Can a model predictive of vitamin D status be developed from common laboratory tests and demographic parameters?

OBJECTIVES: Vitamin D deficiency is highly prevalent and has been linked to increased morbidity and mortality. There has been an increase in testing for vitamin D with a concomitant increase in costs. While individual factors are significantly linked to vitamin D status, prior studies have not yielded a model predictive of vitamin D status or 25(OH)D levels. The purpose of this investigation was to determine if a prediction model of vitamin D could be developed using extensive demographic data and laboratory parameters.
METHODS: Patient data from 6 Veterans Administration Medical Centers were extracted from medical charts.
RESULTS: For the 14,920 available patients, several factors including triglyceride level, race, total cholesterol, body mass index, calcium level, and number of missed appointments were significantly linked to vitamin D status. However, these variables accounted for less than 15% of the variance in vitamin D levels. While the variables correctly classified vitamin D deficiency status for 71% of patients, only 33% of those who were actually deficient were correctly identified as deficient.
CONCLUSION: Given the failure to find a sufficiently predictive model for vitamin D deficiency, we propose that there is no substitute for laboratory testing of 25(OH)D levels. A baseline vitamin D 3 daily replacement of 1000-2000 IU initially with further modification based on biannual testing appears to factor in the wide variation in dose response observed with vitamin D replacement and is especially important in high-risk groups such as ethnic minorities.