Skip to main content

Table 3 Performance Measures of the Prospective Linear and Machine Learning Models on the Test Set

From: Incorporating machine learning and social determinants of health indicators into prospective risk adjustment for health plan payments

Evaluation MetricNo SDHSDH
R2 (95% CI)a
 Linear0.327 (0.300, 0.353)0.327 (0.300, 0.354)
 ML0.388 (0.357, 0.420)0.387 (0.357, 0.419)
MAE (95% CI)b
 Linear6992 (6889, 7094)6991 (6889, 7094)
 ML6637 (6539, 6735)6634 (6536, 6732)
C-statistic (95% CI)c
 Linear0.703 (0.701, 0.705)0.700 (0.699, 0.702)
 ML0.717 (0.715, 0.718)0.716 (0.714, 0.717)
  1. Comparison of performance measures between linear regression and machine learning prospective risk adjustment models, predicting 2017 yearly top-coded spending from 2016 characteristics. The SDH model additionally includes SDH variables obtained from U.S. Census data (see Table 1)
  2. aConfidence intervals for R2 were constructed using the nonparametric bootstrap [21]
  3. bConfidence intervals for MAE were constructed using a paired t-test
  4. cConfidence intervals for C-statistic were constructed using a jackknife procedure [25]