Pilot Study of a Wearable Activity Monitor During Head and Neck Radiotherapy to Predict Clinical Outcomes

PURPOSE

Given the established associations between performance status and survival in a variety of cancers, there is significant interest in using a biometric wearable device (WD) to predict outcomes in the oncology population. In this pilot study, we investigated the ability of a WD to predict meaningful clinical end points in patients undergoing head and neck radiotherapy.

METHODS

Patients receiving head and neck definitive chemoradiotherapy or postoperative radiotherapy/chemoradiotherapy were enrolled in this pilot study, designed to show 90% compliance with using the device. Individuals were asked to wear the WD for 23 hours a day, and hospital admissions, pain medication usage, and FACT-G quality-of-life (QoL) score were prospectively recorded.

RESULTS

Fifty-one patients were enrolled and started using the WD, but eight patients stopped wearing it, resulting in a compliance probability of only 84%. There were 15 hospital admissions, 13 of which were planned for feeding tube placement. There was no step count threshold that predicted the need for admission or more pain medications. However, among the 25 patients with a significant reduction in FACT-G score, the average reductions in daily steps during the week and weekend before the decline were 988 (P = .005) and 1,311 (P = .018), respectively, and the odds of a QoL reduction were more than 4-fold higher among patients experiencing a week-to-week reduction of at least 1,000 daily steps. There was no association between heart rate and any end point.

CONCLUSION

Although not meeting the compliance goal, the majority of patients did use the WD. The WD signal could not identify patients requiring hospitalization or significantly more pain medication, but the finding of reduced step counts before a significant reduction in QoL is provocative.

INTRODUCTION

Head and neck radiotherapy and chemoradiotherapy are associated with a constellation of acute morbidities during treatment, ranging from pain and mucositis to dysphagia, dehydration, and aspiration. This side-effect profile often leads to hospital admission, with two studies finding that more than one third of all head and neck radiotherapy patients experienced an unplanned hospital encounter during treatment.^1,2 Such health care utilization is costly and, more importantly, implies that the patient is at significant risk for worsening complications. Ideally, clinicians would have an indicator of deterioration before clinical decompensation in this population of patients.

CONTEXT

• Key Objective

• Given the significant morbidity of radiotherapy for head and neck cancer, it is important to define predictors of subsequent hospitalization, pain, or declining quality of life. Wearable devices (WDs) that track step counts and heart rate may generate a signal that identifies patients who need more urgent care.

• Knowledge Generated

• This prospective study did show that the majority of patients were able to use the WD over the course of treatment, but neither step counts nor heart rate could predict whether patients were at risk for hospitalization or an increase in pain medication. Patients experiencing larger reductions in their daily step counts were more likely to report worsening quality of life.

• Relevance

• Although WDs provide potentially useful information on activity, their utility in radiotherapy patients remains to be seen. Larger studies are necessary to better define if and how they may have a role in on-treatment and/or surveillance care.

Noninvasive and wearable fitness trackers are now a ubiquitous consumer item with the ability to precisely monitor step counts, heart rate, and sleep activity. These devices have been previously used to assess and monitor patients on radiation treatment, with many studies assessing their utility. Previously, Dr Ohri et al³ performed a prospective study of step monitoring of 38 patients receiving chemoradiotherapy, finding that increasing step counts were inversely proportional to the risk of hospitalization. Moreover, a decrease in step counts over a radiotherapy course was associated with decreasing quality of life (QoL).

In this study, we performed a prospective study of the use of a commercially available, Fitbit 2 wearable device (WD) in patients undergoing head and neck radiotherapy. The goal of this study was to determine whether practically assessable changes in step count and/or heart rate were associated with hospital admission or emergency room visits, a change in pain medication requirement, and/or a change in QoL.

METHODS

UT Southwestern Medical Center's Internal Review Board (ClinicalTrials.gov NCT03574870) approved this prospective study. Enrollment of study subjects took place from April 2017 through November 2018, and included patients treated at UT Southwestern Medical Center and Parkland Hospital. All patients signed informed consent. Eligible patients were required to be treated with head and neck definitive chemoradiotherapy or postoperative radiotherapy or chemoradiotherapy. Patients were included in this study if they had an Eastern Cooperative Oncology Group performance status of at least 2 and were willing to wear the WD and download the accompanying application to their smartphone.

Device

Patients were distributed a WD at least 1 week before the treatment start date, with the device being synchronized with the patient smartphone and Fitabase API (purchased from the company). Patients were instructed to wear the device for 23 hours a day, taking it off only for personal hygiene. The study research coordinator ensured proper synchronization with the Fitabase application during the weekly on-treatment visits.

Activity Metrics

Each device recorded step count, resting heart rate, and sleep habits—total minutes asleep, total minutes in different stages of sleep, and total time in bed. The weekly averages of each measure, as well as the difference in these counts between weeks were calculated. In addition, we also evaluated relative change in step counts in comparison with the baseline measurement, in the event that the patient's baseline activity informed the impact of a subsequent change.

End points

The original end point of this pilot study was to assess the feasibility of wearing the WD for 23 hours each day, with an aim of 90% compliance. This threshold was chosen empirically as we believed that the majority of patients should be willing to use the WD to invest further energy in identifying a signal and deploying it more routinely in clinic. Unfortunately, Fitabase did not report hourly data for steps or heart rate, and thus, we were not able to properly assess hourly end points. We were able to report the number of patients who were willing to continue wearing the device versus withdrawing because of inconvenience or discomfort.

Clinical end points were decided a priori to include hospitalizations and ER encounters, utilization of pain medication using the Medication Quantification Scale (MQS) Version III, and change in QoL as measured by the FACT-G. A reduction in total FACT-G score of 10 points (equivalent to approximately 10% of the baseline score) was considered a minimally important difference.⁴

Statistical Analysis

The majority of hospitalizations were for gastrostomy insertion, a procedure often planned several days in advance. Decisions to place the gastrostomy tube, as well as readjustment of pain medications and QoL assessments, were almost universally performed during the weekly on-treatment visit, which occurred the same day each week. As a result, the specific day of any significant change in an end point was more a function of scheduling rather than the patient's activity a set number of days before that date. Therefore, we made the decision to calculate these measures according to the data from the previous calendar week (previous Monday through Sunday) and/or weekend. For patients who required reactive gastrostomy tube placement, the week of the decision to insert the tube was used as the time of the event. The goal was to identify an activity signature that could be identified at the very beginning of the week that could predict for an adverse outcome over the following several days.

We developed two primary predictors: change in daily steps between the prior week and the week before it, and the change in daily steps between the previous weekend and the weekend before it.

RESULTS

Feasibility and Patient Cohort

A total of 51 patients were enrolled in this study and agreed to use the WD, but eight patients (16%) ultimately refused to wear the device. Since we defined general feasibility as the retention of enrolled subjects at a probability > 90%, we concluded that using such a fitness tracker was not routinely feasible.

Of the eight patients who did not complete the study, two patients complained of skin irritation; one of them stopped wearing the device before treatment started and the other during the third week. One patient stopped wearing the device during the fourth week without an obvious reason beyond she did not like it. One patient did not start wearing it until the third week and then only wore it for 2 additional weeks. Four patients refused to wear the WD before starting radiotherapy without a stated reason, despite informed consent and trial enrollment. Therefore, 96% of patients (43 of 45) who started radiotherapy with the device wore it over the course of treatment. There were no obvious characteristics of these noncompliant patients that separated them from the remaining 43. Notably, five of the eight patients (63%) who ultimately refused the WD were admitted for failure-to-thrive/malnutrition during treatment.

Of the remaining 43 patients, the tracker was used an average 88.1% (standard deviation [SD] 37.4%) of all potential days. On average, patients wore the device for 71.8 days (SD 21.7 days) and averaged 3,765.0 steps (SD 1,637.4 steps) per day over the course of the study.

The patient cohort is given in Table 1. The median age of the population was 61 years, with predominantly male gender (86%). Slightly more than half (51%) of all patients had oropharynx cancer, and most individuals were treated with definitive intent.

TABLE 1.

Patient and Treatment Characteristics

End points

Fifteen patients (35% of followed patients) were admitted to the hospital or seen in the ER, with the majority (n = 13) of the encounters because of malnutrition and the need for a gastrostomy tube. One patient was admitted with an infection and one with a metabolic derangement.

The mean FACT-G score during the first week of treatment was 85.9 (SD 14.0), and the mean FACT-G during the last week of treatment was 76.6 (SD 18.7). The average difference from the first to last week of treatment was –9.5 (SD 16.2, P = .002). Twenty-five patients (58%) experienced a drop of at least 10 points from the first week until the last. The mean FACT-G scores at 1 and 3 months following treatment were 85.9 (19.3) and 90.5 (14.1), respectively.

The median maximum increase in the week-to-week Medication Quantification Scale was 14.5 (interquartile range, 11-23.7).

Step Counts

Figure 1A shows the change in weekly step counts over the course of treatment for all patients, and Figure 1B shows those weekly step counts as a function of any admissions over the course of treatment. Figures 2A and 2B show the same data during the weekends. In comparison with the week before the decision was made for hospital admission, the patients' daily average reductions in weekly and weekend steps were 630 (P = .04) and 1,082 (P = .05) steps, respectively. By contrast, there was no statistical difference between any 2 successive weeks when a patient was not admitted, and there was no weekend in which the daily step count was statistically lower when a patient was not admitted. Figure 3 shows the average difference step counts between 2 successive weekday and weekends with and without an admission.

FIG 1.

(A) Average daily weekday steps over the course of treatment. Error bars reflect standard deviation. (B) Average daily weekday steps for patients as a function of whether they were admitted during treatment.

FIG 2.

(A) Average daily weekend steps over the course of treatment. Error bars reflect standard deviation. (B) Average daily weekend steps for patients as a function of whether they were admitted during treatment.

FIG 3.

Average reduction in daily steps (weekday or weekend) before weeks with or without an admission. For not admitted weeks, the weekly difference was averaged across all patients and all weeks in which there was not an admission. SD, standard deviation.

By contrast, there was no difference in admission risk between patients who experienced at least one 1,000 average step difference between any 2 weeks or weekends, nor was there any difference in the risk of admission between patients who were in the upper half of the largest between-week (1,587 steps) or between-weekend (2,057 steps) decreases. Similarly, there was no significant difference in the risk of admission for those in the upper half of patients experiencing the largest week-to-week (< 62% of baseline) or weekend-to-weekend (< 34% of baseline) decline as a percent of their baseline activity. Baseline activity level was not related to a subsequent risk of admission, as there was no difference in admission among patients in the upper half (4,050 daily steps) or upper quartile (5,955 daily steps) of average daily steps at baseline. Moreover, there was no statistical difference in the odds of a hospital encounter among patients experiencing larger relative changes in weekday (62% or greater) or weekend (34% or greater) steps from baseline.

Among the 25 patients with a significant reduction in FACT-G score at some point during treatment, the average reduction in daily steps during the week before the significant decline in QoL was 988 (SD 1,573, P = .005), and the average reduction in daily steps in the weekend before the decline was 1,311 (SD 2,392, P = .018). There were no significant differences in the week-to-week step counts among patients who did not experience a significant QoL decline. More importantly, the odds of a significant QoL decline were significantly greater among patients in the upper half of maximum weekend step reductions (odds ratio 5.53; 95% CI, 1.46 to 20.90; P < .01) and among patients experiencing at least 1 week of a reduction in 1,000 daily steps from the previous week (odds ratio, 4.20; 95% CI, 1.02 to 17.32; P = .040). On the other hand, there was no statistical difference in the odds of developing a significant reduction in QoL among those in the upper half of weekday or weekend step reductions relative to their baseline activity. There were no correlations between the FACT-G scores at 1 and 3 months and the step count during the last week or the extent of the biggest week-to-week declines among patients. By contrast, there were modest correlations between the baseline FACT-G score and the same score at 1 (r = 0.58, P < .001) and 3 (r = 0.47, P = .009) months following treatment.

There was no association with step counts and significant change in the pain medication intensity. When comparing average daily step counts in the week or weekend before a specified increase in MQS score, there was no significant reduction in daily steps when the threshold was the median value of 14.5, 23.7 (as an extreme), or 5 (the average change per week).

Heart Rate

Figure 4A shows the change in the average daily heart rate over the course of treatment, and Figure 4B shows the same information stratified by whether patients were admitted or not. The mean resting heart rate across patients was 72.3 (SD 8.1). There were no associations between any heart rate parameter (i.e. baseline or change in heart rate) and hospital admission, FACT-G score, or pain medication intensity.

FIG 4.

(A) Average weekday daily heart rate over the course of treatment. Error bars reflect standard deviation. (B) Average weekday daily heart rate for patients as a function of whether they were admitted during treatment.

Sleep

Patients averaged 413.9 minutes of sleep per day, or 6.9 hours (SD 1.5 hours). However, the tracker compliance decreased during sleeping hours with patients only wearing the tracker 60.5% (SD 28.4%) of all potential sleep time. Only one patient was 100% compliant with using the device for all the assigned days and hours of sleep. The number of days the patients wore the device during sleep time ranged from 8 to 86 days, with the average being 50 days (SD 21 days).

There was no difference in average nightly sleep before the need for hospital admission or before a significant increase in pain medication. However, patients experiencing a significant reduction in their FACT-G score (in comparison with their baseline level) had significantly less sleep in the prior week (37.3 minutes, SD 79.2 months, P = .039).

Survival

At a median follow-up time of 27.8 months from the start of radiotherapy (interquartile range, 21.7-35 months), the 2-year overall survival probability is 84%, and the median has not been reached. There were no associations between overall survival between patients in the upper half of activity in either the first or last week of treatment, nor was there a survival difference in patients who experienced a large week-to-week reduction in their step count. There were no survival differences between patients with large declines in their FACT-G score over treatment, nor was the 1- or 3-month FACT-G score predictive of survival.

DISCUSSION

In this study, we were not able to identify a biometric indicator that could reasonably identify a patient at risk for hospital admission or need for an increase in pain medication. Although there was a reduction in step counts in the week before the decision for a hospital admission, there was no increased risk of a hospital encounter as a function of the average daily step count since the daily step count naturally went down for all patients over time. The one noteworthy finding was that patients with significantly reduced step count from the previous week were more likely to have a significant reduction in QoL score, and we identified thresholds that may identify this population at greatest risk for a QoL decline.

The primary end point of this pilot study was 90% patient compliance, and per protocol, we did not meet this end point, with only 84% of consented patients using the WD over the course of treatment. This metric does not include whether the device was worn 23 hours per day, and we could not derive this estimate. Only one patient consistently wore the device at night, but patients did have steps recorded 88% of all potential days. Yet, when investigating compliance further, it became clear that the majority of patients (96%) who started using the device continued wearing it. Although our target of 90% may have been too high in retrospect, 84% is likely still sufficient to warrant additional study in the on-treatment or potentially post-treatment setting. Indeed, patients who refused to wear the device after consent appear to be a high-risk cohort with a 63% admission risk, and refusal behavior itself may indicate the need for close on-treatment symptomatic management.

It is recognized that head and neck radiotherapy can cause significant acute toxicity over the course of treatment, leading to hospital admission in approximately one third of cases.^1,2 Patients may require inpatient management for malnutrition and enteral feeding, pain control, infection, or the general multifactorial failure-to-thrive diagnosis. In theory, identifying patients at risk for such complications before their acute presentation should improve health outcomes and moderate the costs of management.

For example, routine collection of patient-reported outcomes has been shown to improve overall survival in a prospective study of patients receiving palliative chemotherapy. In a landmark randomized study of patient-reported outcome measurement in patients receiving outpatient chemotherapy, the use of routine symptom reporting improved QoL, admission risk, and overall survival,^5,6 perhaps either because of prompt management of potentially life-threatening symptoms or prolongation of systemic treatment. A similar trial in France restricted to patients with lung cancer found similar results with respect to overall survival, in part because patients' performance status was improved on detection of recurrence.⁷

In principle, easy-to-wear activity trackers may overcome any limitations in reporting compliance by fundamentally assessing the impact of any symptom on their mobility or heart rate. Changes in step count or heart rate may provide an early trigger to the physician team of patient decompensation which, similar to the patient-reported outcome studies, may allow for an early and successful intervention. Indeed, in a study using a wearable fitness tracker for 38 patients receiving chemoradiotherapy, Ohri et al³ showed that an increase of 1,000 steps (averaged over a 3-day period) was associated with a significantly reduced risk of hospitalization. An additional analysis from this study among patients with lung cancer showed that baseline activity level predicted subsequent hospitalization, treatment break, and death.⁸ These promising data have led to a randomized study by NRG of using a fitness tracker in patients receiving chemoradiotherapy for locally advanced lung cancer, with the primary objective to reduce adverse clinical events during treatment (NCT04878952). In a very small study of 12 patients receiving palliative radiotherapy, there was a correlation between activity before radiotherapy and survival.⁹

Other studies of wearable technology have been less promising when analyzing step counts. Nilanon et al¹⁰ studied the use of an activity tracker in patients receiving outpatient chemotherapy, and only 68% of all patients were compliant with wearing the device, and there was no correlation between step counts and unplanned health encounters. However, these authors did find a significant correlation between fewer nonsedentary hours and a reduced incidence of health encounters, suggesting that impaired activity level is associated with increased care utilization. The remaining question may be how such activity is parameterized and whether step counts are a sufficient summary measure to identify the at-risk population. In a heterogeneous prospective study of patients with colorectal cancer either undergoing colectomy or starting chemotherapy, there was no significant association between change in step counts in the pretreatment or post-treatment periods and the development of protocol-specified toxicity.¹¹

In considering possible explanation for our negative results, the overall very small number of acute hospital admissions was, of course, problematic. Most encounters were planned, minimizing any acute decompensation one might see reflected in step counts. With a much larger study, we may have seen more unexpected hospital admissions and been able to discern a biometric signal. In addition, a larger cohort would have provided more data to analyze the QoL results and perform multivariable analyses. Moreover, as opposed to outpatient chemotherapy monitoring, radiotherapy patients are typically seen daily, and the treating team may identify at-risk patients (with subsequent management) before seeing a change in their activity level.

This latter reality does highlight the key conclusions from this study. Patients are in clinic every day, so if a WD is going to improve patient management, it needs to be able to predict adverse outcomes before the patient presenting with a complaint. Unfortunately, we did not see an indicator that changes in step counts or pulse over time could identify patients who may decompensate in the coming days. Activity declined for most patients, but most patients did not require hospitalization. Most patients experienced progressive increases in pain medication, but bigger reductions in activity did not predict which patients needed more aggressive pain control.

That said, patients who were taking fewer steps did report greater reductions in their QoL, and we found that a simple cut point (average reduction in 1,000 daily steps from the previous week) was sufficient to highlight this at-risk cohort. Since pretreatment and post-treatment QoL outcomes do predict for survival following head and neck chemoradiotherapy,^12-14 there may be a place for WD to monitor patients, especially after treatment when they are not interacting with the health care system on a daily basis. Indeed, remote monitoring of patient activity may work in concert with telemedicine to manage patients after radiotherapy in a cost-effective manner.

In conclusion, although we did not meet our prespecified goal of 90% compliance, our 84% outcome (and 96% among those who started wearing the device during treatment) suggests that this tool may be used by the clear majority of the patient population. This study did not find strong evidence that an activity tracker could facilitate an early intervention for patients at risk for a hospital encounter or pain during head and neck radiotherapy, but the relationship between activity change and QoL is provocative. Overall, these results should be considered in how to evaluate the potential utility of a WD in a larger number of patients, including a more nuanced approach to the metric reflecting activity and defining outcomes beyond hospital encounters.

AUTHOR CONTRIBUTIONS

Conception and design: David J. Sher, Steve Jiang, Dat Vo, Baran D. Sumer

Financial support: Steve Jiang

Provision of study materials or patients: Nhat-Long Pham

Collection and assembly of data: David J. Sher, Nhat-Long Pham, Jennifer L. Shah, Dat Vo, Andrew T. Day

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/cci/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).