In recent years, interest in diagnosing vitamin D deficiency by measuring 25-hydroxycholecalciferol (25(OH)D) in serum has increased significantly. A significant proportion of the general population has low levels of 25(OH)D,1,2 but deficiency is found in most patients with chronic conditions, especially chronic kidney disease (CKD).3–8

The cause of the deficiency would seem to be multifactorial. Factors related to exposure to ultraviolet light (UV),7,9,10 and comorbidity such as diabetes11–13 may explain much of the deficit of vitamin D. In patients with CKD, the reduction in glomerular filtration rate,14–16 proteinuria3,17–19 and commonly used drugs20 may also have a significant influence. However, the relative individual importance of all these factors in vitamin D deficiency is unknown.

Multiple linear regression analysis is the method of choice to establish the degree of variation of a dependent variable using the information offered by independent variables.

One of the assumptions that must be met for prediction models to be reliable and accurate is that independent variables have no significant correlation between each other.21 This requirement is very difficult to meet when analysing the potential determinants of vitamin D levels in a population with CKD. There is a high degree of association between factors such as physical activity or exposure to ultraviolet radiation (UV) and age, gender and comorbidity, or between proteinuria and the prescription of certain drugs such as angiotensin inhibitors and receptor blockers. This complicates the interpretation and significance of the regression coefficients and the relative importance of each of these potential determinants.

One way of overcoming the difficulties generated by the multicollinearity in a multivariate linear regression model is determination, in addition to the standardized coefficients (beta), of the structure coefficients which provide additional information about the bivariate relationship between a predictor and the observed effect, without the influence of other predictors in the model.21–23

“Weighting of the relative importance” is an accredited method for establishing the proportional contribution of each predictor to the coefficient of determination (R2), correcting the effects of the cross-correlation between predictors.21–23

The aim of this study was to analyze the determinants of 25(OH)D levels in a cohort of patients with advanced CKD not on dialysis, and establish the relative importance of the variables found to be part of the best predictive model.

The total number of patients studied was 397, but 30 were excluded because they were already receiving vitamin D supplements (27 patients) or were being treated with carbamazepine (3 patients).

The characteristics of the remaining 367 patients who made up the study group are shown in Table 1. The total number of patients with serum 25(OH)D levels below 20ng/ml was 298 (81%).

No significant differences were found in the 25(OH)D concentrations according to the time of year the extraction was made (winter 13.8±8.1ng/ml; autumn 12.9±6.5ng/ml; spring 13.7±7.2ng/ml; and summer 14.1±6.7ng/ml; p=0.886, ANOVA).

Table 1 also shows the differences between patients with sufficiency or lack of vitamin D. The most notable characteristics of those with vitamin D deficiency were as follows: older women; more comorbidity; less physical activity; a more pronounced reduction in average serum albumin concentration and increased proteinuria, with no significant differences in the estimated glomerular filtration rate; and, to a lesser extent, being treated with xanthine oxidase inhibitors.

By multivariate linear regression, the best predictive model for 25(OH)D levels consisted of age, gender, decreased physical activity, proteinuria and serum bicarbonate (Table 2). Variables that were not part of the best equation but came close in their significance were the comorbidity index, serum uric acid and treatment with xanthine oxidase inhibitors (p=0.05).

Analysis of the multicollinearity of the model showed a condition number of 19.57, i.e. at the limit for considering the model to have severe multicollinearity.

The independent variables that showed significant bilateral correlation were age and physical activity – the greater the age, the less physical activity – (R=0.357; p<0.0001), and age and gender – males younger than females – (R=−0.121; p=0.021).

Analysis of relative weights (Table 2) showed proteinuria to be the variable with the greatest relative importance, followed by age. The sum of both variables contributed almost 50% to multiple R. The other 50% was divided among the following independent variables: decline in physical activity; gender (more vitamin D deficiency in females); and serum bicarbonate concentrations (the lower the bicarbonate levels, the more vitamin D deficiency) (Table 2).

With the best predictive model variables transformed into categoricals, a score was set for prediction of severe 25(OH)D deficiency (concentrations <15ng/ml). Each of the following categories added one point: aged over 70; being female; decreased physical activity; proteinuria >1g/g of creatinine; and serum bicarbonate below 22mEq/l.

The cut-offs for transforming continuous variables into discrete variables were estimated by observing points with greater sensitivity and specificity in receiver operating characteristic curves (Youden index).

Fig. 1 shows the percentage of patients with severe 25(OH)D deficiency according to the predictive score ranging from 0 to 5. Approximately 90% of patients with scores of 4 or 5 had severe vitamin D deficiency, while only 25% of those with none of these characteristics had vitamin D deficiency (Fig. 1).

Fig. 1.

Graph showing the percentage of patients with vitamin D deficiency (serum 25-hydroxycholecalciferol levels <15ng/ml) according to predictive score based on age, gender, physical activity, proteinuria and serum bicarbonate. The total number of patients (n) in each scoring subgroup and the percentage with vitamin D deficiency are specified.

(0.07MB).

The results of this study show that over 80% of patients with advanced CKD have vitamin D deficiency. The predictors of vitamin D concentrations with the most relative importance were proteinuria, age, gender, reduction in physical activity and serum bicarbonate.

A previous study on a large Spanish population, the OSERCE Study,3 describes the main parameters of mineral and bone disorder associated with CKD, including 25(OH)D levels. However, in that study, it was not among their objectives to identify the determinants of 25(OH)D levels, and even if it had been, they would not have been able to. In fact, they concluded that 25(OH)D levels are associated only with PTH and, to a small extent, with proteinuria.3

The design of our study left aside the relationship between 25(OH)D and bone mineral metabolism parameters (probably more effect than cause), and focused on establishing what was or were its determinants. Therefore, unlike the OSERCE study, the predictive information provided by our study on 25(OH)D levels is much more rigorous and thorough, and the analysis included the majority of the known potential determinants. The main original aspect of our study though is the statistical analysis for the “relative importance” weights, i.e. we not only determined which factors were associated qualitatively with 25(OH)D levels, but we also quantified the importance of each factor to explain the association model and compensate for the collinearity inherent to these studies.

Due to the frequent cross-correlation between the variables proposed as determinants of a dependent variable, the resulting models can be inaccurate and the results can lead to confusing interpretations.21 This could be the case when analysing the determinants of 25(OH)D concentrations in CKD.

Use of the relative weights statistical method makes it possible to estimate the relative importance of each independent variable in a prediction model, computing the effect of multicollinearity.21–23

In this study, proteinuria was the variable with the greatest relative importance in the prediction model.

Other studies have also shown an association between the magnitude of proteinuria and the degree of vitamin D deficiency in patients with CKD3,17,18 or in renal transplant recipients.19

In some studies in which proteinuria was not measured, an association was found between vitamin D deficiency and hypoalbuminaemia,4,7,18 or the presence of glomerular diseases that tend to present with severe proteinuria.11,13,25

Since most of the studies on this subject have been cross-sectional, we cannot conclude that there is a causal relationship between proteinuria and vitamin D deficiency.

A high percentage of circulating vitamin D is bound to proteins (DBP and albumin), which are filtered in large quantities through the sick glomeruli and this can lead to significant losses in urine.14 A plausible hypothesis to explain this association would therefore be a vitamin D depletion effect caused by proteinuria.

Further data to support the hypothesis that vitamin D deficiency is associated with protein depletion are the lower 25(OH)D levels in patients on peritoneal dialysis; these patients tend to have higher protein loss than those treated with haemodialysis.26,27

However, there might also be a hypothetical proteinuric effect associated with vitamin D deficiency. Although it has been found that vitamin D analogs (calcitriol, paricalcitol) have a reducing effect on proteinuria, a study has also shown significant reductions in albuminuria after repletion of vitamin D levels with cholecalciferol.28 Proteinuria might therefore not only be a cause but also an effect of vitamin D deficiency, which would reinforce the interpretation of its greater relative importance.

Other aspects such as older age, female gender or decreased physical activity are more predictable factors that contribute to lower vitamin D levels.13,29

One finding difficult to interpret in this study is the significant association, although with less relative importance, between the concentration of serum bicarbonate and 25(OH)D levels.

Acidosis does not cause any effects on the synthesis or degradation of 25(OH)D.30,31 Ketoacidosis32 or ketogenic diets33 could cause reductions in 25(OH)D levels, but this assumption does not seem relevant among our patients.

A common link between acidosis and 25(OH)D reduction in the clinical context of proteinuric kidney disease could be proximal tubule dysfunction.

A more dysfunctional proximal tubule, in addition to possibly causing a more severe acidosis (type 2 renal tubular acidosis) could lead to increased urinary losses of 25(OH)D due to lack of reabsorption of proteins (DBP, albumin), requiring tubular expression of multiligand endocytic receptors, such as megalin and cubilin.34 Vitamin D deficiency has even been identified as direct cause of proximal tubule dysfunction.35 The proximal tubule–proteinuria–vitamin D interaction, though speculative, could help explain these findings.

The time of year the sample for 25(OH)D measurement is taken has been shown to influence the results in some studies.6,7 In very northern continental climates in the northern hemisphere, there is a huge difference between the cold seasons and the hot seasons in terms of the amount of sunlight. In contrast, the climate in Extremadura in Spain is southern continental, with quite sunny winters, which encourage people to go out in the sun, clement springs and autumns, but very hot summers, when it is necessary to shelter from the sun. These aspects may help explain the lack of significant differences in 25(OH)D levels according to the time of year in our patients.

In this study, we found no significant association between low levels of 25(OH)D and statins or ACE inhibitors and/or ARAs. In another recent study20 in a smaller group of patients with predialysis CKD and severe vitamin D deficiency, Yuste et al. found even lower 25(OH)D levels in those treated with statins, and higher levels in those treated with ACE inhibitors or ARAs. In the multivariate analysis of this study, however, the authors did not include proteinuria in the model fit, a fact that might have been of great interest for the interpretation of the results.

In that same study,20 Yuste et al. detected higher 25(OH)D levels in patients treated with allopurinol. In our study, treatment with XO inhibitors (allopurinol or febuxostat) was associated with higher levels of 25(OH)D compared to no such treatment, although this covariate narrowly missed being statistically significant in the multivariate analysis (p=0.05).

There is a relationship between purine metabolism and vitamin D, with an inhibitory effect of uric acid and xanthine on 1-α-hydroxylase having been demonstrated.36–38 However, what role the inhibition of xanthine-oxidase may play in 25(OH)D metabolism is uncertain.

The anti-inflammatory effect of XO inhibitors has been suggested as cause of the higher levels of 25(OH)D.20 In our study, however, the 25(OH)D levels were not correlated with any inflammatory markers, such as C-reactive protein, and nor did those treated with XO inhibitors have lower C-reactive protein concentrations (data not shown). In contrast, we did find that patients treated with XO inhibitors had significantly less proteinuria than those not treated (1.39±1.38 vs. 2.13±2.46g/g creatinine; p=0.005). The collinearity with proteinuria may therefore at least in part explain the association of treatment with XO inhibitors and 25(OH)D levels.

This study has limitations, particularly in relation to the cross-over design, which prevents us from firmly establishing causal inferences.

In conclusion, vitamin D deficiency is very common in CKD, and the magnitude of proteinuria is a factor with high relative importance in predicting reduced 25(OH)D levels. In view of the strong association with proteinuria, any study on the adverse effects of vitamin D deficiency in CKD (e.g. progression of renal failure) should take this relationship into account, since the magnitude of proteinuria and associated factors can also be very important as determinants of the clinical progression and prognosis of this disease.

The authors declare that they have no conflicts of interest.