Evaluation scale to assess the accuracy of cuff-less blood pressure measuring devices

OBJECTIVE: The call for early detection of hypertension and cardiac events creates a heavy demand for devices that can be used for blood pressure (BP) monitoring at home and in ambulatory settings. An emerging type of BP monitors without an occluding cuff has drawn great attentions for this application because it is comfortable and capable of providing continuous readings. For the development the cuff-less devices, it is crucial for the clinicians and engineers to joint efforts in establishing an evaluation standard. METHODS: This study attempts to contribute to its initiation in two ways. First, a new distribution model for measurement differences between the test device and the reference was proposed. We verified the model using evaluation results from 40 devices, of which 80% of the American Association for the Advancement of Medical Instrumentation and British Hypertension Society reporting results were in agreement, as compared with 50%, if the original normal model was used. We further tested a cuff-less device on 85 patients for 999 datasets and found that the differences between the proposed distribution and that of the device were nonsignificant for systolic BP measurements (Kolmogorov-Smirnov = 0.036, P = 0.15). Second, some evaluation scales were studied for their capability to assess the accuracy of cuff-less devices. For mean absolute difference, a map was developed to relate it with the criteria of American Association for the Advancement of Medical Instrumentation, British Hypertension Society, and European Society of Hypertension protocols, on the basis of the proposed distribution model; for mean absolute percentage difference, it is prominent in evaluating devices that have measurement errors often increasing with BP, which is an issue has not been fully explored in existing standards. CONCLUSION: This study focused on the statistical aspect of establishing standard to assess the accuracy of cuff-less BP measuring devices. The results of our study on the validation reports of various cuff-based devices and an experimental study on a cuff-less device showed that the t4 distribution is better than the normal distribution in portraying the underlying error distribution of both kinds of devices. Moreover, based on both the theoretical and experimental studies, mean absolute difference or mean absolute percentage difference is recommended as continuous scale to assess the accuracy of cuff-less devices for their own distinctive advantages.