Clinical measures identify vitamin D deficiency in dialysis

Abstract

Background and objectives: Vitamin D deficiency (defined by serum levels of 25-hydroxyvitamin D) is common in patients with ESRD on hemodialysis, but risk factors are unknown. This study was conducted to determine whether routinely measured clinical and demographic parameters could identify dialysis patients who are vitamin D deficient.

Design, setting, participants, & measurements: Nine-hundred eight patients with 25-hydroxyvitamin D levels were identified from the Accelerated Mortality on Renal Replacement (ArMORR) cohort of incident U.S. dialysis patients and were divided into training (60%) and validation (40%) sets. Predictive models were generated from routinely assessed clinical and demographic data in the training set using logistic regression modeling, neural networks, and decision trees with vitamin D deficiency as the dependent variable. Models underwent progressive variable reduction to identify the simplest model that remained predictive.

Results: Seventy-nine percent of the population was vitamin D deficient (25-hydroxyvitamin D <30 ng/ml). Black race, female sex, winter season, and hypoalbuminemia (serum albumin < or =3.1 g/dl) were the strongest predictors of vitamin D deficiency. In the validation set, the presence of hypoalbuminemia and winter season increased the likelihood of vitamin D deficiency in black women (from 90% to 100%), black men (from 85% to 100%), white women (from 82% to 94%), and white men (from 66% to 92%).

Conclusions: Deficiency of 25-hydroxyvitamin D is nearly universal among patients with hypoalbuminemia initiating chronic hemodialysis in winter.

Figure 1.

Distribution of 25-hydroxyvitamin D levels. The median 25-hydroxyvitamin D for the entire population (red) and for individuals with specific characteristics (blue) are presented along with the interquartile range for each group. Subgroups for continuous variables are defined based on population medians.

Figure 2.

Regression models for vitamin D deficiency generated from the training set using thresholds defining deficiency as 25-hydroxyvitamin D <30 ng/ml, <20 ng/ml, and <10 ng/ml for models 1, 2, and 3, respectively. Black race, albumin, and season were identified as predictors in each of the three models.

Figure 3.

Decline in predictive power of regression models to predict vitamin D deficiency after stepwise removal of the least predictive variables. At each stage, the weakest remaining predictor was removed from the model and its predictive power was reassessed. Model 2 proved most resilient to variable reduction and its top four predictors (race, sex, season, and albumin) were used in the final model.

Figure 4.

Likelihood of vitamin D deficiency at baseline, with hypoalbuminemia, and with hypoalbuminemia and winter season. Regardless of 25-hydroxyvitamin D level used to define deficiency (<10 ng/ml, <20 mg/ml, or <30 ng/ml), individuals who initiated dialysis in winter and were hypoalbuminemic (serum albumin ≤3.1 g/dl) were more likely to be vitamin D deficient.