Prognostic profiles for discharge destination identified during intensive care unit admission: a decision tree analysis.

Single centre historical study, so results may not apply to other ICUs or countries. No external validation was performed, so the profiles may be overfitted to this dataset. The decision tree produced 5 terminal nodes but only 3 broad prognostic groups, and cut-offs such as 8.3 days ICU stay should be read as probabilities, not strict thresholds. Missing data were common, including 130 cases for step lifts and 156 for home modifications, which may affect model stability. The study excluded patients who died or had non-independent pre-admission ADLs, so it does not apply to those groups. PT sessions were higher in the non-home discharge group, but this is observational and cannot show whether PT helped or reflected greater need. The model includes only recorded variables, and other social or functional factors may matter.

The authors declared no conflicts of interest. The funding statement does not name a funder.

Do not use these profiles as a deterministic prediction for one patient. The study was single-centre, historical and not externally validated, and the cut-offs are probabilistic.