Assessment of Tree-Based Statistical Learning to Estimate Optimal Personalized Treatment Decision Rules for Traumatic Finger Amputations

Abstract

Importance: Optimal treatment for traumatic finger amputation is unknown to date.

Objective: To use statistical learning methods to estimate evidence-based treatment assignment rules to enhance long-term functional and patient-reported outcomes in patients after traumatic amputation of fingers distal to the metacarpophalangeal joint.

Design, setting, and participants: This decision analytical model used data from a retrospective cohort study of 338 consenting adult patients who underwent revision amputation or replantation at 19 centers in the United States and Asia from August 1, 2016, to April 12, 2018. Of those, data on 185 patients were included in the primary analysis.

Exposures: Treatment with revision amputation or replantation.

Main outcomes and measures: Outcome measures were hand strength, dexterity, hand-related quality of life, and pain. A tree-based statistical learning method was used to derive clinical decision rules for treatment of traumatic finger amputation.

Results: Among 185 study participants (mean [SD] age, 45 [16] years; 156 [84%] male), the median number of fingers amputated per patient was 1 (range, 1-5); 115 amputations (62%) were distal to the proximal interphalangeal joint, and 110 (60%) affected the nondominant hand. On the basis of the tree-based statistical learning estimates, to maximize hand dexterity or to minimize patient-reported pain, replantation was found to be the best strategy. To maximize hand strength, revision amputation was the best strategy for patients with a single-finger amputation but replantation was preferred for all other injury patterns. To maximize patient-reported quality of life, revision amputation was the best approach for patients with dominant hand injuries, and replantation was the best strategy for patients with nondominant hand injuries.

Conclusions and relevance: The findings suggest that the approach to treating traumatic finger amputations varies based on the patient's injury characteristics and functional needs.

Figure 1.. Screening and Enrollment for the FRANCHISE (Finger Replantation and Amputation Challenges in Assessing Impairment, Satisfaction, and Effectiveness) Study

MCP indicates metacarpophalangeal.

Figure 2.. Propensity Score Diagnostics Including Distribution and Covariate Balance

A, Distributions of the estimated propensity scores for each patient are plotted by the patient’s actual treatment assignment. B, Covariate balance across patient demographic and injury characteristics was assessed before and after propensity score adjustment using absolute mean differences. A value of 0.1 or less is generally considered to be adequately balanced. We observed balance across most covariates, although adjusted absolute mean differences for number of fingers amputated, age, US site, race/ethnicity, and work-related injury demonstrated poor covariate balance. There was a high degree of pairwise correlation between the variables US site, race/ethnicity, and work-related injury (all P < .001).