An AI-Powered Decision Support Tool for Early Detection of MIS-C
I am excited to share that I have submmitted a project proporsal to the Clinicians Leading Ingenuity IN AI Quality (CLINAQ) Fellowship Program. If accepted, I will be joining other fellows to work on this incredible project and see what new insights we're able to find.
About the AIM-AHEAD CLINAQ Fellowship Program
The AIM-AHEAD CLINAQ Fellowship Program is designed to foster leadership and significant contributions to the field of AI in medicine. The program seeks proposals that align with its objectives and have the potential to impact healthcare delivery broadly. It emphasizes engagement and collaboration within the AIM-AHEAD community, encouraging fellows to contribute to documentation and training resources, empower new users, and help cultivate a diverse community. The program also provides access to valuable resources, including datasets and structured courses and training.
My Research Proposal: An AI-Powered Early Warning System for MIS-C
My project addresses a critical unmet clinical need: the early identification of Multisystem Inflammatory Syndrome in Children (MIS-C). MIS-C is a severe, post-infectious inflammatory condition linked to SARS-CoV-2. It presents with diverse symptoms and can progress rapidly to critical illness and even mortality if not recognized and treated promptly.
What is MIS-C?
MIS-C is characterized by systemic inflammation and multi-organ dysfunction. Current diagnostic criteria are often applied after significant disease progression has occurred, highlighting the urgent need for earlier detection.
Symptoms and Diagnostics
MIS-C can manifest with a variety of symptoms, making early diagnosis challenging due to its varied clinical presentation. It is associated with evidence of recent SARS-CoV-2 infection or exposure, severe illness requiring hospitalization, multisystem organ involvement, and elevated inflammatory markers, in the absence of an alternative plausible diagnosis. Key temporal laboratory values, such as inflammatory markers (e.g., C-reactive protein, ferritin, D-dimer), cardiac biomarkers (e.g., troponin, brain natriuretic peptide), and coagulation parameters, are critical in its evolving presentation.
Treatment
While not explicitly detailed in the proposal, the goal of an early warning system is to enable earlier initiation of immunomodulatory therapies and supportive care, which can improve patient outcomes and reduce the burden on pediatric critical care units.
Current Unknowns and the Need for an Early Warning System
Despite increased awareness, early diagnosis of MIS-C remains challenging. Existing AI/ML applications have primarily focused on differentiating MIS-C from other conditions at the point of diagnosis, meaning they are reactive rather than proactive. My project aims to move beyond static risk prediction to dynamic, time-series forecasting by leveraging temporal trends in granular pediatric health data. The central hypothesis is that an AI-powered early warning system can accurately predict the onset of MIS-C prior to clinical diagnosis, enabling earlier intervention and improving patient outcomes.
My Project Aims and Approach
My research design involves a retrospective cohort study within the N3C Data Enclave, which offers a vast, diverse, and longitudinal dataset of COVID-era pediatric health data. This rich temporal data is ideal for capturing subtle, evolving physiological changes that precede a formal MIS-C diagnosis.
The project has three specific aims:
Aim 1: Establish comprehensive, temporally-rich pediatric MIS-C cohort and control groups from N3C Data Enclave. This involves meticulously defining and extracting a robust cohort of pediatric patients (0-21 years) diagnosed with MIS-C, along with appropriate control groups (SARS-CoV-2 positive children who did not develop MIS-C, and healthy pediatric controls). A crucial aspect is the extraction of longitudinal clinical data, including demographics, diagnoses, procedures, medications, vital signs, and especially temporal laboratory values.
Aim 2: Develop and validate advanced temporal AI/ML models for early MIS-C detection. This aim focuses on engineering time-series features from raw longitudinal data, transforming clinical measurements into meaningful predictive signals by capturing trends, rates of change, and variability.
Aim 3: Assess the generalizability and equity of the early warning system and disseminate findings. This final aim focuses on the translational potential and ethical implications. A key goal is to rigorously evaluate model fairness and generalizability across different demographic groups (e.g., age, race, ethnicity) and across diverse N3C contributing sites.
I am enthusiastic about the opportunity to contribute to the field of pediatric healthcare through this fellowship, with the ultimate goal of developing an AI tool that can improve patient care for children at risk of MIS-C.
More information about the program is found here: