An AI-Driven Live Systematic Reviews in the Brain-Heart Interconnectome: Minimizing Research Waste and Advancing Evidence Synthesis

Abstract: The Brain-Heart Interconnectome (BHI) combines neurology and cardiology but is hindered by inefficiencies in evidence synthesis, poor adherence to quality standards, and research waste. To address these challenges, we developed an AI-driven system to enhance systematic reviews in the BHI domain. The system integrates automated detection of Population, Intervention, Comparator, Outcome, and Study design (PICOS), semantic search using vector embeddings, graph-based querying, and topic modeling to identify redundancies and underexplored areas. Core components include a Bi-LSTM model achieving 87% accuracy for PICOS compliance, a study design classifier with 95.7% accuracy, and Retrieval-Augmented Generation (RAG) with GPT-3.5, which outperformed GPT-4 for graph-based and topic-driven queries. The system provides real-time updates, reducing research waste through a living database and offering an interactive interface with dashboards and conversational AI. While initially developed for BHI, the system's adaptable architecture enables its application across various biomedical fields, supporting rigorous evidence synthesis, efficient resource allocation, and informed clinical decision-making.

• The paper presents an AI-driven platform that automates systematic reviews in the Brain-Heart Interconnectome, reducing research waste.
• It employs a Bi-LSTM for 87% accurate PICOS detection and integrates RAG with GPT-3.5 to outperform GPT-4 in specialized queries.
• The system utilizes graph-based analytics and interactive dashboards to enable real-time evidence updates and enhance research quality.

AI-Driven Systematic Reviews in the Brain-Heart Interconnectome

The paper "An AI-Driven Live Systematic Reviews in the Brain-Heart Interconnectome: Minimizing Research Waste and Advancing Evidence Synthesis" introduces an AI-driven platform designed to enhance systematic reviews in the emergent field of the Brain-Heart Interconnectome (BHI). The interdisciplinary domain of BHI integrates cardiology and neurology to explore interactions between cardiovascular and neurological systems. However, the field faces significant challenges due to the redundancy of research efforts, suboptimal quality standards, and inefficient synthesis of evidence. The paper delineates the system's architecture and its practical implications, emphasizing the minimization of research waste and the support of high-quality evidence synthesis.

System Architecture and Methodology

The paper outlines the multipronged system infrastructure, incorporating state-of-the-art AI techniques, NLP, and graph-based analytics. Key components are:

1. Automated Population, Intervention, Comparator, Outcome, Study Design (PICOS) Detection: Utilizing a Bi-directional Long Short-Term Memory (Bi-LSTM) model, the system achieves an 87% accuracy in identifying PICOS-compliant studies. This phase prioritizes studies with rigorous methodological standards.
2. Retrieval-Augmented Generation (RAG) and LLMs: The system integrates RAG with GPT-3.5, achieving superior performance compared to plain GPT-4. This combination is particularly effective for specialized BHI queries, leveraging graph-based insights and semantic retrieval.
3. Graph-Based Querying and Topic Modeling: Neo4j stores complex relationships among interventions, outcomes, and study designs, whereas BERTopic identifies thematic clusters. This approach aides in detecting redundancies and underexplored areas while maintaining a continuously updated ‘living’ database.

The system's workflow is divided into four primary phases, following standard systematic review pipelines: define and search; screen and assess; extract and synthesize; interpret and update. Through this structured methodology, the platform streamlines the evidence synthesis process, thereby reducing the time and effort typically expended on repetitive or low-impact research.

Results and Implications

The system demonstrates substantial efficacy in improving the coherence and utility of systematic BHI reviews. By engaging domain-specific NLP and AI models, the platform mitigates the risk of inclusion of redundant or low-quality studies and facilitates real-time integration of newly published data into existing reviews. The interactive dashboards, supported by conversational AI, further enhance user accessibility, empowering researchers and clinicians to interact dynamically with the evolving knowledge base.

The paper presents notable numerical results, indicating that the RAG framework outperformed GPT-4 in 75% of domain-specific BHI inquiries. The study design classifier achieved a 95.7% overall accuracy, underscoring the system’s ability to prioritize methodologically sound research.

Future Directions and Broader Impact

While initially tailored for the BHI field, the architectural framework and methodologies discussed in this paper reveal promising potential for broader applicability across diverse biomedical domains. As AI systems continue to evolve, the adaptation of this platform to other cross-disciplinary fields encountering large volumes of research data could advance evidence-based practices significantly. This approach underscores the potential shift towards more dynamic, AI-enhanced evidence synthesis methodologies, offering versatility and scalability aligned with evolving clinical needs and research opportunities.

Potential limitations, as mentioned in the paper, include challenges related to data quality, verification processes for newly integrated studies, and scalability issues concerning data privacy and interoperability. Further research will aim to refine these areas, facilitating broader implementation and real-world applicability.

In conclusion, this paper suggests a methodologically robust and AI-enhanced platform that addresses critical challenges within BHI systematic reviews. By leveraging advanced AI and NLP techniques, it paves the way for more efficient, comprehensive, and relevant evidence synthesis, ultimately informing clinical decisions and policy-making in the biomedical field.