Scholarlink

Background <p>Limited data exist on predictive models incorporating patient-reported and claims-based measures to identify older adults at risk for opioid use disorder (OUD) or opioid overdose (OD).</p> Objectives <p>To develop a predictive model and identify predictors of time to first OUD or OD for older adults.</p> Design <p>Prognostic study using data from Health and Retirement Study (HRS) participants with linked Medicare claims between January 1, 2006, and December 31, 2021.</p> Participants <p>Older (≥ 65&#xa0;years) HRS-Medicare participants with chronic pain and prescribed opioids within 1&#xa0;year before their first biennial HRS survey.</p> Measurements <p>Forty potential predictors derived from Medicare claims data and HRS surveys.</p> Main measures <p>Incident diagnosis of OUD or OD was ascertained from Medicare claims data. Four survival models—traditional Cox, Cox with backward variable selection, LASSO–penalized Cox, and survival random forest—were used to account for time-fixed and time-varying predictors to predict time to first OUD or OD.</p> Key Results <p>Of 4190 older adults, 181 experienced incident OUD or OD during a mean (SD) follow-up of 6.1 (4.0) years. All 4 survival models performed equally, with mean C statistics from 0.753 (0.043) to 0.849 (0.018) and from 0.723 (0.054) to 0.790 (0.042) in training and testing sets, respectively, in predicting time to first OUD or OD during follow-up. Across all models, the top leading predictors of OUD or OD for older adults were duration of opioid use, uncontrolled pain, and use of other central nervous system medications.</p> Conclusion <p>In this prognostic study, traditional Cox and machine learning predictive models developed using patient-reported and claims-based measures performed equally well in predicting time to first OUD or OD and identified predictors for older adults. These models may be useful to monitor and identify older adults at risk for OUD or OD for early intervention.</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

Predictive Models for Time to First Opioid Use Disorder or Opioid Overdose Among Older Adults

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月16日

查看全文
  • Chien-Wei Chiang,
  • Guy Brock,
  • Siegfried Schmidt,
  • Roger B. Fillingim,
  • Stephan Schmidt,
  • Yu-Jung Jenny Wei

摘要

Background

Limited data exist on predictive models incorporating patient-reported and claims-based measures to identify older adults at risk for opioid use disorder (OUD) or opioid overdose (OD).

Objectives

To develop a predictive model and identify predictors of time to first OUD or OD for older adults.

Design

Prognostic study using data from Health and Retirement Study (HRS) participants with linked Medicare claims between January 1, 2006, and December 31, 2021.

Participants

Older (≥ 65 years) HRS-Medicare participants with chronic pain and prescribed opioids within 1 year before their first biennial HRS survey.

Measurements

Forty potential predictors derived from Medicare claims data and HRS surveys.

Main measures

Incident diagnosis of OUD or OD was ascertained from Medicare claims data. Four survival models—traditional Cox, Cox with backward variable selection, LASSO–penalized Cox, and survival random forest—were used to account for time-fixed and time-varying predictors to predict time to first OUD or OD.

Key Results

Of 4190 older adults, 181 experienced incident OUD or OD during a mean (SD) follow-up of 6.1 (4.0) years. All 4 survival models performed equally, with mean C statistics from 0.753 (0.043) to 0.849 (0.018) and from 0.723 (0.054) to 0.790 (0.042) in training and testing sets, respectively, in predicting time to first OUD or OD during follow-up. Across all models, the top leading predictors of OUD or OD for older adults were duration of opioid use, uncontrolled pain, and use of other central nervous system medications.

Conclusion

In this prognostic study, traditional Cox and machine learning predictive models developed using patient-reported and claims-based measures performed equally well in predicting time to first OUD or OD and identified predictors for older adults. These models may be useful to monitor and identify older adults at risk for OUD or OD for early intervention.