ISPOR 2026: Identifying Drug Repurposing Opportunities with AI Literature Review

AI-Assisted Drug Repurposing

What if the next breakthrough treatment is already available today? It could be hidden within an existing drug, waiting to be discovered. This idea drives the growing field of AI-assisted drug repurposing. The approach looks for new therapeutic uses for approved medications.

Traditional drug development can cost billions and take more than a decade. In contrast, repurposing builds on drugs that already have established safety profiles, manufacturing processes, and real-world data.

However, finding new uses for existing drugs is difficult. Promising signals may appear as secondary endpoints, subgroup analyses, or unexpected observations. These insights can be scattered across thousands of scientific papers.

At ISPOR 2026 in Philadelphia, MadeAi presented a compelling proof-of-concept study. The study showed how a structured literature review can systematically identify hidden signals. It also demonstrated how this approach can speed up hypothesis generation while maintaining scientific rigor and regulatory standards.

Challenges in Drug Repurposing

Drug repurposing is not new. Sildenafil, for example, was originally developed for angina and later repurposed for erectile dysfunction. Thalidomide was also repurposed for multiple myeloma. These examples show the potential of drug repurposing. However, finding new opportunities remains difficult.

Researchers must keep up with a rapidly growing volume of biomedical literature. PubMed alone adds hundreds of thousands of papers each year. Manual reviews are slow and expensive. They may also miss subtle but important signals, especially when those findings are not the main focus of a study.

Traditional systematic literature reviews (SLRs) for repurposing can take months. Teams must screen large numbers of records and manage differences between reviewers. They must also connect findings across different therapeutic areas. As a result, promising opportunities may take longer to identify. This can slow innovation and delay the development of new treatment options. To address these challenges, the MadeAi team developed and tested an AI-assisted structured literature review methodology.

Iron supplementation was selected as the proof-of-concept. Its use in anemia is well established, but existing research also suggests potential applications beyond anemia.

AI-Assisted Drug Repurposing Methodology: AI–Human Approach

The hybrid methodology combines the speed of AI with the knowledge of human experts. AI can screen large volumes of research and help identify patterns across studies. Meanwhile, specialists review the findings and provide medical judgment.

This approach allows AI to handle repetitive, high-volume tasks while experts focus on interpretation and validation. In this proof-of-concept, the complete review was finished in just 84 hours.

AI – Assisted Literature Review Workflow

Key steps include:

• Protocol Development: Human reviewers established and approved the eligibility criteria, PICOS framework, and search strategy.
• Literature Search: The PubMed search identified 2,140 records.
• Deduplication: AI tools supported duplicate detection. No duplicates were identified in this case.
• Title and Abstract Screening: AI supported study selection, followed by human validation. Of the 2,140 records, 1,287 were excluded. This left 853 records for full-text review.
• Full-Text Screening: AI helped prioritize relevant articles. The process resulted in 46 final studies, including 13 in anemia patients and 33 in non-anemia populations.
• Data Extraction and Synthesis: AI supported the extraction of off-label benefits and risks. Subject-matter experts then reviewed and interpreted all findings.

Speed was one of the key results. The entire process, from protocol development to final qualitative synthesis, took just 84 hours.

AI screening accuracy reached 85%. This result highlights the value of a hybrid approach. AI can manage volume and identify patterns, while experts provide medical judgment and interpretation.

Findings: Uncovering the Therapeutic Signals

The review identified signals across several therapeutic areas.

The findings included improvements in quality of life and clinical outcomes. These signals appeared in areas such as cardiovascular disease, chronic kidney disease, pregnancy, and inflammatory bowel disease. The results show how structured AI-assisted reviews can help identify therapeutic signals that may be difficult to detect through manual review alone.

Safety Profile Consistent With Known Risks. No Widespread Novel Safety Concerns Identified.

In Anemia Patients (13 studies):

• Improvements in fertility, quality of life (QoL), and psychological outcomes.
• Better postoperative recovery, mobility, muscle strength, and exercise capacity.
• Reduced hospital stays, infection risk, and fibromyalgia-related symptoms in certain settings.
• Better outcomes in gynecological surgeries and other procedures.

In Non-Anemia Populations (33 studies):

• Cardiovascular Disease (13 studies): Findings included reduced hospitalization and improved cardiac function, exercise capacity, and ventricular performance. Some settings also showed a lower risk of heart failure severity and all-cause mortality.
• Chronic Kidney Disease (7 studies): Findings included better fatigue management, quality of life, and functional outcomes. Studies also reported fewer cardiovascular events and hospitalizations.
• Pregnancy (5 studies): Reported benefits included fewer complications, better postpartum recovery, and lower rates of depression.
• Inflammatory Bowel Disease (4 studies): Findings included anti-inflammatory effects, improved somatic growth, and reduced disease activity.
• Cancer and Other Areas: Studies reported shorter hospital stays, better functional status, and supportive benefits during treatment.
• Additional signals were identified in HIV, orthopedics, and immunological disorders. These included improvements in infection outcomes and survival measures.

Ferric carboxymaltose emerged as a leading formulation. Across populations, studies reported functional and healthcare utilization benefits.

Importantly, the reported risks were generally mild and consistent with known safety profiles. These included gastrointestinal discomfort and injection-site reactions. This consistency added confidence to the identified repurposing signals.

Performance Table: AI-Assisted Literature Review Efficiency

AI-Assisted Drug Repurposing – Connecting Research to Results

This approach can support pharmaceutical companies, biotech firms, HEOR teams, and medical device developers in several ways.

First, it can shorten the time needed to generate hypotheses. Work that once took months may be completed in days. This can support faster go/no-go decisions for repurposing programs.

Second, AI can help teams review a larger body of evidence. It can surface subtle signals that may be difficult to identify manually. At the same time, source tracking can support review and regulatory scrutiny.

Third, the approach can reduce repetitive work. Experts can spend less time screening and extracting data. Instead, they can focus on interpreting findings, planning follow-up studies, and developing clinical strategies.

The approach may be especially relevant to cardiovascular disease, nephrology, women’s health, and chronic disease research. For example, the iron supplementation case study shows how literature signals can inform research into new indications, combination therapies, or supportive care strategies. The methodology also reflects a broader shift toward responsible AI in regulated environments. AI provides speed and scale, while human experts provide oversight and scientific judgment. Together, these elements can help organizations balance efficiency with trust.

Broader Implications and Future Directions

This proof-of-concept goes beyond a single drug. It presents a framework that could be applied to other compounds and therapeutic areas. As the volume of scientific literature grows, AI-assisted tools may become increasingly useful for evidence generation.

The next step could involve connecting literature insights with other types of evidence. These may include real-world evidence, clinical trial results, and social data. Multimodal analysis could then provide a broader view of potential repurposing opportunities.

At the same time, emerging AI evaluation frameworks can help teams assess these tools. Key areas include accuracy, factuality, reliability, and readiness for real-world use. For organizations, structured AI support can turn literature review into a faster and more scalable process. However, human oversight remains essential when evidence is used to guide scientific and clinical decisions.

Author’s Note: This article was supported by AI-based research and writing, with Claude 4.5 assisting in the creation of text and images.