logologo

Popular Boards

1 Mentions
Jose Benitez-Aurioles, Carlos Raul Ramirez Medina, David Jenkins | PLOS Digital Health | (2026)
Primary Link Pubmed DOI Openalex Full text (OA)

Key Takeaways

Sample Definition And Size

The study included 1,026,139 adult patients (aged ≥18 years) in the UK newly prescribed an opioid without cancer (development cohort), and externally validated the models in a cohort of 337,015 patients. The development cohort contributed 2,350,730 patient-years of follow-up, and the validation cohort contributed 781,362 patient-years. The outcome (opioid-related death) occurred in 1,226 individuals (0.12%) in the development cohort and 293 individuals (0.09%) in the validation cohort. Competing deaths accounted for 5.1% and 5.9% of deaths in the development and validation cohorts, respectively.

Study Type

Retrospective cohort study developing and evaluating three competing risk time-to-event prediction models: a Fine & Gray regression model with LASSO penalization, a competing random survival forest model, and a DeepHit deep neural network model.

Conflicts Of Interest

No conflicts of interest are declared in the available metadata (no COI statement found in the abstract or metadata).

Results Summary

During internal validation, the C-statistics (discrimination) were: regression model 84.3%, random forest 84.4%, neural network 82.1%. During external validation, C-statistics were: regression model 81.8%, random forest 81.5%, neural network 81.5%. Key predictors associated with higher risk included prior substance abuse, lung and liver comorbidities, initiation with morphine, fentanyl, or oxycodone, and co-prescription of gabapentinoids.

Abstract

The global rise in prescription opioid use has contributed to an opioid epidemic, associated harms, and unintentional deaths in several western countries. Opioids however continue to be regularly prescribed for acute pain and in the chronic pain context due to limited treatment options. Currently there are no accurate tools that help predict which patients prescribed opioids may be at risk of death, which depends on the cultural context and varies across countries. Existing models do not account for statistical considerations such as censoring and competing risks. Using nationally representative data from the United Kingdom from 1,026,139 patients newly prescribed an opioid, we developed three competing risk time-to-event models: a regression model, a random forest, and a deep neural network to predict opioid-related deaths using UK primary care records. The models were externally validated in an external cohort of 337,015 patients. The models exhibited good discrimination and positive predictive value during internal validation (C-statistic for the regression model, random forest, and neural network: 84.3%, 84.4% and 82.1% respectively), and external validation (C-statistic for the regression model, random forest, and neural network: 81.8%, 81.5% and 81.5% respectively). Prior substance abuse, lung and liver comorbidities, morphine, fentanyl, or oxycodone at initiation and co-prescription of gabapentinoids were some of candidate predictors associated with a higher risk of opioid-related mortality within the models. These results demonstrate how routinely collected data from a nationally representative dataset may be used to develop and validate opioids risk algorithms to better help clinicians and patients predict risk to this serious adverse outcome.

Referenced In

Mercedes C.
2 months ago
AI Papers and Discussion

🚨 AI Can Predict Opioid Death Risk… So Why Isn’t It Used in Clinics?

What began as an effort to treat PAIN has, in many parts of the world, evolved into a devastating public health crisis — with prescription opioids contributing to a growing number of opioid-related deaths.

A recent review compiled 44 machine learning studies attempting to predict opioid harm — yet almost NONE have made it into real clinical use.

Interestingly, model performance across these papers was moderate to strong, but 41% lacked proper calibration, meaning their risk predictions may not reflect real-world probabilities.

This UK study adds one more to the pile, using competing risk time-to-event models on over 1 million patients. It predicts opioid-related death with ~82% accuracy.

Top predictors include prior substance abuse, lung/liver co-morbidities, strong opioids at initiation, and gabapentinoid co-prescription.

What they did differently:

  • Predicted mortality rather than overdose

  • Implemented competing risks framework accounting for deaths from other causes

  • Tested if deep learning helps: 48,500-parameter neural network under-performed to a simple LASSO regression

  • Acknowledged poor calibration in external validation, where models overestimated absolute risk by 2-7×, designing percentile-based scores as workaround

  • Built for deployment: EHR-native features, SHAP interpretability, no data leakage

This new paper is a more rigorous model. Yet, it still may not reach patients.

Given the high recall, specificity and lower precision, the model works best when used against its design. The "Implementation Irony" holds true as it is trained to flag danger, yet it succeeds only at clearing safety.

It can suggest who probably will not die.

It cannot say who WILL

And in the midst overwhelming clinicians with false alarms.

⸂⸂⸜(രᴗര๑)⸝⸃⸃ Hey everyone!! 👋 Biomed engineering PhD student here — I always enjoy seeing how technology might actually translate into real healthcare impact. Anyways, this study recently caught my attention, and I’m curious to hear what others think.

🤔 Food for thought:

  • If simple models beat deep learning, why do we keep building bigger ones?

  • Is negative screening (identifying those safe to proceed) even useful to clinicians? 

  • Thus far, most models are based in the US or the UK, how far would the prediction shift in a new area / culture?

Read Full Post
3