Scholarlink

<p>The goal of this study is to develop and validate a machine learning approach for predicting opioid overdose risk and identifying associated risk factors among Alabama Medicaid beneficiaries during the period of 2016–2023. Using the administrative claims data, we trained three machine learning models, penalized logistic regression, random forest and gradient boosting machines, on the 2016–2018 data with 168,625 records, and test and validate the models on the 2019–2023 data with 229,212 records. A modern sampling approach, SMOTE, was incorporated into the study to deal with the imbalance in the overdose outcome. The machine learning models demonstrated strong performance assessed by C-statistics, sensitivity and specificity, precision and recall rates, etc. Several risk factors were identified including changes in prescription patterns, beneficiary’s age, and prescription denials. Differences in ROC-AUC and PR-AUC between models with and without SMOTE were modest; interpretation therefore emphasizes improvements in recall rather than overall discrimination. SMOTE-based performance metrics were averaged across 50 resampling iterations to reduce instability from single-run oversampling.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning approach for opioid overdose risk prediction among Alabama Medicaid beneficiaries with opioid prescriptions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月13日

查看全文
  • Jason Parton,
  • Qin Wang,
  • Eric C. Wang,
  • Courtney Hanson,
  • John A. Higginbotham

摘要

The goal of this study is to develop and validate a machine learning approach for predicting opioid overdose risk and identifying associated risk factors among Alabama Medicaid beneficiaries during the period of 2016–2023. Using the administrative claims data, we trained three machine learning models, penalized logistic regression, random forest and gradient boosting machines, on the 2016–2018 data with 168,625 records, and test and validate the models on the 2019–2023 data with 229,212 records. A modern sampling approach, SMOTE, was incorporated into the study to deal with the imbalance in the overdose outcome. The machine learning models demonstrated strong performance assessed by C-statistics, sensitivity and specificity, precision and recall rates, etc. Several risk factors were identified including changes in prescription patterns, beneficiary’s age, and prescription denials. Differences in ROC-AUC and PR-AUC between models with and without SMOTE were modest; interpretation therefore emphasizes improvements in recall rather than overall discrimination. SMOTE-based performance metrics were averaged across 50 resampling iterations to reduce instability from single-run oversampling.