Estimation methods with ordered exposure subject to measurement error and missingness in semi-ecological design.

Area of Science:

• Epidemiology
• Occupational Health
• Biostatistics

Background:

• Epidemiological studies often face challenges with accurate exposure measurement, particularly in occupational settings.
• Group-based strategy (GBS) is used to mitigate measurement error bias by assigning group-level exposure means to individuals.
• Bias can arise in GBS when group means are estimated from small sample sizes, leading to inaccurate exposure-disease associations.

Purpose of the Study:

• To evaluate and compare different methods for exposure assessment in semi-ecological studies.
• To address bias in risk estimates caused by small numbers of exposure measurements and missing data.
• To identify a reliable and cost-effective approach for estimating exposure-disease associations in occupational epidemiology.

Main Methods:

• A simulation study compared four methods: naive complete-case analysis, GBS, constrained GBS (CGBS), and constrained expectation-maximization (CEM).
• The study incorporated naturally ordered groups/jobs using constrained estimation methods.
• Regression models with measurement error and missing values were analyzed using expectation and maximization (EM) algorithms.

Main Results:

• Naive and GBS methods were inadequate with small numbers of exposure measurements.
• The CEM method performed well when moderate numbers of exposures were observed with error.
• CGBS demonstrated superior bias-reducing properties and was easier to implement than CEM, especially with substantial missing exposure data.

Conclusions:

• Constrained group-based strategy (CGBS) is a valuable method for semi-ecological studies with ordered group means and limited exposure data.
• CGBS offers improved bias reduction compared to other methods, particularly when dealing with missing exposure information.
• The findings support cost-effective study designs, enabling reliable exposure-disease association estimates with reduced exposure measurement efforts.