Assessing the Potential for Bias From Nonresponse to a Study Follow-up Interview: An Example From the Agricultural...

Area of Science:

• Epidemiology
• Public Health
• Biostatistics

Background:

• Prospective cohort studies are crucial for disease etiology research.
• Participant attrition, due to non-response or death, is a significant challenge.
• Missing data from attrition can introduce bias in study findings.

Purpose of the Study:

• To investigate the impact of attrition on a large prospective cohort study (Agricultural Health Study).
• To assess whether participant response to follow-up interviews was associated with baseline characteristics.
• To evaluate the effectiveness of inverse probability weighting (IPW) in mitigating attrition bias.

Main Methods:

• Analyzed response rates and baseline characteristics of 52,394 farmers enrolled in the Agricultural Health Study (1993-1997).
• Compared associations between smoking and cancer incidence in the full cohort versus the 2005-2010 respondent subcohort.
• Applied inverse probability weighting (IPW) to the respondent subcohort to adjust for potential attrition bias.

Main Results:

• Participant response was associated with age, race/ethnicity, education, marital status, smoking, and alcohol consumption.
• Differential attrition led to observed bias in exposure-outcome associations, which IPW failed to fully correct.
• Non-differential attrition resulted in similar estimates between the full cohort and subcohort, rendering IPW unnecessary.

Conclusions:

• Attrition in prospective cohort studies can introduce bias, particularly when response is differential by exposure.
• Inverse probability weighting (IPW) may not fully correct bias in all scenarios of differential attrition.
• Investigating attrition's influence using self-reported data is essential for ensuring the validity of cohort study findings.