Predictive accuracy of three scoring systems in a single intensive care unit

Full text

Objective

To evaluate the predictive accuracy of the severity of three scoring systems — SAPS II, MPM II and LODS — in a single Intensive Care Unit (ICU).

Design

Prospective cohort study.

Setting

General medical and surgical ICU in a tertiary teaching hospital in city of São Paulo, Brazil.

Patients

318 consecutive, unselected patients over the period from February to June of 1999. Cardiac surgical and burns patients were excluded.

Outcome measure

Patients vital status at ICU discharge.

Interventions

None.

Measurements and statistical analysis

Data required to calculate the patient's predicted mortality by the considered scoring systems were collected. Area under the receiver operating characteristic curve (ROC curve) was calculated to evaluate discrimination of each scoring system. The goodness-of-fit statistic was used to assess calibration of the models and the strength of the association between the predicted and observed outcome.

Main results

The mean age was 55.2± 19 years. The length of ICU stay was 10.64± 13.09 days, and the mortality rate was 26.1 %. Calibration and discrimination results and the predicted risk of death are shown in the Table.

Conclusions

The results showed good discrimination for all the models.

LODS, MPM 0 and SAPS II showed good calibration, i.e, there were no significant discrepancy between predicted and observed mortality. MPM 24 showed a poor fit.

According to this, MPM0 can be used as a predictor of death at patients admission, SAPS II as a predictor of death in the first 24 h of admission, and LODS as end point in studies of organ dysfunction in our ICU.

Scoring systems need to be evaluated in terms of predictive accuracy within a single institution before applying them to make quality of care assessments.

Table