Investigating and quantifying selection bias in research using placental pathology samples

Abstract Introduction Increasing research interest has focused on placental pathology and pregnancy outcomes. Placental pathology examinations that are requested by the delivering provider are often used for research due to convenience, potentially introducing selection bias. We leveraged a large cohort of prospectively collected and examined placentas to compare prevalences of placental pathology and to quantify potential selection bias. Methods All placentas were prospectively collected from participants and pathology examinations were completed in the Stress, Pregnancy, and Health (SPAH) study, regardless of delivering provider request. In all cases, placental pathology was categorized and graded into four major groups: acute inflammation (AI), chronic inflammation (CI), fetal vascular malperfusion (FVM), and maternal vascular malperfusion (MVM). We compared the distribution of placental pathology in cases with and without pathology examination requests. Odds ratios (OR) for preterm birth (PTB) and small for gestational age (SGA) infants were calculated in the whole cohort and compared with the requested subset. Relative odds ratios (ROR) were used to quantify the magnitude of selection bias. Models were adjusted for sociodemographic and pregnancy characteristics. Results A total of 575 placentas were collected and examined, 287 with delivering provider examination request and 288 without request. The prevalence of AI, CI, and FVM was similar among the two groups. However, the prevalence of MVM was significantly higher in the requested placentas, particularly for high‐grade MVM (15% vs. 8%, p < 0.01). In adjusted models, MVM was significantly associated with increased odds of PTB in the whole SPAH cohort and the requested subset (SPAH OR: 3.07, 95%CI: 1.76,5.35; requested OR: 3.34, 95%CI: 1.74,6.38). MVM was also significantly associated with SGA infants (SPAH OR: 3.72, 95%CI: 1.55,8.95; requested OR: 4.18, 95%CI: 1.36,12.9). RORs were 1.09 for PTB and 1.12 for SGA, indicating 9% and 12% overestimation in a provider‐requested sample. When considering grade of pathology, selection bias had the greatest impact on the association between PTB and high‐grade MVM. Conclusions MVM was the only placental pathology seen more frequently in the delivering provider requested sample. Our RORs quantify differences in the associations, providing an estimate for the potential impact of selection bias. Research using a convenience sample of requested placentas may modestly overestimate associations, though the overall findings and interpretation of associations remain unchanged.