Metabolic inflexibility mediates the association between type 2 diabetes and impaired post-exercise heart rate recovery: a mediation analysis

Background

Coronary vascular disease (CVD) is the leading cause of mortality worldwide, while type 2 diabetes mellitus (T2DM) coexists in up to 40% of patients with established CVD and nearly doubles the risk of death. Heart rate recovery (HRR) is a recognized predictor of adverse cardiovascular outcomes, but the specific influence of T2DM on HRR remains incompletely understood. Metabolic inflexibility—reflected by a reduced peak respiratory exchange ratio (RER)—is characteristic of T2DM and may mediate its impact on HRR. We investigated this potential mediation relationship in post-percutaneous coronary intervention (PCI) patients.

Methods

We performed a single-center, retrospective cross-sectional study of 275 post-PCI patients at Beijing Anzhen Hospital from September 2023 to August 2024. Demographics, traditional cardiovascular risk factors, and physical activity status were recorded. Cardiopulmonary exercise testing (CPET) provided peak RER (RERpeak), ΔRER, and first-minute after peak exercise HRR (HRR1). Using multivariate regression, we assessed associations among T2DM, RER parameters, and HRR1, while adjusting for potential confounders. Mediation analysis evaluated whether RER parameters mediated the relationship between T2DM and HRR1. Subgroup analyses were conducted according to β-blocker use.

Results

A total of 275 post-PCI patients were included (198 without T2DM, 77 with T2DM). Compared with non-diabetic participants, those with T2DM exhibited significantly lower RERpeak (1.12 ± 0.07 vs. 1.17 ± 0.08; p < 0.001) and ΔRER (0.27 ± 0.08 vs. 0.31 ± 0.10; p = 0.001). T2DM was independently associated with worse HRR1 (β = − 2.47, 95% CI: − 4.53 to − 0.41, p = 0.0197) after adjustment. Both RERpeak and ΔRER were negatively correlated with T2DM and positively correlated with HRR1. Mediation analysis showed that RERpeak partially mediated 25.06% of T2DM’s adverse effect on HRR1, while ΔRER accounted for 17.62%. Subgroup analyses revealed that in patients not receiving β-blockers, RERpeak mediated 28.46% of the T2DM-HRR1 association, while ΔRER accounted for 13.51%; however, this mediation effect was not evident in β-blocker users.

Conclusion

In this post-PCI population, metabolic inflexibility partially mediated the negative association between T2DM and impaired post-exercise HRR. These findings highlight the need to improve metabolic flexibility in diabetic patients as part of risk stratification and targeted therapeutic interventions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12986-026-01114-3.

Research Insights

What is currently known about this topic?

• •Heart rate recovery (HRR) is a established independent predictor of cardiovascular events and mortality.

• •Impaired HRR is common in CVD and T2DM patients, but mechanisms are unclear.

• •Metabolic inflexibility is increasingly recognized as a hallmark of T2DM.

• •There’s emerging evidence on metabolic and autonomic pathway crosstalk, but the role of metabolic inflexibility in HRR was unknown.

What is the key research question?

Does metabolic inflexibility mediate the association between type 2 diabetes and impaired heart rate recovery in post-PCI patients?

What is new?

• •This study provides the insight that metabolic inflexibility partially mediates the T2DM-HRR relationship.

• •The mediating effects were significant only in patients not receiving β-blocker but disappeared in those on β-blocker therapy.

How might this study influence clinical practice?

Interventions moving beyond glucose control to include metabolic flexibility optimization in comprehensive diabetes-CVD management.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12986-026-01114-3.

Introduction

Coronary vascular disease (CVD) remains the leading global cause of death, responsible for approximately 9.4 million fatalities each year [1]. Percutaneous coronary intervention (PCI) accounts for 30–40% of all revascularization procedures, with over 3.5 million performed annually [2–4]. Despite symptomatic relief afforded by PCI, major trials have consistently failed to demonstrate a reduction in mortality or myocardial infarction compared to optimal medical therapy, even in patients with moderate to severe ischemia [5, 6], underscoring the need to consider additional prognostic determinants in CVD.

A prevalent comorbidity, type 2 diabetes mellitus (T2DM), affects 20–40% of individuals with established CVD [7, 8], and markedly accelerates atherosclerosis and enhances the risk of myocardial infarction, heart failure, and cardiovascular mortality [9]. The coexistence of T2DM and CVD nearly doubles mortality and reduces life expectancy by up to 12 years [10], highlighting the clinical imperative to understand the impact of T2DM in this population for improved risk stratification and prognostication.

Heart rate recovery (HRR)—the rapidity of heart rate decline following exercise—has emerged as a robust, independent predictor of adverse cardiovascular events [11–14]. Impaired HRR is frequently observed in patients with both CVD and T2DM and has been independently linked to elevated risks of sudden cardiac death and all-cause mortality [15, 16]. While impaired HRR is common in T2DM, the relationship between T2DM and HRR remains incompletely defined.

Metabolic inflexibility, defined by impaired switching between glucose and fatty acid oxidation, is increasingly recognized as a hallmark of T2DM and can be quantified by a reduced peak respiratory exchange ratio (RER) during exercise testing [17, 18]. Emerging data indicate crosstalk between metabolic and autonomic neuroregulatory pathways [19]; nevertheless, whether metabolic inflexibility mediates impaired HRR in diabetes is unknown.

Therefore, we hypothesized that metabolic inflexibility may mediate the association between T2DM and impaired HRR. We aimed to assess this relationship in post-PCI patients using mediation analysis, with the goal of informing risk stratification and guiding therapeutic strategies for this high-risk cohort.

Methods

We performed a single-centre, retrospective cross-sectional study of consecutive post-PCI patients who attended the Cardiac Rehabilitation Centre at Beijing Anzhen Hospital for routine assessment between September 2023 and August 2024. Eligible participants were 20–80 years of age, had completed percutaneous coronary intervention within the preceding three years and were receiving guideline-directed medical therapy. Within one month of cardiopulmonary exercise testing, all patients underwent laboratory evaluation, transthoracic echocardiography and coronary imaging to exclude restenosis. Three independent cardiologists reviewed each case to confirm the absence of absolute or relative contraindications to exercise testing, as specified by the American Heart Association [4].

Patients were excluded if they had: (1) heart failure defined by current guideline definition; (2) prior cardiac surgery other than PCI, including coronary artery bypass grafting or open thoracic procedures; (3) valvular heart disease; (4) uncontrolled arrhythmias (atrial fibrillation, frequent ventricular ectopy or ventricular tachycardia) and current treatment with either ivabradine or non-dihydropyridine calcium channel blockers therapy; (5) hepatic or renal dysfunction; (6) chronic obstructive pulmonary disease or abnormal spirometry before cardiopulmonary exercise testing; (7) lower-limb impairment of any cause, including documented diabetic peripheral neuropathy confirmed by prior nerve conduction studies, symptomatic lower-extremity numbness or paresthesia, orthopedic limitations, or peripheral arterial disease affecting exercise capacity; (8) anaemia or thyroid dysfunction; (9) pulmonary hypertension; (10) active infection; (11) malignancy; (12) severe obesity (body-mass index > 40 kg m⁻²); (13) premature termination of CPET or peak respiratory-exchange ratio < 1.05; (14) severe symptoms during cardiopulmonary exercise testing (chest pain, malignant arrhythmia or ischaemic ST-segment change); or (15) missing data.

Ethical considerations

This study was conducted following the “Declaration of Helsinki” by the World Medical Association and the strengthening the reporting of observational studies in epidemiology (STROBE) statement checklist (available in Supplementary file). Approval and informed consent was waived by the “Clinical Research Ethics Committee of Beijing Anzhen Hospital, Capital Medical University” as the patient identity information has been concealed.

Data collection and definition of variables

At the index visit, each participant underwent a standardised cardiac rehabilitation assessment, and all observations were entered into a structured electronic registry. Demographic details, conventional cardiovascular risk factors (smoking status and alcohol intake), comorbid conditions and current pharmacotherapy were abstracted from the clinical record. Habitual physical activity was quantified with the International Physical Activity Questionnaire–Long Form (IPAQ-LF) and was classified into low, moderate or high activity following the standard scoring protocol [20]. Diagnoses were coded according to the International Classification of Diseases, 10th Revision (ICD-10-CM). T2DM was diagnosed following the American Diabetes Association guidelines (HbA1c ≥ 6.5%, fasting plasma glucose ≥ 126 mg/dL, or current use of antidiabetic medication). Laboratory variables, including hepatic and renal panels, N-terminal pro-brain natriuretic peptide, and hemoglobin, glycosylated hemoglobin (HbA1c) along with a detailed medication inventory, were extracted from the electronic database. To minimise recall bias, two trained investigators independently verified questionnaire responses against the electronic medical record and supplemented missing data.

Cardiopulmonary exercise testing

Cardiopulmonary exercise testing (CPET) was performed on an electronically braked cycle ergometer (CPX-600, Madecare Medical Systems, Hebei, China) equipped with a breath-by-breath metabolic cart. All CPETs were performed under standardized pre-test conditions: participants were instructed to (1) avoid vigorous exercise for 24 h prior to testing; (2) abstain from caffeine and smoking for at least 3 h before testing; (3) participants were instructed to fast for at least 3 h before testing or consume only a light meal at least 2–3 h prior to avoid prior both hypoglycemia and postprandial metabolic effects; (4) continue taking their regular cardiovascular medications (including beta-blockers, as clinically indicated); and (5) wear comfortable clothing and appropriate footwear. All tests were conducted in a temperature-controlled laboratory (20–22 °C, 40–60% relative humidity). Gas analyzers were calibrated before each testing session using standard reference gases (16% O₂, 4% CO₂) and a 3-L calibration syringe for volume calibration, following manufacturer guidelines and laboratory standard operating procedures. The protocol comprised three sequential stages: (1) Seated rest (3 min) for acquisition of baseline variables, including resting respiratory-exchange ratio (RERrest); (2) Unloaded pedalling at 60 rev min⁻¹ (3 min); (3) Symptom-limited ramp exercise with work-rate increments of 10–20 W min⁻¹, individually titrated to obtain an 8–12 min total exercise duration. Twelve-lead ECG, brachial blood pressure and transcutaneous oxygen saturation were monitored continuously. Testing was terminated at volitional fatigue or on attainment of guideline-defined stop criteria [21]. A maximal effort was confirmed when ≥ 2 of the following were present: (1) peak respiratory-exchange ratio (RERpeak) > 1.05; (2) heart rate ≥ 85% of the age-predicted maximum; (3) a VO₂ plateau despite further workload increments; or (4) Borg rating of perceived exertion ≥ 17 (6–20 scale). Maximum oxygen uptake (VO2max) was defined as the highest 30-s rolling average of oxygen uptake during the final minute of exercise, and metabolic equivalents (METs) were calculated as VO₂max (mL kg⁻¹ min⁻¹)/3.5 [22]. RER at rest (RERrest) is defined as the respiratory exchange ratio measured during the resting phase before exercise begins, calculated as the ratio of carbon dioxide output (VCO₂) to oxygen uptake (VO₂) while the subject is at complete rest in either seated or supine position prior to the start of the exercise protocol. All CPET data were acquired and processed using automated computerized algorithms integrated into the metabolic cart system, thereby ensuring objective and standardized outcome ascertainment independent of participant characteristics. Peak RER (RERpeak) and RERrest were stored automatically by the software. Delta RER (ΔRER) was calculated as the arithmetic difference between peak exercise and resting values (RERpeak − RERrest). This parameter represents the dynamic ventilatory response to increasing metabolic demand and reflects the transition from predominantly aerobic to mixed aerobic–anaerobic metabolism during incremental exercise. Heart rate recovery at 1 min (HRR1) is defined as the absolute reduction in heart rate from peak exercise (HRpeak) to 1 min after exercise cessation. Quality assurance was maintained through scheduled equipment servicing, daily calibration logs and strict adherence to a standardised operating procedure.

Statistics

Data are presented as mean ± standard deviation or median (interquartile range [IQR]) according to distributional characteristics; categorical variables are shown as number (percentage). Between-group differences were evaluated using Student’s t-test or Mann-Whitney U-test for continuous variables and χ² or Fisher’s exact test for categorical variables. We constructed multivariable regression models to examine three key relationships: (1) the association between T2DM and HRR1, (2) the relationship between ventilatory parameters (RERpeak and ΔRER) and HRR1, and (3) the independent contribution of T2DM to RERpeak and ΔRER. All models were adjusted for potential confounders including age, sex, body-mass index, smoking status, cardiovascular risk factors, pharmacotherapy and comorbidities. To address missing HbA1c data, we performed multiple imputation and conducted sensitivity analyses adjusting for the imputed values. Detailed results are presented in the Supplementary Materials. Results are presented as standardized β-coefficients with 95% confidence intervals. To investigate potential biological mechanisms, we performed causal mediation analyses to quantify the extent to which ventilatory parameters (RERpeak and ΔRER) mediated the relationship between T2DM and HRR. Total, direct and indirect effects were calculated using bootstrapping with 1,000 replications to generate bias-corrected confidence intervals. We conducted prespecified subgroup analyses stratified by β-blocker usage. To account for multiplicity, post-hoc Benjamini-Hochberg false discovery rate (BH-FDR) correction was applied within each analytically coherent family of hypothesis tests rather than globally, as the study encompassed analyses addressing distinct scientific questions at the primary, secondary, subgroup, sensitivity, and mediation levels. For regression-based analyses, corrected significance was determined by comparing ranked p values against BH-adjusted thresholds at a FDR of 5%. For mediation analyses, mathematical dependency among pathway components (total, indirect, and direct effects) precluded the use of FDR correction within each model; instead, significance was evaluated using bias-corrected bootstrap confidence intervals (1,000 iterations). Across parallel mediation models within the same analytical level, the indirect effect p values were additionally subjected to BH-FDR correction as an independent family. Statistical analyses were performed in R (v.4.3.3) and IBM SPSS Statistics (v.24). Two-sided P < 0.05 was considered statistically significant.

Results

Final analyzed population

As shown in Fig. 1, a total of 328 consecutive post-PCI patients who met the inclusion criteria and had no exclusion criteria were initially enrolled in this study. Prior to cardiopulmonary exercise testing (CPET), 23 patients were excluded for the following reasons: acute respiratory tract infection preceding the scheduled CPET (n = 2), new-onset cardiac arrhythmia preceding the scheduled CPET (n = 4), new-onset cardiac ischemia preceding the scheduled CPET (n = 2), and missing data (n = 15). Consequently, 303 patients successfully completed CPET. Of these, an additional 28 patients were subsequently excluded due to submaximal effort during CPET (n = 20) and exercise-induced adverse events (n = 8). Ultimately, a total of 275 patients with complete data were included in the final analysis.

Fig. 1.

Flowchart of patients enrolled in the study. PCI indicates percutaneous coronary intervention; CPET, cardiopulmonary exercise testing

Baseline characteristics

As shown in Table 1, baseline characteristics of our study cohort (N = 275) comprising 198 non-diabetic and 77 diabetic patients revealed comparable demographic profiles between non-DM patients and DM patients with no significant differences in age, anthropometric measurements, lifestyle factors or echocardiographic findings. Analysis of RER parameters demonstrated that while RERrest remained similar between two groups (0.86 ± 0.08 vs. 0.85 ± 0.07; p = 0.598), significant differences were observed in RERpeak values, which were notably lower in DM patients (1.12 ± 0.07 vs. 1.17 ± 0.08; p < 0.001). Consequently, the ΔRER was significantly attenuated in DM patients compared to non-DM patients (0.27 ± 0.08 vs. 0.31 ± 0.10; p = 0.001). These findings were consistent with other cardiopulmonary exercise testing parameters, including reduced VO₂max (1358.71 ± 356.81 vs. 1497.07 ± 391.64 mL/min; p = 0.007), lower METs (5.30 ± 1.24 vs. 5.69 ± 1.28 mL/kg/min; p = 0.024), and impaired HRR1 (16.31 ± 6.60 vs. 20.00 ± 7.41 beats/min; p < 0.001) in DM patients.

Table 1.

Baseline characteristics of patients according to T2DM status

Variables	Non-DM patients (n = 198)	DM patients (n = 77)	p value
Demographic characteristic
Age(years), mean ± SD	56.86 ± 10.90	57.42 ± 10.78	0.703
Gender (female)	32 (16.16%)	14 (18.18%)	0.687
Height (cm)	170.25 ± 7.08	169.30 ± 6.42	0.304
Weight (Kg)	75.38 ± 12.38	73.14 ± 10.95	0.166
BMI, kg/m2	25.92 ± 3.29	25.47 ± 3.16	0.311
Lifestyle indicators
Smoking, n (%)			0.984
Never smokers	73 (36.87%)	29 (37.66%)
Current smokers	43 (21.72%)	16 (20.78%)
Former smokers	82 (41.41%)	32 (41.56%)
Drinking, n (%)			0.998
Never drinkers	118 (59.60%)	46 (59.74%)
Current drinkers	34 (17.17%)	13 (16.88%)
Former drinkers	46 (23.23%)	18 (23.38%)
PA status, n (%)			0.651
Low	106 (53.54%)	43 (55.84%)
Moderate	66 (33.33%)	27 (35.06%)
High	26 (13.13%)	7 (9.09%)
Time from index PCI, n (%)			0.401
1 month to 1 year	119 (60.10%)	42 (54.55%)
1 year to 3 years	79 (39.90%)	35 (45.45%)
Comorbidities, n (%)
Hypertension	90 (45.45%)	50 (64.94%)	0.004
Hyperlipidemia	94 (47.47%)	42 (54.55%)	0.292
History of myocardial infarction	59 (29.80%)	21 (27.27%)	0.679
DM duration, n (%)			< 0.001
< 1 year	0 (0.00%)	26 (33.77%)
1 year to 3 years	0 (0.00%)	14 (18.18%)
> 3 years	0 (0.00%)	37 (48.05%)
Laboratory results
HbA1c (%)	5.74 ± 0.52	6.51 ± 0.80	< 0.001
Echocardiographic findings
LVEDD (mm)	47.04 ± 4.40	46.74 ± 4.86	0.641
LVESD (mm)	30.75 ± 4.57	30.83 ± 5.22	0.907
LVEF (%)	61.85 ± 5.87	60.49 ± 7.72	0.131
CPET parameters
VO2max (mL/min)	1497.07 ± 391.64	1358.71 ± 356.81	0.007
METs (mL/kg/min)	5.69 ± 1.28	5.30 ± 1.24	0.024
HRrest (beats/min)	69.27 ± 11.45	73.22 ± 12.57	0.013
HRpeak (beats/min)	126.19 ± 17.25	120.79 ± 16.53	0.019
HRR1 (beats/min)	20.00 ± 7.41	16.31 ± 6.60	< 0.001
RERrest	0.86 ± 0.08	0.85 ± 0.07	0.598
RERpeak	1.17 ± 0.08	1.12 ± 0.07	< 0.001
ΔRER	0.31 ± 0.10	0.27 ± 0.08	0.001
Medications
Beta blockers	82 (41.41%)	39 (50.65%)	0.166
Metoprolol	172 (86.9%)	61 (79.2%)	0.113
Bisoprolol	26 (13.1%)	16 (20.8%)	0.113
ARNI	26 (13.13%)	12 (15.58%)	0.597
ARB	32 (16.16%)	18 (23.38%)	0.164
ACEI	16 (8.08%)	3 (3.90%)	0.219
Metformin	0 (0.0%)	28 (36.4%)	< 0.001
Insulin	0 (0.0%)	11 (14.3%)	< 0.001
Statins	180 (90.91%)	71 (92.2%)	1.00
PCSK-9 inhibitor	19 (9.6%)	6 (7.8%)	0.64

Data are presented as mean (standard deviation), median (interquartile range), or proportion (%) as appropriate. BMI indicates body mass index; PA, physical activity; PCI, percutaneous coronary intervention; DM, type 2 diabetes mellitus; HbA1c, glycosylated hemoglobin; LVEDD, left ventricular end-diastolic diameter; LVESD, left ventricular end-systolic diameter; LVEF, left ventricular ejection fraction; CPET, cardiopulmonary exercise testing; VO2max, maximum oxygen uptake; METs, metabolic equivalents(VO2max/kg [mL/kg/min] ÷ 3.5); HRrest, heart rate at rest; HRpeak, heart rate from peak exercise; HRR1, heart rate recovery at 1 min; RERrest, RER at rest; RERpeak, peak RER from exercise; ΔRER, the arithmetic difference between RERpeakand RERrest; ARNI, angiotensin receptor-neprilysin Inhibitor; ARB, angiotensin receptor blocker; ACEI, angiotensin-converting enzyme inhibitor

Associations among T2DM, RER parameters, and HRR1 by multivariate regression analysis

Multivariate regression analysis revealed a significant inverse association between T2DM and HRR1 (Table 2). In the basic model adjusted for demographic and anthropometric factors (sex, age, height, and weight), T2DM was associated with a substantial reduction in HRR1 (β = −3.79, 95% CI: −5.66 to −1.93, p < 0.0001). Notably, this association remained robust in the fully adjusted model that accounted for comprehensive cardiovascular risk factors including sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, LVEF, time from index PCI, DM duration, and β-blocker use (β = −2.47, 95% CI: −4.53 to −0.41, p = 0.0197).

Table 2.

Multivariate logistic regression analysis for assessing the association between T2DM and HRR1

	Model 1			Model 2
	β value	95%CI	p value	β value	95%CI	p value
T2DM	−3.79	−5.66, −1.93	< 0.0001	2.47	−4.53, −0.41	0.0197

Model 1: adjusted for sex, age, height and weight

Model 2: adjusted for sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, β-blocker use, LVEF, time from index PCI, and DM duration

Multivariate regression analysis also revealed significant negative associations between T2DM and RER parameters during CPET (Table 3). T2DM was independently associated with reduced RERpeak, in both minimally adjusted (β = −0.05, 95% CI: −0.07 to −0.03, p < 0.0001) and fully adjusted models (β = −0.04, 95% CI: −0.06 to −0.02, p < 0.0001). Similarly, the change in ΔRER showed significant impairment in diabetic subjects (β = −0.04, 95% CI: −0.07 to −0.02) with comparable statistical significance in both basic demographic-adjusted (p = 0.0007) and comprehensive cardiovascular risk factor-adjusted models (p = 0.0003). The persistence of these negative associations after extensive adjustment for potential confounders, including anthropometric parameters, lifestyle factors, comorbidities, and β-blocker use.

Table 3.

T2DM is associated with impaired RERpeak and reduced ΔRER during exercise

	β value	95%CI	p value
RERpeak
Model 1	−0.05	−0.07, −0.03	< 0.0001
Model 2	−0.04	−0.06, −0.02	< 0.0001
ΔRER
Model 1	−0.04	−0.07, −0.02	0.0007
Model 2	−0.04	−0.07, −0.02	0.0003

Model 1: adjusted for sex, age, height and weight

Model 2: adjusted for sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, β-blocker use, LVEF, time from index PCI, and DM duration

Table 4 demonstrates that both RERpeak and ΔRER are independently associated with improved HRR1. In Model 1, which adjusted for sex, age, height, and weight, RERpeak was significantly associated with HRR improvement (β = 18.29; 95% CI, 7.37 to 29.21; p = 0.0012), as was ΔRER (β = 12.43; 95% CI, 3.14 to 21.73; p = 0.0093). These associations remained robust in Model 2 after further adjustment for METs, hypertension, hyperlipidemia, history of myocardial infarction, LVEF, time from index PCI, DM duration, and β-blocker use, with RERpeak showing a β coefficient of 17.64 (95% CI, 7.10 to 28.17; p = 0.0012) and ΔRER yielding a β of 12.60 (95% CI, 3.75 to 21.45; p = 0.0057).

Table 4.

RER parameters are independently associated with improved HRR1

	Model 1			Model 2
	β value	95%CI	p value	β value	95%CI	p value
RERpeak	18.29	−7.37, 29.21	0.0012	17.64	7.10, 28.17	0.0012
ΔRER	12.43	3.14, 21.73	0.0093	12.60	3.75, 21.45	0.0057

Model 1: adjusted for sex, age, height and weight

Model 2: adjusted for sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, β-blocker use, LVEF, time from index PCI, and DM duration

Mediation analysis of RER parameters

In the RERpeak mediation model (Fig. 2A), T2DM exerted a significant total negative effect on HRR1 (β = −2.67, 95% CI: −4.48 to −0.94, p = 0.004). This effect was partially mediated by reduced RERpeak, with a statistically significant mediation effect (β = −0.69, 95% CI: −1.24 to −0.20, p = 0.006), accounting for 25.06% of the total effect. The direct effect of T2DM that was not explained by RERpeak remained significant (β = −2.00, 95% CI: −3.90 to −0.29, p = 0.022). Similarly, Fig. 2B demonstrated that reduced ΔRER during exercise significantly mediated the relationship between T2DM and impaired HRR1. The mediation effect through ΔRER (β = −0.47, 95% CI: −1.05 to −0.09, p = 0.01) constituted 17.62% of the total effect of T2DM on HRR1.

Fig. 2.

Mediation analysis of RER parameters in the relationship between T2DM and HRR1 in total patients. Panels A and B show mediation analyses in total patients (n = 275), evaluating RERpeak and ΔRER, respectively. Each panel displays the total effect of T2DM on HRR1, the direct effect (unmediated pathway), the indirect effect through the specified mediator, and the proportion of effect mediated, with 95% confidence intervals and p-values. All models are adjusted for adjusted for sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, β-blocker use, LVEF, time from index PCI, and DM duration

Subgroup mediation analysis

Subgroup analyses stratified by β-blocker use were performed. In individuals not taking β-blockers (Fig. 3A and B), the mediating effect of respiratory gas exchange parameters on the relationship between T2DM and HRR1 remained significant. Specifically, RERpeak significantly mediated a substantial portion of the T2DM-HRR1 association (β = −1.02; 95% CI: −1.97 to −0.29; p = 0.004), accounting for 28.46% (p = 0.008) of the total effect of T2DM on HRR1 (β = −3.59; 95% CI: −5.87 to −1.34; p = 0.008). The direct effect of T2DM on HRR1, independent of RERpeak, also remained significant (β = −2.57; 95% CI: −4.86 to −0.15; p = 0.038). ΔRER demonstrated a trend towards mediation, explaining 13.51% (p = 0.04) of the total effect (β = −0.49; 95% CI: −1.27 to −0.02; p = 0.04), with the direct effect of T2DM remaining statistically significant (direct effect: −3.11; 95% CI: −5.40 to −0.64; p = 0.012).

Fig. 3.

Mediation analysis of RER parameters in the relationship between T2DM and HRR1 in subgroups stratified by β-blocker use. Panels A and B show mediation analyses in patients not taking β-blockers (n = 154), evaluating RERpeak and ΔRER, respectively. Panels C and D show corresponding analyses in patients taking β-blockers (n = 121). All models are adjusted for sex, age, height, weight, smoking, drinking, METs, hypertension, hyperlipidemia, history of myocardial infarction, LVEF, time from index PCI, and DM duration

Conversely, in the subgroup of individuals taking β-blockers (Fig. 3C and D), the mediating role of RER parameters became statistically non-significant. While T2DM still showed a negative total effect on HRR1 (β = −3.17; 95% CI: −6.09 to −0.42; p = 0.024), the mediation effect of RERpeak was not significant (β = −0.26; 95% CI: −1.19 to 0.55; p = 0.49), explaining only 8.25% (p = 0.50) of the total effect (Fig. 3C). Similarly, ΔRER did not significantly mediate the relationship (β = −0.41; 95% CI: −1.50 to −0.34; p = 0.25), accounting for 13.62% (p = 0.26) of the total effect (Fig. 3D). The direct effects of T2DM on HRR1 were borderline significant in this group (RERpeak mediation: β = −2.91; 95% CI: −5.88 to −0.08; p = 0.046; ΔRER mediation: β = −2.74; 95% CI: −5.65 to −0.03; p = 0.048).

Stratified multivariate regression analyses were conducted according to β-blocker use status (Supplementary Tables S1-S3). In fully adjusted models, T2DM was significantly associated with lower HRR1 in both β-blocker users (β = −2.008, 95% CI: −4.854 to −0.838, p = 0.0167) and non-users (β = −3.595, 95% CI: −6.050 to −1.140, p = 0.0048) (Table S1). T2DM was also significantly associated with lower peak RER and reduced ΔRER in both groups (all p < 0.05, Table S2). For the associations between metabolic flexibility parameters and HRR1 (Table S3), among non-users, both peak RER (β = 24.830, 95% CI: 11.320 to 38.340, p = 0.0004) and ΔRER (β = 14.168, 95% CI: 2.526 to 25.810, p = 0.018) were significantly associated with HRR1. Among β-blocker users, neither peak RER (β = 7.409, p = 0.39) nor ΔRER (β = 11.426, p = 0.12) showed significant associations with HRR1.

Sensitivity analysis and multiple comparison correction

In sensitivity analyses adjusting for multiply imputed HbA1c values, the associations between T2DM and HRR1 as well as RER parameters remained statistically significant and directionally consistent with the primary analyses (Supplementary Tables S4-S8). Across all analytical families, the principal findings were robust to post-hoc multiple comparison correction (Supplementary Tables S9-S10). In the regression analyses, BH-FDR correction within each family confirmed that the associations between T2DM, RER parameters, and HRR1 remained statistically significant in the primary, secondary, and sensitivity analyses. In the subgroup analyses stratified by β-blocker use, the associations were preserved in non-users but were non-significant in β-blocker users, findings that were non-significant prior to correction and are consistent with the known chronotropic suppression by β-blockers. In the mediation analyses, both RERpeak and ΔRER significantly mediated the T2DM–HRR1 association after BH-FDR correction of indirect effects and bias-corrected bootstrap evaluation of all pathway components, in the overall sample, after additional HbA1c adjustment, and in β-blocker non-users; the indirect effects were non-significant in β-blocker users, consistent with the subgroup regression findings. Overall, the vast majority of tests remained significant after correction, and all non-significant results were confined to β-blocker users.

Discussion

Our findings reveal a novel mechanistic pathway through which T2DM impairs post-exercise HRR1, specifically demonstrating that metabolic inflexibility partially mediates the detrimental effects. T2DM is significantly associated with both slower HRR following peak exercise and reduced metabolic flexibility during exercise, as assessed by RERpeak and ΔRER. Metabolic inflexibility was independently associated with impaired HRR1, with mediation analysis demonstrating that reduced RERpeak and ΔRER each independently accounted for a substantial proportion of the adverse diabetic effect on HRR1. Notably, subgroup analysis stratified by β-blocker use revealed that this metabolic-autonomic mediation was statistically significant only in patients not receiving β-blocker therapy, whereas no mediation effect was observed in those taking β-blockers. This differential pattern suggests that metabolic inflexibility may influence autonomic nervous system function through β-adrenergic signaling, which in turn impairs heart rate recovery, and that β-blocker therapy interrupts this sequential metabolic-autonomic pathway. These findings establish metabolic inflexibility as a potentially modifiable therapeutic target in diabetic cardiovascular autonomic dysfunction, with implications for risk stratification and personalized treatment in high-risk diabetic patients.

HRR impairment in T2DM and cardiovascular prognostic value

HRR1 reflects the complex interplay between parasympathetic reactivation and sympathetic withdrawal in cardiopulmonary exercise testing [23]. Our analysis confirmed that T2DM was independently associated with impaired HRR1 following peak exercise, which represents a clinical manifestation of cardiovascular autonomic neuropathy (CAN) with high prevalence in diabetic populations, consistent with extensive literature documenting autonomic dysfunction in T2DM [24]. Beyond its diagnostic utility for CAN, HRR has emerged as a valuable prognostic indicator with independent predictive value for cardiovascular outcomes. Cole et al. established that abnormal HRR (< 12 beats/min at 1 min) independently predicts mortality with fourfold increased risk [12], while meta-analysis of nine prospective studies demonstrated that each 10 beats/min decrease in HRR confers hazard ratios of 1.13 (95% CI: 1.05–1.21) for cardiovascular events and 1.09 (95% CI: 1.01–1.19) for all-cause mortality [13]. Importantly, abnormal HRR demonstrates independent prognostic value in T2DM populations beyond its association with diabetes per se [16, 25, 26] and predicts silent myocardial ischemia [27]. HRR measurement also serves as a tool for assessing revascularization efficacy, with successful percutaneous coronary intervention consistently associated with improved HRR parameters [11, 28]. However, the therapeutic implications remain uncertain, as current evidence regarding cardiac rehabilitation’s capacity to improve HRR parameters yields conflicting results across studies [29, 30]. These collective findings suggest that while HRR impairment functions as a biomarker of CAN, its clinical significance extends beyond diabetic autonomic dysfunction, potentially reflecting fundamental pathophysiological processes that influence long-term clinical outcomes.

Impaired metabolic flexibility in T2DM during exercise challenge

Metabolic flexibility, defined as the body’s ability to regulate fuel oxidation based on substrate availability, is a fundamental characteristic of healthy metabolism [31, 32]. Under physiological conditions, tissues efficiently transition between glucose utilization during feeding and fatty acid oxidation during fasting. In contrast, T2DM is characterized by metabolic inflexibility, manifesting as persistent reliance on fatty acid oxidation even under glucose-abundant conditions [33]. Exercise represents a physiological challenge demanding metabolic flexibility, with progressive increases in exercise intensity requiring greater glucose oxidation through both oxidative phosphorylation and anaerobic glycolysis. In our study, T2DM patients displayed significantly lower RERpeak and ΔRER during CPET compared to non-diabetic subjects, and multivariate analysis confirmed that T2DM was independently associated with both reduced RERpeak and ΔRER. These findings are consistent with prior observations in diabetic populations demonstrating lower RER at peak exercise and during early recovery [34]. While healthy individuals demonstrate intensity-dependent metabolic switching during exercise, T2DM patients exhibit impaired substrate transition with similar or elevated fatty acid oxidation rates during exercise [35, 36]. The reduction in both RERpeak (reflecting maximal carbohydrate oxidation capacity) and ΔRER (reflecting the magnitude of metabolic transition) indicates that T2DM patients have diminished capacity to shift from predominantly fat oxidation at rest to carbohydrate oxidation during high-intensity exercise, confirming metabolic inflexibility as a key metabolic abnormality in this population.

Metabolic Inflexibility as a Mediator Between T2DM and Impaired HRR

Our mediation analysis revealed that metabolic inflexibility partially mediated the association between T2DM and impaired HRR1. This metabolic-autonomic relationship is supported by emerging evidence demonstrating bidirectional interactions between metabolic flexibility and autonomic regulation. In active postmenopausal women, greater metabolic flexibility was significantly associated with improved resting autonomic function [37], while higher metabolic flexibility has been shown to increase energy expenditure and enhance the capacity to utilize carbohydrates as substrate at higher exercise intensities [38].These observations reinforce the intimate coupling between metabolic substrate dynamics and autonomic nervous system function.

The subgroup analysis stratified by β-blocker use provides mechanistic evidence that metabolic inflexibility influences HRR1 through modulation of cardiac autonomic function. A critical question is whether metabolic inflexibility mediates the T2DM-HRR1 relationship, or whether both represent parallel consequences of autonomic dysfunction. HRR1 is primarily mediated by vagal reactivation following exercise [39], which is impaired in T2DM patients regardless of exercise intensity compared to adults without T2DM [40], yet mechanisms underlying this diabetic parasympathetic withdrawal remain incompletely defined. If T2DM’s effect on HRR1 operated solely through autonomic dysfunction, β-blockade should uniformly attenuate this relationship. Similarly, if T2DM’s impact on metabolic flexibility were mediated through autonomic mechanisms, β-blocker use should diminish these associations. However, our findings reveal a differential pattern. In the β-blocker subgroup, T2DM remained significantly associated with reduced RERpeak, ΔRER, and HRR1, yet the mediation effect of metabolic inflexibility was abolished, with no significant associations between metabolic flexibility (both RERpeak and ΔRER) and HRR1 (Supplementary Tables S1-S3). This indicates that while T2DM directly impairs both metabolic flexibility and autonomic recovery independent of β-adrenergic signaling, the coupling between metabolic inflexibility and HRR1 requires intact β-adrenergic pathways.

While limited sample size may contribute to this finding, our subgroup analysis provides new insights into conflicting evidence regarding β-blocker effects on heart rate recovery. Medeiros et al. demonstrated that β-blocker therapy combined with exercise training improved HRR and aerobic capacity in post-myocardial infarction patients, whereas long-term β-blocker monotherapy achieved neither outcome [41]. Studies in heart failure patients show mixed results, with some reporting no β-blocker effect on HRR [42, 43] while others document significant alterations [44, 45]. Since our post-PCI group excluded patients with heart failure, arrhythmias, and other conditions affecting HRR, and the analyses adjusted for confounding variables such as exercise habits, the diminished mediating role of metabolic inflexibility on HRR1 in the β-blocker group may indicate the beneficial prognostic effects of β-blockers.

Mechanisms Linking Metabolic Inflexibility to Autonomic Neuropathy

An important question is whether metabolic inflexibility represents an early event in T2DM that contributes to HRR impairment, or whether both emerge as parallel consequences of autonomic dysfunction. Accumulating evidence have suggested bidirectional interactions between metabolic inflexibility and autonomic nervous system dysfunction [46, 47]. Autonomic alterations may contribute to insulin resistance pathogenesis through multiple pathways, while insulin resistance reciprocally impairs autonomic function via hemodynamic and metabolic disturbances, thereby creating a vicious cycle that amplifies cardiovascular risk factors including hypertension, dyslipidemia, and further metabolic deterioration. However, emerging evidence suggests that metabolic perturbations may occur early in T2DM and subsequently exacerbate autonomic neural injury. T2DM affects autonomic function through both indirect mechanisms (e.g., cardiac remodeling) and direct neural damage [48]. Feldman et al. comprehensively described how chronic hyperglycemia, dyslipidemia, and aberrant insulin signaling precipitate pathological cascades including DNA damage, endoplasmic reticulum stress, mitochondrial impairment, and neurotrophic deficits, mediated by advanced glycation end products, free fatty acids, oxidized lipoproteins, reactive oxygen species, and inflammatory pathways [49]. These observations indicate that metabolic abnormalities can directly initiate processes compromising autonomic function. Given that metabolic inflexibility represents a fundamental metabolic derangement in T2DM, we hypothesize that it may contribute to CAN development through several interconnected mechanisms.

First, the temporal pattern of CAN development suggests involvement of upstream metabolic factors. The prevalence of CAN demonstrates a positive correlation with disease duration, ultimately affecting up to 65% of patients with T2DM [50, 51]. This progressive pattern indicates that CAN is not always present at diabetes onset, suggesting that initiating factors beyond hyperglycemia may be required. Second, as a core component of metabolic inflexibility, insulin resistance typically precedes overt hyperglycemia, and substantial clinical evidence demonstrates that metabolic inflexibility can be detected during the prediabetic stage [52]. Supporting this temporal sequence, one study found that both T2DM and prediabetic patients had significantly lower RERpeak than healthy controls, further underscoring the early emergence of impaired metabolic flexibility in T2DM progression [18]. Third, mitochondrial dysfunction, a hallmark of metabolic inflexibility, directly compromises neuronal energetics. Studies have identified various mitochondrial abnormalities in individuals with T2DM and insulin resistance, demonstrating that reduced mitochondrial capacity correlates with decreased resting lipid oxidation and subsequently promotes muscle lipid accumulation [53, 54]. Mitochondrial dysfunction leads to diminished oxidative ATP synthesis, altered metabolite homeostasis, and impaired glycolysis. Dysfunctional mitochondria generate insufficient energy and lose their ability to transport normally along axons, exacerbating axonal damage and providing a direct mechanistic link between metabolic inflexibility and autonomic neural injury [55]. Fourth, chronic low-grade inflammation (“metaflammation”) may bridge metabolic inflexibility and autonomic injury [56–58]. T2DM upregulates inflammatory pathways involving pattern recognition receptors, oxidative stress, nuclear factor signaling, advanced glycation products, and pro-inflammatory cytokines implicated in diabetic autonomic neuropathy [59]. Metabolic inflexibility perpetuates inflammation through substrate oxidation abnormalities and mitochondrial dysfunction, creating an environment where inflammatory molecules damage autonomic tissue. Fifth, impaired AMPK signaling may contribute to both metabolic inflexibility and diabetic neuropathy. AMPK serves as a critical regulator of metabolic flexibility, promoting β-cell survival and counteracting insulin resistance, thereby facilitating glucose metabolism and uptake [60, 61]. Under conditions of metabolic inflexibility, elevated glucose levels compromise mitochondrial function and diminish normal AMPK pathway activity, further mediating the metabolic alterations and neuropathic complications associated with diabetes mellitus [62].

Study limitations

Several limitations should be acknowledged. First, the single-center retrospective nature and specific patient population may limit the generalizability of our findings. Longitudinal studies are needed to establish whether metabolic inflexibility precedes autonomic dysfunction or both emerge as parallel consequences of T2DM. Second, we did not perform comprehensive autonomic nervous system assessment (e.g., heart rate variability, baroreflex sensitivity, cardiovascular autonomic reflex tests). While HRR is a well-validated marker of cardiac autonomic function with established prognostic significance, and our β-blocker-stratified analyses provide indirect mechanistic support, direct autonomic testing would have strengthened causal inference. Third, rigorous exclusion criteria (e.g., COPD, severe anemia, severe obesity) and the RER peak > 1.05 threshold, while necessary to ensure accurate CPET interpretation and minimize confounding, may limit generalizability to more heterogeneous or frail post-PCI populations. Fourth, although formal multiple comparison correction was not pre-specified, post-hoc BH-FDR correction applied within each analytical family, combined with bias-corrected bootstrap confidence intervals for mediation pathway components, confirmed that the vast majority of associations remained statistically significant; the few non-significant results were confined to β-blocker users and were biologically expected, collectively supporting the robustness of the principal findings. Nevertheless, given the exploratory nature of the subgroup and mediation analyses, independent prospective validation remains warranted. Fifth, we did not account for antidiabetic medication use, which may differentially influence metabolic flexibility. Similarly, the absence of inflammatory biomarker data (e.g., C-reactive protein, interleukin-6) prevented exploration of inflammation-mediated pathways. Future prospective, multi-center studies incorporating comprehensive autonomic testing alongside metabolic assessments, with objective physical activity monitoring and longer follow-up periods, are warranted to validate these findings, establish causality, and explore clinical implications for cardiovascular risk stratification and targeted interventions in T2DM patients.

Conclusions

This study demonstrates that metabolic inflexibility, assessed by RER parameters, serves as a significant mediator in the relationship between T2DM and impaired post-exercise HRR in post-PCI patients. The mediation effect was particularly pronounced in patients not receiving β-blockers, that β-blockers may have a protective role in mitigating the adverse impact of metabolic inflexibility on HRR1. These findings provide novel insights into the pathophysiological link between T2DM and cardiac autonomic dysfunction, highlighting metabolic flexibility as a potential therapeutic target for improving cardiovascular outcomes in diabetic patients.

Abbreviations

CVD

Cardiovascular disease

PCI

Percutaneous coronary intervention

T2DM

Type 2 diabetes mellitus

HRR

Heart rate recovery

RER

Respiratory exchange ratio

CPET

Cardiopulmonary exercise testing

BMI

Body mass index

CPET

Cardiopulmonary exercise testing

PA

Physical activity

VO2max

Maximum oxygen uptake

METs

Metabolic equivalents

HRpeak

Heart rate from peak exercise

HRR1

Heart rate recovery at 1 min

RERrest

RER at rest

RERpeak

Peak RER from exercise

ΔRER

The arithmetic difference between RERpeak and RERrest

LVEDD

Left ventricular end-diastolic diameter

LVESD

Left ventricular end-systolic diameter

LVEF

Left ventricular ejection fraction

ARNI

Angiotensin receptor-neprilysin Inhibitor

ARB

Angiotensin receptor blocker

ACEI

Angiotensin-converting enzyme inhibitor

CAN

Cardiovascular autonomic neuropathy

Author contributions

JH Wu and YT Liu conceptualized the study. JH Wu and Y Feng screened patients in the Cardiac Rehabilitation Center. N Li, Y Shao, and Y Zhang performed the screening and collecting of medical records from the questionnaires and the medical database. YT Liu and SH Zhang performed pulmonary function testing and cardiopulmonary exercise tests. YT Liu performed the statistical analysis, wrote the manuscript, and ideated and produced the tables and figures. JH Wu reviewed and edited the manuscript.

Funding

This study was funded by Beijing Traditional Chinese Medicine Science and Technology Development Fund Project (Project No.BJZYYB-2023-22) and Noncommunicable Chronic Diseases-National Science and Technology Major Project (2025ZD0551808).

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to them containing information that could compromise the privacy of patients but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted following the “Declaration of Helsinki” by the World Medical Association and the strengthening the reporting of observational studies in epidemiology (STROBE) statement checklist (available in Supplementary file). Approval and informed consent was waived by the “Clinical Research Ethics Committee of Beijing Anzhen Hospital, Capital Medical University” as the patient identity information has been concealed.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.