. Author manuscript; available in PMC: 2021 Nov 16.

Published in final edited form as: Adv Neural Inf Process Syst. 2020 Dec;33:17976–17986.

Table 2:

Value function under the fitted ITRs, mean (standard deviation), evaluated with respect to four post-treatment outcomes (lower values indicate healthier states and thus better performance) from 100 cross validations on the STAR*D data.

Outcome measure	Proposed	CausalForest	TARNet	Deep-O	Deep-Q	Deep-A
HAMD	12.23 (0.38)	12.62 (0.45)^*	12.37 (0.51)^*	13.59 (0.38)^*	13.55 (0.50)^*	12.32 (0.54)
QIDS	8.33 (0.32)	8.97 (0.30)^*	8.63 (0.38^*	9.66 (0.28)^*	9.52 (0.36)^*	8.57 (0.42)^*
CGI	2.54 (0.08)	2.72 (0.08)^*	2.61 (0.11)^*	2.92 (0.08)^*	2.89 (0.12)^*	2.60 (0.12)^*
WSAS	20.27 (0.46)	19.98 (0.72)	20.05 (0.77)^*	22.16 (0.73)^*	21.99 (0.83)^*	20.14 (0.73)

An asterisk (*) indicates that our method is significantly better using pairwise t-test (p-value < 0.05).