A sampling strategy for longitudinal and cross-sectional analyses using a large national claims database

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Medicare claims files are crucial for national healthcare utilization data.
• Obtaining and analyzing large Medicare datasets for longitudinal studies presents significant challenges.
• A well-documented method for creating representative samples is needed.

Purpose of the Study:

• To present a method for constructing longitudinal patient samples from Medicare claims files.
• To ensure these samples are representative of the overall Medicare population annually.
• To facilitate retrospective cohort studies with multi-year follow-up.

Main Methods:

• Utilized Medicare Master Beneficiary Summary Files over a 10-year period.
• Targeted ~900,000 patients annually, stratified by county and race/ethnicity with minority oversampling.
• Retained patients based on continuous enrollment and geographic stability, replacing non-retained patients to maintain sample representativeness.

Main Results:

• The final sample averaged 899,266 patients annually, closely mirroring population demographics (age, race, sex).
• Chronic condition prevalence in the sample closely matched population rates (within 0.12% for 21 comorbidities).
• Longitudinal cohorts derived from the sample accurately reflected population cohorts after 5- and 10-year follow-up.

Conclusions:

• The described sampling strategy effectively generates representative longitudinal Medicare patient samples.
• This method is adaptable for other national claims databases and specific sub-population oversampling.
• The approach supports robust retrospective cohort studies using large-scale healthcare data.