Impact of season on the association between vitamin D levels at diagnosis and one-year remission in early Rheumatoid Arthritis

The study evaluates associations between serum vitamin D metabolites at diagnosis and one-year remission, in early diagnosed rheumatoid arthritis(RA). The CIMESTRA-cohort comprised 160 newly diagnosed RA patients, treated aiming at remission. Vitamin D supplementation was recommended according to national guidelines. Dtotal(25OHD2 + 25OHD3) was dichotomized at 50 nmol/L, 1,25(OH)2D was categorized in tertiles. Primary outcome was remission(DAS28-CRP ≤ 2.6) after one year. Associations were evaluated using logistic regression, further adjusted for pre-specified potential confounders: Age, sex, symptom-duration before diagnosis, DAS28-CRP and season of diagnosis. Results are presented as Odds Ratios(OR) with 95% Confidence Intervals(95%CIs). In univariate analyses, neither Dtotal nor 1,25(OH)2D were associated with remission. In adjusted analyses, low Dtotal was associated with higher odds for remission; OR 2.6, 95%CI (1.1; 5.9) p = 0.03, with season impacting results the most. One-year remission was lower in patients with diagnosis established at winter. In conclusion, low Dtotal at diagnosis was associated with increased probability of achieving one-year remission in early RA when adjusting for covariates. Diagnosis in winter was associated with lower odds for one-year remission. Results suggest that season act as a contextual factor potentially confounding associations between vitamin D and RA disease-course. The finding of low Dtotal being associated with higher one-year remission remains speculative.


Results of univariate logistic regression.
Diagnosis established at winter compared to summer (OR 0.4, 95% CI (0.2; 0.9) p = 0.02), Number of Tender Joints (NTJ) (OR 0.9; 95% CI (0.88; 0.98) p = 0.01) and DAS28-CRP (OR 0.7, 95% CI (0.5; 0.9) p = 0.01) at time of diagnosis were significantly inversely associated with one-year remission, independently of vitamin D metabolite levels. This indicates that it is less likely that patients diagnosed during the winter season and in patients with high NTJ or DAS28-CRP at time of diagnosis, will achieve remission. [See appendix Table 1].
Primary analysis, evaluating D total groups. At year one, 96 (61%) of the 143 patients were in remission, 45 (71%) from the group with low D total at diagnosis, and 51 (64%) from the group with normal D total at diagnosis. (See Table 2).
No significant association was found between baseline D total group and remission rates in the crude model; OR 1.4, 95% CI (0.7; 2.7) p = 0.33, whereas the fully adjusted model showed statistical significantly increased odds for one year remission; OR 2.6, 95% CI (1.1; 5.9) p = 0.03, in patients with low D total at time of diagnosis compared to patients with normal D total . (See Table 2) For details on the definition of crude and adjusted models, see the method section. Propensity score analysis including several other factors, potentially affecting the results, did not significantly alters the estimate; OR 2.5, 95% CI (1.1; 5.8) p = 0.03, indicating that the further included factors have only negligible influence. In the fully adjusted model, remission was further inversely associated with diagnosis established in winter; OR 0.3 95% CI (0.1; 0.7) p < 0.01) and DAS28-CRP at diagnosis; OR 0.6 95% CI (0.4; 0.9) p = 0.01 and directly associated with age at diagnosis; OR 1.04, 95% CI (1.0; 1.1) p = 0.03. Stepwise inclusion of the pre-specified variables sex, age, symptom duration, DAS28-CRP at diagnosis and season at diagnosis indicates that especially adjustment for season impacts the estimates in the fully adjusted model. (See Table 3). There was no significant interaction between D total levels and season of diagnosis (p = 0.7).
Model fit was tested to evaluate which model best describes the association between vitamin D metabolites, one year remission and the chosen, potential confounding variables: A likelihood-ratio test showed that fully adjusting the model improved the model fit significantly, whereas the partially adjusted model (including only sex and age at inclusion) did not improved the model fit compared to the crude model. In addition, Akaike Information criterion (AIC) decreased and Bayesian Information Criterion (BIC) increased when including all confounding variables. As we put emphasis on AIC, this is interpreted as the fully adjusted model being the model which most adequately describes the real world, resulting in the best balance between number of included variables and precision.
To estimate whether missing data for one-year remission affects the results, best-and worse case-scenarios were evaluated. These suggested potentially consequences of missing data, as estimates became insignificant in best-case scenarios. (See Table 2). Secondary analysis, evaluating 1,25(OH) 2 D tertiles. Associations between the active vitamin D metabolite and one-year remission were evaluated similar to the primary analyses for D total . No difference existed in odds of remission between patients with the lowest 1,25(OH) 2 D tertile compared to highest, neither in the univariate model; OR 1.3, 95% CI (0.5; 3.2) p = 0.59, nor in the fully adjusted model; OR 1.8, 95% CI (0.6; 4.9) p = 0.28.
Remission was negatively associated with diagnosis established at winter; OR 0.4 95% CI (0.2; 0.9) p = 0.03, and DAS28-CRP at diagnosis; OR 0.6 95% CI (0.4; 0.9) p = 0.01. [See appendix Table 2].Applying likelihood-ratio-test, both the partially and fully adjusted model showed significantly better fit than the crude model. AIC decreased with each successive adjustment, while BIC decreased from crude to fully adjusted model.

Discussion
Main finding is that the inverse association between serum D total status at diagnosis and one year remission exists only in the model adjusted for season. Although largely unexplained, being diagnosed with RA in the winter is independently associated with lower odds for remission after one year of intensive treatment, suggesting that season confounds the association between vitamin D and RA disease activity.
The immunomodulatory properties of vitamin D are exerted through local conversion of 25OHD into 1,25(OH) 2 D 1,5 . Subjects with low serum D total lack substrate for the local production of 1,25(OH) 2 D 16,17 , increasing the risk of an auto-aggressive Th1/Th17 response 18 . We surmise that our results support the potential existence of a "low vitamin D RA-subtype", which upon anti-inflammatory treatment and relevant vitamin D supplementation is easily treated to remission. On the contrary, subjects with normal D total levels and thus sufficient substrate for local 1,25(OH) 2 D production, who develop RA may be representative of an immune system being more imbalanced and distressed, and thus less susceptible to achieve remission, despite relevant treatment. Alternatively, our results indicate the presence of a "Winter-diagnosed RA-subtype". Patients diagnosed at winter will inevitably have one-year remission evaluated the following winter, where the seasonal impact on disease activity will decrease probability of remission. Though, our study-design does not allow us to speculate whether "winter-diagnosed RA" is less prone to achieve remission in the summer, as we have no half-yearly evaluations.
Our results conflict with other studies evaluating vitamin D and remission in early RA: As opposed to the current study, the comparable studies did not adjusted for season in the statistical analyses, but handled potential annual impact through study design: One study is located in Columbia, a tropical country with no seasons 15 , while the Italian study 14 only evaluated serum 25OHD in the autumn-winter-period, to minimize the seasonal Low D total (<50 nmol/L) N = 67 (42%) Normal D total (≥50 nmol/L) N = 91 (58%) impact. However, RA is diagnosed all year round, and circannual changes in light and darkness is unavoidable in Scandinavia. We find the current study to be representative of a Scandinavian, early diagnosed RA-cohort, and further sheds light on an important contextual factor in evaluation of disease activity in early RA. The impact of season on vitamin D levels is well-known 3,4 Similarly, season impacts the incidence and course of RA [19][20][21] , likely owing to circadian rhythms affecting pro-inflammatory cytokines 22,23 . This may be explained by effects of seasonal changes in vitamin D affecting the immune system 24 , perhaps through seasonal impact on relevant genes, independent of vitamin D metabolism 25,26 .
Our study design implies that one year remission is evaluated the same season as diagnosis was established, and the impact of season on disease activity at diagnosis will be mirrored in the remission-criteria applied one year after. Thus, the probability to observe one-year remission will be lower in patients diagnosed in winter. As D total levels are lower in winter, we suggest this being a plausible explanation to our main finding: season is a strong confounder in the relationship between vitamin D and disease activity in RA. As season independently predicted one-year remission, we further suggest our results to be indicative of season of diagnosis being an important contextual factor in evaluation of RA disease-course. We advocate future research to take season into account when evaluating disease activity in early RA.
The primary outcome; remission, is evaluated as DAS28; a composite of subjective and objective disease-activity markers weighting pain and inflammation. Pain may persist in DAS28-remission 27 , whereas 1,25(OH) 2 D supplementation is reported to significantly relieve pain in early RA 28 . As the CIMESTRA strategy included increased awareness of vitamin D intake, we assume that an eventual painful proximal myopathy owing to low vitamin D was improved during follow-up. In RA, 25OHD levels may be correlated to dietary intake 29 , while improvement of low serum vitamin D upon supplementation in RA is associated with improved treatment response 30 . Moreover, RA patients with low 25OHD and no vitamin D supplementation are less frequently in remission, compared to patients receiving relevant supplementation 31 . Likewise, abstaining from correction of low vitamin D levels is associated with increased recurrence-rates in some 32 , but not all studies 33 . However, the current study is not an interventional study investigating effects of vitamin D supplementation on remission. Patients were supplemented after individual physicians´ evaluation of vitamin D and calcium intake, with no knowledge of actual serum D total level. Suggesting that sufficient vitamin D intake, concomitant to aggressive anti-inflammatory treatment, might partly explain the results, is only hypothesis-generating, despite plausible underlying biological associations supporting the hypotheses 1 . Though, data concerning which patients received supplementation is not available.
If a causal association between low vitamin D levels and high RA disease activity exists, it seems reasonable that patients with more active disease shows better treatment response upon the aggressive treatment-regimen and relevant vitamin D intake, thus explaining the higher remission-rate in the group with lower D total . Furthermore, if a "low vitamin D RA subtype" exists, and low vitamin D levels are associated with higher disease activity and pain 11 , such patients may receive higher doses of anti-inflammatory medication, and be more likely to achieve remission, partly explaining the results.
Limitations of the current study include restrictions owing to the use of post-hoc analyses in a parental RCT, designed for other purposes: The study population is rather small, supported by several estimates having large 95% CIs. However, to the best of knowledge, our study almost doubles observations upon which estimates can be made, compared to other studies evaluating vitamin D and remission in early RA 14,15 . The parental CIMESTRA www.nature.com/scientificreports www.nature.com/scientificreports/ RCT investigated the additive effect of ciclosporine or placebo to an aggressive, treat-to-target strategy, and 50% received ciclosporine. Vitamin D metabolites evaluated in the current study were obtained prior to treatment, and were not affected by ciclosporine treatment. No adjustment for ciclosporine is done, as no significant association was found between ciclosporine and one year remission 34 .
Possible unmeasured residual-confounders must be considered: Relevant markers of calcium metabolism, such as calcium, phosphorous and parathyroid hormone are not evaluated. Neither, information concerning sun-exposure, the main-cause to cutaneous vitamin D production, is assessed. Besides, vitamin D levels have been reported to vary according to meat consumption [35][36][37] . However, we are unable to account for meat-consumption in the current study.
Another potential limitation is the dichotomization of D total, applied in the current study: Cutpoint defining sufficient vitamin D levels varies greatly, depending on e.g. age, race, weight, sun-exposure and diet 3 . The internationally accepted cut-point defining low vitamin D is 50 nmol/l 38 . We acknowledge that dichotomization might lead to loss of more subtle details, and that it has been argued that higher levels of vitamin D might be necessary for better health outcomes 39 . However, we find that the use of a well-known cut-point, and the comparability to the few similar studies increase the external validity, interpretation and clinical use of our results.
To avoid overfitting our statistical models, we evaluated model fits by AIC and concluded that the fully adjusted model was the most realistic model. As different modelling scenarios resulted in inconsistency among results, we advocate for reproduction of our results before firm conclusions are made We believe that our data can be applied to a Scandinavian population with newly diagnosed RA. Despite data collection at the millennium, the aggressive treat-to-target strategy is still the recommendations nowadays, increasing the external validity. Our results need to be reproduced, preferably in sufficiently powered studies with Low D total (<50 nmol/L) Normal D total (≥50 nmol/L) Contrast between groups  Table 2. Achieving remission according to D total at diagnosis. a Crude logistic regression; D total group as independent variable, no other covariates. b Simple logistic regression, further containing age at diagnosis and sex. c Multiple logistic regression, further adjusted for symptom-duration prior to diagnosis, DAS28-CRP at diagnosis and diagnosis established in winter (November to April). For all comparisons, the group having normal D total (i.e defined as "not low") is the reference. Abbreviations: D total : The sum of 25OHD 2 and vitamin D supplementation guided by serum vitamin D measurements, alongside to aggressive treatment and adjustment for season.
In conclusion, low vitamin D total at time of diagnosis in early, treatment-naïve RA is associated with increased odds of achieving remission after one year, when adjusting for relevant confounding variables. Conceivably, this finding is mainly attributable to seasonal variation per se with low vitamin D levels as a potential amplifier. In order to elucidate the relationship between vitamin D status, disease activity and season of the year, follow-up visits at 6 months interval may be useful. In the meantime, clinicians and researchers should be aware of the potential impact by season on RA disease presentation.

Methods patients and outcomes. One-hundred-and-sixty patients were recruited from five Danish University
Clinics from October 1999 to October 2002 in the CIMESTRA-study. Eligibility criteria and treatment-strategy are described elsewhere 34 . One-hundred-and-fifty-eight patients had D total (the sum of 25OHD 2 and 25OHD 3 ) and 1,25(OH) 2 D) measured prior to inclusion and treatment. One-Hundred-and-forty-three patients had DAS28-CRP calculated after one year, and were included in the current study. Patients were allocated to D total group at time of diagnosis, dichotomized as low: D total < 50 nmol/L or normal: D total ≥ 50 nmol/L. Patients who reported insufficient vitamin D and calcium intake at time of diagnosis received daily oral supplements of vitamin D 3 (800 IE/10 mcg) and calcium (1500 mg), according to national guidelines. Treating physician decided eventual need for supplementation unaware of actual serum vitamin D levels, as serum levels were evaluated from biobank for this current study. Measurement of the vitamin D metabolites. 25OHD 2 and 25OHD 3 in serum were analyzed with isotope dilution liquid chromatography-mass spectrometry (LC-MS/MS) using calibrators traceable to international standard reference material NIST SRM 972 40 . Mean coefficients of variation (CVs) of 25OHD 3 were 9.6% and 8.1% at 25 nmol/L and 48 nmol/L, respectively, and for 25OHD 2 , 8.5% and 8.0% at levels of 23 and 64 nmol/L, respectively 41 .
Statistical methods. Differences between D total groups for dichotomous variables at time of diagnosis were evaluated by Chi 2 test. Wilcoxon-Rank-Sum test evaluated differences in continuously variables. Underlying assumptions were evaluated through study-design.
Univariate, logistic regression was performed for D total , comparing levels <50 nmol/L to levels ≥50 nmol/L, in predicting DAS28-CRP ≤ 2.6 after one year. This is defined as the "crude model". In the model defined as the "partially adjusted model", sex and age at diagnosis were included, whereas symptom-duration prior to diagnosis, DAS28-CRP and diagnosis established at winter were further included in the fully adjusted model. These variables were chosen prior to analysis because of their previously reported ability to predict clinical remission 43 . or close physiological relation to vitamin D 4 . Despite patients originally randomized 1:1 to ciclosporine, it was decided not to adjust for ciclosporine, as no association with one-year remission was found for ciclosporine in the original CIMESTRA study 34 .
Further aiming at evaluating the delicate interplay between variables included in the model, a step-by-step approach was used, including sex, age, symptom duration, DAS28-CRP at diagnosis and season of diagnosis in a logistic regression one by one.  Table 3.
Stepwise regression, including a priori chosen variables one by one. Abbreviations: D total : The sum of 25OHD 2 and 25OHD 3 . DAS28-CRP Disease Activity Score, based on 28 joint count and CRP.
The model was subsequently evaluated for interaction between vitamin D and season of diagnosis, as a close relationship exists between season and circulating D total . Likelihood ratios tests were used to evaluate model fit. AIC and BIC assesses the model fit when taking number of co-variates included into account, and thus estimates the risk of over-fitting. Emphasis is put on AIC, while BIC is also presented.
To test the robustness of the model, propensity score analysis was applied to evaluate the potential biasing impact of other variables on the outcome. The propensity score analysis included BMI, ACPA, IgM-RF, NSJ, NTJ, VAS global-patient , HAQ and CRP at time of diagnosis to determine, if confounding by any of these factors was biasing the results.
The potential impact of the missing remission-data of the 15 patients not included in year one was evaluated using best-and worst case scenarios, coding all missing patients as being either in remission ("Best case") or not in remission ("Worst case").
Concerning 1,25(OH) 2 D, similar analyses were performed as for D total , comparing lowest and middle tertile to highest tertile.
Significance is defined as the 5% level. Stata IC 15 44 is used for statistical analyses.

ethics and consent
The study complies with the Oviedo and Helsinki declarations. The protocol was approved by Ethics Committee Region Midt (M-1959-98) and registered at www.clinicaltrials.gov (NCT00209859). All patients received verbal and written information prior to enrolment, and gave written consent at baseline.