Eligibility criteria and recruitment.

The eligibility criteria (Table 1) restrict the cohort to patients undergoing unilateral TKA as a treatment for OA, thereby reducing clinical heterogeneity that could confound recovery trajectories. Exclusion of non-OA indications, revision procedures, and individuals unable to reliably engage with the wearable ensures a sample with sufficient capacity for sustained device adherence and valid longitudinal mobility measurement. To ensure adequate baseline data collection, patients are required to be enrolled at least 7 days prior to surgery. Patients outside a four-week window relative to their surgery date will not be included.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Eligibility criteria.

https://doi.org/10.1371/journal.pone.0345667.t001

Study size.

A study sample of up to 160 participants will be sought over the two-year recruitment period. Given the exploratory nature of this study, no formal sample size or power calculation was performed. The planned study size was determined pragmatically based on feasibility, anticipated recruitment rates and previous orthopaedic studies using wearable devices [17].

Data collection.

Wearable-derived mobility data will be captured using the Garmin Vivosmart 5, which streams minute-level step count data through the Labfront platform. Step counts are generated from Garmin’s proprietary accelerometer-based step-detection algorithms. From these data, higher-order metrics will be derived to characterise post-operative mobility recovery, including daily step count, step-positive minutes (number of minutes with ≥1 recorded step), activity window length (time between first and last step-positive minute), longest inactivity streak (maximum consecutive zero-step minutes), diurnal activity distribution (steps per hour and AM:PM ratio), and day-to-day variability (standard deviation and coefficient of variation of daily steps within rolling 7-day windows).

Sleep duration, resting heart rate (RHR) and heart rate variability (HRV) will be retained as exploratory secondary measures rather than core outcomes. Night-time wear is optional to reduce participant burden. For participants who choose to wear the device overnight, sleep duration will be derived from actigraphy-based algorithms integrating movement and optical heart-rate signals. RHR and HRV will be passively collected through the Vivosmart 5’s wrist-worn photoplethysmography sensor and processed through Garmin’s proprietary algorithms using the Labfront platform. In a laboratory validation study, Garmin devices demonstrated over-estimation of sleep duration by an average of 44 minutes, fair agreement for heart rate (ICC 0.41) with wide error margins (–20 to +30 bpm), poor HRV performance with substantial bias (–22.4 Ms), wide limits of agreement (±92 Ms) and low reliability (ICC 0.24) compared with ECG-derived HRV [18]. These findings indicate that sleep, RHR and HRV lack the stability and validity required for core study outcomes.

However, RHR and HRV provide numeric representations of general physiological function [19], with HRV is recognised as an indicator of autonomic nervous system activity [20]. Therefore, fluctuations in these signals may help contextualise unusual mobility patterns, such as transient reductions in step count during periods of illness or physiological stress. Additionally, poor sleep is associated with mental health disorders and pain sensitization [21,22], both of which may affect post-operative mobility.

PROMs will be collected electronically using the Labfront mobile platform. Onboarding will include installation of the Labfront Companion application and patient education on PROM completion. Schedule and procedure for PROM collection is summarised in Table 2.

From every Saturday post-onboarding to Day 210, participants will complete weekly questionnaires on pain, stiffness, fatigue, and sleep quality using a 0–10 Numerical Rating Scale (NRS). These PROMs will refer to the participant’s experience over the previous 7 days (e.g., “How fatigued have you felt over the last 7 days on a scale of 0 to 10, where 0 indicates no fatigue and 10 indicates maximal fatigue”). Participants will receive a push notification at 9am to complete the four questionnaires, if the questionnaires have not been completed by 7 pm that evening, participants will receive a reminder push notification. Questionnaires close at 9 pm every Saturday evening and will vanish from participant dashboards on the Labfront Companion application.

The OKS will be administered at 30-day intervals, beginning on Day 1 post-onboarding and repeated on Days 30, 60, 90, 120, 150, 180, and 210. The EQ-5D-5L will be administered at 60-day intervals on Days 1, 60, 120, and 180. For both instruments, participants will receive automated push notifications at 9am on the scheduled assessment day, and questionnaires will remain accessible for a 7-day completion window.

This schedule is designed to capture short-term fluctuations in symptoms, medium-term changes in joint-specific function, and longer-term improvements in health-related quality of life, and is consistent with national joint registry practice [23]. All PROMs will be time-stamped and monitored for completion within the Labfront platform, which issues automated reminders and records compliance.

Clinical indicators will be abstracted from the electronic medical record (Meditech). These will include: (i) comorbidities as listed in the preoperative assessment; (ii) surgical method (manual vs robotic assisted) and prosthesis characteristics; (iii) hospital length of stay (LOS) (iv) post-operative complications (superficial or deep infection, wound dehiscence, delayed healing); (v) adverse events (venous thromboembolism [DVT/PE], cardiovascular events [myocardial infarction, arrhythmia], pneumonia, urinary tract infection, in-hospital fall, post-operative delirium); and (vi) reinterventions (joint aspiration or injection, manipulation under anaesthesia, washout or debridement, any reoperation, revision referral, or revision surgery). These variables will be coded as binary or categorical indicators with event dates where applicable. They will be used in both descriptive and adjusted analyses.

Clinical and demographic measures will be collected during onboarding and cross-checked against the electronic medical record. These will include age, sex, height (cm, assessed using a stadiometer), weight (kg, assessed using calibrated scales), body mass index (BMI, calculated as kg/m²) and participation in prehabilitation exercise classes. An overview of data categories and measures collected is presented in Table 3.

Baseline covariates used in this statistical analysis will include age, sex, BMI, baseline OKS, baseline EQ-5D-5L, baseline pain, hospital LOS, surgical method (manual vs robotic assisted) and participation in prehabilitation exercise classes.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Schedule and procedures for PROM collection.

https://doi.org/10.1371/journal.pone.0345667.t002

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Overview of data categories and measures collected in the study.

https://doi.org/10.1371/journal.pone.0345667.t003

Secondary criteria 1: Exploring recovery-phenotype groups.

Daily step count profiles will be smoothed using locally estimated scatter-plot smoothing (LOESS) to reduce noise and emphasise underlying patterns, consistent with prior digital recovery analytics [28]. Shape-based k-means clustering will then be applied to these smoothed trajectories, exploring cluster solutions from 2–5 groups. Clustering results will be evaluated using silhouette indices and clinical interpretability.

Clustering methods will be used to identify descriptive recovery phenotypes. Latent trajectory models such as growth mixture models (GMMs) and group-based trajectory models (GBTMs) will be used to estimate latent classes representing underlying recovery patterns. These approaches are complementary, with clustering providing empirical grouping and latent models characterising probabilistic recovery structures.

Latent class model selection will follow established statistical guidelines, including AIC, BIC, entropy, and minimum class sizes [29]. GMMs will allow within-class variability. GBTMs will model homogeneous within-class trajectories.

Secondary criteria 2: Exploratory early-late temporal association analyses.

Early post-operative wearable-derived step count and PROMs will be examined to determine whether they are temporally associated with corresponding medium-term outcomes at 6 months. Mixed-effects regression models will be specified with fixed effects for time, the outcome measure, and a time × measure interaction to allow the magnitude of associations to vary across the post-operative follow-up period.

For within-variable temporal analyses (e.g., early step count in relation to subsequent step count), mixed-effects autoregressive model structures will be explored, such as stepₜ ~ stepₜ ₋ ₁ with participant-level random intercepts, to explicitly account for temporal dependence in repeated mobility measurements.

For between-variable temporal analyses, models will be adjusted for relevant baseline covariates which are specified in Table 4. Marginal effects will be estimated within the prespecified 7-day analytic windows to characterise how early post-operative outcome measures relate to medium-term outcomes measures.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Overview of statistical analysis plan.

https://doi.org/10.1371/journal.pone.0345667.t004

Analyses will be performed using R (v ≥ 4.3).

Compliance monitoring.

Compliance will be tracked across two key domains (1) wearable adherence (2) PROM completion.

Wearable adherence.

Participants will be instructed to wear the Garmin Vivosmart 5 continuously from up to four weeks pre-operatively to six months post-operatively, with daytime wear expected and night-time wear optional. Adherence will be monitored using the Labfront platform.

A valid day will require >12 hours of wear time between 07:00–22:00, based on evidence that shorter wear leads to biased mobility estimates. A valid week will require ≥3 such days, reflecting Mobilise-D findings that this threshold achieves reliable weekly mobility values without introducing excessive sample loss or selection bias [30]. To ensure adequate alignment between mobility measurement and clinical assessment, this study will incorporate a ± 14-day tolerance window around each PROM collection period. When wearable data do not fully cover the 7-day PROM window, mobility data may be drawn from up to 14 days before or after the scheduled assessment period to complete the sampling frame. This approach follows the methodology outlined in the Mobilise-D study protocol [11], which recommends allowing limited flexibility around clinical assessment dates to ensure sufficient real-world mobility data while maintaining appropriate temporal correspondence between digital and clinical outcomes.

Weeks with no valid days will be treated as missing. PROMs will be multiply imputed under a multilevel framework, while wearable summaries will not be imputed in the primary analysis; sensitivity analyses will include complete-case and model-based imputation approaches.

Implausible values will be excluded according to a priori rules: daily step counts >50,000; heart rate <30 bpm or >220 bpm sustained for >10 minutes; sleep duration <2 h or >14 h on valid-wear days; or HRV values outside the device reportable range or coinciding with non-wear segments. All exclusions will be logged. Sensitivity analyses will repeat primary models using winsorisation (1st-99th percentile) instead of exclusions.

Non-wear will be inferred from data availability within Labfront, defined by absence of step counts and heart-rate recordings over sustained periods, as the platform does not provide direct access to raw accelerometery or proprietary Garmin non-wear algorithms. Device firmware and Garmin app version updates will be documented manually, and analyses will include version-epoch indicators with stratified sensitivity analyses where algorithm changes occur. Periodic adherence feedback will be provided to participants.

PROM compliance.

Compliance will be summarised as the proportion of scheduled PROMs completed. Thresholds are set to guide participant support rather than to exclude data: a weekly target of ≥70% and a monthly target of ≥60% are considered realistic given the duration of follow-up. This aligns with evidence regarding participant adherence to PROMs, which shows completion rates fall between 50% to 80% [31]. All available PROM data will be included in the primary analyses; sensitivity analyses will compare results across adherence bands and under stricter completion thresholds.

Because missingness may be informative (e.g., participants with greater pain, fatigue, or complications may be less likely to complete PROMs), missing PROM data will be addressed using multilevel multiple imputation, which accounts for within-person correlation over time [32,33]. Sensitivity analyses will additionally explore departures from missing-at-random assumptions.

Analytical considerations.

Adherence metrics will be incorporated as covariates and explored in sensitivity analyses to assess their potential influence on recovery trajectories. Sub-group analyses will evaluate whether adherence varies systematically by demographic or clinical factors, recognising that missingness may be informative.

Bias and limitations.

As a single-centre investigation conducted at the Beacon Hospital, findings may have limited generalisability to other healthcare systems or rehabilitation settings. The observation window does not extend to long-term outcomes, where participant dissatisfaction often emerges more prominently [5–7]. Reliance on a consumer-grade device may introduce measurement error, although validation studies support acceptable accuracy for step count and heart rate under free-living conditions [34]. Firmware updates or proprietary algorithm changes may also affect data consistency; device versioning will be documented and sensitivity analyses undertaken where necessary. Finally, providing participants with adherence feedback may introduce reactivity bias, as behaviour can be influenced by awareness of monitoring.

Amendments and study termination.

Any substantive amendments to the study protocol, including changes to eligibility criteria, outcomes, statistical methods or data collection procedures, will be submitted for approval to the Beacon Hospital Research Ethics Committee prior to implementation. Amendments will also be documented and time-stamped on the OSF registration. Minor administrative changes that do not impact participant involvement or study methodology will be recorded in an internal amendment log. Early termination of the study will be considered only in the event of unforeseen safety concerns, logistical infeasibility, or withdrawal of institutional support. Should termination occur, all available data will be analysed to the extent possible, and a final report will be disseminated outlining the reasons for discontinuation and implications for interpretation.