Muscle strength in adolescent men and risk of cardiovascular disease events and mortality in middle age: a prospective cohort study

Muscle strength measurement

As a measure of general muscle strength, we used the mean standardized score of three isometric muscle strength tests; hand grip strength, elbow flexion and knee extension. At baseline, the muscle strength was measured as previously described [16] (personal communication L-O Nordesjö, 19 December 2012). However, the exact test protocol was not available due to military restrictions. In summary, a specially constructed “muscle chair” was used with strain gauges (KRK-2 and KRG-4) and indicator (BKI-1) from AB Bofors modified by AB Medicinska Apparater, Södertälje, Sweden. Hand grip strength was measured with the hand in a vertical position, with 90° flexion at the elbow, and the humerus being parallel with the torso. Elbow flexion and knee extension were measured in a sitting position with 90° flexion over the main joint and with straps over the radial styloid process and 5 cm above the most distal part of the fibular malleolus, respectively. Using percentiles of the mean standardized score (standardized score = (value - mean)/standard deviation), we categorized the cohort into low, average and high strength with the 25th and 75th percentiles as cut offs (that is, the category average strength contained 50% of the sample). By using the mean of the standardized muscle strength scores, the relative strength of each muscle group is equally contributing to the composite muscle strength measure.

Covariates

Weight was measured to the nearest kilogram. Height was measured to the nearest centimeter. We calculated BMI as weight/height². Smoking at baseline was categorized into three groups (non-smokers, 1 to 10 cigarettes per day and >10 cigarettes per day). Self-reported total intake of alcohol was measured in grams per week as previously reported [17] and categorized into four groups (0, 1 to 100, 101 to 250, >250).

As a measure of CRF, we used physical work capacity (in relation to body weight) categorized as low or high. Subjects who were not limited by disease, including a present cardiovascular condition, performed a maximal test (W_max6) on a cycle ergometer (Elema-Schönander Model AM 368 or similar) [18, 19]. The test has been reported to correlate well with VO₂max in young male students (r = 0.9) [20, 21], though the exact test protocol was not available to us due to military restrictions. In short, each subject started on the same load (230 W) and worked until exhaustion or the prescribed pedaling rate (60 rotations/minute) could not be upheld. For a minority still working at 12 minutes, further increments of 30 W were made every 6 minutes. We excluded subjects who did not reach 90% of age-predicted maximal heart rate (MHR) at the end of testing, that is, <175 beats/minute (MHR = 208 - 0.7 * 18.5) [22]. The test results were registered as stanine scores with known corresponding intervals of physical work capacity in W. A stanine scale is based on the normal distribution and usually used for standardized testing. The scale has nine increments, the first and ninth step each delimit 4% of the population whereas step five constitutes 20% [23]. To allow estimation of physical work capacity in relation to body weight as a proxy for CRF, we estimated the exact result for each conscript using the mean of the upper and lower limit of each stanine score interval. Exceptions were subjects with the highest stanine score, which we categorized as having high fitness, independent of their body weight. Otherwise, those with the highest score weighing the most would be systematically categorized as having low CRF.

We collected data on adult and childhood socioeconomic status from Statistics Sweden. As a proxy for the conditions during childhood, we used information on parental occupation and education in 1960. For a minority of subjects with missing values, the self-reported parental socioeconomic status at conscription was used. For categorization into four groups, we used a white/blue collar approach. For adult socioeconomic status, we used attained level of education in 1990.

Statistical analyses

All statistical analyses were made in SAS 9.3 (SAS Institute, Cary, NC, USA). We used Cox proportional hazard models to calculate hazard ratios (HR) for outcomes and control potential confounders. In model I, we adjusted for BMI and smoking. In model II, we further adjusted for alcohol, CRF and socioeconomic status. For the middle age analyses, all men in the cohort living in Sweden on 1 January 1991, were included in model I (37,542 men). For model II (main model), the corresponding number was 27,766 men (74.0%). For the mortality analysis starting in 1970, 30,186 men (78.2%) had complete information in model II.

Our primary endpoint was time to CVD event in middle age. Due to the start of data registration on hospitalization and attained education, the actual follow-up regarding ischemic CVD started when the men were typically aged 39 and continued for 20 years through middle age. In other words, for all men alive 1 January 1991, we collected data on hospitalizations with a CVD diagnosis until 31 December 2010 from the nation-wide Swedish in-patient register [24]. With few exceptions, all diagnoses from in-patient care in Sweden are reported to the National Board of Health and Welfare. We defined CVD events as coronary heart disease or stroke morbidity (including hemorrhagic stroke but excluding transient ischemic attack) according to diagnoses based on the International Classification of Disease (ICD) version 9 (410 to 414, 430 to 434, 436 to 438, 342, 344) and 10 (I20 to I25, I60 to I66).

Starting 1 June 1970, we followed up on all deaths until 2 August 2012. However, the cause of death registered according to the ICD (versions 8, 9 or 10) was not available for deaths in 2012. Thus, the analyses of all-cause and specific mortality differed by seven months in time to follow-up. For the period 1970 to 2011, we separately identified deaths due to CVD (ICD-8, ICD-9: 390 to 459 ICD-10: I00 to I99). We also separately investigated CVD and all-cause mortality in middle age (1991 to 2011), the period for which morbidity data was available.

Sensitivity analyses

To investigate if the measure of socioeconomic status influenced the results, we also investigated socioeconomic status as the socioeconomic index (SEI) 1990. Furthermore, we investigated if year of birth, marital status, height, exclusion due to heart rate <175 at the end of CRF testing, the conscription office or blood pressure (potential mediator) had any influence on HR estimates. Using the first quintile of mean muscle strength as reference, we also tested trend across groups of muscle strength. In separate analyses, we also excluded subjects with CVD (ICD 8: 390 to 459) or psychiatric disease (ICD 8: 290 to 315) at baseline.

Sensitivity analyses

Year of birth and conscription office did not alter the pattern of HR estimates. In general, being unmarried was associated with an increased risk of the outcomes, although we observed no major changes in the HR estimates of muscle strength. High diastolic blood pressure was typically associated with having a cardiovascular outcome but had only minor effects on risk estimates. Adjusting for SEI instead of education or including subjects independently of the heart rate level at the end of CRF testing did not affect the overall pattern of association. Adding height to the fully adjusted models, the observations regarding all-cause and CVD mortality for the low strength group remained essentially unchanged, whereas the association with CVD events in the high strength group were slightly attenuated (HR 0.92, 0.80 to 1.04). We observed no trend across muscle strength groups for any of the outcomes. When we excluded men with a CVD (n = 1,271) or a psychiatric diagnosis (n = 4,324) at baseline, we observed only minor changes of HRs and the overall patterns remained.

Muscle strength as a risk factor for cardiovascular disease

With observations that suggest an inverse association between muscle strength in young men and later CVD, the pathways of potential causality remain as a central question. Does high muscle strength add CVD protection apart from CRF or is it merely a proxy for a more favorable future CVD risk profile? Notably, there are previous studies supporting a biological link between muscle strength and later CVD, such as obesity protection [6] through increased resting energy expenditure [25] and lower arterial stiffness in young men with high muscle strength [9]. Early developmental aspects might also play a role as birth size is associated with both low adult muscle strength [26] as well as ischemic heart disease [27]. Whereas we observed a decreased risk of CVD events in men with high muscle strength, the group had no decreased risk of CVD mortality. In contrast, having low muscle strength was associated with increased risk of CVD mortality but not for CVD events. Potentially, this could be explained by muscle strength having different thresholds of protective effect depending on the severity of the outcome measurement. In the context of this study, there is scarce support in the literature that any of the muscle strength tests would be particularly associated with CRF and CVD. Thus, we have not separately investigated each muscle strength test and its association with the outcomes to reduce the number of primary analyses and avoid a data mining approach. We chose to focus the main analyses on detecting non-linear associations as we found it more feasible to use the average muscle strength group as reference from a public health perspective. Furthermore, the known variability in muscle adaption following training would support a non-linear association [28].

Low muscle strength in adolescent men has previously been linked to an increased risk of later suicide mortality [29]. It is plausible that subjects with psychiatric disease or CVD at baseline perform less well on physical capacity test in general, possibly introducing bias. However, when we excluded subjects with CVD or psychiatric disease at baseline, the risk estimates were only marginally altered. When we adjusted for conscription office, a few offices appeared statistically significant without altering the risk estimates substantially.

Strengths and limitations

The study includes a large population-based sample, includes prospectively ascertained exposure information and baseline data on known confounders, has a long follow-up of four decades, and small loss during follow-up using comprehensive registers. However, there are also limitations that should be considered when interpreting the results. First of all, we do not know how the motivation for military service may have biased our observations. However, as few men were fully discharged, most hypothetical underachievers would still have to enlist, possibly raising motivation for testing. Although morbidity data for CVD events are missing in young adulthood, the incidence of coronary heart disease and ischemic stroke is low prior to middle age [30, 31]. Thus, the large majority of cases should still have been identified with the present study design. Also, if our hypothesis is correct, early cases not registered would at most dilute our results. It should also be noted that all conscripts included in the study did not perform a maximum work capacity test. As non-performers typically have lower muscle strength than performers, this should also at most dilute our results. Furthermore, CRF was not directly measured but approximated from a physical work capacity test on a bicycle. Compared to the treadmill testing used in other studies in the area [6, 8], the CRF variable might be slightly inferior. However, CRF had a statistically significant effect on the HR estimate of total CVD mortality (low vs. high 1.33, 1.10 to 1.60). Noticeably, the study design also allows us to adjust the risk estimates for the mediating effects of attained education. Previous studies have reported that changes of behavior or physical fitness during adulthood affect the mortality risk [32, 33], data not available for the cohort. However, when we adjust for the attained education of subjects in their late 30s, we take some of the environmental differences in early adulthood into account. Although widely used in both primary and secondary screening of CVD, we do not have any data on biochemical markers of CVD risk, such as cholesterol levels. However, we first and foremost regard the metabolic profile of the individual as an intermediate variable between fitness and later CVD and not as a confounder. We have not been able to study the effects of physical activity level on our observations as these data were not available for the cohort. Thus, physical activity level might mediate unmeasured confounder effects. From a statistical viewpoint, it is important to note that Cox proportional hazard models might produce biased results when studying other outcomes than all-cause mortality due to competing risks (that is, death) [34]. However, we did not observe any major differences between subgroup curves of cumulative incidence compared to the corresponding curves adjusted for competing risks.

Generalization and future perspectives

In brief, as long as the limitations of the study are acknowledged, we find our observations generalizable to a male population of European ancestry. The study does not include women as only men were subjected to conscription testing. Furthermore, as the muscle mass and isometric muscle strength greatly differ among young males and females, our study does not give sufficient evidence to suggest a similar association in women as in men. Although the source population was ethnically homogenous and predominantly white, muscle strength has been shown to be an important predictor of mortality also in other ethnicities, such as men of Japanese ancestry [35].

We consider muscle fiber type distribution (type I or type II) to be an area of interest for future research. Muscle fiber type distribution has a major congenital component [36], is affected by physical activity [37], and is associated with isometric muscle strength [38] as well as obesity and type 2 diabetes [39, 40]. With promising techniques that measure muscle fiber type non-invasively [41], large cohort studies on how muscle fiber type distribution is associated with CVD are possible. It is also important to better understand how fitness interventions alter the CVD risk for individuals with different levels of baseline fitness.