Abstract
OBJECTIVE: To identify multivariable subgroups of patients with differential responses to a nurse-delivered care transition intervention after an emergency department (ED) visit in a randomized controlled trial (RCT) using an emerging data-driven method. DESIGN: Secondary analysis of RCT. PARTICIPANTS: 512 individuals enrolled in an RCT of a nurse-delivered care transition intervention after an ED visit. All 512 participants were included in a pre-specified subgroup analysis, and 451 of these had sufficient complete case data to be included in a model-based recursive (MoB) partitioning analysis. METHODS: The primary outcome was having at least one ED visit in 30 days after the index ED visit. Two analytical methods explored heterogeneity of treatment effects: data driven model-based recursive partitioning (MoB) using 37 candidate baseline variables, and a contextual point of comparison with prespecified subgroups defined by ED super-user status (≥ 3 ED visits in previous 6 months or not), sex (male/female), and age, individually examined via treatment arm by subgroup interaction terms in logistic regression models. Internal validation of the MoB analysis via bootstrap resampling with an optimism corrected c-statistic was conducted to provide a bias-corrected estimate. RESULTS: MoB detected treatment effect heterogeneity in a single subgroup, marital status. Unmarried patients randomized to the intervention had a repeat ED use rate of 22% compared to 34% in the usual care group; married patients randomized to the intervention had a 27% ED return rate compared to 12% in the usual care group. Internal validation demonstrated an optimism corrected c-statistic of 0.54. No treatment-by-covariate subgroup interactions were identified among the 3 prespecified subgroups. CONCLUSION: Although exploratory, the results of the MoB analysis suggest that patient factors related to social relationships such as marital status may be important contributors to differential response to a care transition intervention after an ED visit. These were characteristics that the investigators had not anticipated or planned to examine in the individual prespecified subgroup analysis. Data-driven methods can yield unexpected findings and contribute to a more complete understanding of differential treatment effects in subgroup analysis, which can inform further work on development of effective care transition interventions in the ED setting.