BACKGROUND AND OBJECTIVE
It is still lacking real-world evidence based on large-scale studies to substantiate and explain the survival benefit of multidisciplinary care and delineate how different specialties contribute to that benefit. This study is to leverage causal inference techniques with machine learning to investigate the utilization and timeliness of multiple providers in the treatment of patients with ALS and its impact on survival outcome under the real-world clinical setting.
METHODS
We integrated multi-year Medicare claims with multi-site EHRs and extracted a longitudinal dataset of 4,717 confirmed ALS patients following CDC algorithm combining diagnosis codes, drug use, Medicare entitlement and neurologist visits. We adopted marginal structural modeling technique with extreme gradient boosting (MSM-XGB) to systematically examine time-varying causal effects (every 6) of 20 specialties on overall survival. We extensively included an adjustment set of 520 potential confounders spanning demographic (6), socio-economic (16), and clinical (56 labs, 255 disease phenotypes, 187 drug ingredients) domains.
RESULTS
The median (IQR) age of the study cohort is 64 [54, 72], with 55% males, 13% non-white, and 5% Hispanic. Our model showed that the involvement of neurologist, physical therapist, primary care physician, ophthalmologist, psychiatrist could significantly reduce hazard ratio persistently over time but with diminishing effect after reaching optimal point of 0.80 [0.79 – 0.81] (+4m, i.e., 4 months since index date), 0.86 [0.85 – 0.87] (+10m), 0.87 [0.86 – 0.88] (+5m), 0.90 [0.89 – 0.91] (+5m), 0.97 [0.96 – 0.98] (+12m), respectively. The involvement of respiratory therapist showed worsening hazard in first 6 month but reaching minimal hazard ratio of 0.89 [0.88 – 0.90] at +16m.
CONCLUSIONS
Our work has shown promising results in using MSM-XGB to quantify causal effects of multidisciplinary involvement on survival in the care of ALS patients. Future work will involve reducing temporal window to every 3m and evaluate cumulative causal effects of multiple disciplines.