Identifying Key Predictors of Smoking Cessation Success: Text-Based Feature Selection Using a Large Language Model

Comprehensive Summary

This study used a large language model to find key predictors of smoking cessation. The data for this study was pulled from waves 5 and 6 of the Population Assessment of Tobacco and Health (PATH) study from December 2018 to November 2021. Using Gpt-4.1, the top 45 variables from these waves were identified. In the results, the top variables that were found to influence smoking cessation were past 30-day smoking frequency, time passed from waking up to smoking first cigarette, influential people’s views on tobacco, use of tobacco among close associates, emotional dependence, and health harm concerns. In the discussion, the authors talked about how the data acquired could be used to design more effective questionnaires to improve data collection.

Outcomes and Implications

Although no cause-and-effect relationship was established, identifying these predictors still gives a valuable insight into what factors are correlated with smoking cessation success. This study also shows the ability of GPT-4.1 to select features based on textual descriptions of the variables. The successful use of GPT-4.1 also highlights the potential it has in the tobacco research field to create more resource-efficient and targeted intervention strategies.