Associations of handgrip strength with morbidity and all-cause mortality of cardiometabolic multimorbidity

Between March 2006 and July 2010, the UK Biobank recruited > 500,000 community-dwelling individuals, aged 37–73 years, from the general population [20, 21]. This prospectively nationwide population-based study collected extensive baseline information via questionnaires, physical measurements, and biological sample testing from 22 dedicated assessment centers across the UK [22, 23]. Details of the UK Biobank have been reported in previous studies [20, 23]. The UK Biobank study was approved by the North West Multi-Centre Research Ethics Committee (REC reference: 11/NW/03820), and all participants gave written informed consent before joining this study.

In the current study, participants who met any of the following criteria at baseline were excluded: (1) withdrawn from the survey, (2) suffered from type 1 diabetes or gestational diabetes at baseline, and (3) had missing data on HGS. Finally, 441,868 individuals having no cardiometabolic disease, 45,312 individuals with 1 cardiometabolic disease, and 6594 individuals with at least 2 cardiometabolic diseases at baseline were included (Fig. 1).

HGS was assessed using the Jamar J00105 hydraulic hand dynamometer, with each of the dynamometer calibrated before measurement. Participants were seated upright with their elbows at 90° angles so that their forearms were facing forwards and placed on the armrest. Both left and right handgrip strengths were accurately measured through squeezing the dynamometer as strongly as possible for about 3 s. Participants were also asked about their handedness, and the score from their dominant hand was used in analyses. When the participants failed to indicate a dominant hand, the highest score across both hands was selected and used in analyses. Then, we used the combined quartiles (synthesizing quartile standards for men and women) for analyses on account of the different HGS levels between the two sexes (Additional file 1: Table S1) [24].

Cardiometabolic diseases were identified from self-reported medical conditions, primary care, linked inpatient hospital data, and death registry records using the tenth edition of the International Classification of Disease (ICD-10) from “first occurrence fields (Category ID: 1712)” as the criteria for the entire cohort from England, Scotland, Wales, and the national health registrations. Specific information can be found online (https://​biobank.​ndph.​ox.​ac.​uk/​showcase/​label.​cgi?​id=​1712). In this study, cardiometabolic diseases included type 2 diabetes (E11), stroke (I60-64, I69), and CHD (I20-25), and detailed disease definitions can be seen in Additional file 1: Table S2. The time of CM onset was determined as the date when the second cardiometabolic disease was diagnosed. Date of death was received from death certificates held by the National Health Service (NHS) for participants from England and Wales, and the NHS Central Register (NHSCR), Scotland, for participants from Scotland. At the time of analysis, mortality data were available to 23 March 2021. Further detailed information for the outcomes on the UK Biobank can be found at http://​www.​ukbiobank.​ac.​uk. The primary outcomes were CM (at least two from type 2 diabetes, CHD, and stroke) and all-cause mortality.

Covariates included sociodemographic characteristics and lifestyle factors. Sociodemographic covariates included age (continuous), sex (men and women), ethnicity (classified as White, mixed, Asian/Asian British, Black/Black British, and others), and the Townsend deprivation index. The Townsend deprivation index, derived from the postcode of residence, was used to describe the area-based socioeconomic status through the quartiles of indices [25]. Lifestyle covariates were measured by self-report, including smoking status (classified as never, former and current smoking) and frequency of alcohol intake (divided into never, special occasions only, one to three times a month, once or twice a week, three or four times a week, and daily). Physical activity was assessed with the International Physical Activity Questionnaire (IPAQ) short form [26]. The volume of physical activity was calculated as the sum of walking, moderate and vigorous activity over the previous week, measured as metabolic equivalents (METs min/week) [27]. A MET-minute is computed by multiplying the MET score of an activity by the minutes performed. As recommended by IPAQ guidelines, the MET scores were 3.3 for walking, 4.0 for moderate physical activity, and 8.0 for vigorous physical activity. Physical activity was categorized into three groups: low (MET < 600 min/week), moderate (600 to 3000 MET), and high (MET > 3000) [26]. Hypertension was defined using the combination of the self-reported history of antihypertensive medication use, medical diagnosis records, and clinical measurements of blood pressure (systolic blood pressure (SBP) ≥ 130 mmHg or diastolic blood pressure (DBP) ≥ 80 mmHg). Furthermore, height and body weight were measured by well-trained nurses during the initial assessment center visit. We calculated body mass index (BMI) as weight (kg) divided by the square of height (m) and divided BMI into four categories according to the recommendations of the World Health Organization (WHO): underweight (< 18.5 kg/m²), normal weight (18.5 to < 25), overweight (25 to < 30), and obesity (≥ 30). Further details of these measurements can be found in the online UK Biobank Showcase (http://​biobank.​ndph.​ox.​ac.​uk/​showcase).

For participant baseline characteristics, we described continuous data as mean and SD and calculated the frequencies and percentages for categorical variables.

Follow-up time was from the baseline date to the diagnosis date of the outcomes, date of death, or the censoring date (March 23, 2021), whichever came first. Kaplan–Meier curves stratified by HGS quartiles are showed in Additional file 1: Fig. S1. The proportionality of hazards assumption of Cox regression model was tested by Schoenfeld residuals method. Multivariable cubic regression splines were used to evaluate the linearity assumption for the continuous exposure (per 1 SD), and we found no evidence of deviation from linearity.

The associations of HGS (per 1 SD decrease and in quartiles, Q1-Q4) with CM and all-cause mortality were estimated by Cox proportional hazards models. For participants who have none cardiometabolic disease at recruitment (N = 441,868), we examined the association of HGS with onset CM. The multistate model was performed to evaluate the associations of baseline HGS with risks of all transitions from a “healthy” state to one cardiometabolic disease, then to subsequent CM and ultimately to all-cause mortality (Fig. 2). And the multistate model included five transitions: (A) from baseline to one cardiometabolic disease, (B) from one cardiometabolic disease to CM, (C) from baseline to death, (D) from one cardiometabolic disease to death, and (E) from CM to death. In subsequent analyses, we also examined the risk of CM associated with HGS among individuals with only one cardiometabolic disease (type 2 diabetes, stroke or CHD, N = 45,312) at baseline. Finally, the risk magnitude of HGS for all-cause mortality was estimated among baseline CM participants (N = 6594). All results are presented as hazard ratios (HRs) and 95% confidence intervals (CIs).

Multivariable-adjusted models included age, sex, socioeconomic status, ethnicity, smoking status, alcohol drinking, physical activity, BMI, and hypertension. Blood lipids were not included in the adjusted model, since they may be potential mediators that are caused by the exposure and in turn cause the outcome. We used multivariate imputations by chained equations (MICE) based on random forest to fill in the missing values of the covariates [28]. The imputation models used waist circumference, sleep duration, average total household income before tax, forced vital capacity, albumin, and creatinine as the auxiliary variables. We performed 20 iterations and generated 5 datasets to account for the missing data at baseline (Additional file 1: Tables S3 and S4) and combined the estimates of the 5 datasets according to Rubin’s rules [29]. Potential sex, age, or physical activity interactions with HGS were computed by the likelihood ratio test to compare models with and without cross-product interaction terms (Additional file 1: Table S5). Subgroup analyses by age (< 60 vs. ≥ 60), sex (men vs. women) and physical activity (low vs. moderate vs. high) were performed to evaluate the robustness of the findings. The subgroup analyses by physical activity levels were post-hoc analyses following peer-review without a pre-specified hypothesis.

Furthermore, we conducted a sensitivity analysis by adding a new variable = entry_time + 2 (years) as the entry time and removing the participants with a follow-up time less than zero to minimize the potential reverse causation and induction time bias.

All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA) and R software (4.0.5).

A total of 441,868 participants without any cardiometabolic disease at baseline were included in this stage. Over a median follow-up period of 12.1 years, 4701 (1.06%) participants developed CM. The main characteristics of the participants without any cardiometabolic disease at recruitment are summarized in Table 1.

As shown in Table 2, after multivariable adjustment (model 3), per 1 SD lower HGS was positively associated with the risk of CM (HR: 1.17, 95% CI: 1.14–1.21). Compared with the fourth quartile (Q4), the HR (95% CI) values of Q1 of HGS for developing CM were 1.46 (1.34–1.60). The multistate model revealed that HGS (Q1 vs. Q4) was associated with risks of all transitions from baseline to individual cardiometabolic disease, then to subsequent CM and ultimately to death, except the transition from CM to all-cause mortality (transition E, Fig. 2 and Additional file 1: Table S6). The multivariable adjusted HR (95% CI) values of HGS were 1.34 (1.30–1.38) for transition A, 1.16 (1.05–1.28) for transition B, 1.35 (1.29–1.42) for transition C, 1.32 (1.21–1.44) for transition D, and 1.15 (0.94–1.42) for transition E, respectively.

Among 45,312 individuals with one cardiometabolic disease at baseline, 3505 (19.23%) out of 18,224 participants developed CM from type 2 diabetes; 1258 (20.65%) out of 6092 participants had CM from stroke; and 3301 (15.72%) out of 20,996 participants experienced CM from CHD. Baseline characteristics of the study population are presented in Table 1 by different baseline disease conditions. The mean age of the CHD group was older than the other two groups (61.71 years old for CHD vs. 60.12 years old for stroke vs. 59.62 years old for type 2 diabetes). Likewise, the proportion of men among patients with CHD was the highest (CHD 67.00%; type 2 diabetes 60.32%; stroke 53.81%). However, levels of physical activity, BMI, HDL, SBP, DBP, and triglyceride did not meet this pattern.

For individuals with type 2 diabetes at the time of inclusion, the multivariable adjusted HRs (95% CI) the HGS of Q3-Q1 (Q4 as the reference) were equal to 1.01 (0.91–1.11), 1.15 (1.05–1.26), and 1.35 (1.23–1.49), respectively (Table 2).

Patients with stroke at baseline were then followed up for CM outcomes. Lower levels of HGS were associated with an increased risk of CM and the HR (95% CI) value of Q1 of HGS for CM was 1.23 (1.04–1.46) after full adjustment when compared to Q4 (Table 2).

In the fully adjusted model (model 3), positive associations were observed between low HGS and the risk of new-onset CM in patients with CHD at recruitment. Individuals in Q3-Q1 of HGS had 1.14 (1.03–1.25), 1.15 (1.05–1.27), and 1.23 (1.11–1.36) times the risk of developing CM than those in Q4, respectively (Table 2).

Among 6594 individuals with CM at baseline, the mean age was 62.09 ± 5.84 years, and 1826 (27.69%) died over a median follow-up period of 11.7 (10.8–12.6) years. Of those, men were more likely than women to experience CM, with 72% of men and 28% of women, and the average HGS values were 36.39 ± 9.19 for men and 20.90 ± 6.69 for women (Table 1).

When compared to Q4, the unadjusted HR (95% CI) of Q1 of HGS for all-cause mortality (model 1) was 1.87 (95% CI: 1.64 to 2.14). This association was slightly attenuated, but remained after full adjustment (HR: 1.57, 95% CI: 1.36–1.80; Table 3).

In the subgroup analyses by age (< 60 vs. ≥ 60), sex (men vs. women), and physical activity (low vs. moderate vs. high), low HGS still possessed a high risk of CM among participants with none or only one cardiometabolic disease (Additional file 1: Figs. S2-S7), with several exceptions among patients with stroke (e.g., in the group of age under 60 years). Meanwhile, associations of HGS with all-cause mortality among patients with CM persisted in nearly all sub-groups (Additional file 1: Figs. S8-S10). In addition, we also observed a similar pattern of results in the sensitivity analyses for HGS after adding a new variable = entry_time + 2 (years) as the entry time (Additional file 1: Tables S7 and S8).