Timely recognition of deterioration in hospitalized patients is important because early intervention improves clinical outcomes of mortality and unplanned intensive care unit (ICU) admissions and reduces length of stay [1]. Especially in perioperative care, complications related to surgery limit effectiveness of the surgery and are associated with increased mortality and costs [2,3]. From previous studies, it is known that vital signs such as heart rate (HR) and respiratory rate are important indicators of critical illness and are often altered long before a deterioration is clinically apparent [4-6]. In general, patients’ vital signs are assessed multiple times a day in general wards. However, patients may deteriorate between the scheduled measurements [1]. Therefore, both remote and continuous monitoring of HR and respiratory rate is considered a promising tool for early detection of patient deterioration in the low acuity or home setting.

The gold standard for measurement of HR in the perioperative setting is the multiple-lead electrocardiogram (ECG). However, there are practical limitations to continuous measurements of vital signs using ECG due to the obtrusiveness and limited mobility of patients. Novel solutions to monitor vital signs have been proposed in the literature [7]. One of these novel solutions is the wrist-based optical heart rate monitor (OHRM). The OHRM has the advantage of offering unobtrusive, remote, and continuous monitoring. The photopletysmography (PPG) sensor in the OHRM has shown potential to provide robust peak detection from which HR may be calculated [8,9]. Validation studies have been presented on the accuracy of these devices in healthy participants [10-17]. However, it remains unclear whether these tools are also reliable for monitoring vital signs in patients during hospital stay. The robustness of an OHRM should be studied in hospitalized patients before it can be reliably adopted in a clinical setting. Few studies have been performed in hospitalized patients, and these included mainly stable ward patients [13,18,19]. To check the accuracy of the OHRM in the acute phases of disease, the study population should ideally experience some deterioration in HR during the study period. Hospitalized patients are a heterogeneous population where HR can be influenced by all kinds of pathologies, particularly during surgery, which induces hemodynamic, metabolic, endocrine, and immunological alterations [20,21]. The objective of this study was to assess the agreement between the HR extracted from a PPG sensor–based OHRM and that of the gold standard 5-lead ECG connected to the patient monitor during surgery and recovery.

A wrist-worn OHRM may be able to provide continuous unobtrusive HR monitoring in the low acuity care or home settings. To determine this, the validity of OHRM-derived HR must first be assessed in a representative target population and compared to the gold standard 5-lead ECG. In this study, the agreement between the HR derived from an OHRM and a 5-lead ECG connected to a patient monitor was assessed for a representative patient population during the perioperative period. The OHRM could provide an accurate HR (–5 bpm to 5 bpm and –10% to 10% compared to the ECG-derived HR) during both the surgical and recovery phase when the PPG signal was of good quality. A vast majority (121.7/159.1 hours, 76.5%) of the PPG signal was good quality.

Given the hemodynamic changes during the perioperative period and the diversity in surgical procedures, a technical validation, as performed in this study, is essential before the OHRM can be introduced into clinical practice. Very few studies were found in the literature that validated wrist-worn OHRMs in hospitalized patients. One study with a goal of early warning detection using an OHRM was performed in patients during and after discharge from the ICU [13]. The OHRM was a personal fitness tracker, and 24 hours of monitoring started in the ICU while patients were still being monitored by means of a continuous ECG. The authors concluded that personal fitness tracker–derived HRs were slightly lower than those derived from continuous ECG monitoring and not as accurate as pulse oximetry-derived HRs. A feasibility study was performed by the same research group regarding bradycardia and tachycardia detection in the same population [18]. The authors stressed in both studies the importance of subgroup analysis of patients not in sinus rhythm since this negatively impacted measurement accuracy. This corresponds to the findings in our study where measurements during arrhythmia were of low quality.

Another study was designed for AF detection, but also showed good results in sinus rhythm in patients undergoing elective cardioversion for AF [31]. There were fewer patients (N=20) included than in our study, and the agreement analysis was based on QRS intervals as the reference, with a mean difference of 1.3 ms being found between ECG and PPG. Other studies were performed in healthy participants and focused on assessing accuracy during physical activity [11,12,14,16,17,32-34]. However, the results obtained in these studies cannot be translated to our results since surgery was the underlying cause for changes in HR in our study and not physical activity. Factors influencing HR during surgery are hemodynamic changes induced by anesthesia, intraoperative factors such as blood loss and hypothermia, or involvement of vital organs in the area of surgery. Results of previous studies did conclude that motion artifacts remain a challenge in OHRMs. In this study, motion artifacts were less likely to occur since patients were mostly immobilized. Nevertheless, motion artifacts are relevant to consider if the OHRM is to be used in the future for remote monitoring of patients.

The agreement between the ECG- and PPG-derived HR was within the LoA of –5 bpm to 5 bpm and –10% to 10% (whichever was largest) both during surgery and recovery. However, this only applied when the quality of the PPG signal was labeled as “good.” Moreover, a vast majority (during surgery: 121.7/159.1 hours; during recovery: 76.5%; 83.8/112.6, 74.4%) of the PPG signal was good quality. Ideally, the coverage should be 100%, but this may not be realistic since a poor signal-to-noise ratio in the PPG measurements can perturb the detection of a sinus rhythm. Arrhythmias such as ectopic beats, AF, premature ventricular or atrial complex, and paced beats also contributed to a reduction of measurement coverage of the OHRM. This is confirmed by the fact that patients with a medical history of AF had lower overall coverage compared to patients without previous diagnosis of AF, resulting in 25% versus 85% overall coverage, respectively. This was also true for those patients with severe congenital heart disease where median coverage was 47% versus 85% for patients without structural heart disease. Finally, a very small group of patients had an extremely low coverage, but a consequently large influence on the mean coverage. Median coverage was higher, with 85% being good quality data. Exclusions of patients in this study should be taken into account as well when clinical applicability of the OHRM is assessed. Furthermore, 3 patients were excluded since <5 minutes of data were captured in total, which could be explained in 1 case by serious congenital heart disease which involved aberrant anatomy. Another 12 patients with <5 minutes of good quality data during surgery or recovery were also excluded. As the reference standard, ECG is considered capable of providing 100% coverage. However, in clinical practice, this is most likely not the case since ECG HR detection can also fail in the presence of the aforementioned abnormalities.

The limitations of this study are the following. Despite a heterogeneous group of elective procedures and hospital setting, no general ward patients were included. Nevertheless, translation of our findings to patients in the general ward is reasonable as patients are transitioning from immobile to a more moveable state during stay in the recovery room. By using a 1-way random-effects model, the between- and within-patient variance was quantified to explore the effect of heterogeneity of the study group. As indicated by Hamilton and Lewis [35], not accounting for repeated measures can lead to a falsely narrow LoA, mainly with a small number of patients and a large number of measurements per patient. Both the mean bias and between-patient variance are weighted according to the number of observations available for each patient. Hence, patients with more observations will contribute more to the final results. As the distribution of observation times was not normal, some patients contributed substantially more than did others, and results could have been biased to these patients. It is also worth mentioning the assumptions underlying the 1-way random-effects model. Specifically, the model assumes that repeated differences on a single patient are independent and that the within-patient variance of these differences is constant and the same for all patients. First, the independence assumption could have been too strong since hemodynamic changes occurred during surgery or recovery which could have led to autocorrelation in the HR and subsequent differences arising between the PPG- and ECG-derived HR. The effect of autocorrelation on the within-patient variance is unknown, and further studies are needed to take autocorrelation into account [28]. Second, the assumption of homoscedasticity was not formally tested, and it could have been the case that the variance of the differences increased with higher HR. Finally, the possible influence of surgery-specific factors, such as electrosurgical instruments causing interference was not investigated.

With this study in hospitalized patients, we gained knowledge on the influence of the oscillometric blood pressure cuff which disturbs the measurements by compromising the blood flow. Although both nurses and patients experience the wireless and unobtrusive wristband as an advantage, the OHRM will still have to find a way into local workflow. Before early warning systems can be incorporated into timely detect deterioration, clinical studies to define the predictive value of continuous HR monitoring for deterioration in hospitalized patients other than the ICU or operating room are first needed. Although our own and previous studies might not have found equal accuracy compared to the gold standard, there is still more opportunity to produce benefit during the acute phases of illness, which otherwise may go unnoticed in the general ward with monitoring of the HR 2 to 3 times daily. Although the use of an ORHM for deterioration detection seems less time-consuming for nurses, it still remains uncertain what the false alarm rate will be or how much time practical procedures, such as charging the battery of the OHRM, will take. From a practical point of view, placement of the wristband is made problematic by intravenous or arterial lines, identification bands, or bandages on the wrists.

In summary, the current study found that the OHRM is clinically acceptable when good quality data are captured and in settings when high-intensity monitoring, such as in the ICU or operating room, is not mandatory. The OHRM seems less suitable for patients with congenital anatomical changes of the heart or patients with arrhythmias. When the OHRM captures a significant amount of low-quality data in a patient, the suggestion would be to use another monitoring type to ensure safe monitoring. Since the OHRM can report the quality of the PPG signal instantaneously, the decision to switch to ECG monitoring can be made immediately. The reliability of an OHRM to measure HR in patients known to suffer from arrhythmias or structural heart disease requires further research.

In conclusion, this study shows that an OHRM equipped with a PPG sensor can measure HR within the ECG reference standard of –5 bpm to 5 bpm and –10% to 10% in the perioperative setting when the PPG signal is of sufficient quality. This implies that an OHRM can be considered clinically acceptable for HR monitoring in low acuity hospitalized patients and may provide the basis for future studies for remote, unobtrusive, continuous monitoring for timely recognition of deterioration.