Skip to main content
. 2021 Feb 11;81(4):471–482. doi: 10.1007/s40265-020-01435-4

Table 3.

Performance of RL algorithms with comparison between RL and clinicians for glycemic control, hypertension control, and CVD prevention

RL–glycemia
Encounters for which the algorithm’s recommendation differed from the observed clinician's prescription [(%)] 15,578 (13.9)
RL–glycemia Clinician's prescription p-Value
A1c [mean (SE)] 7.80 (0.01) 8.09 (0.01) < 0.001
A1c >8% [(%)] 5421 (34.8) 6617 (42.5) < 0.001
RL–BP
Encounters for which the algorithm’s recommendation differed from the observed clinician's prescription [(%)] 20,251 (17.1)
RL–BP Clinician's prescription p-Value
SBP [mean (SE)] 131.77 (0.06) 132.35 (0.11) < 0.001
SBP >140 mmHg [(%)] 3256 (16.1) 5390 (26.6) < 0.001
RL–CVD
Encounters for which the algorithm’s recommendation differed from observed clinician's prescription (N(%)) 946 (1.6)
R–CVD Clinician's prescription p-Value
FHS [mean (SE)] 13.65 (0.26) 17.18 (0.36) < 0.001
FHS >20% [(%)] 237 (25.1) 299 (31.6) < 0.001
RL–multimorbidity
Encounters for which the algorithm’s recommendation differed from the observed clinician's prescription [(%)] 102,184 (28.9)
RL–multimorbidity Clinician's prescription p-Value
A1c [mean (SE)] 7.14 (0.003) 7.19 (0.005) < 0.001
A1c >8% [(%)] 16,436 (16.08) 20,879 (20.43) < 0.001
SBP [mean (SE)] 129.40 (0.03) 129.58 (0.05) < 0.001
SBP >140 mmHg [(%)] 9800 (9.59) 20,957 (20.51) < 0.001
FHS [mean (SE)] 21.89 (0.04) 25.61 (0.05) < 0.001
FHS >20% [(%)] 48,283 (47.3) 55,957 (54.8) < 0.001

RL reinforcement learning, CVD cardiovascular disease, SE standard error, BP blood pressure, SBP systolic blood pressure