Editor—In their article on scoring systems in intensive care, Gunning and Rowan provide a detailed description of the APACHE II mortality prediction model.1 This model was developed with data collected between 1979 and 1982 and was replaced by APACHE III in 1991.2 Unfortunately, there is a substantial charge for using APACHE III; thus many intensive care units continue using the outdated APACHE II, but do you really want to compare your standard of care with that delivered by North American units 20 years ago?
There are several other problems with APACHE II and APACHE III. Firstly, they use the worst value of several physiological variables (such as blood pressure and heart rate) in the first 24 hours in intensive care to calculate each patient's risk of dying. Any score that uses data collected over 24 hours is affected by the quality of care provided2,3— the very thing that units are trying to assess. Patients mismanaged in a bad unit will have higher APACHE scores than similar patients managed in a good unit, and the bad unit's high mortality will be incorrectly attributed to its having sicker patients.
Secondly, using the worst scores in 24 hours gives a spurious impression of accuracy3; many deaths occur during the first 24 hours in intensive care, and during this time the score is diagnosing death rather than predicting it (it is not difficult to detect that something is wrong with a dead patient).
Thirdly, the worst scores in 24 hours depend on the method of data collection, being about 25% higher with continuous computer monitoring than with manual recording.4
Fourthly, collecting the worst value of 15 variables over 24 hours is difficult; either a disproportionate amount of work goes into collecting the information (which reduces the resources available for more creative research) or, as often happens, it is collected inaccurately or is not collected at all.
The mortality prediction model uses data collected during the first hour after admission to intensive care.2 It was developed by Stanley Lemeshow, professor of biostatistics at the University of Massachusetts, who coauthored one of the standard texts on logistic regression. The model has been derived and tested on over 19 000 patients in Europe and North America and is free.5 For children, the paediatric index of mortality model uses data obtained at the time that a child is admitted to intensive care,3 and it too is free (http://pedsccm. wustl.edu/clinical/pim-readme.html).
The mortality prediction model has substantial theoretical, practical, and financial advantages over the APACHE model for use in adults in intensive care.
References
- 1.Gunning K, Rowan K. ABC of intensive care: Outcome data and scoring systems. BMJ. 1999;319:241–244. doi: 10.1136/bmj.319.7204.241. . (24 July.) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Shann F. MPM is preferable to APACHE as an intensive care scoring system. eBMJ 1999 (www.bmj.com/cgi/eletters/319/7204/241#EL2) (accessed 17 February 2000).
- 3.Shann F, Pearson G, Slater A, Wilkinson K. Paediatric index of mortality (PIM): a mortality prediction model for children in intensive care. Intens Care Med. 1997;23:201–207. doi: 10.1007/s001340050317. [DOI] [PubMed] [Google Scholar]
- 4.Bosman RJ, Oudemans van Straaten HM, Zandstra DF. The use of intensive care information systems alters outcome prediction. Intens Care Med. 1998;24:953–958. doi: 10.1007/s001340050695. [DOI] [PubMed] [Google Scholar]
- 5.Lemeshow S, Le Gall JR. Modeling the severity of illness of ICU patients: a systems update. JAMA. 1994;272:1049–1055. [PubMed] [Google Scholar]