Dynamic risk prediction for diabetes using biomarker change measurements

Table 4 Net reclassification improvement^a

Placebo
	Percentage of individuals for whom the dynamic landmark model estimates a higher risk than the static model	Percentage of individuals for whom dynamic landmark model estimates a lower risk than the static model	Overall Net reclassification improvement (95% Confidence Interval)
1 year
Events	26.5%	73.5%	−3.8% (−26.0, 18.4%)^b
Non-events	28.4%	71.6%	−3.8% (−26.0, 18.4%)^b
2 years
Events	4.3%	95.7%	3.5% (−10.4, 17.3%)
Non-events	2.6%	97.4%	3.5% (−10.4, 17.3%)
3 years
Events	1.4%	98.6%	1.9% (−7.3, 11.0%)
Non-events	0.4%	99.6%	1.9% (−7.3, 11.0%)
Metformin
	Percentage of individuals for whom the dynamic landmark model estimates a higher risk than the static model	Percentage of individuals for whom dynamic landmark model estimates a lower risk than the static model	Overall Net reclassification improvement (95% Confidence Interval)
1 year
Events	40.4%	59.6%	4.6% (−15.8, 24.9%)
Non-events	38.1%	61.9%	4.6% (−15.8, 24.9%)
2 years
Events	19.9%	80.1%	18.6% (−5.1, 42.4%)
Non-events	10.6%	89.4%	18.6% (−5.1, 42.4%)
3 years
Events	5.0%	95.0%	7.0% (−12.9, 26.9%)
Non-events	1.5%	98.5%	7.0% (−12.9, 26.9%)

^a Bolding indicates correct risk movement by the dynamic landmark model e.g. individuals who have an event should be given a higher risk
^b This calculation is based on: (26.5–73.5) – (28.4–71.6) = −3.8

ISSN: 1471-2288