ICU sedation, delirium & recovery
Prediction of post-traumatic stress disorder in family members of intensive care unit (ICU) patients: a machine learning approach
Authors' conclusion
In this model, PTSD is mostly influenced by non-modifiable factors.
Study context
- Population
- Family members of ICU patients
- Participants
- 2,374 family members
- Comparison
- Random Forest vs XGBoost
- Outcome
- PTSD (Impact of Event Scale and Impact of Event Scale-Revised)
- Follow-up
- 90 days after ICU discharge
Methodological detailssupervised machine learning, independent held-out test dataset, machine learning interpretability methods
- Supervised machine learning
- Independent held-out test dataset
- Machine learning interpretability methods
Main finding
As reported in the abstract
Prediction of PTSD (held-out test dataset)
| Group | AUC |
|---|---|
| Random Forest | 0.73 [0.68-0.77] |
| XGBoost | 0.73 [0.69-0.78] |
AUC
Random Forest0.73 [0.68-0.77]
XGBoost0.73 [0.69-0.78]
Further findings
Features contributing to post-traumatic stress disorder (PTSD) predicted probability (best algorithm)
- Most important features all non-modifiable: lower patient age, longer ICU LOS, relative female sex, lower relative age, relative spouse/child, patient death in ICU
Sensitivity analysis in bereaved relatives
- Did not alter the algorithm's predictive performance
Original publication
Prediction of post-traumatic stress disorder in family members of ICU patients: a machine learning approach. Intensive Care Med. Published 19 December 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →