A SERIALLY UPDATED PREDICTOR OF RISK

Abstract

Assessments of severity of illness have only utilized single measurements in isolated time periods. Using a multiple logistic approach, a continuous variable (Risk Index: RI) was developed that utilizes serial measurements and characterizes the risk of acute (< 24 hr) and evenutal mortality. The RI is based on the Physiologic Stability Index (PSI), a validated measure of severity of illness. The parameters of the logistic model were determined from 293 consecutive ICU patients. The number of serial observations required was 2 (p < .05). RI_t = - 6.474 + .159 *PSI_t + .042 *PSI_t-1. A prospective validation was done on a group of 345 consecutive ICU admissions. Results: The model reliably predicted acute deaths in the validation set (γ ², p > .8). Examples are: pre = .7%, obs = .5 ± .3%; pre = 4.7%, obs = 4.7 ± 2.3%; pre = 26.9%, obs = 23.3 ± 11.7%. Receiver operating characteristics (ROC) curves demonstrated excellent performance in predicting eventual deaths. There was no loss of performance in the validation set compared to the estimation set (area under ROC curve = A_z: A_z validation = A_z estimation = .91). At a RI cutoff value of -3.25, the truly predicted survivor fraction = .93, truly predicted nonsurvivor fraction = .83 falsely predicted survivor fraction = .17, falsely prediced nonsurvivor fraction = .07. Discussion: The large amount of data recorded in the ICU naturally arise as serial observations and should be treated as such. The RI maximizes predictive power using this approach. The RI can aid decision making, therapeutic evaluations, health care evaluations, and cost containment.

Log in or create a free account to read this content

Gain free access to this article, as well as selected content from this journal and more on nature.com

Access through your institution

or

Sign in or create an account

Continue with Google

Continue with ORCiD