Clinical decision making under uncertainty: a bootstrapped counterfactual inference approach

Table 4 Policy Learning: Rewards (\(\mu \pm \sigma\)) of policies learned using self-normalized inverse propensity scoring (SNIPS) formulation (10 simulations). Optdigits and Letter are two multiclass classification datasets from the UCI repository [30]. LR=Logistic Regression. NN=Neural Network

Dataset	Expert Policy/ Logging Policy	\(\hat{h}_0\) - NN
		Single NN	NN Ensemble		Adversarial
		SNIPS	SNIPS\(_{inv}\)	SNIPS\(_{avg}\)	SNIPS
UCI	OPTDIGITS (10 actions)	0.785 ± 0.113	0.800 ± 0.143	0.805 ± 0.072	0.767 ± 0.093
UCI	LETTER (26 actions)	0.167 ± 0.055	0.139 ± 0.031	0.121 ± 0.047	0.157 ± 0.041
Warfarin	LR (3 actions)	0.602 ± 0.024	0.608 ± 0.024	0.600 ± 0.019	0.597 ± 0.024
	LR (5 actions)	0.522 ± 0.022	0.525 ± 0.021	0.527 ± 0.024	0.532 ± 0.019
	PHARMA (3 actions)	0.657 ± 0.015	0.646 ± 0.019	0.652 ± 0.020	0.672 ± 0.017
	PHARMA (5 actions)	0.602 ± 0.016	0.588 ± 0.018	0.600 ± 0.018	0.628 ± 0.008
Heparin	Clinician (unknown)	0.321 ± 0.030	0.337 ± 0.046	0.336 ± 0.054	0.325 ± 0.039

ISSN: 1472-6947