PC1059: Influence of Race and Ethnicity Analytic Choices on Preterm Birth Disparity Assessments

Background: The gold standard for epidemiologic analyses of race and ethnicity is self-identified designation. When examining racial disparities in preterm birth (PTB, birth prior to 37 weeks’ gestation), researchers typically aggregate racial and ethnic categories into five bins: Non-Hispanic (NH) Asian and Pacific Islander, NH Black, NH White, NH Other, and Hispanic. This architecture may include analytic decisions such as aggregating “other” with “multiple” or “unknown,” or collapsing country of origin data within Asian or Hispanic identifications. Additionally, on Pennsylvania birth certificates, parents can check more than one race box and are also asked to designate a single race for their infant. The impact of race aggregation analytic choices on estimates of racial and ethnic disparities in PTB is unknown.

Methods: We analyzed Pennsylvania birth certificate data from 2006 to 2014. We examined singleton PTB rates by infant race and ethnicity based on parents’ designation. We assessed impacts of aggregation on PTB rates, as well as differences between estimates derived when multiple race selection is an option versus when parents must designate a single race. Modified Poisson regression was used to calculate risk ratios for PTB compared with the largest group (NH White).

Results: There were 1,098,429 infants in the cohort, of which 86,484 (7.9%) were preterm. With typical levels of aggregation, when multiple races were allowed, PTB ranged from 6.6% among NH Asian and Pacific Islander infants to 11.2% among NH Black infants (Table 1). Heterogeneity was evident upon disaggregation; NH Pacific Islander infants (8.8%) had higher PTB rates than NH Asian infants (6.6%) with greater variability by country of origin including highest rates among Filipino infants (9.5%) (P < 0.0001) (Figure 1). PTB rates also varied across country of origin for Hispanic infants with the highest rates among Cuban infants (9.9%) and across “other,” “multiple,” and “unknown” infants (all P< 0.0001). Rates were overall similar when only a single designation was allowed. However, the largest group shift was a near tripling of the number of infants in the NH “unknown” group, with an associated attenuation of PTB risk from 34% to 25% increased risk compared with NH White infants.

Conclusion: Honoring patients’ self-identified race and ethnicity becomes challenging when performing epidemiologic analyses due to small sample sizes in rare populations, unstable estimates, and lack of statistical model convergence. Consistent with prior studies, there was substantial variation in PTB within Asian and Hispanic groups. Additionally, we observed variation in PTB and shifts in disparity with respect to “other,” “multiple,” and “unknown,” race infants highlighting the need to thoughtfully consider these groups separately prior to aggregation.