Skip to main content
. 2022 Jan 13;37(15):3877–3884. doi: 10.1007/s11606-021-07377-1

Table 3.

Observed vs Predicted Mortality in the SMR-30 Validation Cohort Using 10 Equally Size Bins Defined by Decile of Predicted Risk

Risk decile Predicted risk (lowest, highest) Hospitalizations, N Observed deaths, N (%) Predicted deaths, N (%) Difference*, N (%)
1 (0.000–0.002) 37,379 62 (0.17) 47 (0.13) 15 (0.04)
2 (0.002–0.004) 37,379 94 (0.25) 118 (0.32) −24 (−0.06)
3 (0.004–0.007) 37,379 154 (0.41) 211 (0.56) −57 (−0.15)
4 (0.007–0.011) 37,379 252 (0.67) 332 (0.89) −80 (−0.21)
5 (0.011–0.016) 37,379 384 (1.03) 495 (1.33) −111 (−0.30)
6 (0.016–0.023) 37,380 541 (1.45) 725 (1.94) −184 (−0.49)
7 (0.023–0.035) 37,379 926 (2.48) 1077 (2.88) −151 (−0.40)
8 (0.035–0.057) 37,379 1572 (4.21) 1686 (4.51) −114 (−0.30)
9 (0.057–0.114) 37,379 2951 (7.89) 2991 (8.00) −40 (−0.11)
10 (0.114–0.987) 37,379 9100 (24.35) 9775 (26.15) −675 (−1.81)
1–10 (0.000–0.987) 373,791 16036 (4.29) 17458 (4.67) −1422 (−0.38)

*Differences reflect the observed minus predicted mortality. Negative values indicate that the model over-predicted mortality, while positive value indicate that the model under-predicted mortality

The expected (mean) absolute calibration error14 across risk decile was 0.38%, while the maximum calibration error14 (observed in the highest risk decile) was 1.81%