Date of Award
Summer 6-23-2024
Degree Type
Dissertation
Degree Name
DNP Doctor of Nursing Practice
Department
Nursing
Advisor
Mary Ellen Roberts, DNP
Committee Member
Teresa Conklin, DNP
Committee Member
Paul Testa, MD
Keywords
Obstetric Readmissions, Predictive Model, AUROC, AUPRC, Cognitive computing model, Precision and Recall
Abstract
This study aims to develop and validate an OB-readmission cognitive computing model for identifying women at elevated risk of unplanned readmissions after childbirth. The model intends to assist healthcare providers in identifying vulnerable patients and implementing targeted interventions and strategies to reduce readmission rates, improve patient outcomes, and optimize resource utilization.
A comprehensive analysis of relevant literature identified key predictors associated with postpartum readmissions, including maternal age, race and ethnicity, medical comorbidities, obstetric diagnoses and complications, treatment and diagnostic modalities, socioeconomic status, and previous hospitalizations. These factors were incorporated into the proposed OB predictive model, which was based on the electronic health record’s existing predictive readmission risk model (version 2) for acute medical-surgical patients.
Using electronic health records of a large cohort of postpartum women, machine learning algorithms were employed to build and optimize the obstetric predictive model. Performance evaluation metrics, including the area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC), were used to assess the model's effectiveness for predicting readmissions within 30 days.
The developed predictive model demonstrated promise, with an average AUROC of 0.704, indicating its ability to discriminate between positive and negative cases. The model’s average AUPRC however, was 0.163, suggesting a low signal in predictive capability.
Implementing the OB-readmission predictive model within the hospital setting requires multidisciplinary collaboration, engagement with healthcare professionals, and integration into electronic health record systems. User-friendly interfaces and decision support tools are crucial for easily adopting and interpreting the model's predictions.
Beyond the hospital setting, the model has potential applications in community healthcare, facilitating the development of transitional care programs to prevent readmissions and improve postpartum care continuity. Additionally, the model's insights can inform policy discussions and resource allocation decisions, supporting initiatives to reduce healthcare costs and enhance maternal health outcomes at a larger scale.
In its current state, the OB-readmission predictive model demonstrates moderate accuracy in identifying women at elevated risk of unplanned readmissions 30 days after childbirth. It could enable targeted interventions, optimize resource utilization, and improve patient outcomes. However, to ensure maximal and consistent outcomes, further refinement of the model is needed to ensure its predictive accuracy. The model represents a valuable tool for enhancing postpartum care, reducing readmission rates, and improving maternal health outcomes. Successful implementation and sustainability will require additional collaboration, model reconfiguration, and support from various stakeholders.
Recommended Citation
Doty, Glenn Robert, "Advancing Maternal Health: Development and Implementation of a Cognitive Computing Model to Predict Unplanned Obstetric Readmissions within a Multi-Campus Academic Healthcare System" (2024). Seton Hall University Dissertations and Theses (ETDs). 3189.
https://scholarship.shu.edu/dissertations/3189
Signature Sheet
