Updated Multiple Source Method (MSM)

Background:

The Multiple Source Method (MSM) was developed to model dietary intake using repeated short-term measurements. The MSM considers reported days without consumption, estimates within-person and between-person variance, and can be applied to skewed consumption-amount distributions. Empirical and simulated studies showed that the first version of MSM has also comparable performance and features to similar statistical methods (e.g. Verly-Jr Br J Nutr.2016:116(05),897 – 903). In one simulation study MSM showed some unexpected results after adjusting for food frequency information. Therefore, we updated the MSM to improve the algorithm.

Methods:

Compared to the previous version, we changed the calculation of the consumption-day amounts to distinguish between individuals“ and population-based usual intake. Secondly, we changed the calculation of residuals after applying logistic regression from simple residuals (difference between observed and estimated probabilities) to Pearson residuals. The new algorithm was tested to estimate percentiles of usual intake with simulated data and compared to the NCI method.

Results:

The simulated data showed that the results from the updated MSM are comparable to the results from the NCI method and no further unexpected results due to adjustment were produced by the MSM. For instance, the mean bias for estimating percentiles for daily consumed foods was 0.28% and 0.2% for MSM and NCI, respectively. Adjustment didn“t change the percentiles in the daily consumed case. Mean bias for the MSM of an episodically consumed food decreased from 5.25% to 0.12% after adjusting by a predictor variable. Surprisingly, Apply the NCI method the mean bias for the same food increased from -1.4% to -5.7%.

Conclusion:

The updated version of MSM is a promising statistical approach to model nutritional data from repeated short-measurements and is accessible as a new web program. The accuracy of the estimations on the individual level is under investigation.