INTEGRATING MACHINE LEARNING AND GENERATIVE AI ACROSS THE SYSTEMATIC REVIEW WORKFLOW - AN AI-ENHANCED SLR OF POST AMI CARDIOVASCULAR RISK AND CARDIOVASCULAR INFLAMMATION
Author(s)
Adam Nelson, MD1, Julie T. Mortensen, MSc, PhD2, Iram Muhammad, MD2, Jag Chhatwal, PhD3, Carol Kirshner, MS, MSc4, Kevin Bainey, MD5;
1University of Adelaide, Adelaide, Australia, 2Novo Nordisk, Søborg, Denmark, 3Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA, 4Value Analytics Labs, Boston, MA, USA, 5University of Alberta, Edmonton, AB, Canada
1University of Adelaide, Adelaide, Australia, 2Novo Nordisk, Søborg, Denmark, 3Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA, 4Value Analytics Labs, Boston, MA, USA, 5University of Alberta, Edmonton, AB, Canada
OBJECTIVES: Patients surviving a Type 1 acute myocardial infarction (AMI) have substantial residual risk of recurrent cardiovascular (CV) events, but 1- and 5-year estimates vary by population, endpoint, and AMI subtype. Persistent post-AMI CV inflammation, reflected by elevated levels of high-sensitivity C-reactive protein (hsCRP) and interleukin-6 (IL-6), appears prognostically relevant yet insights have been limited to small cohorts limiting generalizability. We aimed to perform a systematic literature review (SLR) to: (1) report 1- and 5-year risks of recurrent CV events after Type 1 AMI; and (2) report hsCRP/IL-6 trajectories post MI and their associations with recurrent CV outcomes. Novel to our approach, we paired machine-learning-assisted screening with generative-AI-supported extraction and reporting under expert validation.
METHODS: We conducted a protocol-driven PRISMA 2020 SLR. MEDLINE and Embase were searched (January 1, 2015-June 28, 2025). Two reviewers screened titles/abstracts and full texts with PICO Portal ML re-ranking and early stopping at a validated 95% recall threshold. After inclusion, GenAI supported structured extraction summaries, draft risk-of-bias text (NOS, ROBINS-I), and reporting. ML was limited to screening and GenAI to synthesis tasks; all outputs were human-validated.
RESULTS: Searches yielded 6,813 records; 138 studies were included. ML predictions achieved 95% sensitivity (95% CI 93-97) and 64% specificity (95% CI 62-65); NPV was >99% (PPV 21.9%). GenAI accelerated and standardized extraction and risk-of-bias write-up. Preliminary synthesis shows substantial recurrent CV risk after Type 1 AMI. Although few studies reported inflammatory factors, hsCRP/IL-6 levels were generally associated with higher recurrent events.
CONCLUSIONS: Evidence to date indicates Type 1 AMI survivors show substantial recurrent CV risk persisting to 5 years. Persistently elevated hsCRP/IL-6 are associated with higher recurrence suggesting a role for CV inflammation. Finally, a modular ML-GenAI workflow improved SLR efficiency and consistency with full expert validation.
METHODS: We conducted a protocol-driven PRISMA 2020 SLR. MEDLINE and Embase were searched (January 1, 2015-June 28, 2025). Two reviewers screened titles/abstracts and full texts with PICO Portal ML re-ranking and early stopping at a validated 95% recall threshold. After inclusion, GenAI supported structured extraction summaries, draft risk-of-bias text (NOS, ROBINS-I), and reporting. ML was limited to screening and GenAI to synthesis tasks; all outputs were human-validated.
RESULTS: Searches yielded 6,813 records; 138 studies were included. ML predictions achieved 95% sensitivity (95% CI 93-97) and 64% specificity (95% CI 62-65); NPV was >99% (PPV 21.9%). GenAI accelerated and standardized extraction and risk-of-bias write-up. Preliminary synthesis shows substantial recurrent CV risk after Type 1 AMI. Although few studies reported inflammatory factors, hsCRP/IL-6 levels were generally associated with higher recurrent events.
CONCLUSIONS: Evidence to date indicates Type 1 AMI survivors show substantial recurrent CV risk persisting to 5 years. Persistently elevated hsCRP/IL-6 are associated with higher recurrence suggesting a role for CV inflammation. Finally, a modular ML-GenAI workflow improved SLR efficiency and consistency with full expert validation.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR151
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory)