Source: Evans RS, et al. A computer-assisted management program for antibiotics and other anti-infective agents. N Engl J Med 1998;338: 232-238.
Evans and colleagues developed a computerized decision-support system for prescribing anti-infective agents and tested it in the critical care unit of a large university-affiliated teaching hospital. The program was linked to an advanced data management system that accessed an impressive database that included computerized patient records, radiology and laboratory data, microbiology results, cumulative antibiograms, patient allergies, and pharmacokinetic data. Physicians prescribed anti-infectives by direct computer order entry. They were provided with clinical information and therapeutic recommendations that included specific agents and dosages (adjusted for renal and hepatic functions). Clinicians were not required to follow the program's recommendations. Data on indices of quality of care were collected for two years before implementation (pre-intervention phase) and one year afterward (post-invention phase).

Post-intervention, there were substantial and statistically significant reductions in orders for drugs to which patients were allergic and for drugs to which the pathogens were not susceptible. There were also significant decreases in excessive drug dosages and in adverse drug effects.

Improvements in other measurements of outcome were less impressive. Mortality was the same (22%) in both periods. Evans and colleagues compared patients in the intervention period in whom the computer regimen was followed with those in whom it was not followed and with those in the pre-intervention period. In the former group, length of ICU stay, total hospital stay, and total cost of hospitalization were significantly lower than in either of the latter two groups. However, comparing all post-intervention patients to pre-intervention patients showed little difference in these measures.


Evans and colleagues used a highly sophisticated decision support program based on a highly advanced hospital data management system. This program was not merely a set of rigid practice guidelines. Rather, it was capable of giving patient-specific recommendations based on real data reflecting the patients' clinical situation, laboratory findings, and local conditions (including cumulative antibiograms). Clinicians were not required to accept the programs recommendations, nor was there any apparent penalty for not doing so.

Evans et al were able to document impressive improvements in some measures of qualtity of prescribing, such as aviodance of drugs to which patients were allergic and appropriate adjustment of dosage. Why were they not able to document an overall improvement in patient outcomes?

Although infection is a major contributor to morbidity and mortality in ICU patients, it is usually only one of many major problems experienced by these patients. Improvements in one aspect of care may not be sufficient to improve overall outcomes in critically ill patients. It is notable that clinicians overrode the program's recommendations about half of the time. This suggests that the program, despite its impressive sophistication, may not have considered, or have given the same weight to, all of the variables used by clinicians in coming to a therapeutic decision.

Despite having some limitations, I believe that such decision support systems will evolve over time and become important tools in patient management. They have the potential to improve patient care and reduce the frequency of adverse occurrences. In the field of infectious disease, such a system could be a powerful tool to reduce the frequency of inappropriate antibiotic administration, a costly problem that directly contributes to the rise of resistant pathogens.

Clinicians should support the development of decision support systems in their respective fields of expertise. Clinician input is critical to develop appropriate clinical decision making algorithms; participation and feedback is critical to the success and continued improvement of the process. Physician participation is also critical to ensure that such systems truly support patient care and are not solely directed toward the financial goals of the institution (e.g., steering the clinician away from costly but necessary and appropriate treatments).

Although it will be a long time before you will be able to buy one of these programs on CD-ROM from you local computer store, computerized decision support systems will be in widespread use in the not too distant future.