Classification of blood pressure in critically ill patients using photoplethysmography and machine learning

https://doi.org/10.1016/j.cmpb.2021.106222Get rights and content

Highlights

  • Estimation of blood pressure values using pulse rate variability features shows promise for the continuous, non-invasive measurement of systolic, diastolic, and mean arterial pressure.

  • Using photoplethysmography-based pulse rate variability features only, it is possible to classify hypertensive events in critically ill subjects with relatively good performance. However, the classification of hypertensive and normotensive events is still a challenge.

  • Using 5-min windows for the classification and estimation of blood pressure in critically ill subjects using solely pulse rate variability features gives a better performance than using 1-min windows.

Abstract

Objective: The aim of this study was to evaluate the capability of features extracted from photoplethysmography (PPG) based Pulse Rate Variability (PRV) to classify hypertensive, normotensive and hypotensive events, and to estimate mean arterial, systolic and diastolic blood pressure in critically ill patients. Methods: Time-domain, frequency-domain and non-linear indices from PRV were extracted from 5-min and 1-min segments obtained from PPG signals. These features were filtered using machine learning algorithms in order to obtain the optimal combination for the classification of hypertensive, hypotensive and normotensive events, and for the estimation of blood pressure. Results: 5-min segments allowed for an improved performance in both classification and estimation tasks. Classification of blood pressure states showed around 70% accuracy and around 75% specificity. The sensitivity, precision and F1 scores were around 50%. In estimating mean arterial, systolic, and diastolic blood pressure, mean absolute errors as low as 2.55 ± 0.78 mmHg, 4.74 ± 2.33 mmHg, and 1.78 ± 0.14 mmHg were obtained, respectively. Bland-Altman analysis and Wilcoxon rank sum tests showed good agreement between real and estimated values, especially for mean and diastolic arterial blood pressures. Conclusion: PRV-based features could be used for the classification of blood pressure states and the estimation of blood pressure values, although including additional features from the PPG waveform could improve the results. Significance: PRV contains information related to blood pressure, which may aid in the continuous, noninvasive, non-intrusive estimation of blood pressure and detection of hypertensive and hypotensive events in critically ill subjects.

Introduction

Blood pressure (BP), the force with which the blood is pumped around the circulatory system [1], is one of the main vital signs measured in clinical and non-clinical environments [2], [3]. It is usually reported as a ratio between the BP during the systole (systolic blood pressure, SBP) and diastole (diastolic blood pressure, DBP) [4]. Ideal BP values are considered to be between 90/60 and 120/80 mmHg, for diastolic and systolic BP respectively, whereas sustained high BP, also called hypertension, and sustained low BP, known as hypotension, are considered to be BP measurements higher than 140/90 mmHg and lower than 90/60 mmHg [1]. Both hypertension and hypotension are abnormal conditions that may affect the blood flow to tissues and, hence, regular BP monitoring is essential for the detection, prevention and treatment of related diseases [2].

Hypotension is related to light-headedness, blurred vision, weakness, confusion, and fainting, but also to pregnancy, medicines, and medical conditions, such as diabetes mellitus [5]. Some kinds of hypotension are now considered clinically significant and recognized as a cause of impairment of quality of life and potentially worse outcomes [6]. Hypertension, on the other hand, is a condition in which the BP values are persistently high, reflecting an increase in the force applied by the heart to pump the blood throughout the circulatory system [7]. Although it usually does not have noticeable symptoms, hypertension increases the risk of various serious disorders, such as heart attacks and strokes [8], [9]. Since most patients are asymptomatic at the early stages of hypertension, it is usually detected only after substantial vascular damage has occurred [4] and when more serious diseases appear, not only in the cardiovascular system but in other vital organs [2], [10], [11].

The measurement of BP has been traditionally performed using invasive, direct methods based on catheters, or non-invasive indirect techniques based on the inflation of a cuff [2]. However, both techniques pose several challenges and limitations. The invasive alternative can only be performed during surgery or in patients in intensive care units and carries with an increased risk of infection [2]. On the other hand, although the cuff-based methods can be applied in any environment, it does not allow for continuous measurements due to the inflation and deflation of the cuff, and having repetitive measures in short periods can be cumbersome and impractical for certain applications, such as in sleep studies [12], [13].

In the last few decades, increased attention has been given to non-invasive, continuous and cuff-less alternatives for the estimation of BP [9], [11], [14]. Most of these novel techniques are based on the analysis of physiological signals, especially photoplethysmography (PPG) [2], [14], [15]. PPG is a non-invasive, optical technique, used for the measurement of blood volume changes in peripheral tissue [16], and due to its widespread use and simplicity, is one of the most commonly used signals in wearable devices [17]. Some proposed strategies for the estimation of BP using PPG are based on machine learning (ML) algorithms and the extraction of PPG features that may reflect BP-related changes in PPG signals [2], [3]. With the increasing availability of data and the development of powerful ML techniques, BP estimation based on this kind of analysis seems like a promising alternative for the ubiquitous, continuous measurement of BP values, as well as for the identification of hypertensive and hypotensive events in a real-time manner.

Blood pressure is primarily regulated by the sympathoadrenal system on a beat-to-beat basis [4] and PPG-based Pulse Rate Variability (PRV) might contain information related to BP that allows for the identification of hypertension and hypotension using ML techniques. PRV refers to the changes in pulse rate over time and has been applied to the indirect assessment of autonomic activity in our previous work [17]. Moreover, PRV has been associated with changes in PTT [18], [19], and the appearance of essential hypertension (i.e. high BP without underlying conditions) has been related with autonomic dysfunction [20]. Therefore, the aim of this study was to evaluate the applicability of PRV for the identification of hypertensive, normotensive and hypotensive events in critically ill patients, and to assess the capability of PRV-based ML algorithms for the estimation of BP values.

Section snippets

Signal acquisition

A subset of 500 records was obtained from the MIMIC-II Waveform Database [21], [22]. The MIMIC-II database is a freely available, de-identified database which contains physiological signal and vital signs time series obtained from critically ill, adult patients receiving treatment in an intensive care unit (ICU). Waveforms available in the records of this database include continuous high-resolution physiological signals and numeric time series of physiological measurements. In the selected

Signal selection and segmentation

The data set was filtered and 230 records with poor-quality ABP signals and duration of less than 5-min were discarded. The signals from the remaining 270 records were used in the subsequent analysis. From these signals, 4937 5-min segments were extracted and merged, of which 54% were labeled as hypertensive, 25% as hypotensive, and the remaining 22% as normotensive events. Similarly, 11417 1-min segments were obtained and merged. From these, 51% corresponded to hypertensive events, 31% to

Discussion

PRV has been used in several applications in recent years, due to its relationship with Heart Rate Variability (HRV) and to the simplicity and ubiquity of PPG. Although it is not yet clear if PRV can be considered as a valid surrogate of HRV, it has been increasingly studied as a marker of several somatic and mental disorders, and showed promise in the identification of cardiovascular changes [17].

In this study, PRV was used for the identification of BP states and the estimation of BP values in

Conclusion

The aim of this study was to evaluate the applicability of PRV-derived features for the classification and estimation of BP values using ML algorithms, using data obtained from critically ill subjects. The results obtained show promise for the use of PRV for both tasks, although several aspects should be further optimised in order to increase performance, especially for the classification of BP states. Specifically, the identification of hypertensive events in critically ill subjects using the

Declaration of Competing Interest

There are no conflicts of interest in this study.

References (53)

  • High blood pressure (hypertension), 2019, (Online), url:...
  • G. Chan et al.

    Multi-Site Photoplethysmography Technology for Blood Pressure Assessment: Challenges and Recommendations

    J Clin Med

    (2019)
  • K. Welykholowa et al.

    Multimodal Photoplethysmography-Based Approaches for Improved Detection of Hypertension

    J Clin Med

    (2020)
  • R. Mukkamala et al.

    Towards Ubiquitous Blood Pressure Monitoring via Pulse Transit Time: Theory and Practice

    IEEE Trans Biomed Eng

    (2015)
  • M. Radha et al.

    Estimating blood pressure trends and the nocturnal dip from photoplethysmography

    Physiol Meas

    (2019)
  • M. Hosanee et al.

    Cuffless Single-Site Photoplethysmography for Blood Pressure Monitoring

    J Clin Med

    (2020)
  • M. Elgendi et al.

    The use of photoplethysmography for assessing hypertension

    npj Digit. Med.

    (2019)
  • J. Allen

    Photoplethysmography and its application in clinical physiological measurement

    Physiol Meas

    (2007)
  • E. Mejía-Mejía et al.

    Pulse rate variability in cardiovascular health: a review on its applications and relationship with heart rate variability

    Physiol Meas

    (2020)
  • E. Gil et al.

    Photoplethysmography pulse rate variability as a surrogate measurement of heart rate variability during non-stationary conditions

    Physiol Meas

    (2010)
  • I. Constant et al.

    Pulse rate variability is not a surrogate for heart rate variability

    Clin Sci

    (1999)
  • A. Johnson

    MIMIC-III, a freely accessible critical care database

    Sci Data

    (2016)
  • A. Goldberger

    PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals

    Circulation

    (2000)
  • Q. Li et al.

    Dynamic time warping and machine learning for signal quality assessment of pulsatile signals

    Physiol Meas

    (2012)
  • M. Elgendi

    Optimal signal quality index for photoplethysmogram signals

    Bioengineering

    (2016)
  • W. Karlen et al.

    Photoplethysmogram signal quality estimation using repeated gaussian filters and cross-correlation

    Physiol Meas

    (2012)
  • Cited by (24)

    View all citing articles on Scopus
    View full text