PURPOSE: Comparison of illness severity for intensive care unit populations assessed according to different scoring systems should increase our ability to compare and meta-analyze past and future trials but is currently not possible. Accordingly, we aimed to establish a methodology to translate illness severity scores obtained from one system into another.
MATERIALS AND METHODS: Using the Australian and New-Zealand intensive care adult patient database, we obtained simultaneous admission Acute Physiology and Chronic Health Evaluation (APACHE) II and APACHE III scores and Simplified Acute Physiology Score (SAPS) II in 634428 patients admitted to 153 units between 2001 and 2010. We applied linear regression analyses to create models enabling translation of one score into another. Sensitivity analyses were performed after removal of diagnostic categories excluded from the original APACHE database, after matching for similar risk of death, after splitting data according to country of origin (Australia or New Zealand) and after splitting admissions occurring before or after 2006.
RESULTS: The translational models were APACHE III=3.08×APACHE II+5.75; APACHE III=1.47×SAPS II+8.6; and APACHE II=0.36×SAPS II+4.4. The area under the receiver operating curve for mortality prediction was 0.853 (95% confidence interval, 0.851-0.855) for the "APACHE II derived APACHE III" score and 0.854 (0.852-0.855) for the "SAPS II derived APACHE III" vs 0.854 (0.852-0.855) for the original APACHE III score. Similarly, it was 0.841 (0.839-0.843) for the "SAPS II derived APACHE II score" vs 0.842 (0.840-0.843) for the original APACHE II score. Correlation coefficients as well as intercepts remained very similar in all subgroups analyses.
CONCLUSIONS: Simple and robust translational formulas can be developed to allow clinicians to compare illness severity between studies involving critically ill patients. Further studies in other countries and health care systems are needed to confirm the generalizability of these results.