hThe continuous NRI quantifies the appropriateness of the change in predicted probabilities of T2D between two models. The ‘events’ NRI is calculated among those who developed T2D, and the ‘non-events’ NRI is calculated among those who did not develop T2D. Both statistics are calculated as the probability of an ‘appropriate’ change in predicted risk (after the addition of metabolic biomarkers to the model) minus the probability of an ‘inappropriate’ change in predicted risk. For those who developed T2D, an appropriate change would be a higher predicted T2D risk after the addition of metabolic biomarkers to the model. An inappropriate change would be a lower predicted T2D risk after the addition of metabolic biomarkers to the model. When metabolic biomarkers were added to the concise model, among those who developed T2D, 15% more were assigned a higher predicted T2D risk than were assigned a lower predicted risk. The overall NRI is the sum of the ‘events’ and ‘non-events’ NRI statistics. Positive NRI values indicate that the addition of metabolic biomarkers results in a superior model
Back to article page

BMC Medicine

ISSN: 1741-7015

Contact us