Background: Since clinical management of heart failure relies on weights that are self-reported by the patient,\nerrors in reporting will negatively impact the ability of health care professionals to offer timely and effective\npreventive care. Errors might often result from rounding, or more generally from individual preferences for numbers\nending in certain digits, such as 0 or 5. We apply fraud detection methods to assess preferences for numbers\nending in these digits in order to inform medical decision making.\nMethods: The Telemonitoring to Improve Heart Failure Outcomes trial tested an approach to telemonitoring that\nused existing technology; intervention patients (n = 826) were asked to measure their weight daily using a digital\nscale and to relay measurements using their telephone keypads. First, we estimated the number of weights subject\nto end-digit preference by dividing the weights by five and comparing the resultant distribution with the uniform\ndistribution. Then, we assessed the characteristics of patients reporting an excess number of weights ending\nin 0 or 5, adjusting for chance reporting of these values.\nResults: Of the 114,867 weight readings reported during the trial, 18.6% were affected by end-digit preference, and\nthe likelihood of these errors occurring increased with the number of days that had elapsed since trial enrolment\n(odds ratio per day: 1.002, p < 0.001). At least 105 patients demonstrated end-digit preference (14.9% of those who\nsubmitted data); although statistical significance was limited, a pattern emerged that, compared with other patients,\nthey tended to be younger, male, high school graduates and on more medications. Patients with end-digit\npreference reported greater variability in weight, and they generated an average 2.9 alerts to the telemonitoring\nsystem over the six-month trial period (95% CI, 2.3 to 3.5), compared with 2.3 for other patients (95% CI, 2.2 to 2.5).\nConclusions: As well as overshadowing clinically meaningful changes in weight, end-digit preference can lead to\nfalse alerts to telemonitoring systems, which may be associated with unnecessary treatment and alert fatigue. In\nthis trial, end-digit preference was common and became increasingly so over time. By applying fraud detection\nmethods to electronic medical data, it is possible to produce clinically significant information that can inform the\ndesign of initiatives to improve the accuracy of reporting.
Loading....