Addressing selection bias in dental health services research

Area of Science:

• Health Services Research
• Biostatistics
• Public Health

Background:

• Randomization is often not feasible in observational studies, necessitating methods to control for non-random assignment.
• Selection bias can confound the estimation of treatment effects when participants are not randomly assigned to groups.
• The Supplemental Program for Women, Infants, and Children (WIC) is a crucial public health initiative, but its impact on health behaviors like dental visits requires rigorous evaluation.

Purpose of the Study:

• To demonstrate a two-stage analytical technique for controlling selection bias in observational research.
• To examine the effects of WIC participation on dental visits among children aged 1-5 years, accounting for non-random assignment.
• To quantify the impact of selection bias on the estimated effects of WIC on dental utilization.

Main Methods:

• Utilized a two-stage analytical technique to address selection bias in a study of WIC program effects.
• Constructed an analysis file from 5 data sources, including 49,512 children aged 1-5 years.
• Performed specification tests to confirm non-random WIC participation and the presence of selection bias.

Main Results:

• Specification tests confirmed that WIC participation was non-random, indicating the presence of selection bias.
• The estimated effect of WIC on dental visits differed by 36% after adjusting for selection bias using the two-stage technique.
• The two-stage method successfully controlled for potential selection bias in the analysis of WIC's impact.

Conclusions:

• The two-stage analytical technique is effective in controlling for selection bias when randomization is not possible in health services research.
• Failure to account for selection bias can lead to significant under or overestimation of program effects, as demonstrated by the WIC example.
• This methodology provides a robust approach for evaluating public health interventions and improving the accuracy of research findings.