From weeks to hours: Driving speed and intelligence in epidemiology-based forecasting with AI agents

Strategic forecasting is the foundation for data-driven decision-making in the life sciences. It allows pharmaceutical companies to predict the potential of pipeline assets by considering epidemiology, the competitive landscape, and patient drug adoption. This approach shapes critical strategies from clinical trial design to market access plans and promotional budgets by analyzing key healthcare metrics like disease prevalence, diagnosis rates, and therapy adoption.

In epidemiology-based forecasting, the accuracy of key measures such as disease prevalence, diagnosis, and treatment rates, also known as the "top of the funnel," dictates the reliability of downstream projections. Traditionally, researchers derive these inputs from secondary research by scouring scientific articles, synthesizing data, and accounting for biases. The issue with these typical secondary research exercises is that they do not provide measures of their comprehensiveness, and they struggle with reconciling numbers from various sources. Systematic literature review (SLR) is a widely used research protocol that ensures research is reproducible and comprehensive. While SLRs bring rigor and structure to epidemiology research, their manual execution introduces several critical challenges that can compromise efficiency. Apart from being time-consuming, SLRs suffer from other challenges, including:

• Resource intensity: Conducting rigorous SLRs demands significant resources, creating barriers for widespread adoption in fast-paced research environments. These include:
  • Time Burden: It can take a team of researchers 2-3 weeks to complete a single SLR, potentially delaying critical decision-making in drug development or public health planning.
  • Specialized Researchers: Interpreting disease-specific nuances in SLRs requires trained epidemiologists, which can create bottlenecks due to limited numbers of domain experts.
  • Cost Constraints: The manual screening of thousands of studies and the associated data extraction inflate project budgets, often making SLRs a cost-intensive exercise for organizations with tight budgets.
  • Opportunity Cost: Teams usually prioritize speed over rigor, sacrificing SLRs for quicker but less reliable alternatives when deadlines are tight.
    
    

  These persistent resource challenges force organizations to make difficult trade-offs between scientific rigor and operational realities, highlighting the need for innovative approaches to evidence synthesis.

• Subjectivity: A fundamental principle of SLRs is that they should yield the same results if repeated. However, when conducted manually, analysts often introduce  variability. Different researchers may:
  • Apply subjective judgments when determining study relevance, leading to inconsistent inclusion or exclusion criteria.
  • Extract different data points from the same paper based on individual interpretations.
  • Weigh biases differently, such as selection or publication bias, thus altering the final output.
    
    

  This inconsistency reduces the reliability of forecasting models, as two teams reviewing identical literature could produce different epidemiological results.

• Comprehensiveness Concerns: Manual SLRs struggle to guarantee exhaustive coverage of existing research due to:
  • Search Limitations: Researchers may rely on a limited group of databases or miss foreign-language publications.
  • Selection Bias: Analysts may unintentionally prioritize studies that align with their hypotheses.
  • Dynamic Evidence Sources: New studies emerge constantly, while manual reviews can’t easily incorporate real-time updates.
    
    
• Time and Expertise Bottlenecks: Conducting a rigorous SLR is resource-intensive and requires experts with deep domain knowledge. This dependence can be limiting for the SLR process in several ways:
  • Scarcity of Specialists: Accurate SLRs require epidemiologists who understand disease-specific nuances such as evolving biomarker-defined cancer subtypes. Such expertise is scarce and costly.
  • Time Delays: A single review can take 2–3 weeks with a team of 2–3 analysts, delaying forecasting cycles.
  • Opportunity Costs: Teams often skip SLRs for time-sensitive projects, resorting to less rigorous methods that increase forecasting uncertainty.
    
    

  Bottlenecks like those listed above hinder agility in making informed decisions, particularly for early-phase drugs or emerging diseases where rapid insights are critical. Errors in the "top of the funnel,” like overestimating diagnosis rates, cascade into flawed revenue or resource projections.

  Systematic literature reviews typically address these issues by enforcing structured review protocols, ensuring consistency, transparency, and the aggregation of findings. But their manual execution remains a barrier. Forecasting timelines often don’t accommodate week-long literature reviews.

  While natural language processing (NLP) and machine learning (ML) have been applied to systematic literature reviews, their effectiveness has been constrained by the need for extensive labeled training data and an inability to adapt to the nuanced, evolving nature of medical research. These approaches often struggle with context-aware reasoning and domain-specific biases, limiting their reliability for comprehensive evidence synthesis. The inherent limitations of conventional NLP/ML methods have prevented them from fully automating SLRs while maintaining the rigor required for epidemiology and clinical research, creating the need for more adaptive AI solutions.

The critical bottleneck in systematic literature reviews (SLRs) has always been the need for human expertise. Agentic AI represents a fundamental breakthrough by breaking the SLR process into specialized tasks, each handled by dedicated AI agents with distinct capabilities. Some fragments of the overall SLR process, which agentic AI can optimize, are illustrated below:

• Question decomposition agents can:
  • Analyze and reorganize the research question into parts.
  • Identify key variables and parameters needed for the review.
• Paper screening agents can:
  • Review abstracts and titles against the inclusion criteria.
  • Assign relevance scores with documented reasoning.
• Data extraction agents can:
  • Pull key numerical data from studies and appropriate data sources.
  • Standardize metrics across different study formats.
• Meta-analysis agents can:
  • Statistically synthesize results from multiple studies.
  • Assess heterogeneity and effect sizes across datasets.
  • Generate forest plots and other visualization outputs.

There are several benefits to using agentic AI for SLRs. First, this framework ensures transparency. Unlike traditional black-box machine learning models, each specialized agent maintains clear documentation of its reasoning process, enabling researchers to trace and validate every decision. Agentic AI offers exceptional flexibility. The system can seamlessly adapt to diverse disease areas and evolving research questions without extensive retraining or recalibration. It also preserves crucial human oversight. Subject matter experts can intervene at any stage to adjust parameters, refine criteria, or validate findings, maintaining scientific rigor while benefiting from automation. These advantages create a robust SLR solution combining AI efficiency with human expertise.

The potential efficiency gains are substantial:

• From Weeks to Hours: Axtria’s internal testing reveals that an SLR process step that could take a team of analysts 2-3 weeks to complete can be finished in under a day using AI agents.
• Reduced Reliance on Niche Expertise: AI agents encode domain knowledge, allowing teams to scale SLR with one expert playing a supervisory role, instead of teams of human experts.
• Transparency and Auditability: Unlike black-box ML models, agents provide reasoning at each step, enabling researchers to trace decision pathways.

These efficiencies produce faster, more reliable inputs for epidemiology forecasting and accelerate strategic decisions on market potential, clinical trial design, and resource allocation.

While AI-powered SLRs represent a leap forward, the broader vision is a suite of specialized agents automating each forecasting step, from patient funnel modeling to competitive intelligence and revenue projections. Traditionally, answering a critical forecasting question could take weeks. With AI agents, the time required could shrink to days.

This shift isn’t just about efficiency; it’s about enabling real-time strategic decision-making processes in fast-moving fields like infectious disease modeling, rare disease epidemiology, and therapeutic market forecasting. As AI agents evolve, they’ll further bridge the gap between data complexity and actionable insights, redefining how strategic forecasting powers the life sciences industry.

Integrating agentic AI into literature reviews marks a paradigm shift in epidemiology forecasting. By automating labor-intensive processes while preserving scientific rigor, these systems unlock unprecedented speed and scalability, turning SLRs from a bottleneck into a strategic accelerator. The future of epidemiology research isn’t just faster, it’s smarter.

FAQs