The Effect of Feedback Density on the Frequency of Error Message Override Behavior

Computerized Physician Order Entry (CPOE) Systems have been proposed as a technologically based approach to addressing medication errors by allowing the clinician to enter medication orders directly into a desktop or mobile computer unit while receiving immediate performance feedback from the computer system. An unforeseen problem that has been encountered by health care organizations is a high rate of error message override behavior by physicians who report low levels of confidence in the comprehensiveness and applicability of the computer-provided error messages, software over-sensitivity and excessive information presented during medication order entry. The purpose of this study was to investigate the relationship between performance feedback presented in the form of a computer delivered error message, and the frequency that subjects override or ignore that performance feedback.;One hundred and twenty subjects completed a chess simulation versus a computer controlled opponent while receiving no feedback, low complexity performance feedback, moderate complexity performance feedback with the option of accessing hints or assistance and high complexity performance feedback with mandatory maximum performance feedback.;Results indicated that there were significant differences (p<.05) between groups in the percentage of error messages overridden. Subjects who received high complexity performance feedback committed the fewest errors, overrode significantly more error messages than subjects who received low complexity performance feedback and displayed more override behavior than subjects who received moderate complexity performance feedback. These results demonstrate that providing computer provided performance feedback that is overly complex can produce high levels of override behavior as subjects actively seek to avoid these error messages rather than utilizing the information provided.;Significant between-group performance differences (p<.05) were also observed with subjects who received moderate complexity performance feedback losing significantly fewer games than subjects who received low complexity performance feedback and fewer games than subjects who received high complexity performance feedback. Subjects who received moderate complexity performance feedback were significantly more likely to access hints or assistance than were subjects who received low complexity performance feedback, and more likely to access hints or assistance than subjects who received high complexity performance feedback. In general, subjects who received feedback in all forms lost significantly fewer games than did subjects who received no feedback, with subjects who received more detailed feedback (moderate and high complexity) losing fewer games than subjects who received little or no feedback (low complexity or no feedback).;These performance differences highlight the relationship between feedback complexity, error message override frequency and overall performance. Although better than no feedback, providing overly complex error messages can negatively affect performance by encouraging higher levels of error message override behavior which reduces the effectiveness of error alerting systems. In this simulation, providing subjects with less information initially along with the option of accessing as much supporting information as needed produced a lower frequency of error message override behavior and fewer negative final outcomes. While allowing organizations to provide comprehensive, detailed and immediate alerts on a truly continuous schedule, computer based error alert systems can also introduce behaviors that lessen the effectiveness of medical error interventions aimed at reducing negative final outcomes.