Predicting Geriatric Rehabilitation Stays of ≤4 Weeks After Hip Fracture Surgery: Machine Learning Approach Using Physical Activity and Patient Data.

42 of 142 patients were excluded because they had incomplete physical activity data or missing patient characteristics, mainly from sensor battery or removal issues. The final analysis had 100 patients, and the authors say the sample size may be relatively small and some models prone to overfitting. The models were tested on a split from the same 100-patient sample, not on a new group, and the authors recommend validation in a larger population before clinical integration. The study used a 4-week cutoff, so patients close to the cutoff can be classified differently; 29-day and 30-day stays were misclassified as 4 weeks or less. Complications after the first week were not included, and one patient with a 42-day stay developed a complication during rehabilitation that may explain misclassification. Patients with severe cognitive impairment were excluded, so the results do not apply to that group. The setting is Dutch skilled nursing homes and a 4-week decision point based on Dutch funding pathways, so it may not transfer to other systems.

The authors declared no conflicts of interest.

Do not treat this as a tested discharge tool. The models were developed in 100 Dutch patients and tested on a split from that sample, 42 were excluded for missing data, and the authors say validation in a larger population is needed before clinical integration.