Table 3—

Sensitivity and predictive values of various targeting strategies, validation data set

ComplicationProportion of population targetedSensitivity for 1996 eventsPositive predictive value
Macro- and microvascular
 Top 30%* from the “best” model307216.4
 Prior events or diagnoses317215.8
 Top 30%* from three-variable model307116.1
 Top 30%* from numerical score337415.3
 Top 30% of 1994–1995 HbA1c levels30317.1
Infectious disease
 Top 30%* from the “best” model30726.4
 Prior events or diagnoses15447.5
 Top 30%* from three-variable model30686.1
 Top 30%* from numerical score30676.0
 Top 30% of 1994–1995 HbA1c levels30383.4
Metabolic
 Top 30%* from the “best” model30831.2
 Prior events or diagnoses1.5338.5
 Top 30%* from three-variable model30751.0
 Top 30%* from numerical score29821.3
 Top 30% of 1994–1995 HbA1c levels30590.9
  • Data are %.

  • *

    * Patients with the highest 30% of predicted risk scores in the validation data set. For the numerical risk score, the proportion selected may deviate slightly from 30% because its seven observed values did not allow categorization by deciles.