[Failure of prediction of results with APACHE II. Analysis of prediction errors of mortality in critical patients]

The evaluation of the prognosis of critically ill patients by the APACHE II method is common in intensive care units (ICU). The aim of the present was to analyze the possible factors associated to errors in prediction.

A prospective study of 564 consecutive admissions in a department of intensive medical care was carried out. Prediction errors were studied by the calculation of the probability of death established after the first 24 hours of admission by means of APACHE II. The factors analyzed in relation to the prediction errors were: the diagnosis or cause of admission to the ICU, the length of the stay in the ICU, the time until possible death, the possible relation of the death with the cause of admission and the treatment given to the patients during the first 24 hours. Statistical analysis was performed with the SPSS software package with significance being determined at p < 0.05.

Mortality was of 20.6% (116 cases) with three cut off points being chosen for probability of death (50, 70, and 90%). Accuracy of precision was 83.5%, 82.8% and 80.1%. There were 64 false survivors (mortality lower than 50%, 13.25%-64/483) and 29 false deaths (survival greater than 50%, 35.8%-29/81). Upon analysis of the cause of admission of these patients in whom there were prediction errors it was found that there were no differences among the false survivors and the false deaths. Significant differences were only detected upon comparison of the false survivors with the verified survivors, however these disappeared when the 136 cases admitted due to myocardial infarction were excluded. Neither did the length of stay in the ICU demonstrate any significant difference except among the verified and false deaths in that the stay was longer in the latter.

The factors analyzed did not demonstrate that they may influence or be associated with errors in prediction of the prognosis of patients admitted to an intensive care unit, with these errors probably being due to errors in the system used.