Dietary Supplement Label Explorer Methodology

The STRIPED research team utilized a rigorous, multi-step methodology to systematically categorize over 180,000 (as of 2/29/2024) supplement labels from the National Institutes of Health Office of Dietary Supplements’ Dietary Supplements Label Database (DSLD). We aimed to identify which, if any, of our three categories of interest–weight loss, muscle building, and/or cleanse/detox–applied to each label. Through our analysis, labels could be assigned to one, more than one, or none of the categories depending on what claims were included on the label.

In order to build the STRIPED Dietary Supplement Label Explorer, we developed a gold standard database of over 5000 labels from the DSLD to categorize labels for weight loss, muscle building, and/or cleanse/detox, using a double-coding process with human coders. Two researchers independently coded the labels and reconciled any discrepancies with a third researcher. We then created deterministic heuristics to automatically identify these label categories in the gold standard database based on keywords and phrases. A deterministic heuristic is an algorithmic approach that systematically makes a series of deterministic choices that are optimal at each step, which we utilized to efficiently categorize the very large number of labels in the DSLD. The heuristics were iteratively tested and refined using rigorous, scientific performance metrics against the gold standard database housed in a custom WordPress tool. Once finalized, the heuristics were integrated into the tool to power a public-facing searchable database of all labels in the DSLD. The results of our performance testing for our deterministic heuristic are as follows:

Performance metrics for the automated classification of dietary supplement labels in DSLD via deterministic heuristic vs. gold standard classification

Label code  Sensitivity Specificity Positive Predictive Value Negative Predictive Value  F1 Score Accuracy Score
Weight loss  0.98 0.97  0.63 0.999 0.77 0.97
Muscle building  0.99 0.99 0.89 0.999 0.94 0.99
Cleanse/Detox 0.97 0.997 0.92 0.999 0.95 0.996

Through our rigorous, scientific process for development, testing, and refinement, our tool achieved excellent performance metrics in categorizing labels. Still, readers should be aware that our tool may miss or miscategorize a small proportion of labels in DSLD. We encourage users to use the keyword search in our tool to further refine searches for their own purposes. Users also should be aware that while the National Institutes of Health Office of Dietary Supplements intended DSLD to contain only Supplement Facts labels, DSLD still inadvertently includes a small number of Nutrition Facts labels.

Now freely and publicly available on our website, the STRIPED Dietary Supplement Label Explorer can be used to browse, search, and filter labels identified through our rigorous development and testing process. This tool will be updated quarterly to align with the records contained in the current version of the NIH’s Dietary Supplements Label Database.

Return to the STRIPED Dietary Supplement Label Explorer

 

Disclaimer

The STRIPED Dietary Supplement Label Explorer (DSLE) is a research tool designed to facilitate the categorization of dietary product labels contained in the NIH Dietary Supplement Label  Database (NIH DSLD) by type, for research use only.  The manufacturer or distributor is responsible for information on product labels; therefore, the labels might be incomplete and/or inaccurate.  The Harvard T.H. Chan School of Public Health is not responsible for the NIH DSLD or any information contained therein. The inclusion of a product and its label in any search results using the DSLE is not an endorsement of that product or its manufacturer.  The description of the DSLE methodology and other information on the DSLE site are not intended as a warranty or guarantee as to the accuracy, completeness, or benefits of the DSLE; we disclaim any and all representations or warranties.  The DSLE may contain errors or defects that may result in miscategorization or inaccurate information; users should use the tool for research purposes only and at their own risk, and should independently verify any categorization of product labels.