359. Stratified Treatment Optimization under Partial Observation Scarcity across Patient Subgroups
Invited abstract in session TA-34: Advancements of OR-analytics in statistics, machine learning and data science 2, stream Advancements of OR-analytics in statistics, machine learning and data science.
Tuesday, 8:30-10:00Room: Michael Sadler LG10
Authors (first author is the speaker)
| 1. | Haiyan Yu
|
| Peking University Chongqing BigData Research Insistute |
Abstract
Stratified medicine presents a promising opportunity to improve clinical outcomes by recommending a possibly distinct treatment to each patient stratum (e.g., contextualized by age group, sex, etc.). However, this encounters the challenge of having inadequate observations from some under-represented patient subgroups (assigned to one treatment) of the stratum as compared to other subgroups in the same stratum. We call this phenomenon partial observation scarcity across patient subgroups, which magnifies the common challenge of limited observations across patient strata, relative to additional patient- level covariates. We propose a weighting-estimation-then-optimization (WETO) approach to optimize the stratified treatment rule (assignment of each stratum to some treatment) under partial observation scarcity. To address the challenge of small sample sizes in underrepresented subgroups, we introduce the concept of subgroup pattern (SP) to leverage the estimates from a few well-represented subgroups specified by the SP. To make fair comparisons between treatments while combating the heterogeneity within each patient subgroup due to additional covariates, we optimize the weights of individual patients in each well-represented subgroup through entropy balancing optimization, to reduce the confounding effects of the covariates. A case study based on real clinical records of hospitalized type-2 diabetic patients confirms that the WETO-based analytics framework with the incorporation
Keywords
- Decision Analysis
- Service Operations
- Medical Applications
Status: accepted
Back to the list of papers