The influence of length of stay in the ICU on power of discrimination of a multipurpose severity score (SAPS). ARCHIDIA.

OBJECTIVE

To assess how the power of discrimination of a multipurpose severity score (Simplified Acute Physiology Score; SAPS) changes in relation to the length of stay (LOS) in the intensive care unit (ICU).

DESIGN

In order to compute the SAPS probability, a model derived from logistic regression was developed in a cohort of 8059 patients. Measures of calibration (goodness-of-fit statistics) and discrimination [receiver operating characteristic (ROC) curve and relative area under the curve (AUC)] were adopted in a developmental set (5389 patients) and a validation set (2670 patients), both randomly selected. Once the logit was developed and the model validated, the whole database (8059 patients) was again assembled. To evaluate the accuracy of first-day SAPS probability over time, area under the ROC curve was computed for each of the initial 10 days of ICU care and for day 15.

SETTING

24 Italian ICUs.

PATIENTS

A total of 8059 patients out of 10065 consecutive admissions over a period of 3 years (1990-1992) were included in this study. Patients whose SAPS was not correctly compiled (n = 687), patients younger than 18 years (n = 442), and patients whose LOS was less than 24 h (n = 877) were excluded from this analysis.

INTERVENTIONS

None.

MEASUREMENTS AND RESULTS

The logistic model gave good results in terms of calibration and discrimination, both in the developmental set (goodness-of-fit: chi 2 = 9.24, p = 0.32; AUC = 0.79 +/- 0.01) and in the validation set (goodness-of-fit: chi 2 = 8.95, p = 0.537; AUC = 0.78 +/- 0.01). The AUC for the whole database showed a loss in discrimination closely related to LOS: 0.79 +/- 0.01 at a day 1 and 0.59 +/- 0.02 at day 15.

CONCLUSION

The logistic model that we developed meets high standards for discrimination and calibration. However, SAPS loses its discriminative power over time; accuracy of prediction is maintained at an acceptable level only in patients who stay in the ICU no longer than 5 days. The stay in the ICU represents a complex variable, which is not predictable, that influences the performance of SAPS on the first day.