Dynamic prediction of mortality among patients in intensive care using the sequential organ failure assessment (SOFA) score: a joint competing risk survival and longitudinal modeling approach

In intensive care units (ICUs), besides routinely collected admission data, a daily monitoring of organ dysfunction using scoring systems such as the sequential organ failure assessment (SOFA) score has become practice. Such updated information is valuable in making accurate predictions of patients' survival. Few prediction models that incorporate this updated information have been reported. We used follow†up data of ICU patients who either died or were discharged at the end of hospital stay, without censored cases. We propose a joint model comprising a linear mixed effects submodel for the development of longitudinal SOFA scores and a proportional subdistribution hazards submodel for death as end point with discharge as competing risk. The two parts are linked by shared latent terms. Because there was no censoring, it was straightforward to fit our joint model using available software. We compared predictive values, based on the Brier score and the area under the receiver operating characteristic curve, from our model with those obtained from an earlier modeling approach by Toma et al. [Journal of Biomedical Informatics 40, 649, (2007)] that relied on patterns discovered in the SOFA scores over a given period of time.