Longitudinal Methods Versus Multiple Imputation to Infer Missing Maternal Data in Registry-Based Pregnancy Studies.

Area of Science:

• Epidemiology
• Biostatistics
• Reproductive Health

Background:

• Birth registries often suffer from incomplete data, leading to missing values.
• Multiple Imputation (MI) is a common technique for handling missing data.
• The utility of using data from multiple pregnancies of the same woman to improve MI accuracy in birth registries is not well-established.

Purpose of the Study:

• To compare the performance of five methods for inferring missing maternal characteristics in a medical birth registry.
• To evaluate longitudinal methods against MI, incorporating data from previous and future pregnancies.

Main Methods:

• Utilized data from the Medical Birth Registry of Norway (2004-2018) for mothers with multiple pregnancies.
• Applied longitudinal methods using past, future, and closest pregnancy records.
• Performed single-pregnancy MI (index records only) and multiple-pregnancy MI (index and closest reference records).
• Assessed validity by comparing inferred/imputed values with actual values using specific metrics for continuous and binary variables.

Main Results:

• Longitudinal methods demonstrated the highest accuracy for continuous variables, followed by multiple-pregnancy MI, and then single-pregnancy MI.
• For binary variables, longitudinal methods generally yielded superior validity parameters compared to MI.
• Multiple-pregnancy MI showed performance comparable to longitudinal methods, while single-pregnancy MI had significantly lower agreement.

Conclusions:

• Longitudinal methods are more effective than MI for inferring missing maternal characteristics in medical birth registries.
• Incorporating data from multiple pregnancies improves the accuracy of missing data imputation.