Design: This study is a secondary analysis of a clinical trial (19). This was a randomised, crossover, controlled trial. Eligible participants were randomly allocated either into a group which received the VR intervention for 12 weeks followed by the control treatment for 12 weeks, ‘VR–Control’ (VRC), or vice versa, ‘Control–VR’ (CVR). The variables were recorded in two periods: 12 months before the implementation of the exercise program and 12 months after the start of the program. The participating patients were recruited from the Haemodialysis Unit at the beginning of 2018.
We included patients aged over 18 years who were clinically stable, on maintenance HD treatment for at least 3 months when starting the study, and who had signed their informed consent to participate in the study. We excluded individuals who had suffered a myocardial infarction in the 6 weeks prior, with unstable angina during exercise or at rest, a lower-limb amputation above the knee (without a prosthesis), cerebral vascular disease such as stroke or transient ischemia, skeletal muscle alterations, respiratory diseases that worsened with exercise, people unable to perform functional tests or the planned intervention, and those with a lower-limb vascular access (VA).
The VR exercise intervention was carried out between April and September 2018, lasted 12-weeks, and was implemented and supervised by a physiotherapist with the support of the HD unit nursing staff. Each session during the intervention period comprised a 5-min warm-up followed by the VR exercise session, which lasted a maximum of 30 min, depending on each patient’s individual fitness level, and was completed intradialysis during the first two hours of the HD session.
The exercise program was adapted from a non-immersive VR system game called Treasure Hunt, with the aim of making the patient intradialysis exercise session experiences more enjoyable. In this game the players must hunt for objects such as virtual coins while avoiding obstacles like virtual explosives by freely moving their legs, especially by raising their lower limbs up (hip flexion with the knee extended and foot in a neutral position) or left or right (hip abduction and adduction with the knee extended and foot in a neutral position). The difficulty level of the game was graduated according to the patient’s characteristics and the participants could change the leg they used to play the game whenever they became tired.
The hardware we used was a standard desktop computer and monitor screen with a Microsoft Kinect® movement-detection camera. Before starting each session, the physiotherapist defined the VR intervention for each patient by selecting the number of exercise sets and rest periods, and their duration. The game difficulty automatically adapted to match the user’s in-game results: an increased difficulty level meant that more objects appeared for the players to ‘catch’ or ‘avoid’, and they appeared and disappeared at a faster rate. Once the game finished, each patient completed five minutes of gentle stretching. The ideal exercise difficulty level was between 12 and 15 out of 20 on the perceived exertion scale. If the session was too easy (6–11 on the scale) or too hard (16–20) the physiotherapist altered the game settings to make it easier or harder.
The following patient descriptive variables were recorded at baseline: age (in years), sex (male or female), race (Caucasian, Black, or Asian), body mass index (kg/m2), diabetes mellitus status (yes/no), current smoking status (yes/no), renal insufficiency aetiology, VA type (arteriovenous fistula [VAF]/central venous catheter [CVC]), HD technique (low-flow HD [LFHD], high-flow HD [HFHD], or online haemodiafiltration (OLFHD), session duration in minutes, blood flow rate (mL/min), dialysate flow rate (mL/min), and the intrasession standardised urea clearance rate (Kt/V) measured using online clearance monitoring (OCM), serum haemoglobin, albumin levels (g/dL) as well as the mean systolic blood pressure and diastolic blood pressure (mmHg), heart rate (in beats per minute [bpm]), for each session.
Comorbidity was quantified using the Charlson Index which assigns weights to different diseases.
Functional capacity was measured using the 6-min walking test (6MWT) in which, when instructed “walk as far as possible for 6 min”, the patient walked up and down a 30-m corridor (which was marked on the ground every two metres) as far as possible over 6 min. The total distance covered (in metres) was recorded.
The result variables were registered during two periods, the first one covered the 12 months prior to the intervention (Period 1) and the second one recorded the 12 months following the start of the intervention (Period 2). Healthcare resource consumption and costs were measured for several variables: the number of episodes and total amount in euros spent on laboratory tests, radiology tests, hospital pharmacy and medical assistance as Outpatient visits, Emergency healthcare provision, and Hospitalization. To measure consumption, the number of episodes was quantified. To measure costs, micro-costing analysis was used to estimate, in detail, the amount of each resource component used. The information used in this study was derived from different databases linked to each patient’s electronic medical record, as well as from the business resource planning software used for financial management by the healthcare department.
In order to obtain the cost of each of the activities, the information was structured into an in-house cost allocation model based on the ABC (activity-based costing) analytical accounting standard. Briefly, this method divides the production of an organisation into core activities, defines the costs for these activities, and assigns these costs to all the products and services according to the actual consumption of each activity [28, 29]. In this case, the fixed and variable costs considered in this study included supply (medical supplies, pharmacy, energy, etc.), human resources (healthcare and support staff), and financial costs (interests, commissions, and other costs).
Statistical analysis
The sample size calculation was performed using the main cross-over clinical trial data and was based on detecting changes in physical function as measured by the gait speed test. Thus, we used G*Power software to calculate that a minimum of 16 participants would be required to detect an effect size of 0.459 via ANOVA repeated measures with ‘within-between’ interactions.
Continuous variables were presented as the mean and standard deviation and categorical variables were shown as the relative and absolute frequencies.
Several linear mixed models, one per dependent variable, were adjusted in order to assess the differences between the two periods of time in terms of healthcare costs (in euros) and medical assistance.
All the models also included age, gender, comorbidity, functional capacity, and adherence as covariates (fixed effects) and no interactions were included in the models. To account for the non-independence of observations in the case of repeated measures per patient (variables measured for the same patient in two different periods of time), a random intercept was added to the linear models with the patient used as a random factor. All the statistical analyses were performed with R software (version 3.4.1) mainly by using the lme4 package (version 1.1–17).