Cross-Domain Text Mining to Predict Adverse Events from Tyrosine Kinase Inhibitors for Chronic Myeloid Leukemia

Area of Science:

• Oncology
• Pharmacovigilance
• Computational Biology

Background:

• Tyrosine kinase inhibitors (TKIs) are crucial in treating chronic myeloid leukemia (CML) and other cancers.
• Identifying and ranking TKI-related adverse events (AEs), including under-reported ones, is essential for patient safety.
• Current methods may not fully capture the spectrum of potential TKI toxicities.

Purpose of the Study:

• To predict and rank tyrosine kinase inhibitor (TKI)-related adverse events (AEs) using novel text mining techniques.
• To identify both known and previously unreported AEs associated with TKIs.
• To develop a tiered surveillance framework for proactive patient monitoring based on AE risk.

Main Methods:

• Utilized k-means clustering on 2575 CML TKI abstracts to initially group AEs.
• Employed a novel cross-domain text mining approach integrating knowledge graphs, link prediction, and network analysis.
• Analyzed over 30 million articles via SemNet to predict and rank TKI AEs.
• Applied unsupervised rank aggregation for feature importance to establish physiology-based tiers.

Main Results:

• Clustering identified distinct AE profiles for specific TKIs (e.g., gastrointestinal for bosutinib, edema for imatinib).
• Cross-domain text mining predicted and ranked known and novel TKI AEs across three tiers based on physiological relevance and frequency.
• Tier 1 (top 1%) included hematologic, glucose, iron, cardiovascular, and thyroid AEs; Tier 2 (top 5%) included inflammatory and kidney AEs; Tier 3 (top 10%) included gastrointestinal, neuromuscular, and other AEs.
• Ranked AEs included anemia, diabetes, hypertension, hypothyroidism, glomerulonephritis, hepatitis, neuropathy, and secondary cancers.

Conclusions:

• The developed text mining methodology effectively predicts and ranks TKI-related AEs.
• The study proposes a tiered surveillance strategy: regular for Tier 1, infrequent for Tier 2, and symptom-based for Tier 3 AEs.
• This approach can enhance proactive patient monitoring and management of TKI toxicities in clinical practice.