Study design

This prospective cohort study is based on the Danish CIMESTRA trial.62 The study design was initially a 2-year randomised controlled trial and thereafter followed in an open design for 14–17 years with a continuous treat-to-target strategy, aiming at remission63 for the whole period.

All data were prospectively registered: RA disease activity parameters in the DANBIO registry and the cardiovascular events in the patients’ records. The current study allocates the patients in two groups, according to baseline D_total levels. This protocol conforms to the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) statement64 because the study, with allocation of the patients in two groups according to baseline D_total levels and subsequent follow-up, will report the probability of having a cardiovascular event in the group with low levels of D_total, compared with the group with normal levels. The subsequent reporting of the cohort study will conform to the ‘STrengthening the Reporting of OBservational studies in Epidemiology’ (STROBE) statement.65 The risk of bias is assessed using the ROBINS-I guidelines.66

Participants

One hundred and sixty treatment-naive patients with an early diagnosis of RA were originally recruited from five Danish University clinics (trial centres) from October 1999 to completion October 2002. Mean age at inclusion was 53 years (range 20.4–75.2). Further baseline characteristics are described elsewhere.62 Inclusion criteria: fulfilling American College of Rheumatology 1987 criteria for RA,67 disease duration < 6 months, two or more swollen joints and age between 18 and 75 years. Exclusion criteria: glucocorticoid treatment 4 weeks before inclusion, previous use of disease-modifying antirheumatic drugs, malignancy, diastolic blood pressure > 90 mm Hg, elevated serum creatinine, infections with parvovirus B19, hepatitis B, C and HIV, and any condition contraindicating the study medication. The first 2 years of the CIMESTRA trial consisted of two treatment arms: (1) methotrexate (MTX) + ciclosporin A (CsA) (initial combination therapy group) and (2) MTX + placebo-CsA (initial monotherapy group). Intra-articular corticosteroids for suppression of joint synovitis were allowed in both arms of the trial from week 76 to week 104; CsA/placebo-CsA were tapered to zero while MTX was continued. From week 68, hydroxychloroquine 200 mg per day was added to both arms. Oral prednisolone was not allowed during the first 2 years. From year 3 to 5, intra-articular glucocorticoid injections and MTX 20 mg per week were continued. If activity persisted, the patients were switched to triple therapy (MTX, sulfasalazine and hydroxychloroquine) and the next possible step-up was MTX + tumor necrosis factor α inhibitor. For further details, see the original article.62 At 5 years, 50% of the patients were treated with MTX or MTX and hydroxychloroquine, 16% received triple therapy, while 17% were receiving biological treatment. Seventeen per cent had withdrawn from the treatment because of ≥12 months’ American College of Rheumatology remission.68 Before final analysis of this study, treatment at year 10 will also be known.

All patients were recommended daily supplementation with folic acid 1 mg per week and daily oral intake of vitamin D₃ (400 IU/10 mg) and calcium (500 mg). As part of the Danish National Recommendations for osteoporosis prophylactic all patients with RA were reminded of the supplementation when seen by the physician during follow-up visits.

Exposure and control group

Baseline vitamin D_total will be dichotomised at 50 nmol/L, based on the international cut-off value.69 The group with vitamin D_total < 50 nmol/L will be evaluated as primary ‘exposure’ and the group of patients with vitamin D_total ≥ 50 nmol/L will be evaluated as having a normal level of vitamin D.

Outcomes and variables

Cardiovascular events, defined as hospitalisation or death due to acute or elective cardiovascular diseases, are diagnosed by the patients' medical records. Likewise, non-cardiovascular reasons for hospitalisations and deaths will be evaluated for future use. The use of patients records in evaluating cardiovascular events were validated by Gaede et al.70–72 The working group members (MH and TE) will not be blinded to the patients' baseline vitamin D status. We assume that non-blinding of the two evaluators to the patients' medical records is unlikely to affect the adjudication of the objective outcome ‘any cardiovascular event’.

The follow-up time will be calculated in months from date of entry (from October 1999—October 2002) and to time of first cardiovascular event, to death owing to cardiovascular or non-cardiovascular reasons, to withdrawal of informed consent or to 10 October 2016, whichever comes first. The result of the classification will be entered in REDCap (project-redcap.org), provided by OPEN (Odense Patient data Explorative Network) via an encrypted connection, fulfilling the demands for data security by logging all data entries and changes.

Evaluation of the primary outcome: any cardiovascular event

The primary outcome is the occurrence of a first hospitalisation due to myocardial ischemia, heart failure, apoplexia cerebri, arrhythmia, procedure-related cardiovascular event, other cardiovascular reasons and supposed cardiovascular reasons, or death from cardiovascular reasons. Patients experiencing any of these conditions will be defined as ‘having a cardiovascular event’, and thereby experiencing the primary outcome. See figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Classification of events in the CIMESTRA cohort.

Adjudication of events

Hospitalisation will be subclassified as follows:

1. Hospitalisation due to acute reasons

2. Hospitalisation due to elective reasons

An acute hospitalisation is defined as an unscheduled hospitalisation, and is further defined as a minimum one overnight stay at the hospital. The hospitalisation will be dated as the date when admitted to the first overnight stay in the hospital.

Acute hospitalisation will be defined as non-cardiovascular or cardiovascular, based on the total hospitalisation, and will further be subclassified as follows:

Non-cardiovascular hospitalisation

• Non-cardiovascular hospitalisation will be subdivided into cancer, infection, respiratory disease, trauma, suicide and other causes of non-cardiovascular hospitalisation.

Cardiovascular hospitalisation

1. Myocardial ischemia:

   1. Non-fatal or fatal myocardial infarction, defined by national and international guidelines.73

   2. Fatal myocardial infarction is defined as a primary fatal event within 7 days, documented by post mortem autopsy, or by the definition of myocardial infarction according to European guidelines.73

   3. Death of myocardial infarction as a consequence of medical examination/procedure/surgery will be classified as procedure-related death.

   4. Acute coronary syndrome includes acute ischaemic symptoms with eventual elevation of biomarkers or electrocardiographic changes which does not fulfil the criteria of acute myocardial infarction.

   5. Angina pectoris.

   6. Revascularisation procedures (percutaneous coronary intervention or coronary artery bypass graft).

2. Heart failure:

   1. Patients with non-elective hospitalisation or death, minimum one overnight stay, with symptoms or findings of heart failure.

   2. Death due to heart failure is defined as escalating heart failure symptoms before death.

3. Arrhythmias:

   1. Atrial fibrillation or flutter, supraventricular tachycardia and others.

   2. Ventricular tachycardia, ventricular fibrillation and others.

   3. Death due to arrhythmia requires documentation—for example, telemetric transcript, pacemaker output or electrocardiogram.

4. Stroke:

   1. Cerebral haemorrhage, cerebral thromboembolism, transitory cerebral ischaemia and others.

   2. Stroke is defined as abrupt severe neurological deficits, eventually with CT documentation. Death within 14 days after symptom-onset of stroke, and without other obvious reasons, is classified as caused by stroke.

   3. Death from stroke as a consequence of medical examination/procedure/surgery, and without other obviously reasons, is classified as caused by stroke.

5. Procedure-related event:

   1. Any cardiovascular event within 24 hours after cardiovascular medical examination/procedure/surgery.

6. Other cardiovascular hospitalisations:

   1. Hospitalisation caused by other cardiovascular events —for example, pulmonary embolism, rupture of aortic aneurism, etc.

7. Supposed cardiovascular hospitalisation:

   1. Hospitalisations without any documented non-cardiovascular cause.

   2. All deaths which are not defined by the cardiovascular reasons mentioned above, and which are not caused by well-documented non-cardiovascular death.

   3. All deaths without known reason.

Elective hospitalisation will be classified as either non-cardiovascular (subclassified as cancer, infection, respiratory diseases, trauma, suicide and other for future use) or cardiovascular, which is further subclassified as follows:

1. Myocardial ischemia

2. Arrhythmia

3. Heart failure

4. Others.

The cause of death will be defined as cardiovascular, if a non-cardiovascular cause of death is not identified. Cardiovascular death will be further classified by time, as either sudden or non-sudden death. Sudden cardiovascular death is described as either witnessed, if death is witnessed and abrupt within 1 hour after symptom-onset, or non-witnessed with no obvious non-cardiovascular reasons (found dead) The remaining cardiovascular deaths will be classified as non-sudden.

Endpoints and time

All endpoints will be classified according to the definitions mentioned above. Therefore the patients might have more than one event, and even during one single hospitalisation, the patient might have more than one event, but for the primary analyses they will only count as ‘having an event’, classified as either cardiovascular or non-cardiovascular.

Other measurements

Baseline disease activity was evaluated by the Disease Activity Score calculated with C reactive protein on 28 joints. Disability status was self-reported according to the Danish validated version of the Stanford Health Assessment Questionnaire.74 Patient-estimated global health was assessed on a Visual Analogue Scale 0–100 mm.75 The number of swollen joints and number of tender joints were evaluated by the treating rheumatologist. Systolic and diastolic blood pressure were obtained on the right arm with the patient in a sitting position after 5 min rest, using the mean value of two repetitive measurements with 1–2 min between them.

Baseline body weight and height were measured, and BMI calculated as BMI=weight/height squared (kg/m²) was categorised as normal (BMI ≤ 25), overweight (BMI >25 < 30) and obese (BMI ≥ 30). Blood samples at baseline are evaluated for 25OHD₂, 25OHD₃, 1,25(OH)₂D, C reactive protein (CRP), anti-citrullinated peptide antibodies (ACPA) and IgM rheumatoid factor (IgM RF). ACPA and IgM RF are dichotomised as positive versus negative, according to national reference values.62

Patients were asked about smoking at baseline, and smoking status will be dichotomised as ever or never.

Data collection, management and confidentiality

The baseline data are electronically stored in clinical databases approved by Danish Health Authorities. The patient consent form is stored in paper form and has been kept locked at the individual inclusion sites. The outcome adjudication will start when chief consultants at the participating clinics approve access to the patient′s journals. The data entry will be performed using the REDCap browser (project-redcap.org) using password-protected areas, thereby logging any admission as well as alterations by using personal log-in and password for any of the adjudicators.

For future use, the dataset concerning the cardiovascular outcome will be stored infinitely at the Danish Data Archive (https://www.sa.dk/en/services/danish-data-archive).

Statistical methods

Prespecification of the analysis

The primary outcome is the occurrence of a first cardiovascular event, including death from any cardiovascular cause; this will subsequently be interpreted as an eventual harm or benefit from low baseline D_total, compared with the group having normal levels of D_total at baseline.

For what we refer to as the crude model, a χ² test will be used to evaluate the cardiovascular risk in the two groups. Comparisons will be presented as risk ratios, and interpreted based on the absolute risk using risk differences. Results will be presented with 95% CIs.

The primary statistical models will be based on logistic regression analysis, with the model including the vitamin D group, original trial group allocation (CsA or placebo), and Trial Centre as fixed effects. The primary outcome will further be analysed using survival analysis, with event curves based on Kaplan–Meier analysis of the time to first event. The hazard ratio for the primary end point will be calculated using the Cox regression model. Differences between numbers of cardiovascular events in the two D_total groups will be analysed using the log-rank test.

Following simple logistic regression analysis, some of the variables collected at baseline will probably be statistically significantly associated with the CVD events. Thus we will use a multivariable logistic regression, enabling two or more independent variables in the model—that is, simultaneously adjusting for more than one factor that has an impact on the outcome. The relationship between the outcome and the potential confounding variables gender, age, smoking status, BMI ≥ 30, ACPA positivity at baseline, antihypertensive treatment initiated after inclusion and statin treatment initiated after inclusion will be explored in further sensitivity analyses, using multivariable logistic regression modelling. Those variables were chosen because of their well-known association with CVD.

Missing outcome data

The adjudication strategy using data from patient journals might introduce some inconsistencies in the final diagnosis, and thereby misclassification of the primary outcome. Though, we will aim to counteract this, by consulting the third member of the adjudication group in any cases of disagreement.

Missing values represent a potential source of bias, and therefore every effort will be undertaken to ensure sufficient follow-up. There are no missing D_total levels at baseline (Herly M, 2010, abstract only), but we expect that the collection of journal data for cardiovascular events or hospitalisation might be insufficient, mainly because participants may move to another region in Denmark. There might be sociodemographic differences in patients moving from one region to another, but we do not expect changes in these characteristics to be so different that they will markedly change the estimates. Moreover, Denmark is a small country, with a relatively homogeneous population, and equal access to all types of health services, further minimising the risk of regional differences in prevention and registration of health outcomes. We will collect follow-up data concerning hospitalisation in all patients moving to one of the other regions where we have access to patient records, to minimise the amount of data missing.

If outcome data are missing, we assume that they will be missing at random, and therefore, data missing before the patient has had an event, will not bias the estimates toward a stronger association, but solely underestimate the event rate, and thereby the risk of cardiovascular hospitalisation in the cohort. If data are missing after the first cardiovascular event, the primary outcome; first cardiovascular event, will be evaluated as specified in the primary analyses.

Data concerning events after patients are lost to follow-up can also be handled as ‘last observation carried forward’. For a patient moving to another region, the last observation will be that the patient is alive when lost to follow-up. This will possibly lead to an overestimation of the survival in the cohort, and an underestimation of the risk of hospitalisation, and ultimately, number of deaths, thereby diminishing the size of the final estimate towards ‘no association’.

To further evaluate the extent of missing data, and especially the assumption of data missing at random, we will quantify the number of patient-years in the two vitamin D groups.

When data are incomplete at the end point, we will perform sensitivity analyses based on an imputation of missing values, using two different approaches:

1. Imputation of ‘best’ versus ‘worst’ case scenarios by replacing missing values with ‘good’ outcomes (event=no) and ‘poor’ outcomes (event=yes).85

2. Multiple imputations with age, disease duration, gender, BMI and smoking at baseline entered as exposure variables.

Where complete cases and different imputation techniques give varying results, we will attempt to understand this, and report it with appropriate moderation of the conclusion made in the publication. Because of the risk of underestimating the event rate when using the last observation carried forward method, we will put key emphasis on the results of the best-versus-worst-case imputation.

Subgroups, interactions and covariates

Stratified analyses will be performed for the patients with baseline D_total <25 nmol/L. Statistical significance will be defined based on a two-sided significance level of 0.05. Thus all results will be expressed as estimates of the group differences, with 95% CIs to represent precision of the estimates. p Values will be reported to three decimals, whereas p values less than 0.001 will be reported as < 0.001.

STATA IC1486 will be used for statistical analysis.

Using propensity score analyses to assess the robustness of the conclusions

An important strength of observational studies like this is the ability to estimate the influence of vitamin D at baseline on cardiovascular events during long-time follow-up in these patients. This is a crucial strength, as we would be unable to use a better experimental design owing to ethical constraints (ie, it is not possible to randomise to low vitamin D). Keeping this in mind, one drawback of observational studies is that they lack the ability to randomise their participants into treatment groups— that is, there is a chance that we infer from selection bias and confounding by indication. One way to adjust for this is through the use of a propensity score analysis. For this analysis, an adjustment will be performed through various forms of propensity scoring (including stratification, matching and regression adjustment). Each form is separately conducted, reviewed and assessed for its effectiveness in improving the model.

Potential limitations

Instead of being part of the pathogenesis in several diseases, vitamin D might merely be a marker of poor lifestyle,87 ,88 eventually due to RA, or it may be an independent risk factor of CVD, thereby confounding our results.

The primary exposure; baseline D_total is not randomised; the two vitamin D groups are simply allocated at baseline, according to the individual D_total level at baseline. The use of vitamin D as a dichotomous variable is based on international cut-off values,69 to ease the clinical interpretation, though we are aware that we might lose more subtle differences when not assessing D_total as a continuous variable.

We will report and discuss eventual differences in the distribution of baseline variables between the two vitamin D groups, and attempt to evaluate to what extent, and in what direction this might affect the final estimates.

All patients in the cohort are treated with vitamin D and calcium supplementation from baseline, in accordance with the National Danish prevention strategy for osteoporosis in patients with RA. Therefore, a patient initially evaluated as having low vitamin D levels at baseline, will presumably achieve normal levels during the follow-up. If causality between low vitamin D levels and development of CVD exists, such patients might be ‘misclassified’, thereby leading an eventual association towards null. Though, we still find it relevant to evaluate if baseline vitamin D levels can predict long-term cardiovascular outcomes, despite ‘blurring’ the results by substitution. We recommend all rheumatologists to substitute their patients with RA with vitamin D and calcium, both because it is known that low vitamin D levels are common in these patients, and as prophylactic treatment for osteoporosis. However, we also acknowledge that the substitution strategy in our cohort might diminish the comparability with studies not using vitamin D and calcium substitution.

A profound association between D_total level and other cardiovascular risk factors may confound the association between baseline D_total and cardiovascular mortality. To date, associations between vitamin D levels and several cardiovascular risk factors are well established, as mentioned in the introduction, and many of those risk factors are also associated with low vitamin D levels.

In general, it seems that associations between vitamin D, RA and CVD are part of an extremely delicate interplay, challenging clarification of causality, and increasing the risk of confounding.

Despite our attempts to handle these confounders, the study lacks baseline data, such as markers of early atherosclerosis, body composition, insulin resistance and sedentary lifestyle, leading to potential residual confounding. Most of this major limitation is due to the study design, with use of secondary analyses from a prior randomised controlled trial, with no influence on the variables collected at the time of inclusion. Those limitations are regrettable, but we attempt to deal with them in the discussion section of the final paper. We accept those limitations, because the use of a cohort established at the millennium has the advantages of long-term follow-up, and therefore probably an ‘acceptable’ amount of cardiovascular events during this follow-up. Moreover, despite potential confounders, we still find that the study design is appropriate for evaluating if low vitamin D levels at baseline can predict long-term CVD in a cohort of patients with RA treated-to-target and receiving relevant anti-osteoporotic treatment according to National Guidelines.

Using a previously designed cohort provides the advantage of long-term follow-up, but also has some major limitations due to the study design—for example, the exclusion criteria; diastolic blood pressure > 90 mm Hg excludes some patients with a great risk of cardiovascular disease, and may therefore bias the results towards fewer incident cardiovascular events. Neither is there information about lipid status, and therefore we cannot adjust for this well-known confounder. An association between vitamin D and LDL and triglyceride46 ,89–91 is likely, as vitamin D is a marker of generally poor lifestyle. This may lead to an overestimation of the association between low baseline vitamin D levels and CVD.

Although this cohort was established at the millennium, the aggressive treatment, with both arms receiving increasing doses of MTX and intra-articular steroid injections for swollen joints in a step-up treat-to-target strategy, this strategy is still the ‘gold standard’ for treatment of RA in Denmark. Additionally, although the eligibility criteria led to selection of patients with RA with few comorbidities at baseline, the long-term follow-up leads to generalisability of the results.

Like the present cohort, many other RA cohorts are strictly selected to avoid comorbidity, and long-term follow-up studies in RA cohorts with several comorbidities are needed. We try to deal with this issue in the nearby future by using a Danish inception cohort with no exclusion criteria, to extensively evaluate cardiovascular comorbidity during follow-up.92

Defining an event as a hospitalisation, does not take outpatient contacts into account, thereby potentially under-rating the incidence of cardiovascular comorbidity. We assume that outpatient contacts are due to minor comorbidities, and if a more severe comorbidity evolves, a hospitalisation will ultimately occur, thereby showing the patient as having an event in the medical records.

The method of adjudicating events using patient′s medical records is evaluated and validated by Gaede.70–72 We assume this method classifies the events very precisely, also when compared with other methods, such as using the Danish National Patient Registry (Lands Patient Register), which has recently been shown to have general high validity.93 ,94 The two methods of classification in patients with RA could be compared, which would indicate the accuracy of the register-based outcome evaluation for CVD in patients with RA.

This study will be a cohort study, and causality underlying an eventual association on the effect of vitamin D in CVD in patients with RA needs to be evaluated in future studies, preferably in randomised controlled trials, measuring the effect on cardiovascular outcomes of varying sufficiently high doses of vitamin D supplements.