Comparative effectiveness of remote perioperative telemonitoring in cancer surgery: a randomized trial

Abstract

This randomized trial compared remote perioperative telemonitoring (RPM) care versus surgeon only care in patients with gastrointestinal (GI), genitourinary (GU), or gynecological (GYN) cancers (N = 293). The RPM care arm wore a wristband accelerometer and reported symptoms via a mobile application (app) before surgery and at days 7, 14, 30, 60, 90 post-discharge. Triage nurses telephoned patients when data deviated from predetermined thresholds. Participants in the surgeon only care arm used the device and app but received a standard institutional message when threshold deviations occurred. A 6% greater functional recovery rate was observed for participants in the RPM care arm (p = 0.036). Change in symptom severity scores was statistically significant at day 90 and for symptom interference with daily activities at days 14 and 90, favoring the RPM care arm. The RPM care arm experienced fewer major postoperative complications (p = 0.004). RPM care produced statistically significant benefits in postoperative functional recovery and symptoms. Trial Registration: ClinicalTrials.gov, NCT04596384, 10/22/2020.

Introduction

Over 60% of persons with cancer undergo surgical interventions¹, and surgical interventions account for the highest rate of cures for treatments after a cancer diagnosis². Historically, surgical outcomes are measured by disease- and health systems-related parameters, such as length of hospital stay, morbidity, readmission rates, and mortality^3,4. While important, these measures do not accurately reflect the surgical care experience from a patient’s perspective. Patient-centered outcomes are increasingly being used in routine cancer care as quality and value indicators. However, these outcomes are not adequately understood or routinely integrated into surgical oncology.

Due to changes in healthcare economics and advances in minimally invasive surgical techniques, patients are now discharged earlier after surgery^1,5. Postoperative complications that traditionally arise in the hospital are developing in the community and at home. Delays in communicating critical conditions can escalate problems beyond outpatient care. In critical situations, emergency room visits within the same healthcare system or at a different institution occur, resulting in inefficient and fragmented care.

Wearables and digital patient engagement technology have the potential to transform the current surgical care paradigm^6–10 As a passive sensor, these technologies allow individuals to collect, track and store digital biomarkers that serve as measurable health parameters. They also allow the surgical team to access patient-centered data that can help with care decisions in a timely fashion¹¹. The technology has the following advantages compared to current care delivery models: 1) they are highly scalable; 2) they do not depend on a patient’s cognition, language, or health status; 3) they serve as an efficient and unobtrusive method for monitoring postoperative recovery; and 4) they are deployable in various geographic locations and communities. Importantly, these digital biomarkers have the potential to identify patients who need interventions to optimize recovery and outcomes^12–16.

There is an evolving body of evidence favoring remote telemonitoring as a method to enhance synchronous patient encounters in cancer care delivery, although the majority of the studies thus far are conducted in patients receiving systemic cancer therapy^17,18. A recent systematic review of ePROs in cancer care confirmed that ePRO monitoring has positive clinical impact on outcomes (symptoms, QOL, urgent care visits, readmissions) across diverse cancer settings that cover multiple cancer types and stages of disease (early stage, metastatic)^19,20. Basch and colleagues conducted one of the first symptom monitoring trials in metastatic cancer populations, and found improvements in QOL and reduced ER admissions, favoring the symptom monitoring arm²¹. Patients in the symptom monitoring arm were able to remain on chemotherapy longer and experienced longer survival (75% versus 69% for usual care arm)²¹. Importantly, 63% of all patients reported severe symptoms while on study, highlighting the continued challenges with toxicities during treatment²¹.

The PRO-TECT cluster randomized trial conducted through the Alliance for Clinical Trials found significant improvements in physical function, symptom control and QOL for patients in the PRO group²². In advanced disease settings, asynchronous symptom monitoring with nursing communications improved symptom severity and overall survival^21,23. In adjuvant settings, digital symptom monitoring plus nursing triage communication improved physical well-being for up to 12 weeks during treatment²⁴. In cancer surgery, the current evidence suggests that remote telemonitoring programs are feasible and acceptable, with preliminary evidence pointing to potential efficacy on improving postoperative recovery time and symptoms^12,25. A common limitation of the current ePRO/symptom monitoring literature is a lack of health data captured through wearable devices and evidence in surgical oncology populations¹².

The overall purpose of this study was to conduct a comparative effectiveness, randomized controlled trial of perioperative telemonitoring in English and Spanish-speaking persons with cancer scheduled to undergo major abdominal/pelvic surgery. Specifically, the trial aimed to compare outcomes (functional recovery, symptom severity, symptom interference with daily activities, postoperative complications, hospital readmissions, early study withdrawal) between two comparators: remote perioperative telemonitoring (RPM) care and surgeon only perioperative care.

Results

The trial opened to accrual in 2021 and closed to randomization in 2023 (accrual of 27 months). The CONSORT flow diagram is presented in Fig. 1. A total of 398 participants were consented. Of those consented, 4% (N = 16) were ineligible for randomization. Reasons included late ineligibility due to surgery cancellation, lost to follow-up, and consent withdrawal. A total of 382 participants were randomized, with 189 allocated to RPM care and 193 allocated to surgeon only care. A total of 72 participants (25 RPM, 47 surgeon only) did not complete the study. Reasons included lost to follow-up, too busy or too sick, post-randomization disqualification, and death. A total of 293 participants were included in the final primary analysis.

Fig. 1. CONSORT participant flow diagram.

The flow chart displays the progress of all patients through the trial process, from enrollment and randomization to follow-up.

Sample characteristics

Participant sociodemographic, clinical, and surgical characteristics are presented in Table 1. The overall mean age was 60 years (SD = 11.4). Most participants were White, male, college educated, and 34.7% participants with Hispanic/Latino backgrounds. Only 7.3% needed a study provided iPad for participation and all others used personal devices. Most participants were diagnosed with GU cancers (35.9%), followed by GI (35.1%) and GYN (29.1%) cancers. The most common types of procedures included other/combined procedures, prostatectomy, hysterectomy, colectomy, and cystectomy. Seventy-five percent of participants had a minimally invasive procedure (robotic, laparoscopic). About 32% of participants had a postoperative event. There was a statistically significant difference in types of procedure distribution by arm but no differences in complications associated with the procedures. There were no other differences by arm in sociodemographic and other clinical/surgical characteristics.

Table 1.

Baseline sociodemographic characteristics (N = 382)

Characteristics	Overall (N = 382)	Surgeon only perioperative care (n = 193)	Remote perioperative telemonitoring care (n = 189)	P-value
Gender				0.60
Male	199 (52.1%)	98 (50.8%)	101 (53.4%)
Female	183 (47.9%)	95 (49.2%)	88 (46.6%)
Age, years	60.6 (11.4)	60.9 (11.2)	60.3 (11.7)	0.67
Ethnicity				0.49
Non-Hispanic	245 (65.3%)	128 (67.0%)	117 (63.6%)
Hispanic	130 (34.7%)	63 (33.0%)	67 (36.4%)
(Not reported)	7	2	5
Race				0.63
White	279 (76.2%)	148 (77.9%)	131 (74.4%)
Asian	30 (8.2%)	16 (8.4%)	14 (8.0%)
African American	15 (4.1%)	7 (3.7%)	8 (4.5%)
Native Hawaiian/other Pacific Islander	3 (0.8%)	2 (1.1%)	1 (0.6%)
American Indian/Alaska Native	2 (0.5%)	2 (1.1%)	0 (0.0%)
More than one race	7 (1.9%)	2 (1.1%)	5 (2.8%)
Other	30 (8.2%)	13 (6.8%)	17 (9.7%)
(Not reported)	16	3	13
Education				0.50
High school or less	74 (20.0%)	35 (18.9%)	39 (21.1%)
College/Vocational School	215 (58.1%)	113 (61.1%)	102 (55.1%)
Graduate school	81 (21.9%)	37 (20.0%)	44 (23.8%)
(Not reported)	12	8	4
Relationship status				>0.9
Single	61 (16.1%)	30 (15.6%)	31 (16.5%)
Married/partnered	257 (67.6%)	132 (68.8%)	125 (66.5%)
Separated/divorced/widowed	60 (15.8%)	29 (15.1%)	31 (16.5%)
Other	2 (0.6%)	1 (0.6%)	1 (0.6%)
(Not reported)	2	1	1
Who lives with patients				>0.9
Live alone	39 (10.3%)	20 (10.5%)	19 (10.1%)
Live with family/friend/significant other	340 (89.7%)	171 (89.5%)	169 (89.9%)
(Not reported)	3	2	1
Primary language				0.38
English	339 (89%)	174 (90%)	165 (87%)
Spanish	43 (11%)	19 (9.8%)	24 (13%)
Diagnosis,n = 382				0.84
Gastrointestinal (GI)	134 (35.1%)	65 (33.7%)	69 (36.5%)
Genitourinary (GU)	137 (35.9%)	71 (36.8%)	66 (34.9%)
Gynecologic (GYN)	111 (29.1%)	57 (29.5%)	54 (28.6%)
Neoadjuvant treatments
Preoperative chemotherapy	95 (24.9%)	52 (26.9%)	43 (22.8%)	0.3
Preoperative radiation therapy	20 (5.2%)	10 (5.2%)	10 (5.3%)	0.8
Type of procedure,n = 366				<0.001
Prostatectomy	84 (23.0%)	41 (22.5%)	43 (23.4%)
Hysterectomy	60 (16.4%)	23 (12.6%)	37 (20.1%)
Colectomy	28 (7.7%)	22 (12%)	6 (3.3%)
Cystectomy	16 (4.4%)	15 (8.2%)	1 (0.5%)
Hepatectomy	12 (3.3%)	7 (3.8%)	5 (2.7%)
Nephrectomy	11 (3.0%)	3 (1.6%)	8 (4.3%)
Gastrectomy	7 (1.9%)	4 (2.2%)	3 (1.6%)
Pancreatectomy/Whipple	7 (1.9%)	6 (3.3%)	1 (0.5%)
Esophagectomy	5 (1.4%)	2 (1.1%)	3 (1.6%)
Low anterior resection	4 (1.1%)	3 (1.6%)	1 (0.5%)
Salpingo-oopherectomy	4 (1.1%)	3 (1.6%)	1 (0.5%)
Abdominoperitoneal resection	3 (0.8%)	0 (0%)	3 (1.6%)
Esophagogastrectomy	2 (0.5%)	0 (0%)	2 (1.1%)
Other	123 (33.6%)	53 (29.1%)	70 (38%)
Not applicable due to no surgery (n)	16	11	5
Surgical approach,n = 366
Minimally invasive (Robotic or Lap)	271 (74%)	140 (77%)	131 (71%)	0.2
Open	95 (26%)	42 (23.1%)	53 (28.8%)	0.2
Postoperative events,n = 366				0.4
Yes	113 (31%)	60 (33%)	53 (29%)
No	253 (69%)	122 (67%)	131 (71%)

Primary outcome

The proportion of days with valid accelerometer wear time did not differ by study arm (Wilcoxon rank sum test p-value = 0.57). Supplemental Table 1 presents the parameter estimates from the linear mixed-effects model. These estimates were used to calculate the average percent change from the preoperative baseline for each group at each time point, as illustrated in Fig. 2. In the RPM care (intervention) arm, patients recovered an average of 31.37% of their baseline step count on postoperative day 0, and 68.31% by day 14. For the surgeon only perioperative care (control) arm, patients on average recovered 31.7% of baseline steps at day 0 and 65.1% at day 14. The treatment effect was 1.06; this suggested that the rate of recovery in the RPM care arm is ~6% higher than in the surgeon only perioperative care arm. A significant interaction of time and treatment arm suggested a statistically significant difference in the recovery rate in step count from discharge to day 14 by arm, with a positive difference favoring RPM care (p = 0.036, Supplementary Table 1).

Fig. 2. Estimated percentage of baseline steps for the first two week after discharge.

The yellow line represents the remote perioperative telemonitoring care arm, and the blue line represents the surgeon only perioperative care arm. Estimated percentage were obtained from mixed effect model to predict the recovery percentage of baseline step at each follow-up day, using time period as an independent variable, subject as a random effect, treatment, age, sex and residential setting as fixed-factors and study arm-by-time period interaction term as the hypothesis variable testing the slope differences over time between study arms. Error bars represent 95% CI of the estimates.

Secondary outcomes

Figure 3 illustrates the trajectory of MDASI symptom severity and interference scores over time for both study arms. In both groups, the two scores showed a similar trend to peak on the day of discharge and gradually declined thereafter until day 90. For symptom severity score, we observed a larger decrease in symptom severity in RPM care at day 90 (Supplementary Table 2). For symptom interference with daily activities, statistically significant differences in changes from baseline were observed at day 14 and day 90, favoring RPM care. There were no significant treatment effects by arm for other follow-up timepoints.

Fig. 3. Symptom outcomes (N = 382).

Panel a: Symptom severity over time as measured by the MD Anderson Symptom Inventory (0–10 scale). Yellow represents remote perioperative telemonitoring care arm and blue represents surgeon only perioperative care arm. Boxplots depict the median and interquartile range, with outliers represented as dots. Panel b Symptom interference with activities over time as measured by the MD Anderson Symptom Inventory (0–10 scale). Yellow represents remote perioperative telemonitoring care arm and blue represents surgeon only perioperative care arm. Boxplots depict the median and interquartile range, with outliers represented as dots.

Most participants had minor postoperative complications (as defined by maximum CCI of <15)²⁶, with 9.8% reporting major postoperative complications (Table 2). Four participants in the surgeon only perioperative care arm died before study completion. There was no statistical significance by arm in the difference of major postoperative complications. Post-hoc exploratory analysis using linear regression (difference in the continuous CCI by treatment arm) revealed that the average maximum CCI score reported by RPM care was 0.55 times that reported by patients in the surgeon only perioperative care; this difference was statistically significant (p = 0.004), suggesting that the RPM care arm experienced fewer major complications. For readmissions, 14% of participants overall had a readmission after surgery (Supplementary Table 3). The most common reasons for readmission included infections/sepsis (2%), uncontrolled pain (1.2%), and other (1.7%). There were no statistically significant differences by arm for readmissions.

Table 2.

Comparing Maximum Comprehensive Complications Index (CCI) score for 30 days post-discharge (N = 382)

	Overall (N = 382)	Surgeon only perioperative care (n = 193)	Remote perioperative telemonitoring care (n = 189)	Difference (95% CI)	P-value
CCI (binary), n = 360					0.2
Major (maximum CCI > = 15)	35 (9.7%)	21 (11.7%)	14 (7.7%)	0.63 (0.31, 1.28)
Minor (maximum CCI < 15)	325 (90.3%)	158 (88.3%)	167 (92.3%)
Missing	22	14	8
Maximum CCI (continuous), n = 356	3.9 (12.79)	5.7 (16.59)	2.2 (6.93)		0.005
Missing	22	14	8

Early withdrawal occurred in 20.9% of participants overall (14.8% for RPM care, 26.9% for surgeon only perioperative care). This difference was statistically significant (p = 0.004), with more participants in the surgeon only perioperative care comparator experiencing early withdrawal. Overall, most alerts were generated at Day 7 post-discharge (86% of total alerts, N = 228) and Day 14 Post-Discharge (73.2%, N = 202). For symptom alerts, the most common symptoms that triggered an alert included fatigue, pain, sleep disturbance, and distress. For RPM care, 659/684 of total triggered alerts (96.5%) received a triage nurse encounter. Average length of nurse encounters was 11.8 min (SD 9.5).

Discussion

In this study comparing RPM care and surgeon only perioperative care, the results suggest that patients that received RPM care experienced a statistically significant 6% greater functional recovery rate at 14 days after surgery. The improvements on functional recovery aligns with findings from the literature, suggesting that telemonitoring results in significant improvements of postoperative recovery time¹² and use of accelerometer parameters (e.g. daily steps, level of physical activity) as potential predictors of clinical deterioration and healthcare resource use (e.g. hospital readmissions)^8,15,27–29. Population-based cohort study findings also suggest that there is a positive association between wearable activity trackers and physical activity in patients with cancer, supporting the promise of sensor-based devices on promoting participation in physical activity³⁰. The study’s observed recovery rate was not clinically meaningful based on the a priori rate of at least a 14% difference. Increases in daily steps have been shown in the literature to improve outcomes postoperatively, although this area of research is still nascent in surgical oncology¹⁵. The first two weeks following hospital discharge are often the most troublesome for patients and families, as transitions from hospital to home are recent. Thus, the time between discharge and day 14 post-discharge is important from a surgical care decision-making perspective. It is important to note that because participants in both arms could see their daily steps, we cannot eliminate the possibility that just having the device on and seeing daily steps acted as an intervention in and of itself. The symptom intensity and symptom interference with daily activities findings did not meet clinical significance, based on published minimal clinically important difference (MCID)³¹. Overall, the rate of postoperative complications were low in both arms, and multiple factors including surgeon technique (75% minimally invasive procedures) and tumor burden could impact postop complications and may not be modifiable by symptom and functional recovery monitoring alone.

Overall, the study findings and process experiences suggest that while RPM care has potential to improve outcomes and facilitate cancer care delivery, a thoughtful process on program development, technology selection, and clinical workflows are necessary for successful integration into routine care^32,33. Our study design of having participants in both comparators use the technology was selected to truly focus on what we believed as the “active ingredient” of nurse-driven action on the data and not the technology in and of itself nor the passive collection of data. While our findings on early study withdrawal suggest that RPM care is not overly burdensome for patients, future studies should consider the length of device wear, the frequency of electronic symptom capture, blinding of the control arm to daily steps data, and monitoring of other relevant and important symptoms such as sleep. Evidence suggests that patient’s self-efficacy related to symptom management was associated with patient readiness for RPM use^29,34, pointing to the importance of including self-efficacy as a component of RPM programs. Finally, the practicalities and potential burden on nursing staff and workflows should be considered^35,36. In this study, the triage nurses were only alerted by the system when thresholds deviated; upon receiving the alerts, the nurses contacted the patient by telephone for further assessment and management.

An important challenge was the amount of missing data over time, especially for the primary endpoint. This can bias results as patients with missing steps data tended to have higher symptom severity, suggesting that the mechanism of missingness might not be missing at random. Additionally, the imbalance in missing values by arm (higher in surgeon only care) may have impacted findings. The inclusion of three cancer types (GI, GU, GYN malignancies) while being cancers that originate in the abdominal/pelvic regions, may require different surgical procedures with varied postoperative complications that impacted study findings. Finally, the study was initiated during the height of the COVID-19 pandemic, and it is not feasible to fully determine its impact on outcomes observed and study withdrawal rates if conducted outside of the pandemic.

RPM care resulted in some statistically significant but non-clinically meaningful benefits in postoperative functional recovery and symptoms. Future wearable sensor-based research should determine whether a specific range of daily steps are needed pre-and post-operatively to prevent catastrophic postoperative complications and events (including readmissions). More research should determine the best range of postoperative monitoring for specific types of surgical procedures by type of cancer. More intensive interventions that are multimodal (e.g. exercise, nutrition) may augment remote telemonitoring programs in promoting symptom management self-efficacy and improve outcomes.

Methods

The primary endpoint of the trial was the percent change from preoperative baseline in accelerometer-measured daily step count during the first 14 days post-hospital discharge. We selected daily step count as a primary endpoint based on previous evidence suggesting the association of postop functional status with postop outcomes and relevance/importance of maintaining functional status postoperatively from the patient’s perspective^37–39. Secondary endpoints included change from baseline scores for symptom severity and symptom interference with daily activities, maximum postoperative complications during the first 30 days post-discharge, time to hospital readmissions, and time-to-early study withdrawal. We hypothesized that remote perioperative telemonitoring care would improve functional recovery, symptoms, postoperative complications, reduce hospital readmission, and not be excessively burdensome for patients and families. The study team included two patient and one family caregiver partner. The partners were formal paid consultants for the study, and provided feedback on study/intervention design initially and participated in data interpretation. During the three year duration of the study, no important changes were made following trial commencement. Benefits and harms were monitored per the institutional Data Safety and Monitoring Board (DSMB) and reviewed annually. There were no harms reported for the duration of the study.

Participants and settings

The study inclusion criteria included the following: 1) patients scheduled to undergo major abdominal/pelvic surgery for the treatment of GI, GU, GYN malignancies; 2) aged 18 years and older; and 3) ability to read and understand English or Spanish. The trial was conducted at a National Cancer Institute (NCI)-designated comprehensive cancer center and three associated clinical practice sites in the Greater Los Angeles area (Trial registration and protocol: ClinicalTrials.gov, NCT04596384, 10/22/2020). Patients were screened and identified by research staff, surgeons, and surgical care teams. Eligible participants were met by site research staff at regular clinic visits and consented for participation. The study protocol, study related documents (including informed consent) and all study procedures were approved by the Cancer Center’s Institutional Review Board, and all participants provided signed informed consent.

Pre-randomization procedures

Research staff assessed consented participants on technology knowledge and preference, including whether the patient had a digital health engagement device and comfort with mobile applications. Patients that had a consistent digital health engagement device used their personal devices for the study. Patients without a digital health engagement device (e.g. smart phones, tablets, laptop) were provided with a study tablet. The tablets were returned to the study team following study completion.

Randomization and assignment

Following completion of baseline assessments, participants were randomized (1:1) to either receive RPM care or surgeon only perioperative care, using a stratified and blocked randomization. Strata were defined by diagnosis (GI, GU, GYN) and surgical technique (minimally-invasive versus open). The study biostatistical team generated the random allocation sequence. Randomization allocation was concealed in sequentially numbered envelopes; all study-related personnel were concealed from allocation and access to envelopes. Patients were informed of their randomization allocation, but their surgeons were blinded from the allocation. Setup and instructions on use of the mobile application (TapCloud™ – WellSky; https://wellsky.com; USA) and Actigraph accelerometer were provided in-person and/or by telephone.

Study conditions

Study triage nurses were trained using institutional triage protocols for delivery of RPM care. Intervention fidelity (the extent to which the study arms were delivered as designed and intended) for the RPM Care arm was monitored via audio-recording of sessions; participants provided informed consent for all sessions to be recorded. Continuous fidelity monitoring for RPM care was initiated through random sampling of 15% of triage encounter calls in the first week of the study and 3% monthly thereafter. Triage nurses, following institutional guidance, documented all triage encounters with RPM care participants in the electronic medical record (EMR) system.

RPM care (intervention)

The intervention (RPM Care) included the following: 1) objective assessment of functional activity using Actigraph GT9X Link accelerometer; 2) subjective electronic symptom monitoring (as measured by the 19 items of the MD Anderson Symptom Inventory - MDASI); 3) real-time, nurse-driven alert/feedback system based on pre-determined outcome thresholds^31,40 (daily steps of ≤1500; 1+ symptoms rated moderate to severe intensity) and triggered based on patient input; and 4) standard post-surgical care. Telemonitoring began before surgery and continued after discharge and up to 90 days post-hospital discharge.

Actigraph GT9X Link (Actigraph Corp) is a research-grade accelerometer that captures multiple activity-related data, including daily steps taken and sedentary time. It uses a solid state 3-axis accelerometer to capture and record data. The device continuously monitors activity trends; data can be stored for up to 180 days on the device. It is water-proof with a battery life of 14 days, and weighs 14 g. The device has a programmable LCD window that can be configured to display date/time and real-time activity feedback. Actigraph is among the most widely used and extensively validated devices for activity monitoring⁴¹. For this study, accelerometer captured daily steps was the primary outcome of interest and this was continuously captured from post-randomization (before surgery) to 90 days post-hospital discharge (after surgery). The device was provided free of charge to participants for the entire duration of study participation. For electronic symptom monitoring, the 19 items of the MDASI were built into the TapCloud/WellSky mobile application platform. Patients checked off the answer for each of the 19 items.

Surgeon only perioperative care (control)

Participants randomized to the control condition (surgeon only perioperative care) also wore an accelerometer and provided electronic symptom data. They received a standard message through the mobile application to call the institutional triage phone number when self-reported data deviated from the predetermined thresholds; participants did not receive triage nurse calls.

Outcome measures

All outcomes were captured at baseline (before surgery) and at days 7, 14, 30, 60, and 90 post-hospital discharge. Functional recovery was measured using the Actigraph GT9X Link as previously described⁴¹. Functional recovery data were collected continuously before and after surgery, and participants were encouraged to wear the watch as soon as possible following surgery.

Symptom severity and interference with daily activities was assessed by the MDASI, a validated measure of 13 common cancer-related symptoms (pain, fatigue, nausea, disturbed sleep, distress/feeling upset, difficulty remembering, lack of appetite, drowsiness, dry mouth, sadness, vomiting, numbness/tingling) and six common functional domains (including walking, activity, working, relations with others, enjoyment of life, mood). A movement of 1.2 points is clinically meaningful. The MDASI has been validated in surgical populations; Cronbach Alpha reliability ranges from 0.82 to 0.94^31,42.

Postoperative complications were assessed using the Comprehensive Complications Index (CCI), which summarizes complications (on a scale from 0 to 100) and is based on the established Clavien-Dindo classification. The CCI was validated in a study with 1299 participants, and external validity was tested in a randomized trial evaluating pancreas, esophageal, and colon resections^26,43. Hospital readmissions data were captured for all postoperative timepoints.

Statistical analysis

The primary endpoint for the trial was the percent change from preoperative baseline in accelerometer-measured daily step count during the first two weeks post-hospital discharge. One hundred sixty-six patients per arm (332 total) provided 80% power to detect the hypothesized 14% difference in functional recovery (based on the study team’s preliminary data⁴⁰), using a 2-sided, repeated measures-difference of slope test with alpha = 0.05. To account for 25% anticipated attrition, the study planned to enroll 199 patients per arm (398 total).

There were no a priori interim analysis or stopping guidelines for the trial. For the primary endpoint, accelerometer data were collected at 30 Hz and aggregated into 1-min epochs using the normal filter. Steps per day were determined using the manufacturer’s step algorithm. Valid wear time was defined with the Choi algorithm⁴⁴ in which a day is considered valid if the device was worn for at least 10 h. A moving average of the valid step data was computed using a window size of 9 (spanning 4 days prior to and 4 days after the day under analysis), or from day 0 to 4 days after the day under analysis if the day assessed was day 0, 1, 2 and 3 after surgery). Only patients with valid step data, or imputed valid step data, at baseline and between day of discharge and day 14 were included in the analysis. Specifically, if a participant had no valid step data between day of discharge and day 14 but did have valid data up to 4 days after day 14, this moving average approach was used.

Baseline daily step count was calculated as the average steps per day before surgery, or the average of the daily steps beyond 40 days from surgery if available. To examine the log transformed percent change from preoperative baseline in daily step count during the first 2 weeks post-discharge, a generalized linear mixed model was employed, using time period as an independent variable, subject as a random effect, treatment, age, sex, and residential setting as fixed-factors and study arm-by-time period interaction term as the hypothesis variable testing the slope differences over time between study arms. The treatment effect, representing the ratio of the rate of step recovery between the RPM care and surgeon only perioperative care over the first two weeks post-discharge, was calculated by exponentiating 14 times the parameter estimate for the interaction between time and treatment arm.

For symptoms, ratings in the MDASI were averaged into two subscale scores: mean core symptom severity and mean interference. Patients who answered 7 of the 13 core symptom severity items or 4 of the 6 interference items were included in the analysis. The mixed-effect model was used to compare the difference in score change from baseline between the two arms. The study was originally designed to use an intention-to-treat analyses. The decision was made to shift to including participant data with 10-day wear time and those who answered 7 of the 13 core symptom severity items and 4 of the 6 interference items of the MDASI due to the challenges with missing data in the study and concerns of bias. Maximum CCI scores were categorized as below vs above and equal to 15²⁶, and logistic regression was used to evaluate the effect of study intervention. The maximum CCI score was also analyzed as a continuous variable and compared between the two arms with Wilcoxon rank sum test. Time to hospital readmission up to 90 days was analyzed with the Fine-Gray model, handling any early withdrawal from study as a competing endpoint. Finally, early withdrawal was defined as attrition before protocol-defined completion or loss of device. Difference in time-to-early withdrawal between the two comparators was analyzed with a proportional hazard model.

Author contributions

V.S. was the primary writer for the manuscript. T.C., L.M., T.D., B.Y., D.R., E.C.R., L.C.R., M.N., B.F., and Y.F. provided feedback on study design. Y.X., S.L., J.P., and C.P.T. analyzed and interpreted the study data and manuscript preparation. All authors read and approved the final manuscript.

Data availability

The minimal datasets generated and/or analyzed during the current study (de-identified participation data, data dictionary) are not publicly available but are available from the corresponding author on reasonable request.

Code availability

The underlying code for this study is not publicly available but are available to qualified researchers on reasonable request from the corresponding author.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-025-01961-z.