Disparity in outcome prediction between APACHE II, APACHE III and APACHE IV

Introduction

The critically ill obstetric population still search for a model that accurately predicts mortality. The study hypothesis was that APACHE IV [1] predicts ICU mortality better than APACHE III [2] and APACHE II [3].

Methods

A prospective collection of data concerning APACHE II and APACHE III, and a retrospective analysis of complimentary data necessary for APACHE IV mortality calculation. Discrimination was assessed by the area under the receiver operator curve (ROC) and calibration by the Hosmer–Lemeshow (HL) goodness-of-fit test. Results are expressed as the mean ± SD. P < 0.05 was considered significant.

Results

The mean age was 31.2 ± 5.9 years. Seventy-five percent were delivered by caesarean section. Seventy-eight percent needed mechanical ventilation. Overall mortality was 11.23% (n = 71/641). Acute physiology scores (APS) of APACHE II and APACHE III were significantly different between survivors and nonsurvivors, respectively (7.2 ± 5 vs 20 ± 9 and 23.5 ± 18 vs 76 ± 39) (P < 0.001). See Table 1.

Table 1 Performance of the scores concerning mortality prediction formulas and acute physiology scores

Conclusion

APACHE II mortality prediction is out of date. APACHE III and APACHE IV mortality have excellent discrimination but poor calibration. Considering the APS alone, the APACHE systems discriminate and calibrate well. APACHE IV can therefore be considered the best mortality prediction model. Incorporation of new predictor variables such as mechanical ventilation and importance of respiratory dysfunction explains part of this improvement. Regular recalibration of mortality prediction formulas is important and helps improve calibration for aggregate patient samples. For specific subgroups of patients, however, this measure is probably insufficient; we need to incorporate new specific variables.