Obstetric Quality and Safety
Poster Session 4
Nikolina Docheva, MD
NorthShore University HealthSystem
Evanston, IL, United States
Urmila Ravichandran, MS
NorthShore University HealthSystem
Evanston, IL, United States
Marci Adams, MPH
NorthShore University HealthSystem
Evanston, IL, United States
Constandina Kapogiannis-Politis, MS, RN
NorthShore University HealthSystem
Evanston, IL, United States
Ekaterina Livschiz, BS
NorthShore University HealthSystem
Evanston, IL, United States
Richard Chesis, N/A
NorthShore University HealthSystem
Evanston, IL, United States
Nirav Shah, MD
NorthShore University HealthSystem
Evanston, IL, United States
Richard Silver, MD
NorthShore University HealthSystem
Evanston, IL, United States
Beth A. Plunkett, MD,MPH (she/her/hers)
Clinical Professor
Evanston Hospital, Endeavor Health, previously known as NorthShore University HealthSystem
Evanston, IL, United States
To create and validate an automated Peripartum Risk Score Model (PRSM) in the electronic medical record (EMR) to identify patients at highest risk for SMM during delivery admission.
Study Design: Using health-system data from all deliveries/terminations from 1/1/13-10/21/20, a retrospective cohort was used to build/validate the PRSM model. A separate prospective cohort (7/21/22-12/26/22) was used for prospective validation. The 21 CDC-defined SMM were identified by ICD9/10 codes, audited and time-stamped using pre-specified diagnostic criteria. Excluded from both cohorts were patients with an SMM prior to/on admission. Model includes demographic and clinical data. One variable from all pairs of highly correlated variables (Pearson’s r ≥0.6) was removed as were variables without a univariable signal. Logistic regression model was built/ validated to predict SMM using the retrospective cohort and then validated prospectively. SMM was characterized as Composite (any SMM), Minor (transfusion of < 3 units of blood) and Major (transfusion of ≥3 units of blood or any non-transfusion SMM).The area under the receiver operating characteristic curve (AUC) was used to assess goodness of fit. Sensitivity and precision were determined.
Results:
Descriptive characteristics were similar between the retrospective (n=36,581) and prospective (n=2,940) cohorts with some exceptions.(Table) For Composite SMM, the PRSM had an AUC of 0.8 (95%CI: 0.76-0.84) on retrospective data and an AUC of 0.76 (95% CI: 0.7-0.82) on prospective validation. When the top 30% of prospective PRSM scores were flagged, the sensitivity for the prediction of Major SMM was 74% and 69% for both Composite and Minor SMM. The sensitivity increased to 90% for Major and 82% for Composite and Minor SMM when the top 40% of patients were flagged with a precision of 8%.
Conclusion:
We have built and prospectively validated an automated PRSM in the EMR to identify pregnant patients at highest risk for SMM during their delivery admission.