Females Display Lower Risk of Myocardial Infarction From Higher Estimated Cardiorespiratory Fitness Than Males: The Tromsø Study 1994-2014

Objective To examine the dose-response association between estimated cardiorespiratory fitness (eCRF) and risk of myocardial infarction (MI). Patients and Methods Adults who attended Tromsø Study surveys 4-6 (Janurary 1,1994-December 20, 2008) with no previous cardiovascular disease were followed up through December 31, 2014 for incident MI. Associations were examined using restricted cubic splines Fine and Gray regressions, adjusted for education, smoking, alcohol, diet, sex, adiposity, physical activity, study survey, and age (timescale) in the total cohort and subsamples with hyperlipidemia (n=2956), hypertension (n=8290), obesity (n=5784), metabolic syndrome (n=1410), smokers (n=3823), and poor diet (n=3463) and in those who were physically inactive (n=6255). Results Of 14,285 participants (mean age ± SD, 53.7±11.4 years), 979 (6.9%) experienced MI during follow-up (median, 7.2 years; 25th-75th, 5.3-14.6 years). Females with median eCRF (32 mL/kg/min) had 43% lower MI risk (subdistributed hazard ratio [SHR], 0.57; 95% CI, 0.48-0.68) than those at the 10th percentile (25 mL/kg/min) as reference. The lowest MI risk was observed at 47 mL/kg/min (SHR, 0.02; 95% CI, 0.01-0.11). Males had 26% lower MI risk at median eCRF (40 mL/kg/min; SHR, 0.74; 95% CI, 0.63-0.86) than those at the 10th percentile (32 mL/kg/min), and the lowest risk was 69% (SHR, 0.31; 95% CI, 0.14-0.71) at 60 mL/kg/min. The associations were similar in subsamples with cardiovascular disease risk factors. Conclusion Higher eCRF associated with lower MI risk in females and males, but associations were more pronounced among females than those in males. This suggest eCRF as a vital estimate to implement in medical care to identify individuals at high risk of future MI, especially for females.


C
ardiovascular diseases (CVD) are the leading causes of global mortality. 18][9] Young females also have higher in-hospital MI mortality than males. 8The American College of Cardiology and American Heart Association (ACC/AHA) guidelines on CVD prevention highlights missing data on risk assessment in young individuals. 10As such, early identification of these high-risk groups may further improve prevention of MI and aid in lowering the burden of CHD. 11ardiorespiratory fitness (CRF) is consistently found to be an independent predictor of CHD [12][13][14][15][16] and mortality, 13,17 which has led to suggestions that CRF should be included as a vital measurement in routine medical care. 18However, the dose-response association between CRF and CHD appears equivocal between studies.Some studies report a Jshaped association, 12,13,15,16,[19][20][21] whereas others report an inverse, linear association, with the lowest CHD risks at the highest CRF level. 22,23These inconsistent observations may be attributed to categorization of CRF into study-specific centiles, 12,13,15,16,[19][20][21][22][23] which leads to loss of information 24 and complicates translation to clinical decision making. 11Thus, using continuous CRF data may preserve information content and improve statistical power. 24ncluding cardiopulmonary exercise testing for direct CRF assessment in routine care is challenging owing to high costs, time, the need for specialized equipment, and skilled physiologists. 251][22] Therefore, eCRF may be a feasible option for routine assessment when direct cardiopulmonary testing is unavailable or unfeasible. 25][15] In studies examining CRF and MI risk including both sexes, results are inconsistent. 16,19In one study, higher CRF was associated with a lower MI risk in males but not in females, 16 whereas in another study, only in females. 19n this study, we aimed to examine the dose-response association between eCRF in continuous data form and risk of MI in a large cohort of females and males and in subsamples of individuals with 1 or more CVD risk factors.

Study Sample and Design
This is a prospective cohort study with adult participants aged 25-86 years from the Tromsø Study, an ongoing population-based cohort study in Tromsø municipality, northern Norway. 27We included participants attending at least one of the Tromsø4-Tromsø6 surveys (Tromsø4 1994-1995, attendance: 77%; Tromsø5 2001, attendance: 79%; Tromsø6 2007-2008, attendance: 66%) 27 because these include information on variables to estimate CRF (age, sex, waist circumference, self-reported physical activity, and resting heart rate).Additional inclusion criteria were information on education, alcohol, diet, and smoking.We excluded participants with present or previous CVD.If participants attended more than once, their earliest attendance was used.In total, 14,285 participants were included (Supplemental Figure 1, available online at http://www.mcpiqojournal.org), of which 7873 (55%) were females (Table ). 28

Ethical Considerations
The Tromsø Study surveys were conducted according to the Declaration of Helsinki.All participants provided written informed consent.The Regional Ethics Committee for Medical and Health Research Region North approved this study (Ref.: 2016/1792).

Diagnosis of MI
Incident MI diagnosis was identified through linkage to the diagnosis registry at the University Hospital of North Norway, the only hospital serving Tromsø municipality, 29 and the Norwegian Cause of Death Registry, 30 searching for International Classification of Disease (ICD), ninth edition, codes 410-414, 427, 428, 798, and 799 and ICD-10 codes I20-I25, I46-I48, I50, R96, R98, and R99.In addition, manual and/or electronic text searches for notes on MI were completed with paper (used until 2001) and digital hospital records for all participants with a diagnosis of ICD-8 and ICD-9 codes 430-438 and ICD-10 codes I60-I69, G45, G46, or G81. 29Experienced physicians reviewed and validated all diagnoses on the basis of hospital records and, when available, death certificates and autopsy reports.Review of medical records minimizes misclassification in data collection from health registries. 29Emigration and moving date were retrieved from the Norwegian Population Registry.Participants were followed up from  (1994-1995 and 2001, respectively), we used the formula by Nauman et al, 21 being based on the Cohort of Norway physical activity questionnaire 31 that was used in Tromsø4-5 (Supplemental Table 1, available online at http://www.mcpiqojournal.org).For participants in Tromsø6 (2007-2008), we used the formula by Nes et al, 32 which is based on the physical activity frequency, intensity, and duration questionnaire that was used in Tromsø6 (Supplemental Table 2, available online at http://www.mcpiqojournal.org).Different algorithms were fitted because of different physical activity questionnaires in Tromsø4-5 vs Tromsø6.Nevertheless, both formulas are validated in the same cohort sample and should, thus, represent the same CRF values, which both explain w60% of the variance in directly measured CRF from a test to exhaustion using indirect calorimetry. 21,32rom these formulas, we expressed eCRF as maximal oxygen uptake in milliliter per kilogram body weight per minute (mL/kg/min) (Supplemental File 1).

Covariates
On the basis a directed acyclic graph, we identified education, smoking, alcohol intake, diet, age, sex, adiposity, and physical activity as potential confounding sources in the association between eCRF and MI (Supplemental Figure 1).Because age, sex, waist circumference, and physical activity are included in eCRF formula, education, smoking, alcohol intake, and diet were included as potential confounders.Educational level was categorized into primary school, high school, university <4 years, and university !4 years (Supplemental File 2, available online at http://www.mcpiqojournal.org).
Smoking was categorized as current, previous, or never.We harmonized alcohol intake (units/ wk) from multiple questions on alcohol intake (Supplemental File 3 and Supplemental Tables 3-5, available online at http://www.mcpiqojournal.org).Diet quality was harmonized according to national nutritional guidelines 33 on a scale from 0.0 to 4.0 of fruit, saturated fat, fish, and processed meat intake from multiple questions on food intake (Supplemental File 4 and Supplemental Tables 6-8, available online at http://www.mcpiqojournal.org).

Definition of Subsamples With Additional CVD Risk Factors
Hypertension (yes/no) was defined by a combination of questionnaires, reported medicine use (Anatomical Therapeutic Chemical: C02, C03, C07, C08, and C09), and blood pressure recordings (Supplemental File 5, available online at http://www.mcpiqojournal.org).Hyperlipidemia (yes/no) was defined by serum total cholesterol (!5.17 mmol/L), questionnaires, and reported medicine use (Anatomical Central obesity is defined as waist circumference (cm) thresholds at specific body mass index thresholds, as described by Ross et al. 28 .eCRF, estimated cardiorespiratory fitness; MET, metabolic equivalent of task.
Therapeutic Chemical: C10) (Supplemental File 5).Central obesity was defined as specific waist circumference (in centimeters) thresholds at specific body mass index (calculated as the weight in kilograms divided by the height in meters squared) categories (normal weight, overweight, obese, and obese class II) according to Ross et al 28 (Supplemental File 5).Metabolic syndrome was defined according to the International Federation of Diabetes 34 (Supplemental File 5).Physical inactivity was defined as reporting <7.5 metabolic equivalents of tasks (METs) per week of moderate intensity (equivalent to the lower-limit physical activity guideline 35 ), which we calculated from the Cohort of Norway physical activity questionnaire (Tromsø4-5, 1994-1995 and 2001, respectively) (Supplemental Table 7) and the physical activity frequency, intensity, and duration questionnaire (Tromsø6, 2007-2008) (Supplemental Table 8).

Statistical Analyses
To examine the dose-response association between eCRF and MI, we used restricted cubic splines in Fine and Gray regressions 36 to account for competing risks of death from other causes than MI.We performed analyses separately by sex because CRF 21 and MI risk 9 differ by sex.We further examined the associations in subsamples with hypertension, hyperlipidemia, metabolic syndrome, central obesity, being physically inactive, or not meeting any nutritional guidelines and in those having !2, !3 and !4 CVD risk factors.As metabolic syndrome is a composite of multiple CVD risk factors, 34 it was not included in the summation of !2, !3 and !4 CVD risk factors.We adjusted all analyses for education, smoking, diet quality, alcohol intake, study survey (dummy variable), and age as timescale. 37Waist circumference and physical activity were included in the eCRF formulas 21,31 and consequently not additionally adjusted for to avoid multicollinearity.Participants entered the analyses 2 years after study attendance (left truncation).
Using time-dependent weights as described by Lambert, 38 we calculated modified weighted Schoenfeld residuals to test proportional subdistributed hazards by goodness-of-fit tests by Zhou et al 39 ; all covariates indicated proportional subdistributed hazards (all P>.12) except smoking in both models (females: P¼.004; males: P¼.01).However, the log-log survival plot of subdistributed hazards displayed reasonable parallel lines between subgroups of smoking status (Supplemental Figures 2 and 3, available online at http:// www.mcpiqojournal.org).Knots in the restricted cubic splines were placed at the 10th, 50th, and 90th percentiles of the distribution of eCRF.The reference value for doseresponse splines were set at the 10th percentile of the distribution, separately by sex (males: 32 mL/kg/min; females: 25 mL/kg/min).Changing knot placements or knot numbers did not change interpretation of the spline slopes.Wald tests indicated departure from linearity in all models (all P<.004).
For sensitivity analyses, we performed the following: (1) examined the associations in those aged <60 years and >60 years to evaluate whether age had large influence on the association magnitude (age is inversely associated with CRF 31,40,41 ); (2) created agespecific quintiles of eCRF in accordance with recommendations to limit the influence of age in CRF-health outcome associations 40 ; and (3) set study entry 5 years after study attendance to evaluate the influence of reverse causation bias.All analyses were performed using Stata version 17 (StataCorp) with an a at 0.05.Data are shown as subdistributed hazard ratio (SHR) with 95% CIs and as frequency (%) or mean AE SD for descriptive data.
Compared with the total sample (Figure 1), the association magnitudes and spline slopes were similar in subsamples with 1 other CVD risk factor (Figures 2 and 3), and even with 4 or more CVD risk factors, higher eCRF was associated with a lower risk of MI (Figure 4).

DISCUSSION
In this prospective cohort study, higher eCRF was associated with a substantial lower risk of MI in an exponential pattern in both females and males.However, the lower risk was more pronounced in females than that in males.These patterns of association were also evident among those with 1 or more CVD risk factors and in sensitivity analyses of those older than 60 years.
Previous studies including both fatal and nonfatal MI and including both sexes have reported conflicting results regarding the association between CRF and MI separately for females and males, with 1 study reporting an association only in females, 19 whereas another only in males. 16In this study, higher eCRF was associated with a lower MI risk in both females and males.However, using continuous eCRF data in restricted cubic splines, this study illustrated that the dose-response association between higher eCRF and MI was nonlinear and more pronounced in females than that in males.For example, an eCRF corresponding to 32 mL/kg/min in maximal oxygen uptake (equivalent to 9.1 METs in a maximal exercise test) for females and 45 mL/kg/min (equivalent to 12.9 METs in a maximal exercise test) for males would indicate a w40% lower risk of MI; higher eCRF levels would indicate an even lower MI risk.
Although MI incidence and mortality is decreasing in the Western world, 2,4-7 preventive measures can further decrease CHD mortality and morbidity. 11The ACC/AHA guidelines on CVD prevention highlights the importance of identifying those who will benefit most from preventive measures, especially as risk assessment for young individuals are lacking. 10Acute MI more often manifests silently in females than that in males, 9 and more young females than males are hospitalized owing to MI, 8,9 leading to a greater comorbidity burden among females than that in males. 9The dose-response curves for CRF as observed in this study may aid clinicians in evaluating risks of future MI among their patients separately for females and males, with simple use of self-reported physical activity, waist circumference, and resting heart rate.Hence, our findings indicate that eCRF may aid as an essential early MI risk identifier, especially among females.Moreover, although previous studies also have observed an association between CRF and CHD in individuals with established CVD risk factors, such as hypertension, 42 Frequency Estimated cardiorespiratory fitness (mL .kg -1.min -1 )

Hyperlipidaemia Hypertension
Subdistributed hazard ratio hyperlipidemia, 43 smoking, 43 and obesity, 44 this is the first study using eCRF to examine the association with fatal and nonfatal CHD among females and males with CVD risk factors.Thus, even among those with other established risk factors for CVD, eCRF may be able to identify those at even higher risk of MI.Indeed, previous studies have indicated that including directly measured CRF 12 or eCRF 21 improves the risk prediction of CHD beyond traditional risk factors.The lower MI risk with higher eCRF in females compared with that in males may be because females on average experience MI at older age than males 9 ; because age explains most of the variance (w30%) in the eCRF formulas, 21,31 this could inflate the association magnitude.However, the sensitivity analysis including only individuals older than 60 years displayed a substantially lower MI risk in females, indicating that females may derive greater MI risk reduction from a high CRF than males, 19 as also observed for physical activity and CHD. 45lternatively, males overreport their physical activity level to a greater extent than females 46 ; because the eCRF formulas include self- reported physical activity, 21,31 the sex difference in the association magnitude may also partly be explained by greater regression dilution bias in results for males than that for females.

Strengths
In this study, our use of restricted cubic splines illustrates that eCRF is nonlinearly and exponentially associated with a lower risk of MI in both females and males.These substantial magnitudes are likely attributed to our use of continuous eCRF data that preserve information quality and statistical power, 24 which is easier to transfer to clinical decision making than arbitrary study-specific categorized data. 11Moreover, we used Fine and Gray regressions to account for competing risk of other causes of death. 36It is previously shown that traditional time-to-event analysis may overestimate the risk of CVD because it handles death as censoring instead of a competing risk for the outcome of interest. 473]48 However, because higher physical activity levels associate with lower risk of CVD mortality in all world regions, 49 and physical activity improves CRF, 50 influence of CRF is likely of similar biological effect, even in different regions or ethnicities.For example, higher CRF is found to be inversely associated with MI risk in Trinidadian males. 51Nevertheless, research using eCRF in low-income and middle-income countries is still warranted to confirm and test the implementation and feasibility of eCRF in these regions.
Higher age is inversely associated with CRF. 52Because age is also the greatest component of the eCRF formulas used, 31 this may limit the possibility to examine an effect of CRF independent of age. 40When examining age-specific eCRF quintiles, the association magnitudes were attenuated at the highest ends of eCRF but mostly revealed similar findings as the spline modeling (ie, our main analyses).However, grouping continuous data leads to loss of information and statistical power. 24Nevertheless, it is well known and should be acknowledged that CRF declines with increasing age. 53Thus, it is still likely that increasing physical activity levels 50 or engaging in structured exercise 54 of sufficient volume and intensity will maintain or improve CRF at all ages, which potentially can aid in preventing MI and lower CHD burden and mortality.
Physical activity was self-reported, which is influenced by information bias. 46Thus, we likely misclassified some individuals as active and inactive when performing separate analysis of those meeting and not meeting current physical activity guidelines. 55This may also influence eCRF as mentioned earlier.Indeed, the CRF formulas are less precise for outliers (the most-fit and least-fit individuals) when compared against an exercise test to exhaustion using indirect calorimetry. 31This is likely a result of a generally nonlinear association between eCRF and directly measured CRF. 56evertheless, despite some misclassification, the simplicity and reported utility 26 of eCRF make a case for including this measure in routine medical care. 25

CONCLUSION
In this prospective cohort study, a higher eCRF was associated with a lower MI risk in both females and males and among those with other CVD risk factors, but associations were more pronounced among females than that in males.These findings suggest eCRF is a vital estimate to implement in routine medical care to identify individuals at high risk of future MI, especially for females.
POTENTIAL COMPETING INTERESTS M.L.L. has received lecture fees from Bayer, Sanofi, and BMS/Pfizer not related to this study.The remaining authors declare no conflict of interest.

FrequencyFIGURE 1 .
FIGURE 1. Restricted cubic spline Fine and Grey regressions of eCRF and risk of MI among females and males: the Tromsø Study 1994-2014.Data are shown as subdistributed hazard ratio (line) with 95% CIs (shaded area), adjusted for education, smoking, diet, alcohol, and study survey; age, waist circumference (ie, adiposity), and physical activity are adjusted by inclusion in the eCRF formulas.Reference of the spline is set to the 10th percentile of the distribution of the group (ie, among females and males separately), and values are shown between the first and 99th percentile of the distribution of estimated cardiorespiratory fitness.Frequency refers to the frequency of observed eCRF.eCRF, estimated cardiorespiratory fitness; MI, myocardial infarction.

FIGURE 2 .
FIGURE 2. Restricted cubic spline Fine and Grey regressions of eCRF and risk of MI among females and males with (A) hyperlipidemia (n¼2965, MI¼400); (B) hypertension (n¼8290, MI¼834); (C) central obesity (n¼5784, MI¼379); and (D) metabolic syndrome (n¼1410, MI¼181): the Tromsø Study 1994-2014.Data are shown as subdistributed hazard ratio (line) with 95% CIs (shaded area), adjusted for education, smoking, diet, alcohol, and study survey; age, waist circumference (ie, adiposity), and physical activity are adjusted by inclusion in the eCRF formulas.Reference of the spline is set to the 10th percentile of the distribution of the group (ie, among females and males separately), and values are shown between the first and 99th percentile of the distribution of estimated cardiorespiratory fitness.Frequency refers to the frequency of observed eCRF.eCRF, estimated cardiorespiratory fitness; MI, myocardial infarction.

Frequency 40 BFIGURE 3 .
FIGURE 3. Restricted cubic spline Fine and Grey regressions of eCRF and risk of MI among females and males who (A) are smokers (n¼3823, MI¼372); (B) are physically inactive (n¼6255, MI¼443); and (C) do not meet any nutritional guideline (n¼3463, MI¼382): the Tromsø Study 1994-2014.Data are shown as subdistributed hazard ratio (line) with 95% CIs (shaded area), adjusted for education, smoking, diet, alcohol, and study survey; age, waist circumference (ie, adiposity), and physical activity are adjusted by inclusion in the eCRF formulas.Reference of the spline is set to the 10th percentile of the distribution of the group (ie, among females and males separately), and values are shown between the first and 99th percentile of the distribution of estimated cardiorespiratory fitness.Frequency refers to the frequency of observed eCRF.eCRF, estimated cardiorespiratory fitness; MI, myocardial infarction.

TABLE . Descriptive
Characteristics of the Participants: The Tromsø Study 1994-2014

TABLE . Continued
a MI¼293): the Tromsø Study 1994-2014.The CVD risk factors included are smoking, hypertension, hyperlipidemia, obesity, poor diet, and physical inactivity.Data are shown as subdistributed hazard ratio (line) with 95% CIs (shaded area), adjusted for education, smoking, diet, alcohol, and study survey; age, waist circumference (ie, adiposity), and physical activity are adjusted by inclusion in the eCRF formulas.Reference of the spline is set to the 10th percentile of the distribution of the group (ie, among females and males separately), and values are shown between the first and 99th percentile of the distribution of eCRF.Frequency refers to the frequency of observed eCRF.CVD, cardiovascular disease; eCRF, estimated cardiorespiratory fitness; MI, myocardial infarction.