Poster Session 2
Category: Labor
Poster Session 2
Misa Hayasaka, MD
Research scholar
Macon & Joan Brock Virginia Health Sciences at Old Dominion University Eastern Virginia Medical School
Norfolk, Virginia, United States
George R. Saade, MD
Department of Obstetrics and Gynecology, Eastern Virginia Medical School at Old Dominion University
Norfolk, Virginia, United States
Grace Spencer, BS, MS
Eastern Virginia Medical School at Old Dominion University
EVMS OBGYN, Virginia Health Sciences at Old Dominion University, Virginia, United States
Tetsuya Kawakita, MD, MS
Associate Professor
Macon & Joan Brock Virginia Health Sciences at Old Dominion University Eastern Virginia Medical School
Norfolk, Virginia, United States
To identify baseline characteristics associated with heterogeneity in the effect of elective induction at 39 weeks on maternal and neonatal outcomes, utilizing individual treatment effect (ITE) estimates.
Study Design:
We conducted a secondary analysis of a randomized controlled trial involving nulliparous women randomized to either elective induction at 39 weeks or expectant management. A Desirability of Outcome Ranking (DOOR) score, ranging from spontaneous vaginal delivery without severe neonatal morbidity (most favorable) to cesarean delivery with perinatal death (least favorable), was assigned on an eight-level ordinal scale.
We employed a causal forest machine learning model to estimate ITEs for the DOOR score, based on maternal baseline characteristics. Honest splitting was used to divide the dataset into training and validation subsets. Model performance was assessed using the Q-score. Feature importance was derived from the causal forest model. Multivariable regression was performed to enhance interpretability by examining the direction and strength of associations between baseline factors and predicted ITEs.
Results:
Among 6,096 participants, DOOR scores were skewed toward favorable outcomes, with 69% receiving the highest score. The median predicted ITE was greater than 0, and the Q-score was 0.54. Mean differences in DOOR scores between the induction and expectant management groups increased across tertiles of predicted ITEs (Figure 1).
Feature importance analysis and multivariable regression showed that weight gain during pregnancy, BMI at randomization, gestational age at the first prenatal visit, maternal age, and Bishop score were the strongest contributors to treatment effect variation (Figure 2).
Conclusion:
Causal forest modeling, utilizing the DOOR score as a composite and patient-centered outcome, effectively identifies baseline characteristics associated with differential treatment benefit. This approach may facilitate the tailoring of induction strategies to patients most likely to benefit, thereby supporting shared decision-making.