EUVIMEDObjevte EUVIMED
Open access · CC BY-NC 4.0

A Bayesian hurdle Conway–Maxwell–Poisson regression for predicting post-surgical length of stay in adult traumatic brain injury

Pichit Boonkrong, Subij Shakya, Jittrapat Kongsup, Montira Muanjan

Array · 2026-11 · 101434

DOI: 10.1016/j.array.2026.101434 ↗

At a glance

Journal article in Array, 2026-11. Written by Pichit Boonkrong and 3 co-authors. 74 references. The full text is freely available (open access, CC BY-NC 4.0).

AbstraktBibliografické údajeCiteRelated studies

Abstract (original language)

Hospitals cannot plan neurosurgical beds and medication stocks for adults with traumatic brain injury (TBI) without a reliable prediction of how long each patient will stay after surgery, with its uncertainty. Such prediction is difficult because length of stay (LOS) is a count of days that is never zero, highly variable, long-tailed and sharply peaked at two to three days. Existing models capture only part of this pattern and give single-number predictions that cannot be used directly for capacity planning. A two-part Bayesian count model was therefore developed: the first part handles one-day stays separately, and the second, a flexible Conway–Maxwell–Poisson (CMP) count model, describes stays of two days or longer. Its predictions were translated into expected bed-days, occupied beds and stay-length tiers for medication planning. The model was compared with eight alternative count models in 215 adult TBI patients at a Thai trauma centre, using twelve admission predictors. It achieved the best overall fit, outperformed seven of the eight alternatives and performed comparably to a two-part negative binomial model; only the two-part models reproduced the two-to-three-day peak. Admission Glasgow Coma Scale was the only strong predictor: each additional point was associated with a 6.20% lower rate of hospital days (95% CrI 4.80%–7.90%). Severity-matched medication tiers reduced expected provisioning loss by 45.40% compared with fixed 3- and 7-day tiers. This is the first such two-part model for TBI LOS and the first to carry its predictions into hospital-level planning decisions.

Cite

APA

Boonkrong, P., Shakya, S., Kongsup, J., & Muanjan, M. (2026). A Bayesian hurdle Conway–Maxwell–Poisson regression for predicting post-surgical length of stay in adult traumatic brain injury. Array, 101434. https://doi.org/10.1016/j.array.2026.101434

Vancouver

Boonkrong P, Shakya S, Kongsup J, Muanjan M. A Bayesian hurdle Conway–Maxwell–Poisson regression for predicting post-surgical length of stay in adult traumatic brain injury. Array. 2026:101434. doi:10.1016/j.array.2026.101434

BibTeX
@article{Boonkrong2026,
  author = {Boonkrong, Pichit and Shakya, Subij and Kongsup, Jittrapat and Muanjan, Montira},
  title = {A Bayesian hurdle Conway–Maxwell–Poisson regression for predicting post-surgical length of stay in adult traumatic brain injury},
  journal = {Array},
  year = {2026},
  pages = {101434},
  doi = {10.1016/j.array.2026.101434}
}
Start for free →