Skip to main content

Table 2 Statistical performance of Alternative Predictive Models using U.S./Spanish weights.

From: Applying diagnosis and pharmacy-based risk models to predict pharmacy use in Aragon, Spain: The impact of a local calibration

  Alternative Predictive Models
  Dx-PM Rx-PM DxRx-PM
  U.S. Weights Spanish Weights U.S. Weights Spanish Weights U.S. Weights Spanish Weights
Variance explained. R 2 18.9% 29.4% 22.2% 40.6% 23.5% 42.6%
Area Under ROC Curve* .868 .902 .900 .941 .903 .949
Sensitivity* 30.6% 39.4% 27.5% 52.3% 31.2% 53.2%
Specificity* 96.3% 96.8% 96.2% 97.5% 96.4% 97.5%
Mean pharmacy expenditure Year-2 (€)*       
   True positives 3,059 3,076 3,244 3,236 3,248 3,244
   True negatives 233 233 236 234 235 234
  1. *Outcomes refer to top 5% Year-2 pharmacy cost group. The Area Under ROC Curve ranges from 0.5 (model no better than the flip of a coin) to 1.0 (perfect true positive and true negative classification).
  2. Dx: physician assigned diagnosis. Rx: pharmacy prescriptions filled by physicians. PM: predictive model.