Assessing the ability of an instrumental variable causal forest algorithm to personalize treatment evidence using observational data: the case of early surgery for shoulder fracture

Table 2 Distribution of Instrumental Variable Causal Forest Algorithm Early Surgery Absolute Effects on the Detriment Outcome^a for Medicare Patients with Proximal Humerus Fractures in 2011 by Number of Trees in IV-CFA Forest and Minimum Leaf Node Population Size in Each Tree

Trees in IV-CFA Forest	Minimum Leaf Node Population Size	Mean	St Dev	Percent of Patients with Positive Effect	Min	10^th	25^th	50^th Median	75^th	90^th	Max
3000	50	.139	.265	72%	-1.369	-.186	-.021	.143	.304	.468	1.107
	100	.137	.172	77%	-.376	-.086	.013	.138	.254	.354	.679
	200	.137	.115	87%	-.207	-.014	.048	.146	.218	.288	.457
	300	.136	.093	92%	-.158	.012	.067	.141	.202	.255	.383
	400	.137	.077	96%	-.105	.032	.082	.143	.191	.239	.351
4000	50	.135	.264	72%	-1.254	-.188	-.020	.136	.297	.462	1.090
	100	.136	.171	77%	-.357	-.086	.015	.136	.251	.350	.703
	200	.135	.115	86%	-.200	-.017	.047	.142	.219	.283	.430
	300	.136	.093	92%	-.145	.009	.067	.142	.203	.256	.373
	400	.135	.078	95%	-.106	.031	.081	.140	.190	.236	.338
5000	50	.136	.263	71%	-1.209	-.189	-.023	.139	.300	.462	1.152
	100	.137	.171	77%	-.381	-.081	.013	.136	.251	.350	.697
	200	.138	.114	87%	-.194	-.014	.048	.144	.218	.286	.435
	300	.136	.091	93%	-.142	.012	.069	.142	.202	.254	.384
	400	.136	.078	95%	-.089	.031	.080	.141	.191	.237	.350

^a1 the patient died or had an adverse event during the period 61–365 days following the index PHF, 0 otherwise

ISSN: 1471-2288