Logistic ordinal regression for the calibration of oscillometric blood pressure monitors

Oscillometric Blood Pressure (BP) monitors are omnipresent and used on a daily basis for personalized healthcare. Nevertheless, physicians generally approach these devices cautiously since the mercury Korotko sphygmomanometer remains the golden standard. Various reasons explain the hesitating attitude of the medical world towards automated BP monitors: (i) its principle is based on the pressure pulsations arriving at the cu by the cardiac cycle instead of an audio wave used by physicians triggered by the turbulences in the artery, (ii) the actual computation of the systolic and diastolic BP from the measured oscillometry is manufacturer dependent and not based on general scienti c principles, (iii) the quality of the oscillometric monitors is labeled by a trial such that the devices correspond well to the Korotko method for the average healthy patient but deviates for patients su ering from hypoor hypertension. In this paper, we develop a statistical learning technique to calibrate and correct an oscillometric monitor such that the device better corresponds to the Korotko method regardless of the health status of the patient. The technique is based on logistic regression which allows correcting and eliminating systematic errors caused by patients su ering from hyper -or hypotension. No user interaction is required since the technique is able to train and validate the calibration procedure in an unsupervised way. In our case study, the systematic error is reduced by nearly 50% corresponding to the performance speci cations of the device.