In a retrospective study, the machine learning tool was able to screen for potential risks of cardiovascular disease nearly 60 days before the patient's medical record showed any signs of a related condition or before they were officially diagnosed or treated for it.
Invaryant's machine learning (ML) technology facilitated the early identification of high-risk cardiovascular conditions during pregnancy by detecting signals and trends in patients’ medical records before they were recognized by health care providers.
The ML algorithm, Healthy Outcomes for all Pregnancy Experiences-Cardiovascular-Risk Assessment Technology (HOPE-CAT), was able to screen for potential cardiovascular disease risks by a mean (SD) of 56.8 (69.7) days earlier than the first date of diagnosis or intervention of a related condition reported in a patient’s electronic health record (EHR). These findings were published in JMIR Cardio.
The Georgia-based health technology company’s ML algorithm analyzed retrospective data from a large health care system's EHRs in a virtual server environment, leveraging deidentification and standardization to ensure unbiased data analysis. Clinical experts in cardio-obstetrics selected risk factors, which were used to train HOPE-CAT iteratively, incorporating relevant literature and current risk identification standards. Following refinement, HOPE-CAT generated risk profiles for each patient, distinguishing between standard and high risk, and these profiles were matched with clinical outcomes related to cardiovascular pregnancy conditions, calculating the time difference between the risk profile date and the actual diagnosis or intervention in the EHR.
The analysis included 6069 patients aged between 18 and 40 years at their initial pregnancy-related visit, each having more than 1 pregnancy-related medical encounter. Patients were excluded if they had limited EHR data, only 1 pregnancy-related visit, or if the first medical encounter in the EHR lacked corresponding pregnancy information due to being solely a birth record.
A total of 604 pregnancies leading to childbirth had records or diagnoses that could be compared with the risk profile, with most patients being identified as Black (79.8%) and aged between 21 and 34 years (84.4%). Preeclampsia was the most prevalent condition among the cohort (90.6%), followed far behind by thromboembolism (2.7%) and acute kidney disease or failure (2.2%). On average (SD), there was a 56.8 (69.7) day gap between the identification of risk factors by HOPE-CAT and the initial diagnosis or intervention of a related condition recorded in the EHR. Notably, HOPE-CAT demonstrated its highest effectiveness in early detection of myocardial infarction, with a mean (SD) delta of 65.7 (81.4) days.
According to the authors, this retrospective investigation contributes to the growing body of literature on ML applications in clinical settings and addresses the scarcity of ML research in obstetrics, with previous reviews highlighting limited focus on this field. While ML has shown promise in predicting conditions like preeclampsia and hypertensive disorders in obstetrics, the current study offers a unique approach by examining HOPE-CAT's timeliness in risk assessment rather than focusing solely on predictive parameters. By introducing a novel tool for real-time monitoring of cardiovascular risk during pregnancy, the authors said these findings underscore the potential of ML to enhance perinatal care for high-risk patients.
“Pregnancy-related disorders are not only associated with complications during pregnancy, but they also portend future cardiometabolic and long-term cardiovascular-related morbidity,” the authors noted. “Implicit racial biases contribute to these health inequities, resulting in increased maternal morbidity and mortality when Black women with valid and important health concerns are dismissed. Enabling early and effective screening for pre-existing comorbidities and early identification of risk with enhanced technological applications, like HOPE-CAT, independent of patient characteristics and descriptors or provider bias, has the potential to mitigate factors leading to racial biases.”
This study faced several limitations, including that it relied solely on EHRs, lacking access to patients' complete health histories and potentially omitting valuable unstructured data such as clinical notes, which could have impacted result interpretation. Additionally, the need to clean and standardize data from multiple EHRs highlighted the nonstandardized nature of EHR data, despite not diminishing the efficacy of the ML algorithm. Further, the retrospective nature of the study—confined to data within one health care system and spanning the COVID-19 pandemic—may have introduced biases such as delayed diagnosis due to pandemic-related care barriers, and the inability to analyze data by race, ethnicity, or age could have overlooked important demographic differences in risk assessment.
According to the authors, future ML endeavors may incorporate various data sources like unstructured data from natural language processing, wearables, and remote patient monitoring devices, enabling a more comprehensive understanding of patient health. Additionally, integrating social determinants of health factors into ML models could aid in addressing health equity issues and combatting the widening racial and ethnic disparities seen in US maternal mortality rates.
“To facilitate the reversal of this trend, it is imperative that risk identification occurs earlier in the pregnancy trajectory to allow for increased monitoring and referral to more specialized care and that ML technology is leveraged to support maternal health screening in routine appointments,” the authors concluded. “The results from this study of ML through HOPE-CAT provide foundational evidence to develop solutions to mitigate the harmful impacts of pregnancy and improve maternal health for all.”
While Invaryant's HOPE-CAT has undergone technical validation, the authors also said further research is necessary to establish its clinical validity and ensure its effectiveness in improving obstetric outcomes.
Reference
Shara N, Mirabal-Beltran R, Talmadge B, et al. Use of machine learning for early detection of maternal cardiovascular conditions: retrospective study using electronic health record data. JMIR Cardio. 2024;8:e53091. doi:10.2196/53091
Elevated Stress Hyperglycemia Ratio Linked to Higher All-Cause Mortality After Cardiovascular Events
November 20th 2024Evidence highlights the predictive value of the stress hyperglycemia ratio for mortality risk in patients with acute myocardial infarction, ischemic stroke, and acute heart failure.
Read More
New Study Finds Risk Groups, Outpatient Care Barriers in Chronic Liver Disease
November 20th 2024Patients with chronic liver disease who were unable to establish care were 85% more likely to require recurrent hospitalizations. This group included a disproportionate number of women and individuals with physical limitations affecting their health.
Read More
Why Right Heart Catheterization Confirming PAH Diagnosis May Be Underperformed
November 20th 2024Professional guidelines say that when pulmonary arterial hypertension (PAH) is diagnosed, right heart catheterization should be performed, but a quarter of the time, it isn’t—so investigators set out to discover why.
Read More
OS Better With Belantamab Mafodotin Triplet vs Daratumumab in R/R MM
November 19th 2024The key secondary end point of overall survival (OS) was met in the DREAMM-7 trial of belantamab mafodotin (Blenrep; GSK) for the treatment of patients with relapsed/refractory multiple myeloma (R/R MM).
Read More