Evaluating the healthiness of chain-restaurant menu items using crowdsourcing: a new method.

OBJECTIVE: To develop a technology-based method for evaluating the nutritional quality of chain-restaurant menus to increase the efficiency and lower the cost of large-scale data analysis of food items.

DESIGN: Using a Modified Nutrient Profiling Index (MNPI), we assessed chain-restaurant items from the MenuStat database with a process involving three steps: (i) testing 'extreme' scores; (ii) crowdsourcing to analyze fruit, nut and vegetable (FNV) amounts; and (iii) analysis of the ambiguous items by a registered dietitian.

RESULTS: In applying the approach to assess 22,422 foods, only 3566 could not be scored automatically based on MenuStat data and required further evaluation to determine healthiness. Items for which there was low agreement between trusted crowd workers, or where the FNV amount was estimated to be >40 %, were sent to a registered dietitian. Crowdsourcing was able to evaluate 3199, leaving only 367 to be reviewed by the registered dietitian. Overall, 7 % of items were categorized as healthy. The healthiest category was soups (26 % healthy), while desserts were the least healthy (2 % healthy).

CONCLUSIONS: An algorithm incorporating crowdsourcing and a dietitian can quickly and efficiently analyze restaurant menus, allowing public health researchers to analyze the healthiness of menu items.