A Novel Approach for Prediction of Vitamin D Status Using Support Vector Regression

Abstract

BackgroundEpidemiological evidence suggests that vitamin D deficiency is linked to various chronic diseases. However direct measurement of serum 25-hydroxyvitamin D (25(OH)D) concentration, the accepted biomarker of vitamin D status, may not be feasible in large epidemiological studies. An alternative approach is to estimate vitamin D status using a predictive model based on parameters derived from questionnaire data. In previous studies, models developed using Multiple Linear Regression (MLR) have explained a limited proportion of the variance and predicted values have correlated only modestly with measured values. Here, a new modelling approach, nonlinear radial basis function support vector regression (RBF SVR), was used in prediction of serum 25(OH)D concentration. Predicted scores were compared with those from a MLR model.MethodsDeterminants of serum 25(OH)D in Caucasian adults (n = 494) that had been previously identified were modelled using MLR and RBF SVR to develop a 25(OH)D prediction score and then validated in an independent dataset. The correlation between actual and predicted serum 25(OH)D concentrations was analysed with a Pearson correlation coefficient.ResultsBetter correlation was observed between predicted scores and measured 25(OH)D concentrations using the RBF SVR model in comparison with MLR (Pearson correlation coefficient: 0.74 for RBF SVR; 0.51 for MLR). The RBF SVR model was more accurately able to identify individuals with lower 25(OH)D levels (<75 nmol/L).ConclusionUsing identical determinants, the RBF SVR model provided improved prediction of serum 25(OH)D concentrations and vitamin D deficiency compared with a MLR model, in this dataset.

Predicted vitamin D status during pregnancy in relation to offspring forearm fractures in childhood: a study from the Danish National Birth Cohort

“…In an exposure-disease approach using predicted vitamin D scores instead of biomarker measurements, we must expect random measurement error to attenuate the measure of association; however, the potential loss in precision may be recovered by the larger study sample available when using predicted vitamin D status (30) . Other studies defining prediction models of vitamin D status showed 25(OH)D concentrations to be predicted quite accurately by a range of demographic and lifestyle factors (30)(31)(32)(33) . Several studies have been published using predicted vitamin D scores when testing exposuredisease hypotheses (30,(34)(35)(36)(37) , and the method seems to be broadly accepted for an epidemiological approach in large cohort studies.…”

Hypovitaminosis D and Calcium Intake in Adult Population

“…our research, did not take into account the usual calcium intake or sun-exposed body parts (33). This research did not measure Vit-D intake within the usual diet nor the consumption of antacids or laxatives, factors directly related to serum levels or absorption of Vit-D. One of the strengths of the design is the consecutive and random selection of participants, which took into account the ses and allowed including equitably subjects belonging to low, medium, and high socioeconomic levels (34).…”

Development and Validation of a Predictive Model of Hypovitaminosis D in General Adult Population: SCOPYD Study

“…The number of participants to be included in the sample used to construct the multiple linear model was determined so that the relative reduction of the model predictive ability for new participants, measured by the coefficient of determination (R-squared or R 2 ), would not be more than 2.5%. Previously published predictive models had a R 2 between 0.13 and 0.42 [ 7 , 11 , 18 , 29 , 30 ]. We hypothesized that the predictive ability of the model would be at least 0.4.…”