Accelerometry Assessed Physical Activity of Elderly Adults Hospitalized with Acute Medical Illness - an Observational Study

Background: In a hospital setting and among elderly patients, inactivity and bedrest are associated with a wide range of negative outcomes such as functional decline, increased risk of falls, longer hospitalization and new institutionalization. Our aim was to assess the distribution and determinants of physical activity (PA) levels using wrist-worn accelerometers in elderly adults hospitalized with acute medical illness. Methods: Observational study conducted from February to November 2018 at an acute internal medicine unit in the University hospital of Lausanne, Switzerland. We enrolled 177 patients aged ≥65 years, able to walk independently prior to admission. PA during acute hospital stay was continuously recorded via a 3D wrist accelerometer. As previously suggested, PA was defined as an acceleration ≥12 mg. Clinical data was collected from medical records or by interview. Autonomy level prior to inclusion was assessed using Barthel Index score. Results: Median interquartile range - IQR age was 83 74-87 years and 60% of participants were male. The median IQR time spent inactive and in PA was 485 404-563 and 177 114-255 minutes/day, respectively. PA was distributed into three peak periods during the day: between 8 and 10 am, at 12 am and at 6 pm. Approximately 50% of patients were considered physically active. Stepwise logistic regression identified Barthel index [odds ratio and (95% confidence interval) for increase in category 1.85 (1.09-3.14), prescription of sedatives 0.31 (0.12-0.80) and prescription of physiotherapy 2.16 (1.08-4.58) as significantly and independently associated with inactivity. Conclusion: Elderly hospitalized patients are physically active only one quarter of the day and concentrate their PA around eating periods. A Barthel Index below 91 could be used to identify patients at risk of physical inactivity.

Extended bedrest has been described as toxic for elderly patients (1). In a hospital setting and among elderly patients, inactivity and bedrest are associated with a wide range of negative outcomes such as functional decline, increased risk of falls, longer hospitalization and new institutionalization (2)(3)(4)(5). Assessment of PA relies mostly on qualitative nursing observation (6) and remains poorly documented in hospital electronic records (7). Therefore, there is an urging need to objectively monitor and quantify inpatient PA levels.
Most PA levels are assessed based on observational data such as periodic nursing reports (2,7,8) or on nurse standardized functional assessment (6,9). These qualitative assessments are less accurate compared to objective measures of PA such as accelerometry (10). Indeed, accelerometers allow the collection of objective and continuous PA data and have been tested and validated in elderly patients (11)(12)(13)(14).
Further, the data collected can be analyzed with algorithms that classify locomotion and non-locomotion periods in everyday life (15). Still, there is a paucity of studies assessing PA levels by accelerometry in hospitalized elderly patients (12)(13)(14)(16)(17)(18)(19)(20)(21)(22). In a previous paper, Lim et al. reported that the PA levels of 38 hospitalized elderly patients were very low and that most PA was sustained over short periods (12). However, the sample size was small, and the results were not replicated in other settings.
Therefore, the aim of this study was to assess the distribution and determinants of PA levels in elderly hospitalized patients, by means of a wrist-worn accelerometer.

Setting
We conducted this study from February 2018 to November 2018 in a 21-bed internal medicine ward of the Lausanne university hospital (CHUV), in canton Vaud, in the French speaking part of Switzerland. The CHUV has over 1500 beds and admits over 50,000 patients per year.

Recruitment
Patients were recruited on a daily basis, from Monday to Friday. All patients aged ≥65 years admitted directly to the study ward or via the emergency unit were considered eligible. Participants were excluded if they: a) had a probable life expectancy of less than 30 days, based on clinical judgement; b) had insufficient comprehension of French language, c) were unable to stand within the week before hospitalization, as assessed by interview, or d) were forced to bedrest by factors not directly related to the disease (e.g. fracture). The selection procedure was applied within the first three days of hospitalization. If exclusion criteria were not met, patients were invited to participate and received an explanation of the study procedure. If the patient accepted, a written informed consent was signed before the start of the study.
All investigators had previously been trained regarding screening and recruiting methods.

Ethical statement
The study was approved by the Swiss Ethics Committee on research involving humans using BASEC ( www.cer-vd.ch), reference 2017-01907 (decision of 21 December 2017). The full decision of the CER-VD can be obtained from the authors upon request. The study was performed in agreement with the Helsinki declaration and its former amendments, and in accordance with the applicable Swiss legislation. All participants or their legal representatives (in case of confusion or cognitive impairment) provided a signed informed consent before entering the study. If a participant decided to withdraw from the study, data collected until the moment of withdrawal was used.

Physical activity assessment
We assessed PA levels using a wrist accelerometer (GENEActiv Original, ActivInsights Ltd, UK), parametrized at 50 Hz. These accelerometers have been shown to provide a reliable and valid measurement of physical activity in adults (23) and hospitalized elderly patients (12). We provided the patients with a device immediately after inclusion and they could choose on which wrist to wear it. Previous studies have shown that wrist side does not influence measurements (24). Patients were asked to wear the device continuously (day and night, including showering). The observation period was limited to the index hospitalization in internal medicine. Upon discharge or transfer to another department (for e.g. intensive care, surgery), the accelerometer was removed by a nurse or one of the investigators.
Accelerometry data was extracted and analyzed using version 9.1 the GGIR package for R (23). This package estimates time spent in different levels of PA according to predefined thresholds, overall and separately for day and night (R-script provided in annex 2). A valid day was defined as at least 10 hours of daytime wear. Moreover, at least 24-hours of valid data were required for analysis (25). As suggested in a previous study, we used the threshold of ≥12 mg to define PA (12). Patients were considered as physically active or inactive if their average daytime PA was ≥12 mg or <12 mg, respectively, nighttime not considered.

Covariates
Investigators extracted covariates from the hospital electronic database. This included demographics; reason for hospitalization; comorbidities via the Charlson comorbidity index (26); use of sedative drugs at admission; prescription (yes/no) of physiotherapy.
During the baseline interview, investigators collected self-reported physical function 2 weeks before admission (i.e. use of walking aids and history of falls), medical equipment upon inclusion (i.e. urinary catheter or oxygen therapy) and isolation precautions (i.e. for infection control and patient protection).
Prior research suggests that admissions for gait problems/fall, general state of health alteration, and neurological deficit, are predictors of functional decline in hospitalized elderly patients (27,28). Hence, for our analysis, we created a variable based on these conditions, also including musculoskeletal pain, and named it "reason for admission associated with functional decline".
Autonomy was assessed using the Barthel Index score, reported as being the best scale to assess activities of daily living (ADL) (29) and with a widespread use. The modified version improves the internal consistency and provides better discrimination of functional ability.
For patients with cognitive impairment or confusion, the level of autonomy before hospitalization was assessed by interviewing their relatives or caregivers, in face-to-face interviews or by phone call. The patient's ability to perform different ADLs was rated as follows: fully independent; with minimal or moderate help; attempts task but unsafe; and unable to perform. Maximum score was 100. A total Barthel Index score of 0-20 suggests total, 21-60 severe, 61-90 moderate and 91-99 slight dependence. A score of 100 indicates that the patient is independent of assistance from others.
Skin status and risk of bedsores was assessed using the Braden score upon inclusion (30).
The Braden scale rates patients using six subscales: sensory perception, moisture, activity, mobility, nutrition, and friction and shear. The maximum score is 23; a score ≤18 indicates a high risk of sore development.

Statistical analysis
Statistical analysis was conducted using Stata v15.1 (Stata Corp, College Station, TX, USA). Results are expressed as number of patients and (percentage) for categorical variables and as average ± standard deviation or as median [interquartile range] for continuous variables. Between-group comparisons were performed using chi-square or Fisher's exact test for categorical variables and analysis of variance or Kruskal-Wallis nonparametric test for continuous variables. Variables significantly and independently associated with PA status were identified by stepwise forward logistic regression, using the category physically inactive coded as a binary (0/1) variable, and a p-value for entry of 0.05. Variables significantly associated with physical inactivity in the bivariate analysis were included and the results of the logistic regression were expressed as multivariable adjusted odds ratio (OR) and 95% confidence interval (CI).
Sensitivity analyses were conducted after excluding participants with over 20 percent accelerometer non-wear time. Statistical significance was assessed for a two-sided test with a p-value <0.05.

Sample selection
Of the 377 patients screened, 274 were eligible for the study and invited to participate.
Among the 274 eligible patients, 211 (77%) accepted to participate, of whom 177 (84%) had valid accelerometry data. The selection procedure is summarized in Supplemental

Physically active versus inactive patients
Approximately 50% of patients were considered physically active. The characteristics of physically active and inactive patients are summarized in table 3. Inactive patients more frequently reported the use of walking aids within 2 weeks before hospitalization, were more frequently admitted for a reason associated with functional decline, were at higher risk of bedsores at admission, and had a higher dependency level according to Barthel Index score. Inactive patients were also less often prescribed sedative drugs, more often involved in physical therapy program and had a higher number of comorbidities than physically active patients. No difference in PA was noted according to gender.
Stepwise logistic regression identified Barthel index, prescription of sedatives and prescription of physiotherapy as significantly and independently associated with physical inactivity (Table 4). Using the threshold of 91 points for the Barthel Index, the sensitivity and specificity (and corresponding 95% CIs) to detect a physically inactive patient were 47% (36%-58%) and 71% (61%-80%), respectively. Similar findings were obtained when analyses were restricted to patients with less than 20% of accelerometer non-wear time (supplemental tables 2 and 3).

Discussion
There is little information regarding physical activity of hospitalized patients. To our knowledge, this is one of the largest studies measuring physical activity by accelerometry in older patients hospitalized with acute medical illness. According to our results, elderly patients are inactive most of the time, and their PA is distributed into daily patterns.
Physical activity levels and distribution throughout the day Patients spent a median of approximately 3 hours/day on PA, a value slightly lower than pre-admission. Another study also reported that 30 elderly patients (median age 83.6 years) spent less than an hour between 9 am and 5 pm in an upright position and nearly 50% of the day lying down (19). Finally, a study reported even lower durations spent walking (7 min/day) and standing up (35 min/day) in 100 elderly patients (median age 84 years) (17).
Nevertheless, inter-study comparison and reproducibility is very difficult because of the use of different thresholds and different PA metrics. Elderly hospitalized patients are characterized by very low PA levels, and thresholds to define PA in this population are rare and differ according to studies. In Lim et al.'s study, 1-min mean acceleration ≥12 mg was selected to define periods of PA, and corresponded to participants taking at least four steps in one minute (12). Similar thresholds have been reported in free-living elderly (median age ≥65 years) (13,32) and younger (33,34) populations. Still, the ≥12 mg threshold is very low compared to thresholds for light PA developed and validated in laboratory calibration or in free-living populations, which are usually above ≥40 mg (35)(36)(37)(38). For example, analysis of our recordings with a threshold of ≥85 mg as proposed by

Physically active versus inactive patients
Compared to physically active patients, pphysically inactive patients more frequently reported the use of walking aids 2 weeks before hospitalization, were more frequently admitted for a reason associated with functional decline, were at higher risk of bedsores at admission, and had a higher dependency level according to Barthel Index score. These findings are consistent with other studies (8, 39 , 40 ) and suggest that initial evaluation of patients using these metrics could help to identify patients in need of mobilization (9).
Physically inactive patients also had physiotherapy prescribed more frequently. This finding suggests that these patients are correctly identified upon admission and given more attention regarding their low mobility status. Still, the prescribed physiotherapy was not enough to change their PA behavior. Our results suggest that increased efforts are necessary to mobilize hospitalized patients. Nevertheless, the magnitude of efforts needed to achieve an adequate amount of PA during hospitalization may exceed the existing resources of the hospital. Future studies should try to estimate not only the minimum amount of PA needed to prevent increased in-hospital morbidity or length of stay, but also the optimal conditions necessary to deliver mobility interventions.
Physically inactive patients were more likely to have cognitive impairment/confusion and were less often under sedative drugs upon admission. A possible explanation for this last finding could be that this class of medication is avoided in older patients with cognitive impairment as it tends to impair balance and increase fall risk (41).

Implications for clinical practice
Our results strengthen the available evidence that hospitalized elderly patients move very little and that some indicators such as the Barthel Index assessed at admission are relevant to identify patients at risk of inactivity during hospital stay. An interesting finding were the peaks of PA during mealtimes, also reported in another study (12). Hence, a possible way to favor patients' locomotion would be to serve meals in common rooms instead of in hospital beds. Other alternatives include mobilization by family members or volunteers (if the patient's condition allows it) to compensate for lack of resources (42).
This alternative was also recently used in an interventional study aiming to reverse the functional decline associated with acute hospitalization in very elderly patients (43).

Strengths and limitations
The strengths of this study are its large sample size, its broad inclusion criteria, the use of accelerometers to assess PA and the large number of PA hours recorded. Regarding sample size, this study is the largest when compared to other studies in Europe (12,13,(16)(17)(18)44), the USA (20,21,31) and Australia (14,19). Regarding inclusion criteria, and contrary to previous studies (12,31,45,46), we included patients with cognitive impairment/confusion as they are at increased risk of post-hospitalization functional decline (27,39). Finally, PA was assessed using accelerometers, which are considered superior to pedometers (47) and allow data analysis using different algorithms.
This study also has some limitations. First, the study was conducted in a single university hospital, which might limit generalizability, as patients attending a university hospital may present more comorbidities or more severe diseases than patients attending a general hospital. Hence, it would be important that this type of study be implemented in other settings. Second, some recordings were very short (<24 hours) and could not be used, or had a high percentage of accelerometer non-wear time. Still, our findings are similar to another study that used the same methodology and included all available data, and findings were similar when patients with a high non-wear time were excluded (supplemental tables 2 and 3). Finally, our approach did not allow to differentiate between physical activity in bed or elsewhere. Further studies could implement PA assessment by combining accelerometry with manual or automatic recording of location.

Conclusion
Elderly hospitalized patients are physically active only one quarter of the day and concentrate their PA around eating periods. A Barthel Index below 91 could be used to identify patients at risk of physical inactivity.

Consent to publish
There are no details on individuals reported within the manuscript.

Availability of data and materials
All data generated or analyzed during this study are included in this published article and its supplementary information files.

Competing interests
The authors declare that they have no competing interests.

Funding
This study was not funded.

Authors' Contributions
All authors have read and approved the manuscript.
TP collected and analyzed data, and wrote most of the manuscript. PMV analyzed data and wrote part of the manuscript. VK entered patient data and reviewed/edited the manuscript. GR, MM and PV reviewed/edited the manuscript and contributed to the discussion. PMV had full access to the data and is the guarantor of the study.     Results are expressed as odds ratio and (95% confidence interval -CI) Additional File Legend