Exploring the impact of missing data on maternal tobacco program results: an analysis of self-efficacy on smoking cessation

Background: Tobacco use is a major contributor to chronic illnesses worldwide, leading to significant morbidity and mortality. Effective tobacco cessation programs are crucial in reducing the health risks associated with smoking. This study aims to investigate the relationship between confidence scores, indicating individuals’ confidence in quitting smoking on a scale of 0 to 10, and carbon monoxide (CO) levels over time among participants in the North Dakota State Department of Health Maternal Tobacco Program. Methods: We analyzed data from 226 participants, who enrolled in the program between August 2020 and February 2023, using linear mixed-effects models and accounted for missing data using complete case analysis, multiple imputation under a missing at random (MAR) assumption, and multiple imputation under a missing not at random (MNAR) assumption with various delta-adjustment values. This MNAR assumption is considered because we believe patients who resume smoking may be less likely to return for follow-up visits. Results: The findings suggest that higher confidence scores at the first prenatal visit were associated with lower CO levels, and this effect attenuated over time. This indicates that participants with higher confidence scores were less likely to smoke. These results remained consistent across different missing data assumptions, demonstrating the robustness of our findings under both MAR and various MNAR scenarios. The postpartum dataset yielded similar results. Conclusion: Our study highlights the importance of evaluating participants’ self-assessed confidence during their initial baseline visit. The findings indicate that patients with lower confidence scores are at an increased risk of discontinuing their participation in the program.