Patterns of noncompliance with recommendations from a computerized insulin protocol

Detailed research protocols standardize the interventions for clinical studies, but the protocols can be complex. Clinical decision support systems (CDS) can interpret detailed protocols for care providers. Nurses are expected to comply with research protocols; however, nurses sometimes reject the CDS recommendations. This study evaluated patterns of noncompliance with recommendations from a computerized insulin-titration CDS (eProtocol-insulin). The primary research question was: Are there patterns to noncompliance with recommendations from eProtocol-insulin? The study used a descriptive/exploratory design; with data from a nurse questionnaire supplementing data from a retrospective patient cohort reflecting 2 years of software use. A content analysis of text entered by nurses into the software described reasons for noncompliance. Patient data, such as trends over an extended time, were the reason for almost half (49.2%) of declines. The nurse disagreed with knowledge base rules for one quarter (27.3%) of declines, and nurses experienced barriers to compliance for 12.3% of declined recommendations. Organizational factors, usability, and attitudes were not reasons for any declined recommendations in this analysis. A statistical analysis evaluated the relationship between patient variables and noncompliance with recommendations. Demographic characteristics were not associated with declined recommendations. Associations were found between specific patient conditions and declines. However, no clear pattern of predictors emerged. It is likely that more complex models would be needed to reliably predict noncompliance. Nurses described factors that they perceived as important for compliance. The nurses understood the overall protocol but wanted explanations available on demand. Other factors perceived as important included help files accessible through the software interface, maintaining the ability to decline recommendations, and having trust in the protocol development process. Noncompliance was a relatively uncommon event, occurring less than 6% of the time. However, patterns of noncompliance revealed areas where the CDS could be improved. Patients experienced events not accommodated by protocol rules, which the nurses appeared to perceive as safety issues. Even with high levels of overall compliance, evaluation of declined recommendations can lead to improved decision support systems and better patient safety.