A simple clinical predictive index for objective estimates of mortality in acute lung injury

Abstract

Objective: We sought to develop a simple point score that would accurately capture the risk of hospital death for patients with acute lung injury (ALI).

Design: This is a secondary analysis of data from two randomized trials. Baseline clinical variables collected within 24 hours of enrollment were modeled as predictors of hospital mortality using logistic regression and bootstrap resampling to arrive at a parsimonious model. We constructed a point score based on regression coefficients.

Setting: Medical centers participating in the Acute Respiratory Distress Syndrome Clinical Trials Network (ARDSnet).

Patients: Model development: 414 patients with nontraumatic ALI participating in the low tidal volume arm of the ARDSnet Acute Respiratory Management in ARDS study. Model validation: 459 patients participating in the ARDSnet Assessment of Low tidal Volume and elevated End-expiratory volume to Obviate Lung Injury study. Model Validation: 459 patients participating in the ARDSnet Assessment of Low tidal Volume and elevated End-expiratory volume to Obviate Lung Injury trial.

Interventions: None.

Measurements and main results: Variables comprising the prognostic model were hematocrit <26% (1 point), bilirubin >or=2 mg/dL (1 point), fluid balance >2.5 L positive (1 point), and age (1 point for age 40-64 years, 2 points for age >or=65 years). Predicted mortality (95% confidence interval) for 0, 1, 2, 3, and 4+ point totals was 8% (5% to 14%), 17% (12% to 23%), 31% (26% to 37%), 51% (43% to 58%), and 70% (58% to 80%), respectively. There was an excellent agreement between predicted and observed mortality in the validation cohort. Observed mortality for 0, 1, 2, 3, and 4+ point totals in the validation cohort was 12%, 16%, 28%, 47%, and 67%, respectively. Compared with the Acute Physiology Assessment and Chronic Health Evaluation III score, areas under the receiver operating characteristic curve for the point score were greater in the development cohort (0.72 vs. 0.67, p = 0.09) and lower in the validation cohort (0.68 vs. 0.75, p = 0.03).

Conclusions: Mortality in patients with ALI can be predicted using an index of four readily available clinical variables with good calibration. This index may help inform prognostic discussions, but validation in nonclinical trial populations is necessary before widespread use.

Figure 1. Calibration plot

For each patient in the validation dataset, the predicted mortality estimated from the Acute Physiology Assessment and Chronic Health Evaluation III (APACHE) score is plotted against the point total from the custom model developed from the ARDSnet low tidal volume study. Patients with an APACHE III predicted mortality overlapping with the custom model predicted mortality in the validation cohort are shown using triangles (within). Patients where APACHE III predicts a greater rate of death than predicted by the simple model are shown as closed circles (above). Patients where APACHE III predicts a lower rate of death than predicted by the simple model are shown in open circles (below).

Figure 2. Receiver operating characteristic (ROC) curves for the custom model

Panel A – comparison of the ROC curves for the custom model and APACHE III score (area 0.72 vs. 0.67, p=0.09) in the development cohort. Panel B – comparison between custom model and APACHE III (area 0.68 vs. 0.75, p= 0.03) in the validation cohort.

Figure 3. Example calculation of post-test probability of death

Confidence intervals for the post-test probability integrate uncertainty in the likelihood ratio. LR+, positive likelihood ratio; CI, confidence interval.