Using pattern-recognition software to evaluate intrapartum fetal heart (FHR) tracings

OBJECTIVE: Visual interpretation of FHR is vulnerable to human error. Our purpose was to evaluate the association between neonatal outcome and intrapartum FHR characteristics using computerized pattern-recognition. STUDY DESIGN: Secondary analysis of data from a multi-center trial of intrapartum fetal monitoring in nulliparous women with singleton ≥36 weeks. The final 60 mins of stored digital FHR (scalp electrode) were analyzed by PeriCALM Patterns. Files missing >30 mins before delivery were excluded. The outcome was a composite of either 5- min Apgar <4, umbilical artery pH <7.0, seizure, intubation at delivery, stillbirth, neonatal death, or NICU >48 hrs. Prediction of outcome was evaluated using ROC curves with incremental addition of FHR characteristics. Classification and regression tree (CART) analysis was used to segment the population into meaningful subgroups. Adjustment for the presence of preeclampsia was included as surrogate for magnesium use. RESULTS: 4,208 patients were included. Several characteristics were associated with the outcome (Table). While very few had variability <5 bpm, average variability was lower in cases vs. controls. Accelerations were associated with better outcomes, while prolonged decelerations were associated with worse outcomes. Late and variable decelerations were common and not associated with the outcome. Prediction was significantly improved after addition of prolonged deceleration (p=0.0007), followed by %time in acceleration (p=0.006), and variability in the last 15 mins (p=0.02; Figure). CART analysis revealed that only prolonged deceleration provided significant differentiation. CONCLUSION: FHR pattern-recognition software discriminates between fetuses with vs. without adverse outcome. Criteria used for visual interpretation of FHR, such as minimal variability, may not be applicable when using computerized assessment. Decelerations are common and not discriminatory, unless prolonged. Accelerations and variability are other useful features. (Figure presented).