(1073) Data Missingness as a Risk Factor for Preterm Birth

Objective:
Complications from preterm births (PTB) remain the leading cause of death among children below 5. Predictive models of PTB aim to estimate risk to enable early, evidence-based interventions. These models require complete or imputed data—assuming similar relationships between predictive factors and PTB across patients with complete and incomplete data—which may compromise predictive accuracy. The aim of this study is to assess missingness in data collected from a prospective cohort study and whether this impacts estimation of PTB risk.

 

Study Design:
A retrospective analysis on 8955 patients from the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be (nuMoM2b) with documented gestational age. We identified variables previously associated with PTB or considered social determinants of health and would be available for screening at early prenatal visits (Table 1). 5133 patients had complete data across all fields in Table 1 (complete), and 3822 were missing data in one or more fields (incomplete). We estimated a logistic regression model from complete records, including all predictors in Table 1, and separate models for each predictor for complete and incomplete records.

Results:
Incomplete data was associated with 1.6 times the odds of PTB, p < .001. Among participants with complete data, obesity, tobacco use, and poverty are predictive of preterm birth while alcohol and college are protective (all p < .05), consistent with previous studies (Figure 1). By contrast, each of these effects was no longer statistically significant among those with incomplete records (p > 0.05) suggesting missingness may attenuate these associations.

Conclusion:
Even in a rigorously collected prospective study, data missingness persists. Our findings suggest that data missingness itself may signal an elevated PTB risk, acting as a proxy for structural determinants of health. Recognizing the implications of this inequity moving forward is critical for training models and future research should focus on mitigation strategies in parallel with implementation efforts to prioritize equitable data collection.