Table 3.

Evaluation of clinical models for predicting hospitalisation-associated functional decline

Model evaluation	Prediction models
	Developed model	Inouye et al. 1993	Mehta, et al. 2011b	Sager et al. 1996
OR (95% CI)	1.12 (1.08–1.17)	2.44 (1.56–3.81)	1.26 (1.12–1.41)	1.62 (1.24–2.12)
Discrimination a	0.75 (0.68–0.83)	0.67 (0.59–0.74)	0.68 (0.61–0.76)	0.65 (0.57–0.73)
Calibration	Chi2 = 6.35, p = 0.499 c	Chi2 = 0.11, p = 0.948 d	Chi2 = 11.80, p = 0.0667 e	Chi2 = 11.80, p = 0.0667 f
Clinical usefulness	Cutoff value = 13 Sensitivity = 71% Specificity = 70% PPV = 54% NPV = 83% Correctly classified = 70%	Cutoff value = 1 Sensitivity = 46.0% Specificity = 80.2% PPV = 54% NPV = 75% Correctly classified = 69%	Cutoff value = 4 Sensitivity = 65% Specificity = 60% PPV = 45% NPV = 78% Correctly classified = 62%	Cutoff value = 4 Sensitivity = 60% Specificity = 61% PPV = 44% NPV = 76% Correctly classified = 61%

Abbreviations: PPV Positive Predictive value, NPV Negative Predictive Value, CI Confidence Interval;

^a Discrimination was assessed using the C-index, and models were compared using the chi² test for equality for two or more Receiver Operating Characteristic areas: chi² = 12.8, p = 0.005;

^b Sensitivity analysis for Mehta et al. 2011 using a complete case analysis instead of multiple imputation: OR = 1.18 (1.03–1.36); Discrimination: 0.63, 95% CI (0.54–0.73); Calibration: chi² = 4.27, p = 0.640;

Calibration was assessed using the Hosmer – Lemeshow goodness of fit test. The goodness of fit could only be assessed in quantiles of ^c 9 groups, ^d 4 groups, ^e 8 groups and ^f 5 groups because of ties in the data;