Real-Time Expert-System Identification of Blood Pressure Measurement Accuracy During Renal Dialysis Treatment

Abstract

Abstract — Objective: Current practice relies on intermit- tent occluding arm cuff measurements to monitor blood pressure during hemodialysis and to detect hypotension. However, systematic reviews report measurement accuracy challenges associated with brachial cuff measurements observed in the general population, and the factors con- tributing to inaccuracy are likely to be accentuated during dialysis treatment. There is currently no formal process to identify unreliable cuff BP measurements, and staff gen- erally rely on ad hoc extra measurements and averaging readings. The objective of the activity described in this paper was to design a computational method to identify unreliable cuff measurements as they are taken and thus provide decision support to practitioners on dialysis units. Reliable intermittent systolic measurements are fundamen- tally important to both the calibration of continuous blood pressure measurements, and methodologies to predict the onset of hypotension.

Methods: Patient studies with concurrent measurements of real-time continuous dialysis line pressure and intermit- tent systolic brachial cuff pressure during typically 4-hour, dialysis treatment sessions, revealed that some cuff mea- surements lay outside the prediction bounds associated with the expected quasi-linear (time-varying) relationship between arterial line and brachial pressure measurements. An AI expert system was designed, which embodies the mathematical relationships predicted by a system model, and a further complex rule-set which is able to discriminate between reliable and unreliable cuff measurements in real time based on sparse intermittent incoming data. The de- veloped system was deployed on an observational patient study during hemodialysis treatments, outputting recom- mendations and justifications for accepting/rejecting cuff measurements. The accepted measurements were fed into a continuous, non-invasive systolic pressure estimator as calibration, enabling the reliability of the decisions made in the arterial line / systolic pressure domain to be verified in the systolic pressure / time domain. Results: Data collected from a prospective, observational patient study exhibited robust identification of unreliable arm cuff measurements, with the system operating as decision support. Continuous, non invasive, SBP predictions exhibited enhanced accu- racy, in a typical example case, reducing mean error from 16.7mmHg to 6.8mmHg

Conclusion: A hybrid hardware/software system has been designed which utilises non-invasive continuous measurement of arterial dialysis line pressure to improve intermittent arm cuff measurements in order by identifying unreliable arm cuff measurements. The expert system com- putational core showed robust operation in accepting or excluding incoming arm cuff measurements. The devised system can support two requirements in future applica- tions. Firstly, offering a repeatable and robust methodology to identify unreliable arm cuff measurements. Secondly to support the development of reliable SBP prediction algo- rithms to enable early intervention to predict hypotensive episodes and enable early intervention to prevent intradia- lytic hypotension.

Abbreviations: cardiovascular disease (CVD); end-stage kidney disease (EKD); intradialytic hypotension (IDH); blood pressure (BP), systolic blood pressure (SBP), di- astolic blood pressure (DBP), Hemodialysis (HD), expert system (ES)

Index Terms—blood pressure measurement, dialysis, expert system, hypotension, measurement accuracy, real- time