Creating diagnostic scores using data-adaptive regression: An application to prediction of 30-day mortality among stroke victims in a rural hospital in India

Figures & data

Figure 1 Receiver operating characteristic (ROC) curves on the full data for the 4 scores.

Abbreviations: CNS, Canadian Neurological Score; NIH, National Institute of Health.

Table 1 Estimated area under the ROC and inference for the four scoring methods. The standard error was estimated using method proposed by DeLong and colleagues (1988)

Scoring system	Area under ROC	Standard error	95% Confidence interval
POLYCLASS	0.93	0.01	0.90, 0.97
NIHSS Score	0.89	0.02	0.84, 0.94
CNS Score	0.69	0.04	0.61, 0.77
GHS Score	0.84	0.03	0.78, 0.90

Abbreviations: CNS, Canadian Neurological Score; GHS, Guy’s Prognostic Score; NIHSS, National Institute of Health Stroke Scale; ROC, receiver operating characteristic curve.

Table 2 Results from cross-validation showing the specifi city for fi xed sensitivities that are at least 90% on the training data sets for the 3 scores and POLYCLASS. The resulting sensitivities and specifi cities are shown for the validation data sets (those data not used for determining the cut-offs)

Method	Mortality status at 30 days	Predicted dead	Predicted alive	Type of accuracy	Performance(%)
POLYCLASS	Dead	46	5	Sensitivity	90.2
	Alive	30	94	Specificity	75.8
NIHSS Score	Dead	47	4	Sensitivity	92.2
	Alive	44	80	Specificity	64.6
CNS Score	Dead	45	6	Sensitivity	88.2
	Alive	70	54	Specificity	43.6
GHS Score	Dead	47	4	Sensitivity	92.2
	Alive	57	67	Specificity	54.0

Abbreviations: CNS, Canadian Neurological Score; GHS, Guy’s Prognostic Score; NIHSS, National Institute of Health Stroke Scale.

DelongERDelongDMClarke-PearsonDLComparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approachBiometrics198844837453203132

 PubMed Web of Science ®Google Scholar