Clinicians used to prescribe strict bed rest for 2 to 3 weeks after abdominal surgery until pioneers such as Leithauser started challenging this dogma in the 1940s [1]. Through a series of case studies, these pioneers reported that immobility caused harm and that early mobilization was safe and feasible [2-4]. In 2005, the Enhanced Recovery After Surgery (ERAS) Society published its first perioperative guidelines for patients undergoing colorectal surgery [5] and promoted the uptake of early mobilization efforts by clinicians. The guidelines recommended that patients spend 2 hours out of bed on the day of surgery and 6 hours per day out of bed until discharge [5]. Today, ERAS guidelines have expanded to >20 adult specialties, all of them describing early mobilization as a vital component of postoperative care [6].

Despite the widespread acceptance of the ERAS guidelines, the recommendations on postoperative mobilization are built on expert consensus with little to no data supporting the specific mobility goals [7,8]. Early mobilization remains poorly defined in the literature [7-9], and protocols vary substantially between institutions and studies [8,9]. Hence, optimal methods to achieve early mobilization and the impact of specific physical activity components (such as timing, type, duration, frequency, and intensity) [10,11] on postoperative outcomes are still unknown [7-9].

Accelerometers have gained popularity as consumer- and research-grade activity-tracking devices [12]. Their ability to quantitatively measure and summarize physical behaviors has attracted many researchers, and, as a result, the number of publications using accelerometers has grown exponentially in recent years [13]. In this systematic review, we aimed to summarize the current literature on accelerometer-measured postoperative physical activity in the acute inpatient setting and its impact on clinical outcomes after major abdominal surgery.

We first searched PROSPERO [14] to verify the absence of existing or ongoing research on this topic. We then outlined a written protocol (not registered) according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement [15] before conducting the literature search. We adhered to our protocol and the PRISMA statement throughout the review process (Multimedia Appendix 1).

We searched PubMed and Google Scholar using comprehensive search strategies developed with assistance from an institutional librarian. The search strategy included the Medical Subject Headings (MeSH) terms “postoperative period,” “postoperative care,” “accelerometry,” “wearable electronic devices,” “fitness trackers,” and their related terms (Multimedia Appendix 2). The database included all publications up to October 14, 2021. For Google Scholar, we screened the first 100 articles as described previously [16]. Reference lists of related studies were also used to identify relevant articles.

We included studies that used accelerometers to measure physical activity during the hospital stay immediately after major abdominal surgery. We defined major abdominal surgery as any nonobstetric procedures performed under general anesthesia requiring hospital admission. Both open and laparoscopic surgeries were eligible. Studies were eligible only if they evaluated the effects of physical activity on postoperative outcomes. Outcomes of interest included postoperative complications, return of gastrointestinal function, hospital length of stay, discharge destination, and readmissions.

We excluded studies if they (1) involved participants aged <18 years; (2) did not include physical activity measurements during the acute inpatient period; (3) only reported descriptive analysis of physical activity measures without evaluating their impact on clinical outcomes; or (4) were case reports, study protocols, or conference abstracts. Review articles were used to search for additional articles not captured by the database search.

After the literature search, we used Covidence systematic review software (Veritas Health Innovation Ltd) to facilitate study selection, data extraction, and quality assessment. First, 2 independent reviewers (MF and KJR) screened titles and abstracts of all articles identified from the database searches and the studies identified from the reference lists. Studies were excluded if they were not relevant (eg, nonabdominal surgery and use of accelerometers for purposes other than physical activity measurements). The included studies underwent a full-text review by 2 independent reviewers (MF and KJR) using the aforementioned inclusion and exclusion criteria. Any disagreements between the reviewers for each of these steps were resolved by a third reviewer (AFB).

Data extraction was performed by a single investigator (MF). The data collection form contained the following variables: study design, type of surgery, number of participants, patient characteristics, descriptions of interventions (if applicable), device name, device setup (eg, sampling rate, filter, and epoch), device wear location, data collection period, reported measures of physical activity (including but not limited to step count, postural transition, activity duration, time-to-mobilization events, and activity trend over time), and clinical outcomes. We extrapolated the duration of device wear in the hospital based on the methods described in each study.

We used the risk-of-bias assessment tool for nonrandomized studies (RoBANS) [17] to assess the risk of bias in observational studies and the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) [18] to assess the risk of bias in randomized controlled trials (RCTs). For observational studies, we predefined the following factors as confounding variables based on the literature: (1) preoperative level of physical activity [19], (2) American Society of Anesthesiologists physical status classification [19-21], (3) performed procedure [22], (4) open versus laparoscopic approach [20,23-25], (5) duration of surgery [20,21,23], and (6) postoperative intensive care unit admission [20,23]. Two independent assessors (MF and CK) evaluated each study, and conflicts were resolved through consensus. We used R statistical software (version 4.1.1; R Foundation for Statistical Computing) and the R package robvis [26] for data visualization.

Because of the heterogeneity of study populations, methods, and device models, the data were not amenable for a meta-analysis. We performed a qualitative synthesis by summarizing the findings in three themes: (1) device use, (2) metrics used to describe physical activity, and (3) clinical outcomes analyzed in association with physical activity. Observational studies and RCTs were organized separately, given the differences in study designs. The key findings of individual studies were summarized in tables by tabulating the following variables: type of surgery, patient characteristics, main predictor (observational studies), intervention and control (RCTs), and main findings.

We identified 2470 articles: 2446 (99.03%) through the database searches and 24 (0.97%) from the review of reference lists. After screening the titles and abstracts of all 2470 articles, 103 (4.17%) underwent a full-text review. Of these 103 articles, 15 (14.6%) met our selection criteria [27-41]. The reasons for exclusion are detailed in Figure 1.

All articles were published between 2017 and 2021, except for the study by Browning et al [27], which was published in 2007. Of the 15 articles, 9 (60%) were observational studies [27-35], and 6 (40%) were RCTs [36-41]. The median sample sizes were 54 (IQR 50-94) for observational studies and 98 (IQR 64-107) for RCTs. Of the 15 articles, 13 (87%) studied the general surgery population, 1 (7%) was a study of patients who had undergone gynecologic surgery [40], and 1 (7%) included a mixed (abdominal, thoracic, gynecologic, and orthopedic) surgical population [30]. The results of individual studies are summarized in Tables 1-3.

Figure 2 summarizes the risk-of-bias assessment for the observational studies (n=9), all of which used an accelerometer to collect physical activity data and were at low risk of bias for domain 3 (measurement of exposure). By contrast, none adequately controlled for confounders and were determined to be at high risk of bias for domain 2 (confounding variables). Of the 9 studies, only 3 (33%) clearly stated their study objectives [28,30,35]; the remaining 6 (67%) were exploratory, putting them at high risk of bias for domain 6 (selective outcome reporting) [27,29,31-34].

Figure 3 shows the summary of the risk-of-bias assessment for the RCTs (n=6), all of which had at least some methodological concerns and were determined to have an overall high risk of bias, except for the study by Fiore et al [36], which was the most rigorous among these studies.

The device choice, use, and reported outcomes are summarized in Multimedia Appendix 3. Of the 15 studies, 3 (20%) used research-grade accelerometers [27,29,36], whereas the remaining 12 (80%) used consumer-grade devices [28,30-35,37-41]. Among the commercially available devices, the Fitbit series (Google LLC) was used most frequently (7/12, 58%) [28,30,34,35,37,39,41]. Of the 3 studies using research-grade accelerometers, only 1 (33%) study, which used the ActiGraph GT3X25 (ActiGraph LLC), described the accelerometry setup (such as sampling rate, filter, epoch, and analysis algorithm or software) [36].

All studies used accelerometers during the acute inpatient period (per our inclusion criteria). Of the 15 studies, 12 (80%) applied the device within a day after surgery (n=7, 58%, studies on the day of surgery [30-32,35,36,38,39] and n=5, 42%, studies on postoperative day 1 or within 24 hours after surgery [27,29,33,37,40]). Of the 15 studies, 2 (13%) started the device wear after patients were transferred to the floor from the intensive care unit, which occurred on postoperative day 2 or 3 on average [28,41], and 1 (7%) did not report the timing of initiation [34]. The mean in-hospital wear duration was 5.2 (SD 2.0; range 3-10) days. Most of the studies (11/15, 73%) described continuous 24-hour device wear with or without brief interruptions for battery charging or patient showering [27-30,32-34,36,38,39,41], whereas others implied continuous measurement but did not explicitly describe it [31,35,37,40]. Of the 15 studies, 3 (20%) obtained preoperative baseline data, ranging from 2 to 30 days before surgery [31,35,40], and 1 (7%) followed patients after discharge until postoperative day 30 [35].

In this systematic review, we found 15 articles that used accelerometers to evaluate the effects of postoperative physical activity on outcomes after major abdominal surgery, with 14 (93%) published within the last 5 years. Although the observational studies (9/15, 60%) consistently showed that increased physical activity during the immediate postoperative period was associated with improved patient outcomes, only 1 (17%) of the 6 RCTs demonstrated that a mobility-enhancing intervention was beneficial compared with usual care. These findings confirm that physical activity is an important predictor of outcomes, but leave important questions unanswered—what is the optimal postoperative mobilization strategy or the dose of mobilization associated with better outcomes? Because of the high risks of bias, we could not synthesize specific mobility recommendations. However, our study illustrates how accelerometers can be a powerful tool for quantifying objective, continuous measures of physical behaviors in the hospital and provides guidance for future research to improve methodological rigors and study design.

We found from this systematic review that physical behaviors follow certain patterns after abdominal surgery. First, surgery causes a steep drop in physical activity from the preoperative baseline [31,35]. This effect is more significant after open abdominal surgery than after laparoscopic surgery [33] and varies by procedure type [30]. Second, the recovery of physical activity is slow, often requiring >1 month to return to baseline [30,33,35,39]. The recovery speed is also different, depending on the procedure [30-32], which is consistent with previous literature [24,25]. In the observational studies (n=9), increased physical activity during the immediate postoperative period was associated with improved clinical outcomes regarding surgical complications, return of gastrointestinal function, postoperative pulmonary complications, hospital length of stay, and hospital readmission [27-35]. These findings suggest that physical behaviors are important predictors of outcomes. In more practical terms, clinicians could use certain physical behaviors to predict or identify patients at risk for adverse outcomes after surgery.

Notably, 4 (67%) of the 6 mobility-enhancing interventions used in the RCTs did not increase postoperative mobilization compared with usual care [38-41]. The mobility-enhancing interventions ranged from step count feedback with encouragement to designated study physiotherapists assisting patients to achieve set mobility milestones. We speculate several reasons why many of these interventions (4/6, 67%) did not enhance mobility performances beyond usual care: (1) the selected interventions were simply ineffective, (2) the selected activity measure (step count was the most commonly used) was not sensitive enough to detect changes in mobility performances, and (3) mobility performances were nonmodifiable. Furthermore, the RCTs (2/6, 33%) that successfully enhanced physical activity showed conflicting effects on clinical outcomes. In the study by Ni et al [37], patients in the intervention arm achieved higher step counts from postoperative days 2 to 5 and had a faster return of gastrointestinal function and shorter hospital length of stay. By contrast, in the study by Fiore et al [36], more patients in the intervention arm were out of bed (sitting or standing) from the day of surgery through postoperative day 2 and took more steps on postoperative days 1 and 2 but had similar outcomes on return of gastrointestinal function and hospital length of stay.

Several factors could explain why postoperative mobilization had little effect on clinical outcomes when studied prospectively in the RCTs. First, the sample sizes of these RCTs were relatively small (median 98, IQR 64-107). Therefore, they could have lacked the statistical power to detect differences in clinical outcomes. Second, postoperative physical activity may be a prognostic indicator of outcomes rather than a modifiable factor. This theory is plausible, given that the factors associated with reduced postoperative mobilization and worse clinical outcomes often overlap, such as preoperative physical activity level [19-21], open versus minimally invasive approach [20,23-25], and duration of surgery [20,21,23]. Third, it is possible that the achieved differences in mobilization dosage (such as timing, type, duration, frequency, and intensity) [10,11] were not significant enough to affect clinical outcomes. Fourth, routine care that involves basic mobility may be sufficient to prevent immobility harm. Fifth and last, the effects of specific physical activity measures on postoperative outcomes remain unknown [8]; for example, it is unclear whether sitting out of bed (static positioning) is as effective as standing and walking (active mobility) in improving clinical outcomes. Thus, the choice of reported mobility metrics could have affected the researchers’ ability to detect clinically meaningful differences in activity exposures.

The particularly well-conducted study by Fiore et al [36] is worth special attention. The authors defined physical activity as “out of bed at all on the day of surgery and out of bed for at least 6 hours on postoperative day 1-3,” which directly reflects the recommendation described in the original ERAS guidelines [5]. This RCT found no benefit from the authors’ facilitated mobilization intervention, including the 6-minute walk test at 4 weeks (primary outcome), time to gastrointestinal recovery, time to readiness to discharge, length of stay, and 30-day complications. The negative result may be partly due to patient selection because 80% (80/99) of the study participants received laparoscopic surgery. Laparoscopic surgeries have been shown to expedite recovery [24,25,38], and treatment effects from mobility-enhancing interventions may be less pronounced in patients undergoing laparoscopic surgeries than in those undergoing open surgeries, especially in environments with optimal usual care. In the case of the study by Fiore et al [36], patients in the usual care group reached activity levels similar to those reached by patients in the intervention arm by postoperative day 3.

We found that most of the studies (12/15, 80%) used consumer-grade accelerometers to characterize physical behaviors. Commercially available devices have appealing features such as patient familiarity, user-friendly interfaces, and fashionable designs, all of which could improve wear compliance. In addition, measures such as step count are intuitive and easy to interpret among many users. However, consumer-grade accelerometers are different from research-grade accelerometers in that they use proprietary algorithms to compute and report physical behavior measures such as step count and energy expenditure. Furthermore, they do not give researchers access to accelerometer settings such as filter, sampling rate, epochs, and software algorithm. As patients who are hospitalized are distinct from the free-living population in that they spend most of their wakeful time sedentary or in bed [42-44], walk significantly slower, and may hold on to an intravenous pole or an assistive device when ambulating [43], researchers may benefit from using research-grade accelerometers because of their flexibility in terms of data collection and analysis [45].

Importantly, most validation studies of accelerometer devices are derived from laboratory and free-living conditions. These studies show that measurements can vary substantially by device manufacturer [46-50], wear location [47,49-52], and data-processing algorithm [46,48]. For step count, the most commonly reported physical activity outcome in our review (12/15, 80%), the discrepancy in measurement can be as much as 120%, depending on the device and wear location [50]. Moreover, the study population and setting can affect device accuracy; for example, older adults, who tend to walk slower than younger adults, walked at a speed of 0.74 meters per second as outpatients but recorded a speed of 0.46 meters per second as inpatients [42]. One study found 20% absolute percentage errors in step counts at a gait speed of 0.42 meters per second and 45% errors at an even slower pace [49]. Many accelerometers available in the market, including research-grade devices, still await validation in acute inpatient settings [42,53]. As is the case with laboratory biomarkers, digital biomarkers derived from biometric monitoring technologies require multistep validation before they can be applied reliably to a specific patient population and clinical setting [54]. It is critical to be mindful of these limitations when interpreting the results or conducting research using accelerometers because the reported outcomes, particularly step count, are not directly comparable [50-52].

There are 2 major limitations related to the conduct of this systematic review. First, database searches were limited to PubMed and Google Scholar owing to our time constraints; therefore, we could have missed articles available in other databases. To supplement this, we used the reference lists of the included studies and related review articles to identify relevant studies. Second, the heterogeneity of study designs, patient populations, and accelerometer use made it difficult to compare study findings. To minimize the risk of bias resulting from data synthesis, we developed and followed a written protocol using rigorous systematic review processes.

Overall, the quality of available evidence was poor, and we could not synthesize specific recommendations for postoperative mobilization. On the basis of the limitations we identified in the included studies, we recommend that researchers (1) select a patient population that is more likely to benefit from mobility-enhancing interventions (eg, patients undergoing open abdominal surgery rather than laparoscopic surgery and patients with frailty rather than those who are young, healthy, and fit); (2) clearly define and measure timing, type, duration, frequency, and intensity of a mobility-enhancing intervention to delineate the differences in mobility performances achieved by patients in different treatment groups; (3) measure all relevant data (such as patient, surgical, and postoperative factors) to control for confounders adequately; and (4) measure physical behaviors beyond step counts (such as static positioning and in-bed activities) because patients are highly sedentary after surgery [55,56], and step counts only capture snapshots of patients’ mobility status.

In conclusion, although observational studies showed strong associations between postoperative physical activity and outcomes after major abdominal surgery, RCTs have not proven the benefit of mobility-enhancing interventions compared with usual care. To understand the optimal postoperative mobilization strategy or the impact of individual physical activity components such as timing, type, duration, frequency, and intensity, future accelerometer research would benefit from improved study designs, increased methodologic rigor, and more consistent reporting of accelerometer methods [57].