TL;DR
This paper outlines the protocol for the ambitious LABOR study, a multi-country, prospective observational cohort designed to collect extensive physiological and clinical data on nearly 12,000 women during labor and delivery. The goal is to build advanced statistical algorithms to identify mothers and newborns at high risk for adverse outcomes, ultimately informing the development of tools to reduce preventable deaths and complications in low-resource settings.
What Problem Does This Paper Address?
Globally, a staggering number of women and newborns die during childbirth or shortly after, with the vast majority of this burden concentrated in low- and middle-income countries (LMICs). These deaths are often preventable through early identification of those at risk and timely intervention. However, current methods for predicting intrapartum complications are insufficient, particularly in resource-limited settings where access to advanced monitoring and diagnostic tools is scarce. The LABOR study aims to bridge this gap by creating a comprehensive dataset that can power the development of sophisticated predictive algorithms tailored for these challenging environments.
Methods
The LABOR study employs a multi-country, prospective, observational cohort design. Approximately 12,000 women presenting to four hospitals across Ghana, India, and Zambia will be enrolled. A key methodological innovation is the extensive collection of real-time physiological data using wearable sensors (fetal ECG, maternal vital signs, uterine activity) and the deployment of "scribes" – trained data collectors who meticulously document clinical findings, interventions, and diagnoses. This approach aims to capture a level of detail far beyond routine clinical records. The study also incorporates a pilot phase to refine procedures and ensure data quality.
Datasets
The study is designed to generate a large, novel dataset. This dataset will comprise:
- Fixed case report forms and real-time scribe data: Collected via a tablet-based data management system.
- Laboratory data: Including maternal and neonatal blood, urine, and swab samples.
- Fetal and maternal physiological sensor data: Continuously collected during labor and delivery.
- Ultrasound data: Including general surveys and fetal biometry measurements.
This comprehensive data collection is intended to create a rich resource for developing predictive algorithms. No existing benchmarks are directly mentioned as being used for comparison, but the study aims to create a new, extensive observational cohort.
Key Insights and Ideas
- The Power of Real-Time, Multi-Modal Data: The study's core innovation lies in its ambitious plan to integrate continuous physiological sensor data with detailed, real-time clinical observations from scribes. This multi-modal approach is expected to capture nuances of labor and delivery that are often missed in standard data collection.
- Bridging the Gap in Resource-Limited Settings: By focusing on LMICs, the study directly addresses a critical global health disparity. The development of algorithms tailored for these settings is crucial for equitable improvements in maternal and neonatal outcomes.
- Machine Learning for Predictive Power: The ultimate goal is to leverage this rich dataset to build machine learning, deterministic, and statistical algorithms that can prospectively stratify risk. This moves beyond reactive care to proactive identification and intervention.
- The "Scribe" Model: The use of dedicated data collection specialists (scribes) is a novel approach to overcome the challenges of accurate and complete data capture in busy clinical environments, aiming to improve data quality significantly.
Key Results
As this paper describes the study protocol, no results are presented. The study is designed to generate data that will be analyzed to achieve the following:
- Develop statistical algorithms to prospectively stratify women and fetuses by their risk of adverse outcomes.
- Identify actionable intrapartum diagnoses.
- Inform the development of new tools to reduce peripartum morbidity and mortality.
Limitations and Future Work
- Generalizability of Hospitals: The reviewer questioned whether the chosen hospitals are representative of typical facilities in LMICs, and whether the analysis will control for site-specific differences. The authors acknowledge that the four sites are urban, low-resource settings, with three being tertiary teaching hospitals and one a busy second-level public hospital. They plan to explore differences in site data.
- Inclusion/Exclusion Criteria: Concerns were raised that exclusion criteria (e.g., severe medical conditions, imminent delivery) might inadvertently exclude women at highest risk. The authors note this as a limitation for generalization and explain their approach to consent and enrollment.
- Sensor Impact: The potential impact of sensors on maternal comfort, movement, and labor progress was raised. The authors state they did not enroll women without sensors for comparison but plan to publish feasibility results and acknowledge this as a limitation. They also note sensors were removed shortly after delivery, not interfering with breastfeeding.
- Scribe Workload: The feasibility of scribes accurately collecting data from up to three women simultaneously was questioned. The authors agree to present data on scribe assignment and acuity and will report on maternal position data from sensors.
- Composite Outcomes: The use of composite outcomes was noted as potentially obscuring individual risk profiles due to the varied etiologies of component events. The authors clarify their focus on decision-specific outcomes for hypothesis generation and optimal treatment regimes.
- Maternal Death Sample Size: The sample size may not provide sufficient power to estimate the risk of maternal death as a single outcome, though these participants will be included in the composite outcome analysis.
- Future Work: The authors emphasize the need for implementation science research to assess the acceptability, feasibility, affordability, equity, and potential unintended consequences of any developed tools before widespread adoption. They also plan to investigate variations in care between sites and from established protocols in future manuscripts.
Summary updated