Laboratory test results predict death

Hospital mortality rates provide information for evaluating the relative quality of the health care delivered. However, interpretation of hospital mortality rates is fraught with difficulty, not least because such bald data does not take account of the severity of the illness of patients, whose outcome is being measured. 

Risk-adjusted mortality rates that reliably take account of the severity of the patient’s condition (i.e. risk of death) on admission would allow fairer comparison. This poses the problem of how to assess risk of death at the time of admission. A recently published Israeli study suggests that the results of routine laboratory tests at the time of admission predict death reliably and could therefore be used to prepare risk-adjusted mortality rates.

The authors of this study retrieved admission blood test results of all 10,308 patients admitted to four internal medicine wards of their hospital during 2003; it is routine procedure at this hospital for blood to be collected for full biochemical profile (12 measured parameters) and full blood count at the time of admission. 

 Of the 10,308 patients, 573 died during hospital admission and the rest were discharged home. The mean age of those that died (81.1 yrs) was, not unexpectedly, greater than those that survived (69 yrs). 

Statistical analysis of blood test results revealed that for all parameters measured there was a statistically significant difference between the mean value of those who subsequently died and the mean value of those who survived. Moreover, the difference was always in the expected direction.

Further statistical analysis revealed that age and eight of the laboratory tests (albumin, urea, glucose, alkaline phosphatase, aspartate transferase, lactate dehydrogenase, white cell count and neutrophil count) all independently and significantly contributed to the logistic regression model that was "excellent in predicting patient mortalities" with area under ROC curve of 90.4 %. 

The model was tested for all patients admitted during the next year (2004) and proved just as effective in predicting death.