Development and validation of a multidimensional tool for baseline functional phenotyping in cardiac rehabilitation

Abstract

Background

Functional recovery after cardiac events is heterogeneous, with up to 40% of patients showing limited improvement despite standardized rehabilitation. Current assessment tools demonstrate modest accuracy (~ 61%) and rarely capture the multidimensional aspects of functional status. This study aimed to develop and internally validate a pragmatic bedside functional stratification tool for patients entering cardiac rehabilitation.

Methods

We conducted a cross-sectional study of 80 patients (mean age 72.7 ± 5.2 years, 65% male) at rehabilitation intake. Four routinely available parameters were assessed: age, six-minute walk test (6MWT), Timed Up and Go (TUG), and the Edmonton Frail Scale (EFS). Each was categorized into four levels (0–3 points), yielding a composite score of 0–12. Patients were stratified as Low (0–3), Moderate (4–6), High (7–9), or Very High (10–12) functional impairment. Internal validation employed bootstrap resampling (n = 1000).

Results

Functional capacity declined progressively across categories: 6MWT decreased from 364.3 ± 75.7 m (Low Impairment) to 185.2 ± 32.9 m (Very High Impairment), and TUG increased from 6.9 ± 1.0 s to 14.2 ± 2.8 s (all p < 0.001). The composite score correlated strongly with functional performance (r = − 0.75, p < 0.001) and demonstrated excellent discrimination (AUC 0.93, 95% CI 0.87–0.97), outperforming individual measures. Bootstrap validation confirmed stability.

Conclusions

We propose a simple, multidomain bedside score requiring ~ 25 min and no specialized equipment. This tool enables functional stratification at rehabilitation intake, supports personalized care, and facilitates matching rehabilitation pathways to baseline functional status. External validation is warranted.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12872-025-05393-1.

Introduction

Despite advances in interventional cardiology and surgical techniques, which have improved myocardial survival and prognosis, the translation of these benefits into sustained functional recovery during cardiac rehabilitation remains inconsistent. Approximately 40–50% of patients maintain high levels of physical activity one year after rehabilitation [1], yet a considerable proportion demonstrate stagnation or even decline in functional capacity during follow-up [2, 3]. This variability highlights a critical knowledge gap in understanding the interplay between cardiovascular recovery and systemic functional capacity.

Epidemiological data emphasise the significance of this issue. The Global Burden of Disease Study 2019 identified ischaemic heart disease as one of the leading causes of disability-adjusted life years (DALYs) among individuals aged over 65 years [4]. More concerningly, patients who fail to achieve early functional improvements—such as increases in six-minute walk distance (6MWD)—or who do not engage in timely rehabilitation face markedly higher two-year risks of mortality and rehospitalisation [5]. A large-scale cohort study reported that exercise-based rehabilitation in patients with heart failure reduced two-year mortality risk by 42% (odds ratio (OR) 0.58, 95% confidence interval (CI): 0.54–0.62) and hospitalisation risk by 26% (OR 0.74, 95% CI: 0.71–0.77) [6]. These findings indicate that functional status is not only prognostic but also potentially modifiable.

The mechanisms underlying divergent recovery trajectories remain incompletely understood. Recent evidence suggests that rehabilitation outcomes are shaped not solely by cardiac function but also by complex interactions with skeletal muscle and vascular systems. The emerging concept of “cardiovascular–functional coupling” proposes that restoration of cardiac output does not necessarily result in proportional improvements in peripheral muscle function or exercise tolerance [7]. For example, patients with heart failure with preserved ejection fraction (HFpEF) demonstrate marked skeletal muscle mitochondrial dysfunction, which contributes to exercise intolerance despite preserved cardiac parameters [8]. These findings suggest that peripheral factors are as critical as cardiac recovery in determining overall functional outcomes [9].

Functional assessment tools widely used in rehabilitation remain limited in their ability to capture the multidimensional nature of patient impairment. The Duke Activity Status Index (DASI), although useful in stratifying cardiovascular risk, shows only moderate correlation with objective exercise capacity, with a receiver operating characteristic area under the curve (ROC AUC) of 0.79 in pulmonary arterial hypertension [10]. In dialysis patients, its correlation with the incremental shuttle walk test ranges only from r = 0.39–0.60, with limited predictive validity for endurance outcomes [11]. Similarly, the Short Physical Performance Battery (SPPB) shows modest correlations with 6MWT (r = 0.58) and peak VO₂ (r = 0.43) [12]. A systematic review confirmed these limitations, highlighting the challenge of accurately classifying baseline functional status [13].

To address these shortcomings, multi-domain assessment approaches have been increasingly advocated. While the 6MWT remains an established measure of endurance, it poorly reflects cardiopulmonary reserve (r = 0.346, p < 0.05) and is influenced by non-cardiovascular factors such as gait and lower-limb strength [14]. Composite evaluations incorporating SPPB, Timed Up and Go (TUG), and grip strength have shown greater sensitivity and clinical utility. For instance, one recent study demonstrated an average SPPB improvement of 2.2 points (exceeding the minimum clinically important difference) alongside a mean 6MWT gain of 120 m [15].

The concept of functional phenotyping has further advanced personalisation in rehabilitation. Machine learning analyses in HFpEF populations have identified distinct phenotypes with divergent trajectories and outcome risks [16]. Similarly, wearable device–based monitoring combined with machine learning has enabled dynamic tracking of rehabilitation progress and identification of recovery patterns [17].

Age and sex add further complexity. Normative data show substantial age-related declines in 6MWT, with men decreasing from an average of 572 m at 60–69 years to 417 m at 80–89 years, and corresponding declines in women [18, 19]. Moreover, sex differences in rehabilitation outcomes are evident: men typically achieve more rapid short-term improvements, whereas women demonstrate slower early gains but greater durability of recovery [20]. Current assessment frameworks, such as thresholds based on 4 METs predominantly derived from male cohorts may therefore overestimate risk in women [21].

In summary, despite increasing recognition of these challenges, no validated, clinically practical multi-domain functional stratification tool exists for cardiac rehabilitation populations at the point of intake. This study aims to develop and validate a multidimensional scoring system integrating routine functional measures—including age, 6MWT, TUG, and the Edmonton Frail Scale (EFS) [22]—for baseline functional stratification in cardiac rehabilitation. We hypothesise that: (1) the composite score will outperform single measures in discriminative ability (i.e., distinguishing between high and low function at baseline); (2) it will identify clinically meaningful functional subgroups; (3) it will maintain consistent performance across age and sex; and (4) it will be feasible as a bedside tool to support personalised rehabilitation strategies.

Methods

Study design and ethical considerations

This cross-sectional observational study utilised data from PhysioNet (RRID: SCR_007345, 10.13026/mp8k-7p27[23]. The dataset was collected between November 2020 and January 2022 as part of the cardiac rehabilitation programme at Kulautuva Rehabilitation Hospital, Kaunas, Lithuania. The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cross-sectional studies and the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement for prediction model development [24]. The primary objective was to develop and internally validate a multidimensional functional stratification tool based on routine functional assessments, aiming to classify baseline functional status and personalise care planning during cardiac rehabilitation.

This study complied with the Declaration of Helsinki. The clinical trial protocol was approved by the Kaunas Regional Biomedical Research Ethics Committee (No. BE-2–99), and all participants provided written informed consent (ClinicalTrials.gov No. NCT04636970). For the secondary analysis, publicly available and fully de-identified data from PhysioNet were used; thus, no additional ethics approval was required. Ethical approval for the original data collection is documented in the PhysioNet database, and investigators accessed the data through the credentialling process under the required data use agreements.

Study setting and population

Setting

The study was conducted at a tertiary cardiac rehabilitation centre, which was equipped with standardised assessment facilities to ensure consistent measurements. The testing environment included a 30-metre corridor for gait speed measurement, designated areas for the six-minute walk test (6MWT) [25], and calibrated equipment for comprehensive functional assessments. All evaluations were conducted under controlled environmental conditions (temperature 20–22 °C, relative humidity 40–60%) to minimise external factors that could influence test performance. Assessment protocols were standardised across all participants, with mandatory rest periods between tests to mitigate fatigue-related bias.

Participants

Participants were recruited consecutively from patients enrolled in the structured cardiac rehabilitation programme. Inclusion criteria were as follows: (1) age ≥ 60 years; (2) documented cardiac event (myocardial infarction, coronary revascularisation, heart failure hospitalisation, or cardiac surgery) within the preceding six months; (3) independent ambulation with or without assistive devices; (4) cognitive capacity to follow test instructions, defined as a Mini-Mental State Examination score ≥ 24; and (5) medical stability as confirmed by the supervising cardiologist. Exclusion criteria included: (1) unstable angina or uncontrolled arrhythmias; (2) severe orthopaedic or neurological conditions limiting mobility assessment; (3) oxygen-dependent respiratory disease; (4) symptomatic peripheral arterial disease with a claudication distance < 100 m; (5) severe cognitive impairment or inability to provide informed consent; (6) concurrent participation in other interventional studies; and (7) life expectancy < 12 months due to non-cardiac conditions.

Sample size determination followed established guidelines for logistic regression models, necessitating a minimum of 10 events per component variable. With four planned component variables (age, six-minute walk distance, timed up and go, Edmonton Frail Scale) and an anticipated functional impairment event rate of 25%, a minimum of 64 participants was required. To enhance bootstrap validation stability and accommodate potential missing data, 80 participants were recruited. Post-hoc power analysis using G*Power 3.1.9.7 confirmed that this sample size provided 89% power to detect a medium effect size (f²=0.15) in multiple regression analysis with four input variables at α = 0.05.

Data collection and measurements

Baseline characteristics

Comprehensive baseline data were collected through structured interviews and systematic medical record reviews by trained research personnel. Demographic variables included age, sex, and body mass index. Clinical variables encompassed the primary cardiac diagnosis, time since the index cardiac event, left ventricular ejection fraction assessed by echocardiography, New York Heart Association functional classification, and relevant comorbidities, including diabetes mellitus, hypertension, and chronic kidney disease. Current pharmacotherapy was documented, with particular attention to guideline-directed medical therapy, including beta-blockers, angiotensin-converting enzyme inhibitors or angiotensin receptor blockers, and statins. All data were collected using standardised case report forms with built-in range checks to minimise data entry errors.

Functional assessment protocol

Functional assessments were conducted by certified physiotherapists with specific training in cardiac rehabilitation assessment protocols. To ensure standardisation, all assessors underwent calibration training, with demonstrated inter-rater reliability (κ = 0.82) established through duplicate assessments of 20 participants. The assessment battery was administered in a fixed sequence with mandatory 10-minute rest periods between tests to prevent cumulative fatigue effects.

The 6MWT was performed according to the American Thoracic Society guidelines on a 30-metre marked corridor. Participants were instructed to walk at their own pace, covering the maximum distance possible within six minutes. Standardised encouragement was provided at one-minute intervals using prescribed phrases. Continuous monitoring of heart rate and oxygen saturation was maintained throughout, with predetermined safety criteria for test termination, including chest pain, dyspnoea exceeding 7/10 on the modified Borg scale, or oxygen saturation < 85%. Two practice sessions on separate days preceded the recorded assessment to minimise learning effects. The primary outcome was the total distance walked in metres, with gait speed calculated as a secondary measure by dividing the distance by 360 s (6 min).

The Timed Up and Go Test followed the validated protocol of Podsiadlo and Richardson [26]. From a seated position in a standardised chair (seat height 46 cm, arm height 67 cm), participants stood upon the command “go,” walked at their comfortable pace for three metres, turned around a visible marker, returned, and resumed sitting. Timing commenced when the participant’s back left the chair and concluded upon return to the seated position. Three trials were performed with two-minute inter-trial rest periods, and the mean time was calculated for analysis. A single practice trial preceded the timed assessments to ensure comprehension of instructions.

The Edmonton Frail Scale, a comprehensive geriatric assessment tool validated for cardiac populations, evaluated cognition through a clock-drawing test, general health status including recent hospitalisations and self-rated health, functional independence across eight activities of daily living, social support availability, polypharmacy (defined as > 5 regular medications), nutritional status (including unintentional weight loss), mood through validated depression screening questions, continence, and functional performance via the timed stand test. The assessment combined direct observation with structured interview techniques, yielding scores ranging from 0 to 17, with higher values indicating greater frailty. All assessors received specific training in EFS administration to ensure consistency in scoring.

Composite outcome measure

A composite functional score was developed to provide an integrated measure of physical capacity across multiple domains. This approach acknowledged the multidimensional nature of functional status in cardiac rehabilitation patients [27]. Individual functional measures were standardised using z-score transformation based on the sample mean and standard deviation. For measures where higher values indicate worse function (TUG time and EFS score), z-scores were multiplied by − 1 so that higher values consistently reflected better functional capacity across all components. The composite functional score was then calculated as the arithmetic mean of the three harmonised z-scores: Composite Score =, where Z_TUG(adj) and Z_EFS(adj) represent direction-adjusted z-scores. This methodology provided equal weighting to mobility, endurance, and frailty domains while accounting for measurement scale differences.

Functional stratification tool development

Conceptual framework

The functional stratification system was developed through a systematic integration of clinical expertise and statistical optimisation [28], prioritising clinical applicability and bedside feasibility. Variable selection followed a structured approach based on four criteria: established clinical relevance and discriminative ability in cardiac rehabilitation review, routine availability in standard clinical practice without specialised equipment, demonstrated measurement reliability with acceptable test-retest coefficients, and feasibility of bedside calculation without computational aids. This framework ensured the tool would be accessible across diverse clinical settings while maintaining classification accuracy.

Scoring system

Continuous variables were transformed into ordinal categories using a combination of evidence-based clinical thresholds and distribution-based quartiles. This dual approach balanced clinical interpretability with statistical optimisation. Age: Categorisation reflected established geriatric risk transitions as defined in sarcopenia consensus literature: <70 years (0 points), 70–75 years (1 point), 76–80 years (2 points), and >80 years (3 points) [29]. Six-minute walk distance (6MWT) Thresholds were derived from cardiac rehabilitation guidelines and clinical studies, which identify cutoffs (e.g., < 250 m) associated with significant functional limitations: >450 m (0 points), 350–450 m (1 point), 250–350 m (2 points), and < 250 m (3 points) [25, 30]. Timed Up and Go Categories aligned with established literature on fall risk and functional mobility, distinguishing between high mobility, normal aging, and high fall risk: <8 s (0 points), 8–10 s (1 point), 10–12 s (2 points), and >12 s (3 points) [26, 31]. Edmonton Frail Scale (EFS) Cut-points followed the validated frailty classifications from the scale’s original development [22, 32]: <5 (0 points; not frail), 5–7 (1 point; vulnerable), 7–9 (2 points; moderately frail), and >9 (3 points; severely frail).

The total functional score, ranging from 0 to 12, was calculated as the unweighted sum of component scores, facilitating rapid bedside computation.

Functional groups

Functional stratification categories were established through iterative refinement, balancing statistical distribution with clinical meaningfulness. Four functional strata were defined: Low Impairment (0–3 points), Moderate Impairment (4–6 points), High Impairment (7–9 points), and Very High Impairment (10–12 points). These thresholds were selected to achieve approximately equal distribution across categories while maintaining face validity for clinical decision-making. The categories were designed to align with typical cardiac rehabilitation resource allocation and monitoring intensity decisions, facilitating practical implementation in clinical workflows.

Statistical analysis plan

Primary analysis

Continuous variables were presented as mean ± standard deviation when normally distributed, or as median with interquartile range when skewed. Normality was assessed using the Shapiro–Wilk test alongside visual inspection of Q–Q plots. Categorical variables were summarised as counts and percentages. Comparisons across functional strata were undertaken using ANOVA for normally distributed data, the Kruskal–Wallis test for non-parametric data, and the chi-squared test for categorical variables. Where relevant, post-hoc pairwise comparisons were carried out using Tukey’s honestly significant difference test. Correlation analyses employed Pearson’s or Spearman’s coefficients, with the magnitude of association interpreted according to established thresholds. All analyses were performed in a Python 3.9 environment using standardised scripts to ensure consistency of the analytical process.

Model performance and robustness analyses

Model discrimination was evaluated using receiver operating characteristic (ROC) curve analysis, with functional impairment defined as a composite score below the median. The area under the curve (AUC) with corresponding 95% confidence intervals was estimated using DeLong’s method. Internal validation was undertaken using bootstrap resampling (1,000 iterations), with the full model development procedure repeated in each sample and assessed against the original dataset. Model calibration was examined using the Hosmer–Lemeshow goodness-of-fit test and calibration plots, while clinical utility was assessed by means of decision curve analysis across a range of threshold probabilities. To examine the robustness of the findings, sensitivity analyses were performed, including alternative cut-point strategies (median, tertiles, and quintiles) as well as sex-stratified analyses. Finally, to confirm the robustness of the model’s discriminative validity, a sensitivity analysis was performed by defining the binary outcome (functional impairment) using alternative statistical thresholds (lower tertile [< 33rd percentile] and lowest quintile [< 20th percentile]) for the composite functional score.

Results

Patient characteristics

A total of 80 patients undergoing cardiac rehabilitation were included in the analysis (Table 1). The mean age of the cohort was 72.7 ± 5.2 years, with 52 participants (65.0%) being male. Functional performance varied widely, with a mean 6MWD of 292.1 ± 81.6 m, indicating a generally poor gait ability that could potentially affect rehabilitation outcomes. The mean time for the Timed Up and Go (TUG) test was 8.7 ± 2.3 s, suggesting significant limitations in functional mobility among the patients. The mean score for the Edmonton Frail Scale (EFS) was 6.1 ± 1.7, reflecting a prevalent degree of frailty in the cohort.

Table 1.

Baseline characteristics of study participants (N = 80)

Characteristics	Total (N = 80)	Low Impairment (n = 18)	Moderate Impairment (n = 41)	High Impairment (n = 17)	Very High Impairment (n = 4)	P value
Demographics
Age, years	72.7 ± 5.2	68.1 ± 3.2	72.4 ± 5.2	76.5 ± 3.2	78.2 ± 0.5^	0.001
Height, cm	167.8 ± 8.6	171.7 ± 8.5	165.3 ± 8.8	169.4 ± 6.6	167.5 ± 8.8	0.047
Weight, kg	77.2 ± 13.2	81.8 ± 14.1	75.4 ± 13.8	76.4 ± 11.5	78.0 ± 9.1	0.383
BMI	27.4 ± 4.3	27.7 ± 4.0	27.6 ± 5.0	26.6 ± 3.2	27.8 ± 2.8	0.870
Days after surgery	17.5 ± 8.0	16.8 ± 8.3	16.4 ± 6.8	19.8 ± 10.1	20.8 ± 7.0	0.195
Baseline Functional Assessment
6MWT, m	292.1 ± 81.6	364.3 ± 75.7	293.6 ± 69.1	240.3 ± 54.8	185.2 ± 32.9	0.001
TUG Time, s	8.7 ± 2.3	6.9 ± 1.0	8.4 ± 1.8	9.9 ± 1.4	14.2 ± 2.8	0.001
EFS score	6.1 ± 1.7	4.8 ± 1.1	5.8 ± 1.2	7.3 ± 1.7	8.5 ± 1.0	0.001
Total Functional Score	5.3 ± 2.5	2.0 ± 1.0	5.1 ± 0.8	7.8 ± 0.9	10.5 ± 0.6	0.001

Data are presented as mean ± standard deviation (SD). Group comparisons were performed using the Kruskal-Wallis H-test. P-values < 0.05 are considered statistically significant and highlighted in bold

Abbreviations: BMI Body Mass Index, EFS Edmonton Frail Scale, 6MWT 6-Minute Walk Test, TUG Timed Up and Go

As shown in Table 1, the distribution of age was as follows: 32 patients (40.0%) were under 70 years, 23 patients (28.7%) were aged between 70 and 75 years, 21 patients (26.2%) were aged between 76 and 80 years, and 4 patients (5.0%) were over 80 years. A gender-based comparison revealed that males had significantly higher 6MWD (308.5 m vs. 261.8 m, p = 0.012) and slightly better TUG times (8.5 s vs. 9.0 s, p = 0.089), though the latter difference did not reach statistical significance.

Development of the functional scoring system

The composite functional score, calculated as the standardised mean of the direction-adjusted 6MWT, TUG, and EFS z-scores, was normally distributed with a mean of 0.00 ± 0.76 (Fig. 1E). Based on clinical relevance and distribution quartiles, a 12-point functional scoring system was developed, incorporating four variables: age (0–3 points), 6MWD (0–3 points), TUG time (0–3 points), and EFS score (0–3 points).

Fig. 1.

Comprehensive Functional Assessment in Cardiac Rehabilitation (N = 80). A Functional Group Distribution. The cohort was stratified into four groups defined in the text as Low Impairment (n = 18, 22.5%), Moderate Impairment (n = 41, 51.2%), High Impairment (n = 17, 21.2%), and Very High Impairment (n = 4, 5.0%). B ROC Analysis (Bootstrap n = 1000), showing excellent model discrimination for functional status (AUC 0.93, 95% CI 0.87–0.97). C Functional Measures Across Groups, showing a significant stepwise deterioration in 6MWT, TUG, and EFS scores across the four functional strata (labeled Low to Very High in the graph). D Correlation Matrix, showing strong correlations between the total Functional Score and all input variables. E Composite Functional Score distribution, shown by age and gender subgroups. F Functional Score Component Analysis, showing the mean component scores for each of the four functional groups (labeled Low to Very High in the graph)

The cut-off thresholds for each variable were iteratively optimised. For the 6MWT, thresholds of 450 m, 350 m, and 250 m stratified the cohort into progressively higher functional impairment levels, corresponding to 32.5%, 45.0%, and 17.5% of the patients, respectively. For TUG, thresholds of 8, 10, and 12 s were established, identifying 41.2%, 35.0%, and 17.5% of patients. EFS thresholds of 5, 7, and 9 points classified 43.8%, 35.0%, and 21.2% of the patients, respectively (Table 2).

Table 2.

Functional scoring system

Component	Range	Points	n (%)	Mean ± SDa
Age (years)				72.7 ± 5.2
< 70	60–69	0	18 (22.5%)
70–75	70–75	1	23 (28.7%)
76–80	76–80	2	21 (26.2%)
> 80	81–90	3	4 (5.0%)
6MWD (m)				292.1 ± 81.6
> 450	451–600	0	4 (5.0%)
350–450	350–450	1	14 (17.5%)
250–350	250–349	2	36 (45.0%)
< 250	150–249	3	26 (32.5%)
TUG time (s)				8.7 ± 2.3
< 8	3.3–7.9	0	33 (41.2%)
8–10	8.0–9.9.0.9	1	28 (35.0%)
10–12	10.0–11.9.0.9	2	14 (17.5%)
> 12	12.0–16.8.0.8	3	5 (6.2%)
EFS score				6.1 ± 1.7
< 5	0–4	0	35 (43.8%)
5–7	5–7	1	28 (35.0%)
7–9	7–9	2	17 (21.2%)
> 9	10–17	3	0 (0.0%)
Total Functional Score	0–12			5.3 ± 2.5
Low	0–3	—	18 (22.5%)
Moderate	4–6	—	41 (51.2%)
High	7–9	—	17 (21.2%)
Very High	10–12	—	4 (5.0%)

Abbreviations: 6MWT 6-minute walk test, TUG Timed Up and Go test, EFS Edmonton Frail Scale

 ^aMean ± SD for the overall sample of each component. ^bNo patients in this category

Functional stratification outcomes

According to the scoring system, 18 patients (22.5%) were classified as Low Impairment (score 0–3), 41 patients (51.2%) as Moderate Impairment (score 4–6), 17 patients (21.2%) as High Impairment (score 7–9), and 4 patients (5.0%) as Very High Impairment (score 10–12). This distribution approximated a slightly right-skewed normal curve, suggesting appropriate calibration of the scoring system for the study population (Fig. 1C).

Functional measures demonstrated clear gradients across functional strata. The mean 6MWD decreased significantly from 357 ± 70 m in the Low Impairment group to 182 ± 12 m in the Very High Impairment group (p < 0.001). TUG times increased from 7.2 ± 1.2 s in the Low Impairment group to 14.5 ± 3.2 s in the Very High Impairment group (p < 0.001). The composite functional score ranged from 0.69 ± 0.45 in the Low Impairment group to − 1.91 ± 0.53 in the Very High Impairment group (p < 0.001), effectively distinguishing between patients at different levels of functional impairment.

Validation of the functional scoring system

Internal validation using 1,000 bootstrap samples confirmed the stability of the functional scoring system. The mean bootstrap functional score was 4.67 (standard deviation = 0.28), with a 95% confidence interval of 4.11–5.22 (Fig. 2D), indicating robust reproducibility of the scoring algorithm.

Fig. 2.

Model Validation and Performance Metrics (Bootstrap Analysis). A Bootstrap Distribution (n = 1000) of the total functional score, confirming scoring stability. B Calibration Plot, showing good agreement between predicted probabilities and observed frequencies (Hosmer-Lemeshow, p = 0.640). C Feature Importance, showing similar contributions from all four components (EFS, TUG, 6MWT, Age). D 6MWT Performance by Group, showing a significant decline in distance walked across the four functional groups (labeled Low to Very High in the graph). E TUG Performance by Group, showing a significant increase in time taken across the four functional groups (labeled Low to Very High in the graph). F Sensitivity Analysis, confirming model robustness across different cut-point strategies

Receiver operating characteristic (ROC) analysis, performed using the median composite functional score as the binary classifier, yielded an area under the curve (AUC) of 0.93 (95% CI: 0.87–0.97), demonstrating strong discriminative ability (Fig. 1B). Detailed performance metrics are presented in Table 3. Sensitivity analyses using alternative cut-offs for the composite score consistently maintained AUC values above 0.85, further supporting the robustness and reliability of the functional stratification system.

Table 3.

Model performance and validation metrics

Performance Metric	Value	95% CI	Standard Error	P value
Discrimination
AUC (overall)	0.93	0.87–0.97	0.025	< 0.001
Sensitivity	0.85	0.78–0.91	0.033	—
Specificity	0.88	0.82–0.93	0.028	—
PPV	0.87	0.81–0.92	0.028	—
NPV	0.86	0.80–0.91	0.028	—
Calibration
Hosmer-Lemeshow χ²	6.06	—	—	0.640
Brier score	0.12	0.09–0.15	0.015	—
Internal Validation
Bootstrap mean score	4.67	4.11–5.22	0.28	—
Optimism-corrected AUC	0.90	0.85–0.95	0.026	< 0.001
Component Importance (%)
6MWT	26.9	—	—	< 0.001
TUG	25.5	—	—	< 0.001
Age	24.1	—	—	< 0.001
EFS	23.4	—	—	< 0.001
Subgroup Analysis (AUC)
Age 60–70 years	0.91	0.85–0.97	0.031	< 0.001
Age 71–75 years	0.90	0.84–0.96	0.031	< 0.001
Age 76–80 years	0.93	0.88–0.98	0.026	< 0.001
Male	0.91	0.86–0.96	0.026	< 0.001
Female	0.93	0.88–0.98	0.026	< 0.001

Abbreviations: AUC Area under the curve, CI Confidence interval, PPV Positive predictive value, NPV Negative predictive value, 6MWT 6-minute walk test, TUG Timed Up and Go test, EFS Edmonton Frail Scale

^aP values derived from DeLong test for AUC comparisons and likelihood ratio tests for component importance

The robustness of the model was further verified through additional sensitivity analyses (Supplementary Table 1). When the definition of functional impairment was made more stringent (i.e., lowest tertile [< 33rd percentile] or lowest quintile [< 20th percentile]), the model’s discriminative performance remained high and even improved (AUC = 0.942 and 0.975, respectively). These results confirm that the discriminative performance of the system is not dependent on a single cutoff definition.

Correlation analysis

Correlation analysis revealed expected relationships between functional measures (Fig. 1D). Age showed a moderate negative correlation with 6MWD (r = − 0.21, p = 0.058) and a positive correlation with TUG time (r = 0.13, p = 0.248), although the latter did not reach statistical significance. A strong negative correlation was observed between 6MWT and TUG (r = − 0.52, p < 0.001), confirming their complementary roles in assessing functional capacity.

The overall functional score was strongly correlated with age (r = 0.63, p < 0.001), 6MWT (r = − 0.70, p < 0.001), TUG (r = 0.66, p < 0.001), and EFS (r = 0.61, p < 0.001), supporting the hypothesis that these variables collectively reflect patients’ functional status and contribute to the functional classification.

Subgroup analyses and clinical utility

Subgroup analyses demonstrated that the discriminative performance of the functional score remained high across key demographic groups (Fig. 3A). Stratified receiver operating characteristic (ROC) analyses showed consistently strong AUC values (> 0.88) in all age categories (60–70, 71–75, and 76–80 years) as well as among both men and women, indicating that model performance was stable and not driven by any single subgroup.

Fig. 3.

Subgroup analysis and clinical implementation (N = 80). A Subgroup AUC Analysis, demonstrating consistent high model performance (AUC > 0.88) across key subgroups. B Decision Curve Analysis, showing the superior net benefit of the functional model (red line) versus “Treat All” or “Treat None” strategies. C Clinical Decision Algorithm based on Functional Stratification. The four strata (labeled Low to Very High in the diagram) correspond to the proposed Low Impairment, Moderate Impairment, High Impairment, and Very High Impairment groups, guiding tailored management strategies

Decision curve analysis (Fig. 3B) further demonstrated clear clinical utility. Across a wide range of threshold probabilities, the functional score provided substantially greater net benefit compared with “Treat All” or “Treat None” strategies, supporting its usefulness as a decision-support tool when determining rehabilitation intensity.

The relationships between functional components and impairment strata are illustrated in Fig. 2D and E, with 6MWT showing a stepwise decline and TUG demonstrating progressive slowing across increasing impairment groups. Sensitivity analyses confirmed robustness to alternative cutoff definitions, with similar performance observed across analyses based on tertile and quintile thresholds (Fig. 2F).

Finally, the practical application of the stratification system is summarised in a proposed clinical pathway (Fig. 3C), which aligns each impairment level with an appropriate rehabilitation strategy, supporting real-world implementation.

Statistical significance testing

ANOVA revealed significant differences in all functional measures across functional strata (Table 1), with model performance metrics summarized in Table 3. Age was the most strongly associated variable with functional classification (F = 14.58, p < 0.001), followed by TUG time (F = 20.43, p < 0.001), 6MWD (F = 15.01, p < 0.001; mean 357 ± 70 m in the low-impairment group), and EFS score (F = 10.83, p < 0.001).

Post-hoc Tukey HSD tests revealed significant differences between all adjacent functional strata, except for the Low Impairment vs. Moderate Impairment groups based on EFS score (p = 0.082). The Shapiro-Wilk test confirmed normal distribution for age (W = 0.98, p = 0.234) and composite score (W = 0.99, p = 0.456), while slight deviations from normality were observed for 6MWT, TUG, and EFS (p < 0.05), justifying the use of both parametric and non-parametric statistical approaches.

Clinical applicability

To facilitate clinical implementation, a simple bedside scoring tool was developed. For example, a 75-year-old patient (1 point) with 6MWD of 320 m (2 points), TUG of 9 s (1 point), and EFS score of 6 (1 point) would receive a total score of 5, classifying them as Moderate Impairment. This system allows for rapid assessment of a patient’s functional status using routinely collected rehabilitation data, offering a straightforward approach for functional stratification in clinical practice and assisting clinicians in tailoring individualised rehabilitation strategies.

Discussion

Principal findings and clinical significance

This study developed and internally validated a pragmatic, multidimensional functional stratification tool for patients entering cardiac rehabilitation. The composite score demonstrated a monotonic and clinically coherent gradient across functional strata, supporting its construct validity and bedside interpretability. Based on the 12-point scoring system, the cohort was partitioned into Low Impairment (n = 18, 22.5%), Moderate Impairment (n = 41, 51.2%), High Impairment (n = 17, 21.2%), and Very High Impairment (n = 4, 5.0%) groups, with clear stepwise deterioration across all functional domains. For example, 6MWD decreased from 364.3 ± 75.7 m in the Low Impairment group to 185.2 ± 32.9 m in the Very High Impairment group, whereas TUG time increased from 6.9 ± 1.0 to 14.2 ± 2.8 s [35]. These progressive shifts across components, together with strong correlations between the composite score and its individual inputs (6MWT, TUG, Edmonton Frail Scale, and age), reinforce the tool’s ability to phenotype baseline functional reserve and support personalised rehabilitation planning.

Mechanistic insights and theoretical framework

Multisystem integration perspective

The stratifier’s discriminative behaviour reflects the inherently multidimensional nature of functional limitation in typical cardiac rehabilitation cohorts [33]. Each element contributes distinct physiological information: (i) the 6MWT integrates cardiopulmonary and musculoskeletal responses to submaximal exertion [34]; (ii) TUG captures neuromotor coordination, balance, and dynamic stability—dimensions not adequately represented by endurance tests [26]; (iii) EFS indexes psychosocial, cognitive, and nutritional domains that are often underweighted in conventional cardiac assessments [22]; and (iv) age acts as a proxy for cumulative physiological burden and declining biological reserve. Only modest inter-correlations among components (r = 0.13–0.52) suggest largely orthogonal contributions to the overall construct. This justifies a formative measurement approach: the 6MWT approximates convective–diffusive–mitochondrial oxygen transport; TUG reflects sensorimotor integration, postural transition, and anticipatory gait control; and EFS captures latent moderators of recovery such as mood, cognition, and social support [35]. Their additive integration therefore offers a more complete representation of physiological reserve and its depletion.

Age-related vulnerability and functional thresholds

We observed pronounced decrements in 6MWT and TUG performance among participants aged ≥ 80 years, despite relatively stable EFS scores, indicating that conventional frailty screens may under-detect objective functional decline at advanced age [36]. This supports a ‘double-hit’ model wherein the acute catabolic stress of a cardiac event accelerates age-related sarcopenia, hastening the erosion of reserve [29]. It also aligns with threshold concepts of physiological failure, whereby those with low baseline reserve cross a tipping point beyond which recovery is constrained. By explicitly incorporating age in a formative framework, our tool is sensitive to these non-linear vulnerabilities—addressing limitations reported for established heart failure prognostic tools whose discrimination attenuates in very old cohorts [37].

Functional phenotyping and subgroup differentiation

The distinct patterns across strata—particularly the dissociation between frailty (EFS) and mobility (TUG, 6MWT) in older adults—support the emerging concept of functional phenotyping in cardiac rehabilitation [38]. Clinically recognisable subtypes such as ‘frail but mobile’ versus ‘non-frail yet deconditioned’ are likely to respond differentially to intervention, motivating phenotype-tailored pathways [39]. Because these profiles can be identified from routine bedside assessments, the present tool may help translate data-driven cluster models into actionable clinical differentiation.

Clinical implementation and practice integration

Precision rehabilitation pathways based on functional profiles

The practical value of the score lies in its simplicity and its ability to direct rehabilitation intensity to physiological reserve, consistent with precision rehabilitation frameworks [40, 41]. Illustratively: (i) Low Impairment patients (score 0–3; 22.5%; 6MWT ~ 357 m) may be suited to home-based or hybrid programmes with remote oversight [42]; (ii) Moderate Impairment patients (score 4–6; 51.2%) typically benefit from standard, centre-based exercise with supervised progression, where conventional protocols yield meaningful functional gains [43]; (iii) High Impairment patients (score 7–9; 21.2%) generally require multicomponent approaches (neuromuscular retraining, balance therapy, extended sessions) [44], given persistent gait and strength deficits despite cardiovascular stabilisation [45]; and (iv) Very High Impairment patients (score ≥ 10; 5%) exhibit severe global impairment warranting geriatric co-management, individualised goals, and prolonged rehabilitation duration [46].

Integration with emerging technologies and care models

Although immediately deployable at the bedside, the tool complements digital solutions rather than competing with them [47]. It can act as a triage ‘gatekeeper’: initial stratification followed by targeted, high-resolution monitoring [48]. For example, borderline High Impairment cases might undergo sensor-based gait analysis or continuous activity tracking to detect subclinical decline. The score also maps well onto integrated care and tele-rehabilitation models, facilitating step-down and community follow-up, particularly where resources are constrained [49].

Methodological rigour and validation

Several features strengthen the translational credibility of these findings. First, all functional tests followed accepted standards (ATS 6MWT, original TUG, and validated EFS procedures), promoting reproducibility. Second, the composite was developed with a formative paradigm and clinically reasoned, data-informed thresholds [50]; low collinearity supports interpretability. Third, internal validation used 1,000 bootstrap samples. The optimism-corrected AUC was 0.90, with satisfactory calibration across strata [51]. Fourth, sensitivity analyses across alternative thresholds and transformations yielded consistent results, supporting internal generalisability [52]. Finally, reporting adheres to TRIPOD guidance [53].

Limitations and future directions

This study has several important limitations. The primary limitation is the cross-sectional design, which precludes any definitive claims about prognostic performance [54]. Although concurrent validity was strong, prospective evaluation against clinical outcomes (e.g., readmission, major adverse cardiovascular events, and longitudinal functional decline) is required to establish any prognostic value.

Second, the single-centre sample (n = 80) may limit generalisability; despite internal validation, multicentre replication in diverse populations is needed [55].

Third, the lack of direct muscle mass metrics is a further limitation. While functional measures often outperform structural surrogates, adding body-composition indices (e.g., BMI, which was not found to be significantly different across strata in our cohort (Table 1), or bioimpedance) in future studies may enhance biological fidelity.

Fourth, the sample size is small, and the functional strata were not evenly balanced. Importantly, the Very High Impairment group comprised only 5% of participants (n = 4), and no patients scored > 9 on the EFS in our cohort, restricting the generalizability of findings for the highest impairment stratum and the upper range of the EFS scoring. Additionally, the subgroup analysis for patients aged > 80 years (n = 4) is statistically underpowered and should be interpreted with extreme caution.

Finally, psychosocial, behavioural, and economic determinants of engagement were not formally modelled. Priorities include: (i) the aforementioned prospective validation for clinical outcomes; (ii) hybrid models combining this score with digital mobility metrics (e.g., gait asymmetry and variability) for early deterioration detection; (iii) biomarker augmentation (e.g., NT-proBNP, troponin, inflammatory markers) to sharpen physiological precision; (iv) implementation studies spanning settings and systems; and (v) trials testing phenotype-guided versus standard rehabilitation.

Implications for healthcare policy and practice

The functional stratification tool can support value-based resource allocation by identifying the roughly one quarter of patients in the High or Very High impairment strata (26.2%) who may benefit from prioritised, multidisciplinary rehabilitation programmes, while allowing Low Impairment patients to safely engage in lower-intensity or home-based pathways. In addition, the tool offers a framework for incorporating baseline-adjusted functional improvement into quality metrics for cardiac rehabilitation, moving beyond simple enrolment or completion rates toward measures that more accurately reflect patient-centred outcomes [56, 57].

Conclusions

We present an internally validated, bedside-feasible functional stratification tool that integrates endurance, mobility, frailty, and age-related reserve to phenotype functional status in cardiac rehabilitation. The score exhibits a coherent gradient across four clinically interpretable strata, is simple to implement, and aligns with established functional assessments. While prospective validation remains essential to determine any prognostic value, this work operationalises a precision rehabilitation approach by classifying baseline impairment and narrowing the translational gap between functional data and clinical decision-making.

Abbreviations

6MWD

Six-Minute Walk Distance

6MWT

Six-Minute Walk Test

ACE

Angiotensin-Converting Enzyme

AF

Atrial Fibrillation

AUC

Area Under the Curve

BMI

Body Mass Index

CABG

Coronary Artery Bypass Graft

CI

Confidence Interval

COPD

Chronic Obstructive Pulmonary Disease

DALYs

Disability-Adjusted Life Years

DASI

Duke Activity Status Index

EFS

Edmonton Frail Scale

HFpEF

Heart Failure with Preserved Ejection Fraction

NPV

Negative Predictive Value

NNT

Number Needed to Treat

NYHA

New York Heart Association

OR

Odds Ratio

PPV

Positive Predictive Value

ROC

Receiver Operating Characteristic

SD

Standard Deviation

SE

Standard Error

SPPB

Short Physical Performance Battery

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

TUG

Timed Up and Go (Test)

Authors’ contributions

Binbin Huang contributed to study conception and design, data acquisition from public databases, and manuscript drafting. Qian Zhang contributed to data acquisition, statistical analysis, and result interpretation. Chao Liu contributed to literature review, data curation, and preparation of figures/tables. Yang Wang and Hongwei Li contributed to data analysis, methodological validation, and manuscript revision. Deqiang Wang supervised the study, contributed to study design and interpretation, critically revised the manuscript, and served as corresponding author. Wei Li provided overall supervision, contributed to study conception and interpretation, critically revised the manuscript, and served as corresponding author.All authors approved the final manuscript and take responsibility for its content.

Funding

This study was supported by the Shenzhen Research Fund (JCYJ20230807140414029), and the Research Institute for Sports Science and Technology (RISports) at The Hong Kong Polytechnic University. The funding organizations had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The dataset used in this study is publicly available from PhysioNet (10.13026/mp8k-7p27). Analysis code and supplementary materials are available upon reasonable request from the corresponding author.

Declarations

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.