Edinburgh Napier University

  Objective Simple forms of blood pressure (BP) telemonitoring require patients to text readings to central servers creating an opportunity for both entry error and manipulation. We wished to determine if there was an apparent preference for particular end digits and entries which were just below target BPs which might suggest evidence of data manipulation.
Design Prospective cohort study
Setting 37 socio-economically diverse primary care practices from South East NHS Lothian, Scotland.
Participants Patients were recruited with hypertension to a telemonitoring service in which patients submitted home BP readings by manually transcribing the measurements into text messages for transmission (‘patient-texted system’). These readings were compared to those from primary care patients with uncontrolled hypertension using a system in which readings were automatically transmitted, eliminating the possibility of manipulation of values (‘automatic-transmission system’).
Methods A Generalised Estimating Equations method was used to compare BP readings between the patient-texted and automatic-transmission systems, while taking into account clustering of readings within patients.
Results A total of 44,150 BP readings were analysed on 1,068 patients using the patient-texted system compared to 20,705 readings on 199 patients using the automatic-transmission system. Compared to the automatic-transmission data, the patient-texted data showed a significantly higher proportion of occurrences of both systolic and diastolic BP having a zero end digit (OR 2.1, 95% CI 1.7 to 2.6) although incidence was less than 2% of readings. Similarly, there was a preference for systolic 134 and diastolic 84 (the threshold for alerts was 135/85) (134 systolic BP OR 1.52, 95% CI 1.28 to 1.82; 84 diastolic BP OR 1.54, 95% CI 1.28 to 1.86).
Conclusion End-digit preference for zero numbers and specific value preference for readings just below the alert threshold exists among patients self-reporting their BP using telemonitoring. However, the proportion of readings affected is small and unlikely to be clinically important.