Association between serum vitamin D metabolite levels and disease activity in patients with early inflammatory polyarthritis

Authors


Abstract

Objective

Previous in vitro and animal studies have suggested that vitamin D, in particular, its metabolite 25-hydroxyvitamin D (25[OH]D), may have immunomodulatory effects. To study further the potential immunomodulatory effects of vitamin D in humans, we explored the hypothesis that serum vitamin D metabolites may be inversely associated with current disease activity, severity, and functional disability in patients with early inflammatory polyarthritis (IP).

Methods

We studied 206 consecutive patients with IP who were enrolled in the Norfolk Arthritis Register between January 2000 and November 2003 inclusive. Patients were studied within 6 months of symptom onset. None of the patients was taking steroids, and all had received <6 weeks of disease-modifying therapy. Associations between serum levels of 25(OH)D and 1,25-dihydroxyvitamin D (1,25[OH]2D) at baseline and the swollen and tender joint counts, Health Assessment Questionnaire (HAQ) scores, C-reactive protein (CRP) levels, and the Disease Activity Score 28-joint assessment (DAS28) scores at baseline and 1 year were assessed.

Results

The median age at symptom onset was 59 years (range 20–88 years), with a median disease duration of 4 months. At baseline, there was an inverse relationship between 25(OH)D levels and the tender joint count, DAS28 score, and HAQ score. The only inverse relationship with 1,25(OH)2D was with the HAQ score. Each 10-ng/ml increase in the level of 25(OH)D was associated with a decrease in the DAS28 score of 0.3 and in the CRP level of ∼25%. At 1 year, the only significant result was an inverse association between baseline vitamin D metabolite levels and the HAQ score; that is, those with higher metabolite levels had lower HAQ scores.

Conclusion

These data provide further support that vitamin D plays an immunomodulatory role in inflammatory arthritis. This association needs to be examined in other cohorts of patients with early IP, as well as in longitudinal studies. If confirmed, the clinical response to vitamin D supplementation should be examined in early IP.

Vitamin D is widely recognized as a hormone that is important for calcium homeostasis and maintenance of skeletal health. Vitamin D also plays a role in the function of the immune system (1–4). In vitro data have shown that 1,25-dihydroxyvitamin D (1,25[OH]2D) inhibits T cell proliferation and decreases the production of Th1 cytokines interleukin-2, interferon-γ, and tumor necrosis factor α (5). Further, in vivo studies suggest that 1,25(OH)2D supplementation prevents the initiation and progression of inflammatory arthritis (collagen-induced arthritis) in rodents and prevents experimental autoimmune encephalomyelitis (a murine model used to determine the efficacy of drugs for the treatment of multiple sclerosis) (6, 7).

Additional evidence of an immunomodulating role of vitamin D in rheumatoid arthritis (RA) includes the localization of vitamin D receptors to macrophages, chondrocytes, and synoviocytes in tissues, such as rheumatoid synovium, and at sites of cartilage erosion in humans with RA, but not in normal human tissues (8). Low baseline intake of vitamin D in the Iowa Women's Health Study was related to the subsequent development of RA (9). Interestingly, no relationship between vitamin D receptor genotypes and susceptibility to disease was found (10).

While these data suggest that vitamin D status may be important in the inception of disease, it is also logical to suggest that vitamin D may play a role in disease activity. We therefore tested in the present study the hypothesis that vitamin D metabolites are associated with levels of disease activity in an unselected cohort of patients with early inflammatory polyarthritis (IP).

PATIENTS AND METHODS

Patient population.

Study subjects were recruited from the Norfolk Arthritis Register (NOAR), a primary care–based incidence registry of patients with IP, the details of which have been described elsewhere (11). Briefly, all subjects residing within the NOAR catchment area who had been seen by a primary care physician and had synovitis affecting ≥2 peripheral joints that had lasted for ≥4 weeks were eligible for inclusion. Patients with definite inflammatory synovitis that was not subsequently explained by a diagnosis other than RA, psoriatic arthritis, postviral arthritis, or undifferentiated IP were eligible for followup.

The cohort analyzed in the present study were those who were enrolled in the NOAR between January 2000 and November 2003 inclusive, had been seen by the research nurse within 6 months of symptom onset, were corticosteroid naive, and were either disease-modifying antirheumatic drug (DMARD) naive or had taken DMARDs for fewer than 6 weeks. All consecutive patients with an available blood sample that was suitable for analysis were included.

Data collection.

A trained research nurse evaluated all patients at baseline, when an interview and clinical examination were performed. The assessment included swollen and tender joint counts (51 joints evaluated). The presence of rheumatoid nodules was also recorded. Blood was drawn for subsequent analyses, including estimation of rheumatoid factor (RF) levels. All subjects completed the British modification of the Stanford Health Assessment Questionnaire (HAQ) (12). Radiographs of the hands and feet were obtained at baseline and were scored by 2 readers for the presence and severity of erosions using the Larsen method, as described previously (13).

At 1 year, the patients were reassessed clinically. A further blood sample was not obtained. The American College of Rheumatology (ACR; formerly, the American Rheumatism Association) 1987 criteria for RA (14) were applied at baseline and at 1 year, based on cumulative acquisition of relevant features, as previously described (15).

Biochemical measurements.

Vitamin D metabolites were measured by in-house assays as described previously (16–18). Briefly, serum samples were extracted using acetonitrile and applied to C-18 Silica Sep-Pak columns (Waters, Elstree, UK). Metabolites were separated by straight-phase high-performance liquid chromatography (HPLC; Waters, Milford, MA) using a Hewlett-Packard Zorbax-Sil column (Hichrom, Reading, UK) and elution with hexane:propan-2-ol:methanol (92:4:4). Serum levels of 25-hydroxyvitamin D2 and 25-hydroxyvitamin D3 were measured separately by application to a second Zorbax-Sil column and elution with hexane:propan-2-ol (98:2). Levels were quantified by ultraviolet light absorbance at 265 nm and corrected for recovery. The results are expressed as total levels (in ng/ml) of 25-hydroxyvitamin D (25[OH]D). The sensitivity of the assay is 2 ng/ml. Intraassay and interassay coefficients of variation (CVs) were 3.0% and 4.2%, respectively (18). Following separation by HPLC, 1,25(OH)2D was quantified by an in-house radioimmunoassay as described in detail elsewhere (17). The adult reference range is 20–50 pg/ml, with a sensitivity of 1.25 pg per assay tube. Intraassay and interassay CVs were 7.8% and 14.5%, respectively.

Statistical analysis.

Cross-sectional (baseline) analysis.

Vitamin D metabolite levels were compared between patients who did and those who did not meet the ACR criteria for RA and between patients who were positive and those who were negative for RF, using Wilcoxon's rank sum test.

Associations between the 2 vitamin D metabolites and specific markers of disease activity, severity, and functional disability were assessed using regression models. Markers consisted of the number of swollen joints, number of tender joints, HAQ score (categorized into 3 severity groups: mild = <1, moderate = 1–2, and severe = >2), Larsen score (categorized into tertiles), and erosion status (erosion[s] present or absent based on a Larsen score of ≥2 in any single joint). In addition, we assessed the influence of the vitamin D metabolite levels on the CRP level (natural log–transformed; ln[x+1]) and the (CRP-derived) Disease Activity Score 28-joint assessment (DAS28) score (for calculation of the DAS28 using the CRP, see http://www.das-score.nl/www.das-score.nl/index.html). We used negative binomial models for the number of swollen/tender joints, linear models for the CRP level and the DAS28 score, ordered logit models for the HAQ and Larsen scores, and a logistic model for the erosion status. All models were adjusted for age at registration, sex, and the season during which the baseline blood sample was taken. The regression models were used to investigate changes in the selected disease marker for each 10-ng/ml increase in the 25(OH)D level or for each 10-pg/ml increase in the 1,25(OH)2D level.

Prospective analysis at 1 year.

The regression models, as outlined for the cross-sectional analysis, were also used to investigate the relationship between vitamin D metabolite levels at baseline and disease severity at 1 year. The associations between baseline vitamin D levels and the swollen/tender joint counts, HAQ score, Larsen score, and RA status (cumulative ACR criteria) at 1 year were examined in a similar way and adjusted for DMARD use (ever used by the time of assessment at 1 year).

All statistical analyses were conducted using Stata version 9 software (StataCorp, College Station, TX).

RESULTS

Characteristics of the study population.

A total of 206 IP patients were identified who fulfilled the inclusion criteria. These patients were comparable with other cohorts from the NOAR that have previously been reported. Demographic and disease characteristics of the IP patients at baseline and 1 year are shown in Table 1. The median age at symptom onset was 59 years, and just under two-thirds of the patients were women. The median duration of symptoms at presentation was 4 months (interquartile range [IQR] 2, 5). At the baseline assessment, 35.4% of patients were classified as having RA according to the ACR criteria, whereas at the 1-year assessment when the criteria were applied cumulatively, this had increased to 45.2%. However, by the 1-year assessment, indicators of disease activity, severity, and functional disability had shown improvement.

Table 1. Demographic and disease characteristics of the 206 IP patients at baseline and 1 year*
 IP patient cohort (n = 206)
No. of patients evaluatedNo. (%) of patients or median (IQR)
  • *

    IP = inflammatory polyarthritis; IQR = interquartile range; CRP = C-reactive protein; DAS28 = Disease Activity Score 28-joint assessment; HAQ = Health Assessment Questionnaire; RA = rheumatoid arthritis (met the American College of Rheumatology 1987 criteria); RF = rheumatoid factor; 95% CI = 95% confidence interval; DMARDs = disease-modifying antirheumatic drugs.

Baseline assessment
 Sex, no. (%) female206132 (64.1)
 Age at symptom onset, median (IQR)20659 (48, 73)
 Symptom duration at presentation, median (IQR) months2064 (2, 5)
 Swollen joint count, median (IQR) of 51 joints assessed2063 (0, 7)
 Tender joint count, median (IQR) of 51 joints assessed2062.5 (0, 9)
 CRP, median (IQR) mg/dl1797 (3, 24)
 DAS28 score, median (IQR)1793.28 (2.44, 4.25)
 HAQ score, median (IQR)2000.875 (0.25, 1.625)
 Larsen score, median (IQR)1354 (0, 12)
 No. (%) of patients with eroded joints13559 (43.7)
 No. (%) of patients with nodules2068 (3.9)
 No. (%) of patients with RA20673 (35.4)
 No. (%) of patients with an RF titer ≥1:4018146 (25.4)
 Body mass index, median (IQR)20425.6 (22.8, 28.9)
1-year assessment
 Swollen joint count, median (IQR) of 51 joints assessed1551 (0, 3)
 Tender joint count, median (IQR) of 51 joints assessed1551 (0, 4)
 HAQ score, median (IQR)1560.875 (0.25, 1.375)
 Change in HAQ score from baseline, mean (95% CI)153−0.013 (−0.114, 0.088)
 No. (%) of patients with eroded joints15582 (52.9)
 No. (%) of patients with RA20693 (45.2)
 No. (%) of patients with an RF titer ≥1:4018769 (36.9)
 No. (%) of patients treated with DMARDs by the first year206116 (56.3)

Findings of the cross-sectional (baseline) assessment.

Levels of vitamin D metabolites.

The mean levels of both vitamin D metabolites were lower in IP patients who satisfied the ACR criteria for RA at baseline than in those who did not. However, only the 1,25(OH)2D levels were statistically significantly different (P = 0.01) (Table 2).

Table 2. Serum vitamin D metabolite levels in the study patients with IP, by RA and RF status at baseline and 1 year*
 Serum vitamin D metabolite levelsP
25(OH)D, ng/ml (n = 183)1,25(OH)2D, pg/ml (n = 177)25(OH)D1,25(OH)2D
  • *

    Rheumatoid arthritis (RA) was diagnosed according to the American College of Rheumatology 1987 criteria. Values are the median (interquartile range). P values were determined by Wilcoxon's rank sum test, comparing positive versus negative groups. IP = inflammatory polyarthritis; RF = rheumatoid factor; 25(OH)D = 25-hydroxyvitamin D; 1,25(OH)2D = 1,25-dihydroxyvitamin D.

All study patients20 (16, 28)34 (25, 47)  
RA status
 At baseline
  Positive19 (14, 26)31 (22, 40)0.060.01
  Negative21 (16, 30)36 (26, 54)  
 At 1 year
  Positive19 (14, 26)31 (23, 40)0.020.006
  Negative21 (16, 32)38 (26, 54)  
RF status    
 At baseline
  Positive (titer ≥1:40)18 (15, 26)32 (22, 43)0.080.13
  Negative (titer <1:40)21 (16, 30)35 (25, 50)  
 At 1 year
  Positive (titer ≥1:40)19 (15, 27)31 (22, 43)0.160.07
  Negative (titer <1:40)21 (16, 30)36 (25, 50)  

Association between baseline vitamin D metabolite levels and baseline disease activity and severity markers.

Baseline levels of 25(OH)D were associated (inversely) with more markers of disease activity and severity (tender joint count, CRP level, DAS28 score, and HAQ score) than were baseline levels of 1,25(OH)2D (HAQ score) (Table 3). A 10-ng/ml increase in the 25(OH)D level was associated with a decrease in the DAS28 score of 0.30 and in the CRP level of ∼25% (Table 3).

Table 3. Association between disease status at baseline and baseline levels of vitamin D metabolites in 183 IP patients*
Regression modelDisease status assessment at baselinePredicted outcome change (95% CI)
Per 10-ng/ml increase in 25(OH)DPer 10-pg/ml increase in 1,25(OH)2D
  • *

    Data were adjusted for age at baseline, sex, and season during which the blood sample was obtained. IP = inflammatory polyarthritis; 95% CI = 95% confidence interval; 25(OH)D = 25-hydroxyvitamin D; 1,25(OH)2D = 1,25-dihydroxyvitamin D; CRP = C-reactive protein; DAS28 = Disease Activity Score 28-joint assessment; HAQ = Health Assessment Questionnaire (grouped as mild = <1, moderate = 1–2, and severe = >2).

  • Model 1 is a negative binomial regression model for the percentage change in joint counts per each 10-ng/ml or each 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively. Model 2 is a linear regression model for the percentage change in the log-transformed CRP level as a result of a 10-ng/ml or a 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively. Model 3 is a linear regression model for change in the DAS28 score per each 10-ng/ml or each 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively. Model 4 is an ordered logit regression model for the odds of being in a higher HAQ group/Larsen score tertile per each 10-ng/ml or each 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively. Model 5 is a logistic regression model for the odds of having an eroded joint(s) per each 10-ng/ml or each 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively.

  • Significant change in outcome.

1Swollen joint count−16.3 (−41.2, 8.5)−11.6 (−23.5, 0.3)
1Tender joint count−42.7 (−71.1, −14.2)−0.6 (−15.1, 13.8)
2CRP (natural log–transformed; ln[x+1])−26.2 (−50.2, −2.1)−10.9 (−23.3, 1.4)
3DAS28 score−0.30 (−0.53, −0.08)−0.08 (−0.2, 0.04)
4HAQ score (groups)0.61 (0.42, 0.89)0.79 (0.65, 0.96)
4Larsen score (tertiles)0.88 (0.57, 1.35)0.82 (0.66, 1.03)
5Presence of eroded joint(s)0.87 (0.52, 1.46)0.99 (0.76, 1.29)

Associations between the 25(OH)D level and the DAS28 score and the CRP level at baseline are shown in Figure 1 as quintile plots. While there does not appear to be a threshold effect between the 25(OH)D level and the DAS28 score and CRP level as shown in Figure 1, this was tested more formally by comparing the linear regression line estimated between the 2 variables and the fractional polynomial (with 95% confidence intervals), which determines the nonlinear shape of the function between 2 variables. No threshold effect was found between the concentration of either of the 2 vitamin D metabolites and either the DAS28 score or the CRP level.

Figure 1.

Relationship between 25-hydroxyvitamin D levels and A, scores on the Disease Activity Score 28-joint assessment (DAS28) and B, levels of C-reactive protein (CRP; natural log–transformed), by quintiles of 25-hydroxyvitamin D. 95% CI = 95% confidence interval.

Findings of the prospective analysis at 1 year.

Levels of vitamin D metabolites.

The mean baseline levels of both vitamin D metabolites were significantly lower in IP patients who satisfied the ACR criteria for RA by 1 year than in those who did not (Table 2). In other words, subjects with a lower baseline level of vitamin D were more likely to satisfy the ACR criteria for RA by 1 year.

Association between baseline vitamin D metabolite levels and disease outcome measures at 1 year.

There was a significant inverse association between each vitamin D metabolite level and the HAQ score at 1 year; that is, IP patients with higher vitamin D metabolite levels had lower HAQ scores (Table 4). For each 10-ng/ml increase in the baseline 25(OH)D level, there was a 59% chance of being in a lower HAQ score category at 1 year (adjusted for DMARD treatment). Furthermore, baseline 25(OH)D level also showed a significant inverse association with the tender joint count at 1 year. Since blood samples were not taken at 1 year, the relationship between baseline vitamin D metabolite levels and the DAS28 score at 1 year could not be assessed.

Table 4. Association between disease outcome at 1 year and baseline levels of vitamin D metabolites, by adjustment or nonadjustment for DMARD use*
Regression modelDisease outcome at 1-year assessmentPredicted outcome change (95% CI)
Adjusted for DMARD useNot adjusted for DMARD use
Per 10-ng/ml increase in 25(OH)DPer 10-pg/ml increase in 1,25(OH)2DPer 10-ng/ml increase in 25(OH)DPer 10-pg/ml increase in 1,25(OH)2D
  • *

    Data were adjusted for age at baseline, sex, and season during which the blood sample was obtained. Metabolite 25-hydroxyvitamin D (25[OH]D) was measured in 140 patients, and metabolite 1,25-dihydroxyvitamin D (1,25[OH]2D) was measured in 135 patients. DMARD = disease-modifying antirheumatic drug; 95% CI = 95% confidence interval; HAQ = Health Assessment Questionnaire (grouped as mild = <1, moderate = 1–2, and severe = >2).

  • Model 1 is a negative binomial regression model for the percentage change in the joint counts per each 10-ng/ml or each 10-pg/ml increase in the 25(OH)D or 1,25(OH)2D level, respectively. Model 2 is an ordered logit regression model for the odds of being in a higher HAQ group as the result of a 10-ng/ml or a 10-pg/ml increase in 25(OH)D or 1,25(OH)2D level, respectively. Model 3 is a linear regression model for the difference in HAQ score per each 10-ng/ml or each 10-pg/ml increase in 25(OH)D or 1,25(OH)2D level, respectively. Model 4 is a logistic regression model for the odds of having an eroded joint(s) per each 10-ng/ml or each 10-pg/ml increase in 25(OH)D or 1,25(OH)2D level, respectively.

  • Significant change in outcome.

1Swollen joint counts−28.8 (−67.8, 10.3)−0.5 (−19.0, 18.1)−32.3 (−70.3, 5.7)−2.9 (−21.6, 15.7)
1Tender joint counts−46.3 (−90.4, −2.3)1.8 (−16.7, 20.2)−45.5 (−88.4, −2.6)1.8 (−17, 20.5)
2HAQ score (groups)0.59 (0.37, 0.96)0.74 (0.58, 0.94)0.52 (0.33, 0.83)0.72 (0.56, 0.91)
3Difference in HAQ score from baseline−0.11 (−0.25, 0.04)0.01 (−0.06, 0.09)−0.08 (−0.22, 0.06)0.02 (−0.06, 0.09)
4Presence of eroded joint(s)0.79 (0.48, 1.31)0.94 (0.73, 1.22)0.77 (0.47, 1.27)0.95 (0.74, 1.22)

There was no evidence of bias in loss to followup, since there was no difference in baseline vitamin D metabolite levels between patients who were seen only at baseline and those who were also seen at 1 year.

DISCUSSION

In this study, we examined both cross-sectionally and prospectively associations between levels of vitamin D metabolites and measures of disease activity, severity, and functional disability in an unselected cohort of patients with early IP. Cross-sectionally, the associations with the various markers of disease activity and severity were stronger for 25(OH)D than for 1,25(OH)2D. This finding partially supports our current understanding of the way in which vitamin D exerts immunomodulatory effects. The prevailing hypothesis is that circulating 25(OH)D acts as a substrate for conversion to 1,25(OH)2D by activated macrophages, with the 1,25(OH)2D acting in a paracrine manner on immune cells through vitamin D receptors (which are widely expressed on most immune cells) (1, 3, 4, 19). Individuals who are vitamin D replete will have adequate substrate (i.e., 25[OH]D) for 1,25(OH)2D synthesis, whereas those who are vitamin D deficient will have inadequate substrate to support local 1,25(OH)2D synthesis.

Whether circulating 1,25(OH)2D, which acts in an endocrine manner for calcium homeostasis, also has an “endocrine influence” on immunomodulation is less certain. We can speculate from our findings of a weak association between circulating levels of 1,25(OH)2D and disease activity that there is also an endocrine mechanism for immunomodulation. In support of this are intervention studies of oral and parenterally administered 1,25(OH)2D in animals (7) and humans (20) that have shown benefit. It remains uncertain whether increasing levels of circulating 25(OH)D would have greater effects through a paracrine action of 1,25(OH)2D as compared with increasing the levels of circulating 1,25(OH)2D by pharmacologic treatment.

Two previous clinical studies have examined vitamin D levels in patients with RA, but unlike the present investigation, those studies examined patients with established disease of long duration (21,22). Neither study measured joint counts; they used only the CRP level or only the erythrocyte sedimentation rate (ESR) as a measure of RA disease activity. In both studies, no association between 25(OH)D and the CRP level or the ESR was found. An inverse association between the 1,25(OH)2D level and the ESR, but not the CRP level, was demonstrated, but only in blood samples obtained during the winter (21). Conversely, an inverse association between the 1,25(OH)2D level and the CRP level was observed (Spearman's r = 0.52, P < 0.001), but again, only in samples obtained during the winter (22). The interpretation of these data is difficult, not least because of the long duration of RA (0.4–18 years in one study and 0.5–38 years in the other), as well as because patients were taking DMARDs and other treatments that would confound any relationship to disease activity. No correction was made for renal function, and data were not available for the effect of physical incapacity on sunlight exposure, and therefore vitamin D levels. This is important, because physical disability in these patients with a long duration of RA would itself reduce vitamin D metabolite levels.

Our study has a number of strengths. The patients were recruited from a primary care–based study, thereby increasing the generalizability of the data. They were evaluated early after disease onset (median of 4 months) and had not been treated with steroids (which may reduce nonrenal hydroxylation of 25[OH]D) and were not treated with DMARDs by the time of the baseline assessment or had taken DMARDs for <6 weeks. The short disease duration is critical in avoiding the potential confounding influence of prolonged IP on both sunlight exposure and dietary intake of vitamin D. However, if disease activity led to reduced levels of 25(OH)D via lack of sunlight exposure, we might have expected a stronger association with the HAQ score and erosion status than was actually seen.

There were a number of limitations of our study. The baseline associations were cross-sectional, and therefore, the direction of associations cannot be conclusively determined. It is possible that patients with greater disease activity had a worse diet or less sunlight exposure. Other potential confounders include physical activity and body mass index (BMI). Previous studies have shown a weak positive association between physical activity and 25(OH)D levels (23). We were unable to adjust for physical activity because the only potential measure of that was the HAQ, and this was an outcome variable. BMI has been shown to be inversely related to 25(OH)D levels (attributed to sequestration of fat-soluble vitamin D in adipose tissue). However, we found no association between 25(OH)D levels and the BMI in our dataset (r = 0.05, P = 0.57) and so did not adjust for the BMI. Also, it is well recognized that vitamin D deficiency can cause diffuse pain and muscle weakness. Effects of vitamin D levels on these symptoms may contribute to some of the associations seen in this study, such as those between the vitamin D metabolite levels and the HAQ score. The followup at 1 year was relatively short, and we cannot address whether baseline vitamin D status has any longer-lasting effects. Finally, we did not have dietary intake data available for assessment, and so, we were unable to explore the influence of calcium intake, which may be an important factor for the effect of vitamin D on the immune system (3).

In summary, we demonstrated that there is an inverse cross-sectional association between vitamin D metabolite levels and disease activity in patients with early IP and that baseline levels of vitamin D are predictors of the HAQ score at 1 year. It is important to know whether the level of vitamin D influences IP disease activity because vitamin D supplementation is relatively cheap and safe. These data provide support for performing intervention studies in the treatment of IP and RA using vitamin D and nonhypercalcemic analogs of vitamin D metabolites.

AUTHOR CONTRIBUTIONS

Drs. Patel, Silman, and Symmons had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Patel, Silman, Symmons.

Acquisition of data. Farragher, Bunn, Symmons.

Analysis and interpretation of data. Patel, Farragher, Berry, Silman, Symmons.

Manuscript preparation. Patel, Farragher, Berry, Silman, Symmons.

Statistical analysis. Patel, Farragher, Silman, Symmons.

Ancillary