The present work was a sub-study of HuLARS (HUNT Longitudinal Ankylosing spondylitis and Rheumatoid Arthritis Study). In the HUNT study , all inhabitants ≥ 20 years old from the northern part of the Norwegian county of Trøndelag were invited. The HUNT study is an open cohort study and data, including results from questionnaires and blood samples from participants from HUNT2 (1995–1997) and HUNT3 (2006–2008), were used in the present observational study.
Power was calculated based on the following assumptions using data from previous HUNT publications [19, 20]: Approximately 33,000 persons participated in both HUNT2 and HUNT3 and the prevalence of RA was ~ 0.75%; we expected ~ 15% missing data for calculation of eCRF; the average 10-year decline in CRF in healthy people would be ~ 3.8 mL min−1 kg−1; we presumed a 20% larger decline in individuals with RA; and used alpha = 0.05 and a two-sided test. The calculated power was 82%, which was considered sufficient to perform the study.
Based on the information in hospital case files and using the standardized 2010 ACR/European League Against Rheumatism classification criteria for rheumatoid arthritis [20,21,22] or for some cases diagnosed before 2010 the 1987 American College of Rheumatology (ACR) classification criteria due to insufficient information , a previous study identified those with a valid RA diagnosis (n = 578) out of all participants in HUNT2 and HUNT3 who self-reported RA. We excluded those who received an RA diagnosis after HUNT3 (n = 32) and participants with ankylosing spondylitis, psoriasis arthritis, juvenile idiopathic arthritis, or other inflammatory arthritis. The remaining participants were included as controls. The primary aim was to investigate the change in eCRF from HUNT2 to HUNT3; thus, we only included controls and RA patients with valid eCRF in both HUNT2 and HUNT3 and with no missing adjustment variables in the regression analysis (188 RA patients and 26,202 controls) in this analysis (Fig. 1). For the secondary aims comparing eCRF levels in controls and RA patients, we included participants attending either HUNT2 only, HUNT3 only, or HUNT2 and HUNT3, resulting in a higher number of participants for these comparisons as detailed in Fig. 1. Method validation was performed in participant subsets as further described below.
Main outcome variable
eCRF (mL kg−1 min−1) was calculated using the previously published eCRF equations for healthy controls [10, 11], and the RA-specific equation for RA patients . Due to collinearity, variables present in eCRF equations cannot be used as explanatory variables for eCRF in a novel regression analysis. This problem was avoided for the primary aim in the present study because the outcome variable was the change in eCRF from HUNT2 to HUNT3.
The primary and secondary outcomes were compared among RA patients and controls as defined above.
Variables known from the literature to be associated with eCRF change and available in the HUNT surveys were used. The following variables and definitions were used: CVD (yes/no)—self-reported prior or present angina pectoris and/or myocardial infarction and/or stroke. Family CVD history (yes/no)—previous/present stroke and/or hypertension and/or myocardial infarction (MI) before age 60 years in a first-degree relative. Hypertension (yes/no)—blood pressure ≥ 140/90 mm Hg and/or self-reported use of anti-hypertensive medication. Hypertension and systolic blood pressure (SBP) are correlated, and only hypertension was used because those treated with anti-hypertensive medication might have normalized SBP despite a diagnosis of hypertension. Smoking (yes/no)—self-reported prior or present smoking. Asthma (yes/no)—self-reported prior or present asthma. Diabetes (yes/no)—self-reported diabetes and/or the use of anti-diabetic medication and/or having a non-fasting blood-glucose level > 11 mmol × L−1. Cancer (yes/no)—self-reported prior or present cancer. Pain (yes/no)—pain and/or stiffness that had lasted for ≥ 3 of the 12 latest months. Body mass index—weight/squared height (kg/m2). High-density lipoprotein (HDL) cholesterol measured in mmol/L.
PA strongly influences CRF [7, 8]. The American College of Sports Medicine and American Heart Association’s (ACSM/AHA) recommendations for aerobic PA are to perform either moderate-intensity physical activity ≥ 30 min on ≥ 5 days each week (≥ 150 min per week) or to perform vigorous-intensity aerobic activity ≥ 20 min ≥ 3 days a week (≥ 75 min per week). PA at these two intensities may also be combined [10, 23]. To describe the level of PA, the proportions of RA patients and controls fulfilling the ACSM/AHA recommendations for aerobic PA at HUNT2 (baseline) and HUNT3 were calculated from responses to questions about frequency, intensity and duration of weekly performed PA [10, 11, 23].
All participants in HUNT2 and HUNT3 provided written informed consent. The present study was approved by The Regional Committee for Medical and Health Research Ethics (4.2009.1068 and 2018/1149) and was performed in compliance with the Helsinki Declaration.
Data are given as counts or mean with percentages or standard deviation (SD) in parenthesis. p values < 0.05 were considered significant. Analyses were performed using STATA (Version 15.0, StataCorp, College Station, TX, USA).
To evaluate the decline in eCRF from HUNT2 to HUNT3 for the primary aim, regression models were performed in steps with different adjustments. In Step 1, we performed multiple linear regression with change in eCRF as the dependent variable and age (continuous), RA status (yes/no), and the interaction term for age and RA status as independent variables, which permitted investigation of whether eCRF reduction by time was different between RA patients and controls depending on age. Inclusion of age in the model ensured that differences in baseline age between RA patients and controls were adjusted for. We also included the following predefined adjustment variables: baseline eCRF, sex (male = 0 and female = 1), and time from participation in HUNT2 to participation in HUNT3 (years). Baseline eCRF, sex and age were included because the change in eCRF may depend on the starting level, and CRF varies with sex and age. Adjustment for time between the HUNT2 and HUNT3 was included because time varied from 10 to 12 years among individual participants.
The Step 1 model was then further modified to investigate other associations to the decline in eCRF from HUNT2 to HUNT3 (Step 2–4). Based upon literature, further baseline variables possibly relevant for the change in eCRF were considered as detailed above (CVD, family CVD history, hypertension, smoking, asthma, diabetes, cancer, pain, BMI, and HDL cholesterol). PA and RHR could not be included in the main analysis of change of eCRF because of collinearity with the dependent variable.
To reduce the risk of overfitting and promote reliable variable selection, the mentioned explanatory variables were first analyzed by Lasso (least absolute shrinkage and selection operator) regression (n = 1000 repetitions). Lasso identifies the smallest useful set of variables among variables that may be highly correlated, and gives irrelevant variables a coefficient of 0 . Variables with a coefficient different from 0 in the Lasso regression were, therefore, added to the Step 1 model to achieve the Step 2 model. The Step 2 model was then reduced to the final Step 3 model by removal of non-significant variables. In Step 4, the Step 3 model was standardized to compare the effect sizes of the predictors.
The models were compared using the R2 (i.e., the variation in the dependent variable explained by the independent variables), root mean square error (RMSE, i.e., standard error of the residuals, which tells how close the data lie around the line of best fit), Akaike information criterion (AIC) and Bayesian information criterion (BIC), where low numbers mean that the model better fits the data. Assumptions were evaluated using residual plots.
For the secondary aims, analysis was performed separately for HUNT2 and HUNT3 and each participant was included wherever she/he had participated (Fig. 1). Linear regression was used to find the mean sex-specific age-adjusted difference in eCRF between RA patients and controls. Mean eCRF of controls and RA patients aged 30–89 years were further compared with two-sample t tests in ten-year age categories for each sex separately.
As a sensitivity assay, we validated whether the eCRF calculation methods used in the study were comparable employing equivalence testing. With this method, the mean and 90% confidence interval (CI) of the difference between two methods, e.g., the calculated eCRF and measured CRF are evaluated against a predefined equivalence region . The equivalence region indicates how big the difference may be for the two methods still to be considered equivalent. As there is no generally accepted equivalence region for eCRF vs. measured CRF, we evaluated against an equivalence region of ± 1 Metabolic Equivalent (MET) (± 3.5 mL min−1 kg−1).
The eCRF equation for the general population was developed from a sub-study of HUNT3 (HUNT3 Fitness) [11, 19], which ensures that the eCRF equation for the general population fits the controls of our study. To evaluate whether the RA-specific eCRF equation would be adequate for the controls, an equivalence test was performed to compare the calculated eCRF by the RA-specific equation to the measured CRF from CPET in 3,294 of the controls in our study (women, n = 1754 and men, n = 1540), who had also participated in the HUNT3 Fitness study.
The equations for estimation of the RA-specific eCRF in HUNT2 and HUNT3 were slightly different due to the registered variables concerning PA in each survey. In a second equivalence test, we, therefore, compared these two RA equations in 189 RA patients where data for both methods were available. There are similar differences in the eCRF equations used in controls in HUNT2 and HUNT3. Thus, a third equivalence test of the general eCRF equations for HUNT2 and HUNT3 in 27,594 controls was also performed.