Resting heart rate and risk of left and right heart failure in 0.5 million Chinese adults

Study design, setting and population

The China Kadoorie Biobank (CKB) is a prospective study of 512 726 Chinese adults who were aged 30–74 years at enrolment in 2004–2008.19 The design of the CKB study has been previously described.19–21 Briefly, participants were recruited from 10 regions in China. All permanent residents with no major disability in each of the 100–150 administrative units (comprising either urban residential areas or rural villages) were identified through official residential records and invited to participate in the study, with a response rate of about 30%.20 Less than 0.5% (n=2470) of individuals who attended the baseline survey were excluded because they withdrew their consent or had missing data. For this study, participants with a self-reported prior history of coronary heart disease (n=16 345) or rheumatic heart disease (n=937) or who reported use of beta-blockers (n=6512) at baseline were excluded. Additional analyses also excluded individuals with a prior history of chronic obstructive pulmonary disease (COPD) and other chronic diseases at baseline.

Exposures, covariates and outcomes

RHR was measured using calibrated UA-779 digital BP monitor (Omron UA-779; Live Source) to the nearest one beat per minute (bpm) after the participant had been at rest in a seated position for at least 5 min.19 RHR was measured twice, and the mean of both measurements was used for the present analyses.

Information on demographic and socioeconomic status (SES), lifestyle, medical history and self-reported health and current use of medication was collected using an interviewer-administered electronic questionnaire at local assessment clinics. All participants had mean levels of weight, height, RHR and blood pressure (BP) recorded. Repeat surveys of random samples of 5% of the CKB population (with about 80% response rates) were conducted at 4 years (2008) and 8 years (2013–2014) after baseline to assess the reproducibility of exposure variables.

Information on vital status and causes of death was collected by linkage, through unique national identification numbers, to death registries at China’s Disease Surveillance Points system, supplemented by annual active follow-up using local residential records, and contacting participants’ family members.19 Information on hospital admissions was collected by linkage to the health insurance (HI) claims system for all hospital admissions.19 Over 97% of the CKB population reported having HI coverage, and study records are updated every 6 months.19 For the small number of uninsured participants, information on hospital admissions was identified through annual active follow-up. All incident fatal and non-fatal disease outcomes were coded, using the International Classification of Diseases tenth revision (ICD-10), by trained staff who were blinded to the baseline information of study participants. In the present report, the analyses were censored at 31 December 2017. The diagnoses of HF and PHD were obtained from HI billing records, death or disease registries that followed contemporary Chinese clinical guidelines for the diagnosis of these diseases. The current clinical guidelines for HF in China include relevant symptoms (dyspnoea on exertion, orthopnoea, fatigue and oliguria) and clinical signs (bilateral pedal oedema, S3 gallop, hepatojugular reflux and pulmonary crepitations) and use of relevant medications (diuretics and angiotensin-converting enzyme II blockers).22 The current clinical guidelines for diagnosis of PHD include relevant symptoms, clinical signs and specific findings of enlarged P waves and right ventricular hypertrophy on electrocardiography and echocardiography.23 The participants with incident HF and PHD were defined as first non-fatal hospital admission or death due to HF (ICD-10: I50) or PHD (I27).

Statistical methods

Mean values and SD were used to summarise normally distributed variables, and medians and IQR were used for non-normally distributed variables. Multivariable Cox PH regression models were used to examine the associations of RHR with incident HF and PHD after stratification by region, age-at-risk, sex and season at recruitment. The multivariable regression analyses were sequentially adjusted for SES (marital status, highest level of education and household income in Yuan) and other established CVD risk factors (alcohol consumption, smoking status, physical activity in Metabolic Equivalent of Task hours per day (MET-hour/day), body mass index (BMI, in kg/m²), hypertension and diabetes mellitus). Detailed information regarding the categorisation of the covariates included in the multivariable Cox regression analysis is provided in online supplemental table. The likelihood ratio tests (LRT) for heterogeneity were used to assess departures from linearity in continuous variables or ordinal categorical variables. The LRT for heterogeneity was used to assess changes in the χ² statistic for the associations of RHR with incident HF or PHD after sequential addition of potential confounding factors.24

All associations of RHR with disease outcomes were corrected for regression dilution bias, by estimating the HRs and their corresponding 95% CI for fifths of baseline RHR and these were plotted against the mean baseline resurvey RHR in the respective baseline-defined groups. The quintile-specific HR and 95% CI were estimated using the floating absolute risk method,25 which enabled comparisons of the risks of HF or PHD between any two quintiles of RHR. Moreover, in the setting of a linear association of RHR with HF or PHD, the log HRs per 10 bpm higher baseline RHR (and their associated SEs) were divided by the regression dilution ratio (RDR) to obtain HRs per 10 bpm higher usual RHR.26 The RDR was computed using the MacMahon Peto method.26 27

Interaction terms were fitted to assess possible effect modification by age (at baseline), sex, smoking, BMI, physical activity, hypertension and diabetes and LRT were used to assess possible effect modification. Sensitivity analyses excluded first events of HF and PHD that occurred during the first 5 years of follow-up or individuals with any prior non-vascular diseases at baseline (chronic obstructive pulmonary disease, emphysema, bronchitis, pulmonary tuberculosis, asthma, any cancer, kidney disease, cirrhosis or hepatitis B or rheumatoid arthritis) or participants with poor self-reported health status at baseline. The impact of additional exclusions of participants with prior stroke or use of BP-lowering medications was also assessed. In addition, we excluded cases of PHD secondary to first HF diagnosis. Similarly, cases of HF after PHD diagnosis were also excluded. Two-tailed p values less than 0.05 were considered statistically significant. Data were analysed using Stata V.16.1 and R V.4.1.

Patient and public involvement

Local community leaders in China were consulted prior to enrolment of study participants in CKB. The findings of the CKB study are reported in peer-review publications and any relevant public health messages are disseminated using local press, television and internet to study participants.

Association of usual resting heart rate with incident heart failure and pulmonary heart disease

During a median follow-up of 11.1 years (IQR: 10.2–12.1) years, 6082 participants had HF and 5572 had PHD, respectively. A total of only 263 (4.7%) of the 5572 PHD occurred following a diagnosis of HF and only 325 (5.2%) of HF cases occurred after a diagnosis of PHD. There was a J-shaped association of RHR with HF, with a weak inverse association at RHR <75 bpm, but an approximately log-linear positive association at RHR ≥75 bpm (figure 2). In contrast, RHR showed strong positive and apparently log-linear association with risk of incident PHD throughout the range studied (figure 2).

Figure 3 shows the effect of sequential adjustment for potential confounding factors on the strength of the associations of RHR with HF (top panel) or PHD (bottom panel). After correction for regression dilution bias, each 10 bpm higher usual RHR was associated with a 25% (1.25; 95% CI 1.17 to 1.34) higher risk of HF for RHR for 75 bpm or greater (figure 3). The strength of the association of RHR with HF was approximately halved after adjusting for confounding due to hypertension and diabetes (as demonstrated by a 50% reduction in the χ²₁ from 83.4 to 41.3). For PHD, each 10 bpm higher usual RHR was associated with a 74% (1.74; 1.67–1.81) higher risk of PHD (figure 3). In contrast with HF, the strength of the association of usual RHR with PHD was largely unaltered by sequential additional adjustment for confounding factors (16.7% reduction in χ²₁: figure 3).

The risks of HF and PHD for each 10 bpm higher usual RHR varied substantially in different subgroups (figure 4). The strengths of the associations of usual RHR with HF were weaker among participants with hypertension and higher BMI, whereas for PHD, they were stronger among ex-smokers, at younger age and in those without diabetes (figure 4).

Online supplemental figure 1 shows the shapes of the associations of fifths of usual RHR with HF and PHD after excluding participants with prior diseases. Table 2 shows the HR (95% CI) of HF and PHD associated with 10 bpm higher RHR after excluding participants with prior diseases. For both HF and PHD, excluding individuals with non-vascular diseases significantly attenuated the associations, while the converse was true after excluding those with prior stroke or BP-lowering medications. Exclusion of non-vascular diseases at baseline reduced the strength of the association of RHR with HF and PHD by about 50% but did not materially change the shape of the associations (online supplemental figure 1 and table 2). In addition, excluding the HF events after a diagnosis of PHD attenuated the association of RHR with HF (table 2 and online supplemental figure 1). However, excluding events occurring during the first 5 years of follow-up did not materially alter the associations.