A machine learning risk prediction model for acute peripartum cardiovascular complications during delivery admissions

Journal club

1. Title of the study Journal – Journal of American College of Cardiology Published on May 20,2024 PARCCS A Machine Learning Risk-Prediction Model for Acute Peripartum Cardiovascular Complications During Delivery Admissions Salman Zahid , Shikha Jha , Gurleen Kaur , Youn-Hoa Jung, Anum S. Minhas , Allison G. Hays, Erin D. Michos

2. Key points Maternal mortality in the US remains high, with cardiovascular (CV) complications being a leading cause (>20%). The CDC estimate that 4 out of 5 pregnancy-related deaths in the US are preventable. The rise in maternal mortality is believed to be a consequence of several factors, including advanced maternal age, pre-existing medical conditions like DM and hypertension more individuals with CHD surviving to childbearing age.

3. Developing accurate risk prediction models is crucial to identify individuals at high risk for adverse peripartum CV events. These models have the potential to enable health care professionals to implement appropriate preventive measures and interventions to reduce maternal mortality due to CV complications.

4. The aim of this study is to develop a machine learning risk prediction model PARCCS (Prediction of Acute Risk for Cardiovascular Complications in the Peripartum Period Score) to predict CV complications during delivery admissions. AIMS AND OJECTIVES:

5. METHODS- Data from the National Inpatient Sample (2016-2020). Data from 7 million inpatient hospitalizations in 47 states and the District of Columbia (97% of US population). Acute CV/renal complications were defined as a composite of pre- eclampsia / eclampsia , peripartum cardiomyopathy , renal complications, venous thromboembolism , arrhythmias and pulmonary edema. A risk prediction model, PARCCS, was developed using machine learning consisting of 14 variables and scored out of 100 points.

6. OUTCOME- The primary outcome was acute peripartum CV complications during delivery admission among pregnant individuals.

7. ANALYSIS AutoScore was utilized to automatically develop clinical scoring models for specific outcomes. Data was split randomly into three subsets: training, validation and testing subsets with a ratio of 0.7, 0.2, and 0.1, respectively. The training set was used to create the scores, the validation set was used for interim evaluation, and the test set was used to determine the final model’s performance metrics.

8. RESULTS- The analysis included a total of 2,371,661 pregnant patients at delivery hospitalization, of whom 93% had no CV complications and 7% had CV complications. Patients with CV complications had a higher prevalence of co-morbidities and were more likely to be of Black race and lower income. Using the validation set, the performance of the model was evaluated with an AUC of 0.68 and a 95% CI of 0.67 to 0.68.

9. As the predicted risk cutoff increases, the proportion of patients identified as high-risk decreases, but the specificity and PPV increase. At the same time, the sensitivity and NPV decrease. Therefore, choosing the appropriate cutoff depends on the balance between the risks of false positives (unnecessary interventions for low-risk patients) and false negatives (missed opportunities to intervene for high-risk patients).

10. A higher AUC value implies better performance, with a perfect classifier having an AUC value of 1.0. The value of 0.68 suggests that the model has a moderate ability to distinguish between the positive and negative classes.

11. DISCUSSION- CDC’s Pregnancy Mortality Surveillance System data (2017-2019) - causes contributing to pregnancy-related deaths. 14.5% attributed to CV conditions 12.1% for cardiomyopathy and hemorrhage each. 14.3% for Infections or sepsis 10.5% for thrombotic pulmonary or other embolism 6.3% for hypertensive disorders 5.8% for cerebrovascular accidents 7% unknown causes.

12. Advantages of model- It incorporates both established and emerging risk factors. It is the first to specifically target CV complications during delivery. It is useful in handling large amounts of complex data and identifying patterns that may not be easily recognizable by human analysis. Incorporating socioeconomic and racial/ethnic variables can improve risk prediction accuracy and address health disparities in maternal health outcomes.

13. 5. Incorporated cardiometabolic risk factors related to pregnancy. 6. This model was developed using an inpatient dataset that is representative of the diverse population of the US. 7. Strength of the model is an NPV of 96% demonstrating its ability to correctly identify individuals who do not have acute peripartum CV complications at delivery.

14. LIMITATIONS: Machine learning algorithms can produce biased predictions as it does not include minority ethnic groups and socioeconomic disparities. Absence of granular data pertaining to the timing of outcomes. The dataset lacks detailed information on the temporal aspect of adverse CV events during pregnancy. It is a binary model, distinguishing only whether a peripartum CV complication occurred or not. Limited sample size .

15. 6. It did not include COVID-19 infection as a variable, although it has been associated with CV complications during pregnancy. 7. Non-elective admission or electrolyte disorder may be surrogates for patients with acute cardiac morbidity at the time of admission and represent a “red flag,” which is less helpful in defining the need for early screening. 8. Some risk factors identified have been recognized from standard retrospective analysis—underlying comorbidities such as diabetes, renal disease, hypertension, pre-existing heart failure, or stroke—and already should have prompted intensive follow-up for CVD during pregnancy.

16. FUTURE SCOPE- Income was a variable included in this model, the scope of socioeconomic characteristics could be broadened. Incorporating educational level attainment and other socio-economic indicators, such as occupation or neighborhood characteristics, may offer a more comprehensive understanding of the multifaceted factors influencing CV outcomes during pregnancy.

17. CONCLUSIONS PARCCS has the potential to be an important tool for identifying pregnant individuals at risk of acute peripartum CV complications at the time of delivery. Future studies should further validate this score and determine whether it can improve patient outcomes.

18. THANK YOU