Hospital length of stay after serious injury

点击翻译按钮获取中文摘要

DESCRIPTION (PROVIDED BY APPLICANT): This project proposes to improve prediction of hospital length of stay (LOS) and mortality in seriously injured patients, using a recently developed piecewise exponential (PWE) model allowing for multiple outcomes and time-varying covariates. LOS can be difficult to predict in patient populations with significant mortality, because the cases with shortest LOS include both the least serious (discharged alive) and the most serious (fatalities). The multiple-outcome PWE model, designed to overcome this difficulty, will be applied to data from the National Trauma Data Bank(r) of the American College of Surgeons. Half of the database will be randomly selected for model development. First, standard regression models for the estimation of mortality and LOS will be derived. Using similar variables, a multiple-outcome PWE model will also be developed: For specified times after admission, selected cases will be compartmentalized as 1) still in hospital, 2) discharged alive, or 3) died. After distinguishing time periods during which the transition rates from the hospitalized state to the other states are relatively constant, the effects of different covariates on these rates will be determined using Poisson regression. Standard classifications of injury severity, injury mechanism, and preexisting health status will be used, along with standard modeling approaches and data transformations. Where possible, the model will be simplified by combining estimates of covariate effects for adjacent time periods. Mathematical formulas for LOS and mortality for any given set of covariates, assuming piecewise constant rates occurring simultaneously for each transition, can be derived analytically from the PWE model. Using the remaining half of the database, multiple-outcome PWE model predictions of LOS and mortality will be compared to the standard regression predictions and to the actual experience. Covariate effects on early and late mortality, as well as LOS, will be contrasted and analyzed. Predictions from the multiple-outcome PWE model may be useful for cost-effectiveness studies, as well as for determining appropriate reimbursement for the care of injured patients. Time-varying multistate models may also be useful in other areas of medical outcomes research.