Do functional status and Medicare claims data improve the predictive accuracy of an electronic health record mortality index? Findings from a national Veterans Affairs cohort

Table 3 Reclassification table comparing the base model with the base + function + Medicare model

	Reclassification table for individuals who did and did not die on follow-up
Individuals who died during the follow-up period	Base model	Base + function + Medicare model
	Base model	< 0.5	≥0.5	Total
	< 0.5	6492	1106	7598
	≥0.5	1038	15,650	16,688
	Total	7530	16,756	24,286
	NRI for events (death)^a		1106 – 1038 / 24,286 = 0.28%
Individuals alive at the end of follow-up	< 0.5	18,959	766	19,725
	≥0.5	1099	4502	5601
	Total	20,058	5268	25,326
	NRI for non-events (survived)^b		1099 – 766 / 25,326 = 1.31%
	Overall net reclassification improvement^c		0.0028 + 0.0131 = 0.0159
	Integrated discrimination improvement		0.0176

A risk threshold of 50% was used to calculate the net reclassification improvement for the base model compared with the base plus functional measures plus Medicare data model (base + function + Medicare model).
Abbreviations: NRI Net reclassification improvement
^aThe improvement in classification among individuals who died during follow-up is defined as the number of deaths correctly reclassified as higher risk (in boldface = 1106) minus the number of deaths incorrectly reclassified as lower risk (in italics = 1038) divided by the total number of deaths (24,286). Therefore, the NRI for events was 1106 – 1038 / 24,286 = 0.28%
^bThe improvement in classification among individuals who survived during follow-up is defined as the number of non-events correctly reclassified as lower risk (in boldface = 1099) minus the number of non-events incorrectly reclassified as higher risk (in italics = 766) divided by the total number of people who survived (25,326). Therefore, the NRI for non-events was 1099 – 766 / 25,326 = 1.31%
^cOverall NRI of 0.0159 is the sum of the net percentages of patients that were correctly reclassified (into higher or lower risk depending on whether they subsequently did or did not die) by the model that incorporated function and Medicare data (base + function + Medicare model) compared to the base model

ISSN: 1471-2318