Setting
Patients were recruited from Cambridge University Hospital NHS Foundation Trust (CUH), a large tertiary university hospital in England with over 1000 beds. In 2018 CUH had 158,399 visits to the emergency department (ED), with 44,120 emergency admissions [16].
Study Design
This was a secondary analysis of a prospective repeated measures cohort study [17]. Ethical approval was granted by the London Queen Square Research Ethics Committee (17/LO/1817). All participants provided written informed consent.
Patient and Public Involvement (PPI)
Before the study began, a PPI panel was convened. The study design reflects amendments and changes suggested by the panel and patients. The panel also reviewed the final versions of the participant information sheet and consent form.
Sample
We included patients admitted to CUH over an 11-month period (Jan 2018 to Dec 2018), who were aged 75 years or older, experiencing an unplanned hospitalisation (i.e. non-elective), able to give informed consent and expected to be hospitalised for at least 48 hours. Exclusion criteria: admitted more than 24 hours before recruitment; unable to provide informed consent based on an assessment of the patient’s mental capacity (a diagnosis of dementia was not in itself an exclusion criterion); receiving end-of-life care or treatment for diagnosed cancer; inability to cooperate in muscle-strength testing (e.g. unable to sit in chair, or skin integrity problem contraindicating the use of a hand-held dynamometer); transferred to or from the intensive care unit; bed-bound or requiring a hoist to transfer from bed to chair within the 2 weeks before hospitalisation; allergy to adhesive dressings; or if the clinical team had any other concerns regarding skin integrity around the proposed accelerometer sites. Sampling was convenience-based in that most screening took place Monday to Friday, 8:00 to 18:00. On recruitment days all patients over the age of 75 and admitted within the last 24 hours were consecutively screened by a member of the clinical team. Patients who met the inclusion criteria were approached regarding participation.
Sample size
The sample size was based pragmatically on maximising the number of recruits over an 11-month period.
Procedures
Participants were recruited during the first 24 hours of their hospital admission. Recruitment was performed by PH (a physiotherapist with 10 years of clinical experience) or a research nurse, all assessments of patient capacity to consent was made by PH. At recruitment, a series of baseline measurements were taken, and the participants were fitted with accelerometers to measure physical activity.
Measurements
All measurements were taken by an experienced physiotherapist. Baseline measurements consisted of: age, sex, weight, frailty, acute illness severity, co-morbidity burden, falls efficacy, cognition, a self-reported measure of functional ability, a measure of functional mobility and objective physical activity levels via accelerometery.
The objective level of in-hospital physical activity was recorded using wearable accelerometers (AX3, Activity, Newcastle upon Tyne, UK), mounted mid-thigh and at the ankle, attached with adhesive dressings [18]. Using a validated method, data collected included the amount of time in a lying position, sitting position, standing position and walking [18]. The accelerometers were worn by participants after they provided informed consent, and were removed on day 7 or discharge, whichever was earliest. Accelerometer sites were checked daily and re-dressed if the dressing was losing adhesion.
The Survey of Health, Ageing and Retirement in Europe Frailty Instrument (SHARE-FI) tool was used to measure frailty. The SHARE-FI tool is a well validated and simple measurement of physical frailty [19]. Five SHARE variables approximating Fried’s frailty phenotype definition are used: fatigue, loss of appetite, grip strength, functional difficulties and physical activity. Scores range between -2.7 and 13.4 (with 13.4 indicating the most severely frail) [19]. As it is routinely measured as part of clinical care, the Clinical Frailty Scale (CFS) score was also recorded [20]. The scoring of the CFS is based on a global assessment of patients’ comorbidity symptoms, cognition, level of physical activity and dependency on activities of daily living. The possible scores range from 1 (very fit) to 9 (terminally ill).
To measure acute illness severity, we used serum C-reactive protein (CRP) levels, and the National Early Warning Score (NEWS), both of which are routinely collected on admission. CRP is an acute phase-reactant protein released in response to injury, infection or inflammation and is a recognised clinical measure of illness severity [9, 10]. This was collected only for clinical reasons, therefore if CRP was not measured on admission but on day 1 of the study, then the day 1 value was used. The half-life of serum CRP in humans is approximately 19 hours [11, 12].
The NEWS was devised by the Royal College of Physicians of London to standardise the assessment and response to acute illness [21] and has been extensively validated [22].
The Charlson Comorbidity Index (CCI) is a method for classifying comorbid conditions for use as a prognostic indicator [23]. The CCI is based on patients’ diagnoses as coded by the World Health Organization’s International Classification of Diseases (10th version).
Falls efficacy is defined as self-perceived confidence in engaging in activities of daily living without falling [24]. The FES-I is a reliable and validated measure in older adults [25].
The Mini-ACE is a 30-point scale used to detect cognitive impairment [26]. The Mini-ACE has been reported to have higher sensitivity and higher ceiling effect than the Mini-Mental State Examination [26] . A score of 25 or less is suggestive of cognitive impairment [26] .
Self-reported general functional ability was measured using the Barthel Index: a 10-item ordinal scale (0–100) of functional independence with activities of daily living, where a score of 100 represents a high level of functional independence [27]. The participants were asked at baseline assessment to base their answers on their functional ability two weeks before admission.
The DEMMI is a 100-point ordinal scale for the assessment of mobility in older acute medical patients [28]. It consists of 15 items ranging from assessing bed mobility to high levels of dynamic balance. A score of 100 represents a high level of functional mobility [28]. The DEMMI provides interval level measurement and does not have floor or ceiling effects in the acute hospital setting.
Knee-extensor HHD (using the microFET 2, Hoggan Scientific, Salt Lake City, Utah) was measured in participants seated with their knee at 90° with the HDD perpendicular to the leg above the superior border of lateral malleolus; patients were asked to push against it with maximum effort [29]. The HHD was tethered to a stationary object whilst the researcher held it in place to prevent the leg from moving. Force was converted to torque (Nm) by multiplying the result by the distance between the superior border of the lateral malleolus and superior border of the lateral femoral epicondyle. Grip strength HHD (using the JAMAR device, Sammons Preston, Bolingbrook, Illinois) was measured with participants seated with elbow at 90° and wrist in neutral position; participants were instructed to squeeze as hard as possible for a few seconds [30]. With both knee-extensor and grip strength dynamometry readings, participants were asked to repeat the procedure three times on both their left and right sides, the highest force measurements from each side were then averaged to provide the score.
In-hospital physical activity was defined as the amount of upright time (standing or walking).
Analysis
Data were analysed with R software [31]. Continuous variables were described as median or inter-quartile range (IQR), and categorical variables as count and percentage. The analysis was limited to the first 24 hours of measured activity due to attrition of data as patients were discharged from hospital.
To examine the predictors of physical activity in the first 24 hours of study participation we used a best subset analysis using the ‘leaps’ R package [32]. To comply with the assumption of normality of residuals, upright time was transformed using a base-10 logarithm transformation. The model specified a maximum of 5 covariates and used an exhaustive search. To assess for overfitting of the data, the Bayesian information criterion (BIC) was extracted. We also performed k-fold cross validation (k = 10) to predict the different models’ ability to generalise to independent data sets.