Wearable Sensors as a Preoperative Assessment Tool: A Review

Abstract

Surgery is a common first-line treatment for many types of disease, including cancer. Mortality rates after general elective surgery have seen significant decreases whilst postoperative complications remain a frequent occurrence. Preoperative assessment tools are used to support patient risk stratification but do not always provide a precise and accessible assessment. Wearable sensors (WS) provide an accessible alternative that offers continuous monitoring in a non-clinical setting. They have shown consistent uptake across the perioperative period but there has been no review of WS as a preoperative assessment tool. This paper reviews the developments in WS research that have application to the preoperative period. Accelerometers were consistently employed as sensors in research and were frequently combined with photoplethysmography or electrocardiography sensors. Pre-processing methods were discussed and missing data was a common theme; this was dealt with in several ways, commonly by employing an extraction threshold or using imputation techniques. Research rarely processed raw data; commercial devices that employ internal proprietary algorithms with pre-calculated heart rate and step count were most commonly employed limiting further feature extraction. A range of machine learning models were used to predict outcomes including support vector machines, random forests and regression models. No individual model clearly outperformed others. Deep learning proved successful for predicting exercise testing outcomes but only within large sample-size studies. This review outlines the challenges of WS and provides recommendations for future research to develop WS as a viable preoperative assessment tool.

Keywords: exercise testing; perioperative pathway; preoperative assessment; wearable sensors.

Figure 4

(a) Figure to show reference axes in a Tri−axes accelerometer. Presents the axes along which acceleration of movement can be measured across x, y and z. (b) The mechanism for HR detection in a PPG sensor by reflection. The LED can be seen emitting light which is reflected and then detected by the photo-detector and converted into a HR signal. This figure was taken from Moraes et al. (2018) with no changes made, Creative Commons Attribution International 4.0 License [92,93].

Figure 1

Figure to show the stages across the perioperative period. The perioperative pathway refers to the period that spans from the first point at which surgery is considered as a treatment option up until the full recovery [19]. This pathway has several sub-stages [20]. The preoperative period represents the period prior to surgery where any preoperative assessment takes place. The intra-operative period is representative of the period whilst the patient is undergoing treatment. The postoperative period relates to any period immediately following the operation and can continue after patient discharge.

Figure 2

Figure to show common preoperative assessment tools used in practice. The top three boxes present common forms of preoperative assessment that are regularly used in practice (see Section 1.1), whilst the last box with a dashed arrow is included to show the potential for wearable sensors to be used alongside common methods in this context.

Figure 3

Figures for sensor modalities. (a) Shows the percentages of sensor modalities used across research. All research employed accelerometer sensors but a further subsection combine this with either ECG or PPG sensors. (b) Figure to show the variations in locations of sensor types. The common locations for sensors used in research applicable to the preoperative period are outlined.

Figure 5

Figures to show the recordings produced from ECG and PPG recordings. (a) A comparison of the cardiac signals produced from a PPG versus ECG sensor over a period of 2 s. This figure was produced by Elgendi et al. (2019) and was taken from a larger figure with no changes made as part of the Creative Commons Attribution International 4.0 License [92,100]. (b) A segment of an ECG graph that has been portioned to show the stages in a normal cardiac cycle including the P wave, the QRS complex and the T-wave.

Figure 6

Figure to show wearable sensor devices used in research across the body (a) and where these are located (b–f). (b) The Hexoskin smart shirt that collects both ECG and activity data, used with permission from Hexoskin [117]. (c) An ECG wearable device that collects recordings from a single-lead ECG device and 3D-accelerometer data, used with permission from [118]. (d) An upper-arm PPG sensor utilising reflective PPG detection, similar to that used in preoperative monitoring research [119]. The figure is taken as part of a larger figure from Wang et al. (2023), Creative Commons Attribution International 4.0 License [92,120]. (e) The Fitbit Inspire collects a combination of accelerometer and PPG data from the user and is commonly used in preoperative research. The figure is taken from Li et al. (2023), Creative Commons Attribution International 4.0 License [92,121]. (f) The OMROM walking style pedometer that utilises a tri-axis accelerometer to collect step data, used in predicting VO2 max [78]. This figure is taken from Bartlett et al. (2017) as part of a larger figure, Creative Commons Attribution International 4.0 License [92,122].

Figure 7

Venn-diagram to present the common methods for handling missing data from WS. The three techniques identified in the literature for handling the missing periods of data are presented in the Venn diagram. At the intersection between ‘delete’ and ‘tolerate’ the implementation of an extraction threshold was identified to delete data below the threshold and tolerate missing data above the threshold. At the intersection between ‘tolerate’ and ‘impute’, imputation on short-term segments of missing periods was identified as a solution that employs that imputation on select segments.

Figure 8

Figure to show imputation using K-nearest-neighbours. Zhang et al. (2023) utilise the KNN technique to impute on short-term segments of missing data under 10 min in length by utilising previous values from both the step count and heart rate signals to calculate missing values. This figure was produced by Zhang et al. (2023) and was taken from a larger figure but had no changes made, taken as part of the Creative Commons Attribution International 4.0 License [71,92].

Figure 9

Graph to show the implementation of a maxima and minima step-counting algorithm that counts step number based on the number of steps windows detected. Each red line indicates the stopping point of each step window and the start of the next corresponding window, the length of time between each red vertical line indicates the step window size [142]. This figure was produced by Ho N et al. (2016) and was taken with no changes made, used as part of the Creative Commons Attribution International 4.0 License [92,142].

Figure 10

Figure to show the prevalence of each model of analysis across research. In research where multiple models are compared, all models are counted.