DrugCombo: Revolutionizing Cancer Drug Combination Trials Through Integrated Knowledge Base

Pioneering the Future of Combination Cancer Trials?

The development of effective anticancer drug combinations represents a crucial frontier in modern oncology, with combination therapies offering enhanced efficacy compared to single-agent approaches. This trend is evident in the significant increase in cancer drug combination trials registered on ClinicalTrials.gov, which nearly doubled from 905 in 2007 to 1844 in 2023. However, the path to successful combination therapies is fraught with challenges, as evidenced by the concerning 58% failure rate of Phase I clinical trials. This high failure rate underscores the critical need for comprehensive, integrated knowledge resources to guide the design of early-phase combination studies. The DrugCombo knowledge base emerges as a valuable solution to address this pressing need in oncology research, offering an unprecedented integration of pharmacokinetic, dosing, and toxicity data for both individual and combination anticancer drugs to support more informed and safer clinical trial design.

What Challenges and Opportunities Emerge from Data Fragmentation?

A fundamental challenge in designing Phase I trials for drug combinations lies in the fragmentation of essential data across numerous sources. Currently, drug toxicity information is scattered across repositories such as the FDA Adverse Event Reporting System (FAERS) and drug labels, while pharmacokinetic data is dispersed among specialized databases including the Drug Combination Database, Certera Drug Interaction Database, DrugBank, and the Pharmacogenomics Knowledgebase. This fragmentation creates significant obstacles for investigators attempting to design optimal Phase I trials, as critical dose-limiting toxicity (DLT) and maximum tolerated dose (MTD) information lacks systematic organization. An internal survey conducted at The Ohio State University Comprehensive Cancer Center revealed that while most physicians (>80%) consult sources like PubMed, meeting abstracts, and drug labels for toxicity information, fewer than 20% specifically search for pharmacokinetic data in public databases. Similarly, biostatisticians primarily rely on information provided by physicians rather than conducting independent searches across multiple data sources, highlighting the inefficiency of current information retrieval processes for trial design.

What Innovations Drive DrugCombo’s Methodology?

The DrugCombo knowledge base was developed through a sophisticated multi-component approach to data integration and curation. At its core, the database employs a tripartite knowledge model for curating Phase I clinical trial data, encompassing clinical trial metadata (trial identifiers, cancer types), design data (recruitment criteria, treatment details, statistical methodology), and outcome data (MTD and DLT findings). The biocuration workflow involved document triage using PubMed queries, followed by rigorous curator training and validation processes to ensure data consistency and quality. Five expert curators with backgrounds in pharmacology and statistics extracted data according to a standardized protocol, with particular attention to verifying ambiguous trial design methods. This meticulous manual curation process was complemented by automated approaches, including the application of a fine-tuned Microsoft Research BiomedBERT language model to extract adverse drug events (ADEs) from structured product labels, achieving an impressive F1 score of 0.90 and recall of 0.91 in ADE recognition tasks, outperforming several other language models tested under identical conditions.

For postmarketing safety surveillance data, DrugCombo incorporates comprehensive FAERS data from 2004 through 2018, processed through a rigorous methodology including drug name normalization, case deduplication, and ADE standardization using MedDRA terminology. Disproportionality analyses, including reporting odds ratios (ROR), proportional reporting ratios (PRR), and a shrinkage observed-to-expected ratio model (omega), were employed to detect statistically significant drug-ADE associations. This approach enabled the identification of nearly 1.95 million significant individual drug-ADE pairs and over 343,950 drug-drug-ADE triplets, providing valuable insights into potential combination toxicities. Additionally, pharmacokinetic knowledge, including cytochrome P450 enzyme metabolism fractions, drug clearance data, and fraction of dose excreted unchanged in urine, was manually curated from published literature and external databases, further enriching the resource's utility for anticipating potential drug-drug interactions in combination therapies.

How Extensive is the DrugCombo Knowledge Base?

The current version of DrugCombo contains an extensive collection of 14,595 drug entries, including 472 anticancer drugs and 239 FDA-approved anticancer therapies. The knowledge base encompasses 2,748 curated Phase I clinical trials covering 52 cancer types, with the majority (>99%) employing conventional 3+3 design methodology. This represents an unprecedented repository of MTD and DLT data from published trials. The database also contains 99,535 ADE terms extracted from the Adverse Reactions sections and 3,995 ADE terms from Boxed Warnings sections of structured product labels, covering 1,533 individual drug ingredients. The pharmacokinetic knowledge includes data on CYP450 metabolism for 42 anticancer drugs, fraction of dose excreted unchanged for 1,319 drugs, and clearance data for 1,638 drugs. This comprehensive integration of diverse data types positions DrugCombo as a unique resource for investigators designing Phase I combination trials, addressing a critical gap in the current research landscape.

Does the Case Study Validate DrugCombo’s Clinical Utility?

To demonstrate the practical utility of DrugCombo in clinical trial design, the authors presented a case study involving a combination of nivolumab and axitinib for advanced renal cell carcinoma. Using the DrugCombo platform, investigators could rapidly access comprehensive safety profiles for both drugs, identifying common overlapping adverse events including hypertension, anemia, and hyperglycemia, while noting the absence of severe overlapping toxicities. The platform also provided pharmacokinetic insights, indicating no significant drug-drug interactions between axitinib (primarily metabolized by CYP3A4/5) and nivolumab (a therapeutic protein degraded in target tissues). Furthermore, DrugCombo supplied historical MTD data from previous trials, showing that axitinib was tolerated at doses up to 5.0 mg and nivolumab at various fixed and weight-based dosing regimens. This integrated knowledge enabled the development of an evidence-based Phase I trial design with appropriate dosing strategies and anticipated toxicity profiles, potentially improving trial efficiency and patient safety.

Can User-Friendly Design Enhance Trial Planning?

The data access and web application features of DrugCombo are designed for user-friendly interaction with the knowledge base. The platform can be accessed via a web browser at https://drugcombo.info, with a primary focus on displaying knowledge relevant to clinical trial design. The search functionality supports standard text queries through a search box on the home page, allowing users to search for individual drugs or combinations with partial matching capabilities. Two types of filters enable users to refine results by cancer type or to display only information specific to the searched drugs. For broader exploration beyond clinical trial design, DrugCombo offers information-browsing functions through "Exploration" buttons, providing summary tables of drug targets, anatomical therapeutic chemical classifications, and mechanisms of action.

Knowledge representation in DrugCombo is structured through summary and detail pages. The summary page presents critical individual and combinational data objectively, utilizing dumbbell plots to illustrate evaluated doses and MTDs from various trials, clearly displaying drug-dose knowledge to assist researchers in selecting starting doses for Phase I trials. For individual drugs, a summary table lists severe ADEs from the Boxed Warning section of structured product labels, while combination queries provide tables summarizing overlapping ADEs between drugs and drug-drug interaction evidence. The detail page offers more comprehensive information organized in summary tables, including extracted ADEs from structured product labels and detailed DLTs with corresponding MTDs and treatment schedules from curated Phase I trial results. A separate "Clinical Trial information" page provides access to detailed trial design and metadata, accessible by clicking on PubMed ID columns in the MTD or DLT tables.

What Future Enhancements Could Propel Safer Trials?

The DrugCombo knowledge base represents a significant advancement in supporting Phase I trial design for anticancer drug combinations, yet several limitations and opportunities for enhancement remain. Currently, the platform lacks standardized quantitative tools for evaluating the toxicity of potential combinations, and could benefit from more detailed stratification of adverse event severity and prevalence, particularly in specific patient subpopulations such as elderly or pediatric cohorts. Additionally, while DrugCombo provides comprehensive information on approved drugs, it has limited capacity to predict the safety of experimental compounds still in preclinical development. Given that nearly 90% of ongoing combination trials include investigational agents, expanding coverage to include preclinical toxicity data would substantially enhance the platform's utility. Future developments could also incorporate more extensive molecular knowledge, including target pathways and functional annotations, to provide deeper insights into drug combinations at the molecular level and further advance our understanding of potential synergistic and antagonistic interactions.

How Might DrugCombo Shape Future Clinical Research?

The development of DrugCombo addresses a critical need in oncology research by consolidating fragmented information crucial for designing safe and effective Phase I combination trials. By integrating pharmacokinetic parameters, dosing data, toxicity profiles, and historical trial outcomes, this comprehensive knowledge base offers investigators an unprecedented resource to guide evidence-based decision-making in early-phase clinical trials. As cancer therapy increasingly moves toward personalized combination approaches, tools like DrugCombo that facilitate the efficient design of well-informed trials will be essential for accelerating the development of novel therapeutic strategies while minimizing patient exposure to potentially harmful treatments. How might this integrated approach to data consolidation influence the future landscape of Phase I trial design beyond oncology? Could similar knowledge integration models transform early-phase trials in other therapeutic areas with complex multidrug regimens, such as infectious diseases or neurological disorders?

As clinical researchers and trial designers begin to incorporate resources like DrugCombo into their workflow, several important questions emerge regarding the implementation and impact of such knowledge bases. How might the systematic integration of historical MTD and DLT data influence dose-escalation strategies in future combination trials? To what extent could more comprehensive pre-trial assessment of potential drug-drug interactions and overlapping toxicity profiles reduce the incidence of severe adverse events in Phase I studies? Furthermore, as artificial intelligence and machine learning continue to advance, what role might predictive modeling play in further enhancing our ability to anticipate complex drug interactions before patient exposure? These questions highlight the transformative potential of comprehensive knowledge integration in clinical research while acknowledging the ongoing need for thoughtful evaluation and refinement of these emerging tools.

Summary

DrugCombo represents a comprehensive knowledge base developed to address the high failure rate of Phase I cancer drug combination trials. The platform integrates extensive data from 14,595 drug entries, including detailed information on 472 anticancer drugs and 2,748 Phase I clinical trials. It combines pharmacokinetic parameters, toxicity profiles, and historical trial outcomes from various sources, including FDA adverse event reports and drug labels. The system employs sophisticated data curation methods, including a fine-tuned BiomedBERT language model, and provides user-friendly access to critical information for clinical trial design. While currently focused on approved drugs, future developments aim to include experimental compounds and enhanced predictive capabilities for drug interactions and toxicity profiles.