Volume 11 Supplement 2
Could we use the admission Acute Physiology and Chronic Health Evaluation II score for outcome prediction in critically ill obstetric patients?
© BioMed Central Ltd. 2007
Published: 22 March 2007
The APACHE II score (APII) has widespread use in ICUs for research and benchmarking. Physiological data for calculation of the APII score derive from worst values in the first 24 hours after ICU admission.
Mortality prediction by the APII system depends on data sampling. Use of ICU admission data (first hour) could be accurate to predict mortality.
An open prospective data-sampling part of the APRiMo study. Included were critically ill obstetric patients, with ICU length of stay (LOS) ≥6 hours. Admission (H1: first-hour worst physiological data) and H24 (worst 24-hour physiological variables including H1 collected data) were used to generate, respectively: the admission APII score (H1-APII) and H24-APII. The formulae to calculate individual mortality for H1 and H24 APII were those validated for H24-APII as stated by Knaus and colleagues , adjusting for admission diagnosis. We compared both scores by discrimination and calibration statistical tests. P < 0.05 was the threshold for statistical significance.
The study period was January 1996–September 2004. We included 541 patients, overall mortality was 10.5%. Mean H1-APII and H24-APII scores. respectively. were 7.6 ± 6.1 and 8.6 ± 7, with derived mean predicted mortality, respectively, of 8.63% and 9.86%. The H24-APII score was higher than the H1-APII score in 135 patients (25%), among those patients 32 died (24% of patients with worsened APII) vs 6.16% if H1 = H24 (P < 0.01). Running a multiple logistic regression with mortality as the dependent parameter, we found that worsening of the APII score over time is not significantly associated with mortality (P = 0.791), whereas the H1-APII score (P < 0.001) and ΔAPII score (H24-APII minus H1-APII score) (P = 0.04) are correlated with mortality. Respective ORs are 1.28 and 1.45. Overall discrimination ability assessed by receiver operating characteristic curves was good for H1-APII (0.78) and H24-APII (0.784) (P = 0.834).
To avoid variation in APII mortality prediction caused by variable sample rates, the admission APII is reliable. Customizing mortality formulae could improve performances of APII-H1.