What makes an effective computerized clinical decision support system? A systematic review and logistic regression analysis of randomized controlled trials.

Abstract

Context: Computerized clinical decision support systems (CCDSSs) give practitioners patient-specific care advice and are considered an important increment to electronic clinical documentation and order entry systems. Despite decades of research on CCDSS, results from rigorous clinical evaluations remain mixed and systems vary greatly in design and implementation. Objective: To identify factors differentiating CCDSSs that improve the process of care or patient outcomes from those that do not. Data Sources: We searched major bibliographic databases and scanned reference lists for eligible articles up to January 2010. Study selection: 162 eligible comparisons from randomized controlled trials of CCDSS to non-CCDSS care. We deemed successful those systems that improved either 50% of reported process of care outcomes or 50% of patient outcomes. We extracted system characteristics hypothesized to impact patient care and tested them for association with system effectiveness in logistic models. Results: Our primary analysis showed that CCDSSs presented in electronic health records or order entry systems were less likely to be effective than their counterparts (OR, 0.37; 95% CI, 0.17 to 0.80). Systems more likely to succeed than their counterparts provided advice for patients in addition to practitioners (OR, 2.77; 95% CI, 1.07 to 7.17), required from practitioners a reason to override advice (OR, 11.23; 95% CI, 1.98 to 63.72), or were evaluated by their developers (OR, 4.35; 95% CI, 1.66 to 11.44). These findings remained consistent across different statistical methods, sensitivity analyses, and adjustment for other potentially important factors. Conclusions: We identified several factors that may partially explain why some systems succeed and others fail. Primary studies should investigate the impact and optimal implementation of advice provided to patients and practitioners and advice that requires reasons to be overridden. Researchers should also address the reasons for failure of advice presented within charting and order entry systems.