Device-measured physical activity in patients with primary liver cancer: protocol for a prospective cohort study

Aims and objectives

The main aim of this study is to investigate the relationship between PA and clinical outcomes in patients diagnosed with primary liver cancer.

The study objectives are as follows:

Patient and public involvement

Patients and members of the public were not directly involved in developing the research questions, designing this sub-study, selecting the outcome measures, developing the recruitment procedures, or planning the analyses described in this protocol. Participants are involved in the study as research participants rather than as patient or public contributors. Findings from the study will be disseminated through peer-reviewed publications and scientific meetings. Opportunities to communicate the findings in an accessible format to patients and relevant patient communities will also be considered as part of dissemination.

Study design

This study is a single-centre, prospective cohort study conducted as part of the Hepatocellular Carcinoma Expediter Network HUNTER, a UK-led international collaboration that aims to improve the understanding and treatment of HCC (IRAS project ID: 266624).

To provide contextual comparison, the study will incorporate a historical independent healthy control group drawn from the ICICLE cohort—a population-based study of community-dwelling older adults (Del Din, 2016; Del Din, 2019; Rehman, 2020; Kirk, 2023). Control participants from ICICLE will be selected using matched sampling based on age and sex, ensuring demographic comparability with the experimental group.

Participant eligibility

The inclusion criteria for the primary liver cancer group are as follows:

Exclusion criteria include:

Recruitment and consent

Potential participants will be recruited from outpatient liver cancer clinics at Freeman Hospital – Newcastle Hospitals NHS Foundation Trust, where the study will be briefly introduced and explained by the consultant hepatologist and a research team member. If the participant expresses interest in the study, they will be provided with a participant information sheet with further details about the study, including its aims, procedures, and potential risks discussed. Informed consent will then be obtained in writing by a research nurse on-site at Freeman Hospital.

Once consent is obtained, participants will be fitted with the IMUs by the study team. If a research team member is not present at the patient consultation, the IMUs will be posted to the participant’s home address. In this case, the study procedures and participation details will be rediscussed over the phone by a member of the research team. During data collection, participants will have the opportunity to contact the research team to discuss any concerns, ask questions about the study or the devices, and report any issues with the equipment.

Exposures

IMU-measured physical activity:

Device-based PA will be measured using the Axivity AX6 (Axivity Ltd, UK) activity monitor, a 6-axis IMU designed to capture detailed movement data over an extended period. AX6 is a reliable and validated tool for capturing accelerations representative of PA (Echevarría-Polo et al., 2025). These IMUs have been widely used in research on older adults (Doherty et al., 2017, Maylor et al., 2023). The Axivity AX6 measures 23 mm x 32.5 mm x 8.9 mm and weighs 11 grams. It features a 6-axis IMU that tracks both linear acceleration and angular velocity at high frequencies, providing a highly granular record of PA. More precisely, the Axivity records movement in brief epochs (e.g., 5-second intervals), allowing for a detailed, continuous assessment of physical activity and sedentary patterns (Doherty et al., 2017). The device is water-resistant (IPX8 rating), allowing participants to wear it while showering or engaging in water-based activities. The Axivity AX6 is also equipped with a rechargeable battery, which can support over seven days of continuous data collection. These devices are configured using the Open Movement software (version 1.0.0.43). The configuration settings for the devices ensure the collection of acceleration data at a frequency of 100 Hz and range set to 8 g, which is optimal for capturing the necessary movement patterns and activity levels in older adults (Gafoor et al., 2024); the accelerometers are calibrated to monitor the X (left-right), Y (superior-inferior), and Z (anterior-posterior) axes of movement. The gyroscope will be disabled, and recording will be initiated immediately after device disconnection from the docking station to ensure synchronisation with device configuration. Device clocks will be time-aligned before deployment, and data will be recorded in the native.CWA file format for subsequent processing.

Participants will be instructed to wear one fully charged Axivity AX6 device on their lower back, near the fifth lumbar vertebra (L5), secured with a custom hypoallergenic hydrocolloid adhesive patch specifically designed for long wear time periods (Design: Newcastle University (Alcock); Manufacturer: Linxens; formerly Nile AB). The device is secured in place with a plaster patch (Hypafix, BSN medical GmbH, Germany). Additionally, participants will wear the second fully charged wrist-band device on their non-dominant wrist. The wrist monitor should be worn snugly, but comfortably, ensuring that it does not slip or twist during wear. The height from the floor to the bottom of the devices (with arms relaxed down the body) will be noted by the researcher using a measuring tape once the devices are fit on the participants. Both devices will be worn continuously, including during sleep, to ensure continuous data collection over the full seven-day period. Participants will receive clear instructions on how to attach the devices at the beginning of the study if they receive them by post, and they will be advised to contact the study team immediately if they experience any discomfort or skin irritation. If a participant needs to remove or reattach the devices during the monitoring period, they will be given step-by-step instructions from the study team to ensure that the devices are positioned correctly and securely. At the end of the seven days, participants will return both devices using a pre-paid self-addressed envelope.

Once the data collection is complete, the raw data files from the Axivity AX6 activity devices will be downloaded using the Open Movement software (version 1.0.0.43) (OmGui, 2025). These files will be saved in the.CWA format and contain detailed time-series data capturing participants’ movements. To ensure participant confidentiality all extracted files will be pseudonymised. Each file will be stripped of any personally identifiable information (e.g., name, address, contact information), and unique identifiers will be used to track the data associated with each participant. Following pseudonymising, the data files will be securely stored using password-protected OneDrive cloud storage, ensuring both access control and data integrity.

International Physical Activity Questionnaire Short Form:

At the end of the seven-day PA monitoring, participants will also be asked to complete the IPAQ-SF. The IPAQ-SF is a widely used measuring tool that assesses an individual’s self-reported levels of PA (Healey et al., 2020). The IPAQ-SF consists of seven items that assess the frequency (days per week) and duration (minutes per day) of PA over the past 7 days. It captures time spent in three distinct intensity categories: walking, moderate-intensity PA (MPA), vigorous-intensity PA (VPA). In addition, the questionnaire records time spent sitting, providing an estimate of sedentary behaviour. A key feature of the IPAQ-SF is that it only records activities lasting at least 10 continuous minutes, distinguishing it from device-based measures such as the Axivity AX6, which capture all movement irrespective of bout length or intensity threshold. While the devices provide continuous activity data, the IPAQ-SF will offer additional insights into how participants perceive their own PA levels and their participation in different types of PA. The IPAQ-SF has been validated for use in diverse populations and has demonstrated good reliability and validity for measuring PA levels across a wide range of countries and cultures (Craig et al., 2003). Specifically, in older adult populations, the IPAQ-SF has shown adequate criterion validity when compared to accelerometry data. In a large validation study involving older Japanese adults (n = 325), Spearman correlation coefficients between the total IPAQ-SF score and total PA measured by accelerometer ranged from 0.42 to 0.53, depending on age and sex group (Tomioka et al., 2011), thus indicating moderate-to-good validity consistent with established thresholds for population surveillance tools (Craig et al., 2003; Lee et al., 2011; Helmerhorst et al., 2012). Weighted kappa coefficients ranged from 0.39 to 0.49, indicating moderate agreement in classification. Although test-retest reliability was modest (ICC ranging from 0.50 to 0.65) (Tomioka et al., 2011), these results suggest the IPAQ-SF remains a useful tool for capturing self-reported PA patterns among older adults. However, it is important to note that the IPAQ-SF is self-reported and relies on participants’ ability to accurately recall and report their PA. This introduces the potential for recall bias, where participants might overestimate or underestimate their activity levels based on memory, personal perception, or social desirability (Prince et al., 2008). To help mitigate these biases, participants will be instructed to complete the IPAQ-SF shortly after the 7-day monitoring period, reducing the time between the activity monitoring and recall.

Demographics and clinical information:

The characteristics to be reported are listed in Table 1. All the characteristics listed will be extracted from the HUNTER Data Directory (IRAS project ID 266624), except for Walking Aid/Mobility notes which will be collected during an on-site hospital visit of the participants or via phone calls. The information to be collected includes age and gender, as well as details on relevant medical conditions such as liver disease comorbidities, including cirrhosis, obesity, and type 2 diabetes (Harman et al., 2018). Further clinical information, including medication use, will be collected to account for any treatments the participants are currently undergoing, which could potentially influence PA levels and clinical outcomes.

Handgrip strength measured using a dynamometer will be extracted from the records. To assess skeletal muscle area at L3—a validated proxy for total body skeletal muscle mass—computed tomography (CT) scans of participants will be analysed (Mourtzakis et al., 2008; Shen et al., 2004a). The cross-sectional muscle area at this level strongly correlates with whole-body muscle mass measured by magnetic resonance imaging and dual-energy X-ray absorptiometry, supporting its use as a standardized method for estimating total skeletal muscle. Together, these measures will be used to assess muscle strength and levels of sarcopenia in the cohort.

External control cohort

Between June 2009 and December 2011, participants were recruited as part of the ICICLE-PD study (Incidence of Cognitive Impairment in Cohorts with Longitudinal Evaluation–Parkinson’s Disease), which aimed to investigate motor and cognitive progression in Parkinson’s disease (PD) (Yarnall et al., 2014). Axivity devices were worn on the lower back for 7 consecutive days. Participants were instructed to continue their usual activities. The device was placed on the trunk (lower back) rather than at a wrist, and data were processed with identical algorithms for walking-bout segmentation and gait-metric extraction across participants. A subset of participants and age- and sex-matched healthy controls took part in the ICICLE-GAIT study, which focused on detailed gait patterns and physical activity assessment using body-worn accelerometers (Del Din et al., 2016). These community-dwelling controls, free from neurological disease, provide a valuable external healthy control cohort for the present study. Although ICICLE-GAIT used Axivity AX3 devices and the present study uses AX6 devices, the sensor configurations are compatible, and identical raw-data processing algorithms will be applied.

Outcomes

Overall survival

Survival status will be collected as an outcome measure via the HUNTER registry. Data on overall survival will be obtained up to 24 months following the assessment. Information extracted will include vital status (alive or deceased) and, where available and if relevant, date of death. Harnessing registry-derived survival data offers a resource-effective method to ascertain mortality outcomes without imposing additional burden on participants or requiring extended follow-up visits (Kumar et al., 2020).

Treatment tolerance

Treatment tolerance will be estimated based on treatment-related symptoms and modifications to the planned treatment course recorded in the HUNTER registry and linked hospital electronic records. Treatment-related symptoms and toxicities will be identified from clinician-documented adverse events and, where available, graded according to the Common Terminology Criteria for Adverse Events (CTCAE v5.0) (National Cancer Institute, 2017; Freites-Martinez et al., 2021). Indicators of reduced treatment tolerance will include: (i) occurrence of grade ≥ 3 treatment-related adverse events; (ii) unplanned dose reductions; (iii) unplanned treatment delays of ≥7 days; (iv) permanent treatment discontinuation attributed to toxicity; and (v) unplanned hospital admissions or emergency department attendances primarily due to treatment side-effects. These markers are commonly used surrogates of suboptimal treatment delivery and lower relative dose intensity, which have been associated with poorer survival outcomes across a range of solid tumours (Lyman, 2009; Nielson et al., 2021; Qi et al., 2020).

For each primary liver cancer participant, a binary variable will be derived to indicate poor treatment tolerance (defined as meeting at least one of the above criteria), alongside secondary measures such as the total number of treatment-related adverse events and the presence of any grade ≥ 3 toxicity. Where dosing information is available, relative dose intensity (delivered/planned dose over the treatment period) will be calculated for systemic therapies as an additional quantitative marker of treatment tolerance, reflecting established methods for summarising chemotherapy dose delivery in oncology (Lyman, 2009; Nielson et al., 2021; Qi et al., 2020).

Data Processing

IMU-measured physical activity:

Accelerometry data processing

The raw triaxial accelerometer data collected from the Axivity AX6 movement devices will be processed and analysed using the validated GGIR package (version 3.2–0) within the R environment (version 4.4.1) (GGIR, 2019; R, 2024). GGIR is an open-source and widely adopted software tool for processing raw accelerometry data. It facilitates signal calibration, detection of non-wear periods, and computation of PA metrics, and has been used extensively in large-scale population studies such as the UK Biobank and Whitehall II cohort (van Hees et al., 2013; Doherty et al., 2017).

Each accelerometer file will be screened for calibration error, signal dropout, and non-wear time. Non-wear time is identified using GGIR’s validated algorithm, which detects sustained periods of low acceleration variability (e.g., standard deviation <13 mg for ≥60 minutes) indicative of device removal (van Hees et al., 2013). To ensure reliable estimation of habitual physical activity, participants must accumulate a minimum of four consecutive full days of wear time, defined as at least 16 hours per day (Trost et al., 2005; Doherty et al., 2017). Files failing to meet this threshold will be excluded from further analysis. Following quality assurance, GGIR will derive several PA and behavioural metrics from the raw acceleration data. A central variable of interest is average acceleration, quantified using the Euclidean Norm Minus One (ENMO). ENMO is a widely accepted metric that reflects overall movement intensity while accounting for gravitational acceleration. It is calculated as follows:

Where:

ENMO values are typically expressed in milligravity (mg). This metric serves as a proxy for overall activity volume and forms the basis for estimating time spent in different intensity domains—such as sedentary, light, moderate, and vigorous activity—using validated threshold values (van Hees et al., 2014; Hildebrand et al., 2014). All output metrics of interest are described in Table 2 below.

Accelerometer data will be processed using the non-dominant wrist cut-off points established for older adults by Fraysse et al. (2021). The thresholds defining activity intensities will be: < 42.5 mg for sedentary behaviour, 42.5–97.9 mg for light activity, and ≥ 98 mg for moderate-to-vigorous physical activity (MVPA).

These cut-off points were specifically derived in a sample of community-dwelling older adults (mean age ≈ 77 years), ensuring the thresholds reflect the lower energy expenditure and movement patterns typical of this age group. Because energy cost for a given activity increases with age, acceleration thresholds corresponding to equivalent metabolic intensities (e.g., ≥ 3 METs) are lower than those observed in younger adults. Adopting these age-appropriate, non-dominant wrist thresholds improves accuracy in classifying activity intensity among older participants.

Gait

Data analysis for gait metrics will be conducted using a standalone executable built from validated MATLAB® scripts (Del Din, 2016; Hickey et al., 2017). This pipeline allows for the segmentation and processing of raw acceleration data by calendar day and incorporates a logical heuristics-based algorithm designed to identify and quantify walking bouts (WBs). The algorithm has been validated for use in free-living conditions and demonstrates high accuracy in detecting walking activity and estimating step counts (Hickey et al., 2017; Lara et al., 2016; Brodie et al., 2015). Ambulatory bouts are defined as periods of continuous walking activity with at least four consecutive steps and are identified by applying selective thresholds based on the magnitude and standard deviation of the triaxial acceleration signal. A maximum resting period of 2.5 seconds between steps is used to determine the continuity of walking.

Once extracted, the gait data will be processed and summarised into two levels of outcomes based on a validated framework: Macro (behavioural) and Micro (performance) gait characteristics (see Table 3 below). Macro gait characteristics describe walking activity over the full seven-day period. These include the total amount of time spent walking per day, the proportion of time walking relative to total wear time, the total number of steps and ambulatory bouts per day, and the average length of walking bouts. A non-linear descriptor known as alpha (α) will be calculated to evaluate the distribution of bout lengths, with higher alpha values indicating that daily walking time is dominated by short bouts, and lower values suggesting the presence of longer sustained walks (Chastin and Granat, 2010; Del Din et al., 2016). Variability in walking behaviour will also be assessed using the S² metric, which captures the within-subject variability in bout length across the monitoring period (Del Din et al., 2016; Mc Ardle et al., 2022; Hinchliffe et al., 2024).

Micro gait characteristics provide detailed insight into the quality of walking and are calculated for each individual ambulatory bout before being averaged across the monitoring period. These characteristics are drawn from five domains: pace, rhythm, variability, asymmetry, and postural control (Lord et al., 2013; Del Din et al., 2016). Using event detection methods, initial and final contacts during each step are identified to estimate temporal gait parameters such as step time, stance time, and swing time. Step length is estimated using the inverted pendulum model, and step velocity is derived as the ratio of step length to step time. Micro gait variability is calculated as the standard deviation of step time across all detected steps, while asymmetry is evaluated as the absolute difference between left and right step parameters, averaged across all bouts (Godfrey et al., 2015). Postural control characteristics, such as the swing-to-stance ratio, are also derived to assess dynamic balance during gait (Lord et al., 2013; Del Din et al., 2016).

All ambulatory bouts consisting of more than three steps are included in the analysis to ensure sufficient data quality for characterising gait parameters. Micro gait metrics are averaged across bouts for each participant, while Macro outcomes are calculated from the aggregated seven-day dataset. As part of an exploratory analysis, gait outcomes will also be stratified based on bout duration to examine differences in gait characteristics across various walking contexts. Ambulatory bouts will be grouped into short-to-medium (10–30 seconds), medium-to-long (30–60 seconds), and long (≥60 seconds) durations, corresponding approximately to 15–50, 50–100, and over 100 steps respectively. This stratification allows for an examination of gait behaviour in both brief and sustained walking episodes.

International physical activity questionnaire short form:

Data obtained from the IPAQ-SF will be processed according to the guidelines developed by the IPAQ Research Committee (2005). Upon collection, IPAQ-SF data will be screened for completeness, plausibility, and internal consistency in accordance with the IPAQ scoring protocol (IPAQ Research Committee, 2005). Responses will be checked for missing or illogical values (e.g., duration reported without frequency, >7 days/week). Implausible values, such as activity bouts exceeding 180 minutes per day or total activity time > 960 minutes per day, will be truncated. For each domain—walking, moderate activity, and vigorous activity—total weekly time will be calculated by multiplying the number of reported days by the average number of minutes per day.

To quantify overall PA, the IPAQ-SF uses the concept of Metabolic Equivalent of Task (MET), which represents the energy cost of physical activities. Standard MET values are assigned to each category: 3.3 METs for walking, 4.0 METs for moderate-intensity activity, and 8.0 METs for vigorous-intensity activity. MET-minutes per week are then computed for each activity type by multiplying the MET value by the minutes of activity per day and the number of days per week. These are summed to produce a total PA score expressed in MET-minutes per week. Sitting time is assessed separately in the IPAQ-SF and will be reported as the average number of minutes spent sitting on a typical weekday, which is a proxy for sedentary behaviour (Aunger and Wagnild, 2022).

Muscle mass

The CT images will be analysed using ImageJ (Version 1.54p), an open-source image analysis software developed by the National Institutes of Health (ImageJ, 2024; Shen, 2004b). ImageJ is widely used in medical imaging research for its flexibility, reproducibility, and capability to process DICOM-formatted images (Ishida et al., 2020).

Axial CT images will be selected at the anatomical level of the L3, where both transverse processes are visible. Using ImageJ software, the cross-sectional skeletal muscle area (SMA) at L3 will be quantified in square centimetres (cm²). DICOM images will first be calibrated using embedded metadata to ensure accurate pixel-to-distance conversion. A predefined Hounsfield Unit (HU) range of −29 to +150 HU will be applied to isolate skeletal muscle tissue based on radiodensity (Mourtzakis et al., 2008). Manual or semi-automated segmentation will then delineate the relevant muscle groups at this level—including the psoas, erector spinae, quadratus lumborum, and abdominal wall muscles. The total muscle area within these boundaries will be computed and normalised for stature to derive the Skeletal Muscle Index (SMI), calculated as:

Normalisation by height squared allows comparison between individuals of different body sizes and aligns with standardised definitions of sarcopenia (Martin et al., 2013; Cruz-Jentoft et al., 2019).

This imaging-based estimate of muscle mass will be paired with average handgrip strength (HGS), a widely accepted measure of muscle strength, to explore the prevalence and severity of sarcopenia within the study cohort (Cruz-Jentoft et al., 2019). Sarcopenia will be defined according to the revised European Working Group on Sarcopenia in Older People (EWGSOP2) criteria, which emphasize low muscle strength as the primary indicator of probable sarcopenia, confirmed by the presence of low muscle quantity as determined from L3 skeletal muscle cross-sectional area (Cruz-Jentoft et al., 2019). Specifically, low muscle strength will be defined as a mean handgrip strength of <27 kg for men and < 16 kg for women, following EWGSOP2 thresholds. Low muscle mass will be defined as a skeletal muscle index (SMI)—the skeletal muscle area at L3 normalized for height squared (cm²/m²)—of <55 cm²/m² for men and < 39 cm²/m² for women, as per established sex-specific cut-offs validated for CT-based assessment (Mourtzakis et al., 2008).

Sarcopenia-related measures, including L3 skeletal muscle area and handgrip strength, will be analysed to characterise muscle health in the cohort. Participants will be classified as sarcopenic or non-sarcopenic using validated cut-points for skeletal muscle index (SMI) and grip strength (Cruz-Jentoft et al., 2019).

Data and statistical analysis

Sample size

This study is embedded within the existing HUNTER HCC cohort and is therefore event-driven and pragmatically constrained by the number of eligible patients who can be approached, consented, and monitored with wearable devices within the study timelines. Based on current recruitment rates, clinic capacity and device availability, we judged that recruiting 100 participants with valid baseline device-measured PA data is feasible and acceptable in the context of an advanced cancer population with high symptom burden and limited prognosis. Contemporary HCC cohorts report approximately 50–70% two-year all-cause mortality, depending on tumour stage and treatment mix (Forner et al., 2018; Yang et al., 2019; Reig et al., 2022). Assuming a conservative 60% mortality over 24 months, a sample of 100 participants is expected to yield around 60 deaths, which is the primary driver of statistical power for the planned Cox proportional hazards analyses (Schoenfeld, 1983). With approximately 60 events and assuming even distribution of participants across PA exposure categories, we anticipate being able to fit a parsimonious multivariable model with around six prespecified covariates in addition to PA, consistent with the commonly cited rule of at least 10 events per variable to reduce overfitting and improve model stability (Peduzzi et al., 1995). Under these assumptions, the study will have limited power to detect small associations, but should provide reasonably precise, hypothesis-generating estimates for moderate associations, defined here as hazard ratios in the region of 0.60–0.75 (i.e. a 25–40% lower hazard of death for higher versus lower PA categories), between higher PA and lower mortality in HCC, complementing evidence that low muscle mass and poor physical function are associated with worse survival in this population (Revoredo & Del Fabbro, 2023; Guo et al., 2023). The sample of 100 participants will also provide adequate precision for the descriptive objectives, including estimation of the distribution of device-measured PA, gait parameters and body composition, and will generate the effect size and variance estimates required to design future, larger-scale observational or interventional studies.

Statistical analysis

All analyses will be conducted using R (version 4.4.1) with a significance level set at p < 0.05 for all inferential tests. Descriptive statistics will be used to summarise the baseline characteristics of the cohort, including demographic variables, clinical measures, self-reported PA, accelerometry-derived metrics, gait parameters, and sarcopenia indicators. Continuous variables will be presented as means and standard deviations (or medians and interquartile ranges, where non-normally distributed), and categorical variables as counts and percentages.

PA data derived from accelerometers will be analysed using metrics computed through the GGIR pipeline. Summary measures such as total PA volume (ENMO), time spent in sedentary, light, moderate, and vigorous activity, intensity gradient parameters, and bouted MVPA will be compared across participant subgroups and negative external control group defined by age, sex, disease stage, and other relevant clinical factors using t-tests, ANOVA, or non-parametric equivalents where appropriate.

Self-reported PA data from the IPAQ-SF will be used to calculate total MET-minutes per week and time spent in various activity domains. Agreement between device-measured and self-reported PA will be assessed using Bland–Altman plots, Spearman’s correlation coefficients, and intraclass correlation coefficients (ICCs) to explore convergence between self-reported and device-based measurements.

Overall survival will be analysed up to 24 months post-assessment using data obtained from the HUNTER registry. Cox proportional hazards regression models (Cox, 1972) will be the primary analytic framework to estimate associations between exposures and overall survival. Multivariable models will estimate the independent association of physical activity (PA) and gait characteristics with survival, adjusting for age, sex, disease stage, body mass index (BMI), and comorbidity burden. Where available, additional clinical covariates (e.g. smoking status, treatment modality) will be included in sensitivity analyses.

Primary PA exposures will be mean daily ENMO (mg) and time in moderate-to-vigorous physical activity (MVPA, min/day), derived from accelerometer data. For descriptive survival curves, each exposure will be dichotomised at the cohort median (bottom 50% vs top 50%). For inferential analyses, PA exposures will be modelled as continuous variables (scaled per 10 mg ENMO and per 30 min/day MVPA). Gait characteristics (e.g. gait speed) will be analysed analogously as continuous variables (scaled per 0.1 m/s), with median splits used only for Kaplan–Meier (KM) plots. In secondary analyses, exposures will be categorised into quartiles (Q1–Q4), and linear trends across quartiles will be tested by modelling the quartile medians as continuous terms (Harrell, 2015).

To assess potential non-linear associations, restricted cubic spline Cox regression models will be fitted with four knots placed at the 5th, 35th, 65th and 95th percentiles of each exposure (reference = median) (Durrleman and Simon, 1989; Royston and Sauerbrei, 2007). Adjusted hazard ratios (HRs) with 95% confidence intervals will be reported across the exposure range, accompanied by a global Wald test for non-linearity. Dose–response relationships will be visualised using spline plots displaying HRs relative to the reference value.

For descriptive purposes, KM curves will be generated for high vs low PA strata (median split) and compared using log-rank tests (Kaplan and Meier, 1958; Mantel, 1966). The proportional hazards assumption will be evaluated using Schoenfeld residuals (Therneau and Grambsch, 2000); where violations are detected, time-dependent interaction terms or stratified Cox models will be applied. Sensitivity analyses will include exclusion of early deaths (≤30 days post-assessment) to mitigate reverse causation, as well as sex-specific median and quartile categorizations of PA and gait exposures.

Comparative analyses with the ICICLE control group will be conducted to contextualise PA and gait profiles observed in the liver cancer cohort. Because ICICLE provides demographic and raw accelerometry data only, comparative analyses will focus on PA volume (ENMO), bout characteristics, and gait metrics. Between-group differences will be assessed using independent samples t-tests for normally distributed variables or Mann–Whitney U tests for non-normally distributed variables (Field, 2013).

Ethical approval

This study has received ethical approval from the appropriate regulatory bodies as a sub-study of HUNTER (Hepatocellular Carcinoma Expediter Network). Specifically, the HUNTER study has been granted ethical approval under REC reference number 19/NE/0251 and is registered with the IRAS project ID 266624. As outlined in Amendment SAM-05, the study protocol, patient information sheets, and consent forms were updated on 08/11/2024 to include the collection of exercise data from a subset of up to 100 HUNTER patients to reflect the study described.