Highlight
This prognostic study utilized both patient-reported and Medicare claims data from the Health and Retirement Study to develop predictive models for time to first opioid use disorder (OUD) or opioid overdose (OD) in older adults. Four survival analysis methods—traditional Cox proportional hazards, backward-selected Cox, LASSO-penalized Cox, and survival random forest—performed similarly, with moderate to good discrimination. Key predictors identified consistently included longer duration of opioid use, uncontrolled pain, and use of concomitant central nervous system (CNS) medications.
Study Background
The opioid epidemic continues to be a critical public health challenge, and older adults constitute a particularly vulnerable population due to frequent chronic pain, polypharmacy, and physiological changes affecting opioid metabolism and sensitivity. Despite increasing opioid prescriptions in patients aged 65 and older, data on effective predictive models for opioid-related adverse outcomes in this demographic are limited. Early identification of older adults at risk for opioid use disorder or overdose is essential to implement targeted interventions and optimize pain management while minimizing harm.
Study Design
This prognostic study leveraged longitudinal data from the Health and Retirement Study (HRS), linked with Medicare claims from 2006 to 2021. Participants included community-dwelling adults aged 65 and above diagnosed with chronic pain and prescribed opioids within one year before their first biennial HRS survey interview during the study period. The analysis incorporated 40 candidate predictor variables drawn from both claims-based data (e.g., medication use, comorbidities) and patient-reported measures (e.g., pain control, functional status).
Incident opioid use disorder or overdose was identified through Medicare claim diagnostic codes. Four survival analysis models were constructed to predict time to first OUD or OD: traditional Cox proportional hazards, Cox with backward variable selection, Cox penalized by least absolute shrinkage and selection operator (LASSO), and survival random forest models. Models accounted for time-fixed and time-varying predictors. Performance was assessed in both training and testing datasets using the concordance (C) statistic for discrimination.
Key Findings
Among 4190 older adult participants with chronic pain on opioids, 181 (4.3%) developed incident OUD or OD over a mean follow-up of 6.1 years (SD 4.0).
All four models demonstrated comparable performance. In training datasets, C statistics ranged from 0.753 to 0.849, and in testing datasets, from 0.723 to 0.790, indicating acceptable to good predictive discrimination.
Across models, the leading predictors of increased risk for OUD or OD were:
- Duration of opioid use: Longer cumulative opioid exposure was strongly associated with higher risk.
- Uncontrolled pain: Self-reported indications of insufficient pain management independently predicted adverse opioid outcomes.
- Concomitant CNS medications: Use of other central nervous system active drugs such as benzodiazepines or muscle relaxants increased risk.
Other variables of interest included demographic factors, comorbid conditions, and health behaviors, but their predictive contributions were less consistent or weaker.
Expert Commentary
This study fills a critical gap by integrating patient-reported outcomes with robust claims data to enhance risk stratification for opioid-related adverse events in older adults. The comparable performance of traditional regression and machine learning models suggests that simpler Cox-based approaches may suffice for clinical implementation, balancing interpretability and predictive power.
The finding that duration of opioid use and uncontrolled pain are major risk factors aligns with existing literature underscoring chronic opioid exposure and inadequate analgesia as central to OUD development. The impact of CNS polydrug use highlights the importance of careful medication reconciliation and deprescription strategies in this vulnerable group.
Limitations include reliance on administrative claims for outcome ascertainment, which may under-capture OUD cases, and potential residual confounding inherent to observational data. The cohort primarily represents community-dwelling US Medicare beneficiaries, which may limit generalizability to other populations or healthcare settings.
Future research should explore integrating these predictive models into electronic health records for real-time risk monitoring and assess the effectiveness of targeted interventions triggered by model outputs.
Conclusion
Predictive modeling combining patient-reported measures and claims data can effectively identify older adults at elevated risk for opioid use disorder or overdose. Recognizing extended opioid use duration, uncontrolled pain, and concurrent CNS medication as key modifiable risk factors offers strategic targets for intervention. These models hold promise for guiding clinicians and health systems in proactive risk mitigation to improve safety in opioid prescribing among older adults.
Funding and Disclosures
The study was supported by relevant academic and governmental funding agencies as referenced in the original publication. No conflicts of interest are noted.
References
- Chiang CW, Brock G, Schmidt S, et al. Predictive Models for Time to First Opioid Use Disorder or Opioid Overdose Among Older Adults. J Gen Intern Med. 2026 Jul 16. PMID: 42463630.
- Bauer AM, et al. Challenges and considerations when screening for opioid misuse in older adults. Subst Abus. 2020;41(4):479-487.
- Krebs EE, et al. Opioid prescribing for older adults with chronic noncancer pain: a systematic review. J Am Geriatr Soc. 2014;62(9):1815-1823.
- Dowell D, et al. CDC Guideline for Prescribing Opioids for Chronic Pain — United States, 2016. MMWR Recomm Rep. 2016;65(1):1-49.
