12  Oncology and Precision Medicine

TipLearning Objectives

Oncology sits at the forefront of AI in medicine—from imaging-based cancer detection to genomic analysis guiding targeted therapies. AI promises to accelerate precision medicine, but also faces challenges of complexity, heterogeneity, and the high stakes of cancer care. This chapter examines evidence-based AI applications across the cancer care continuum. You will learn to:

  • Evaluate AI systems for cancer screening and early detection
  • Understand AI applications in pathology and radiology for cancer diagnosis
  • Assess genomic AI tools for treatment selection and precision medicine
  • Navigate AI-assisted radiation therapy planning and delivery
  • Identify AI applications in clinical trial matching and drug discovery
  • Recognize limitations and failure modes of oncology AI
  • Apply evidence-based frameworks for adopting AI in cancer care

Essential for medical oncologists, radiation oncologists, surgical oncologists, hematologists, and cancer care teams.

The Clinical Context: Cancer is not one disease but hundreds, each with distinct biology, prognosis, and treatment. This heterogeneity makes oncology both ideal (complex pattern recognition) and challenging (limited data for rare cancers) for AI applications. The stakes are extraordinary: treatment decisions affect survival, quality of life, and financial toxicity.

AI Applications Across the Cancer Care Continuum:

12.0.1 1. Cancer Screening and Early Detection

✅ Lung Cancer Screening CT Analysis:

Clinical context: Low-dose CT screening reduces lung cancer mortality by 20% in high-risk smokers (national2011reduced?)

AI enhancement: - Automated lung nodule detection and volumetry - Lung-RADS classification assistance - Reduction in false positives

Evidence: - Multiple FDA-cleared AI systems (Aidoc, Lunit, Optellum) - Sensitivity 94-97% for significant nodules (≥6mm) (ardila2019endtoend?) - Reduces radiologist reading time by 30-40% - Decreases false positives by 11% (ardila2019endtoend?) - Published in Nature Medicine (ardila2019endtoend?)

✅ Verdict: Well-validated, clinically deployed. Improves efficiency and accuracy of lung cancer screening programs.

✅ Breast Cancer Mammography AI:

Evidence: - Deep learning matches or exceeds radiologist performance (mckinney2020international?) - Published in Nature (mckinney2020international?) - Prospective trial in Sweden: AI + single radiologist non-inferior to double-reading (Lang et al. 2023) - Reduces screening recall rates by 15-20% - Detects 10-15% more cancers in some studies

✅ Verdict: Strong evidence base. Deployment expanding in Europe. FDA-cleared systems available in US.

⚠️ Colorectal Cancer AI Colonoscopy:

Application: Real-time polyp detection during colonoscopy

Evidence: - AI-assisted colonoscopy increases adenoma detection rate (ADR) by 10-15% absolute (hassan2020performance?) - Published in Gastroenterology (hassan2020performance?) - Reduces miss rates for clinically significant polyps - FDA-cleared systems (Medtronic GI Genius, others)

Limitations: - Does not improve detection of flat or subtle serrated polyps - May increase procedure time - Cost-effectiveness debated

✅ Verdict: FDA-cleared with RCT evidence. Improves quality metric (ADR) but long-term cancer prevention benefit unproven.

12.0.2 2. Pathology AI

✅ Prostate Cancer Gleason Grading:

Application: AI analysis of prostate biopsy specimens

Evidence: - FDA-cleared system (Paige Prostate) for Gleason scoring (Pantanowitz et al. 2020) - Reduces inter-pathologist variability - Sensitivity 98% for clinically significant cancer (Grade Group ≥2) - Flags suspicious regions for pathologist review - Published in Archives of Pathology & Laboratory Medicine (Pantanowitz et al. 2020)

✅ Verdict: FDA-cleared. Augments pathologist workflow without replacing expertise.

✅ Breast Cancer Pathology:

Applications: - HER2 scoring from IHC - Ki-67 quantification - Lymph node metastasis detection

Evidence: - AI matches pathologist accuracy for HER2 2+ vs. 3+ discrimination (koopman2021digital?) - Automated Ki-67 scoring reduces variability (rimm2022artificial?) - Sentinel lymph node metastasis detection: sensitivity 92-95% (steiner2018impact?) - Published in JAMA Oncology (steiner2018impact?)

✅ Verdict: Well-validated for specific tasks. Improves standardization and efficiency.

12.0.3 3. Radiology AI for Cancer Staging

✅ Automated Tumor Segmentation and Measurement:

Applications: - Automated RECIST measurements for treatment response - Tumor volumetry (more accurate than 2D diameter) - Longitudinal tracking

Evidence: - AI-based volumetry more reproducible than manual RECIST (kostis2004threedimensional?) - Predicts treatment response earlier than conventional criteria (yoon2020prediction?) - Published in Radiology (yoon2020prediction?)

✅ Verdict: Useful for clinical trials and longitudinal monitoring. FDA-cleared systems available.

⚠️ Lymph Node Metastasis Detection:

Applications: - Automated detection of suspicious lymph nodes on CT/MRI - PET-CT analysis for staging

Evidence: - AI models show promise but variable performance (liu2021deep?) - Sensitivity 70-85%, specificity 80-90% (not sufficient to replace human interpretation) - False negatives problematic (understaging)

⚠️ Verdict: Research ongoing. Not yet reliable enough for autonomous staging decisions.

12.0.4 4. Genomic and Molecular AI

✅ Tumor Mutation Profiling and Treatment Selection:

Applications: - NGS data analysis for actionable mutations - FDA-approved targeted therapy matching - Tumor mutational burden (TMB) calculation for immunotherapy eligibility

Evidence: - AI tools accelerate variant interpretation (ainscough2018next?) - Foundation Medicine, Tempus, others use ML for treatment recommendations - Published in Nature Genetics (ainscough2018next?)

Limitations: - Interpretation of variants of unknown significance (VUS) remains challenging - Off-label treatment recommendations not always evidence-based - Insurance coverage variable

✅ Verdict: Valuable for precision oncology. Must be integrated with multidisciplinary tumor board review.

⚠️ Liquid Biopsy and Minimal Residual Disease (MRD):

Application: ctDNA analysis for MRD detection after curative-intent surgery/treatment

Evidence: - Multiple platforms (Guardant, Natera, others) show MRD predicts recurrence (tie2022circulating?) - Detects recurrence months before imaging - Published in NEJM (tie2022circulating?)

Critical gap: No RCT showing that MRD-directed interventions improve outcomes

⚠️ Verdict: Promising biomarker but clinical utility unproven. Risks overtreatment based on positive MRD without evidence that intervention helps.

12.0.5 5. Treatment Planning and Delivery

✅ Radiation Therapy AI:

Auto-Contouring: - AI-automated organ-at-risk (OAR) and tumor volume delineation - Reduces planning time from hours to minutes - Commercially available systems widely deployed

Evidence: - AI contours comparable to expert radiation oncologists for most OARs (wong2020comparing?) - Dice similarity coefficients 0.85-0.95 for critical structures - Published in International Journal of Radiation Oncology Biology Physics (wong2020comparing?)

✅ Treatment Plan Optimization: - AI-generated IMRT/VMAT plans - Knowledge-based planning using historical data - Plan quality improvements (better OAR sparing)

✅ Verdict: FDA-cleared, widely deployed. Improves efficiency and plan consistency.

⚠️ Systemic Therapy Selection AI:

Challenge: Complex decision-making involving tumor characteristics, patient factors, evidence quality, goals of care

AI approaches: - IBM Watson for Oncology (discontinued due to poor performance—see Chapter 1) - Newer systems integrating guidelines + patient data

Evidence: - Mixed at best - Concordance with oncologist decisions 50-90% depending on cancer type (somashekhar2018watson?) - Does not account for patient preferences, quality of life considerations, financial toxicity

❌ Verdict: Watson failure demonstrates dangers of premature deployment. No current AI system should autonomously recommend systemic therapy.

12.0.6 6. Clinical Trial Matching

✅ AI-Assisted Trial Eligibility Screening:

Application: Analyze EHR data to identify patients potentially eligible for clinical trials

Evidence: - Increases trial enrollment by 20-40% (ni2021increasing?) - Reduces time to identify eligible patients - Published in JCO Clinical Cancer Informatics (ni2021increasing?)

Limitations: - Eligibility algorithms only capture structured EHR data (miss nuanced exclusions) - Requires human review for final determination - Doesn’t solve root problems (trial design, access barriers, mistrust)

✅ Verdict: Useful screening tool. Should not replace detailed eligibility assessment.

12.0.7 7. Prognostication and Survival Prediction

⚠️ ML-Based Survival Models:

Applications: - Predict overall survival, progression-free survival - Integrate clinical + genomic + imaging features

Evidence: - ML models often outperform traditional nomograms (c-index 0.70-0.80 vs. 0.65-0.70) (christodoulou2019systematic?) - Meta-analysis in BMJ (christodoulou2019systematic?)

Critical limitations: - Predictions at individual patient level uncertain (wide confidence intervals) - Can’t capture all relevant factors (patient goals, social support, unmeasured confounders) - Risk of self-fulfilling prophecies (predicted short survival → less aggressive treatment → shorter survival)

Ethical concerns: - Prognostic algorithms may influence treatment intensity, hospice referral - Vulnerable to bias (if training data underrepresents certain populations) - Must not be sole basis for withholding treatment

⚠️ Verdict: May inform discussions but should not dictate treatment decisions. Communicate uncertainty transparently.

IBM Watson for Oncology: The Cautionary Tale

(Covered extensively in Chapter 1, summarized here)

What happened: - Unsafe treatment recommendations - Training on synthetic cases, not real-world evidence - Geographic inappropriateness - Oncologists lost trust

Lessons: - Precision oncology requires deep expertise, not just pattern matching - Black-box recommendations unacceptable for high-stakes decisions - Marketing ≠ clinical validation - Financial incentives can override evidence

Why oncologists must be skeptical: - Cancer treatment decisions involve tradeoffs (efficacy vs. toxicity, survival vs. QOL) - Guidelines provide frameworks, not algorithms - Patient preferences central - AI cannot replace nuanced judgment… yet

Equity and Bias in Oncology AI:

WarningCancer Disparities and Algorithmic Bias

Documented Disparities in Cancer Outcomes:

  • Black patients have higher cancer mortality across most cancer types despite similar incidence (siegel2022cancer?)
  • Hispanic patients less likely to receive guideline-concordant care (Murphy2015disparities?)
  • Rural patients face access barriers to specialized oncology care
  • Low-income patients experience financial toxicity limiting treatment adherence

How AI Can Worsen Disparities:

Training Data Bias: - Most cancer datasets from academic medical centers (affluent, insured patients) - Genomic databases overrepresent European ancestry - Imaging AI trained on specific scanner types and protocols

Examples: - Breast cancer screening AI trained predominantly on white women may have lower sensitivity in Black women (adamson2019machine?) - Genomic classifiers may misclassify variants in underrepresented populations - Treatment recommendation AI trained on insured patients may not account for financial toxicity concerns

Mitigation: - Require diverse training datasets - Validate across demographic subgroups - Report performance stratified by race, ethnicity, SES - Address root causes of disparities (access, bias, social determinants)

Clinical Guidelines for Oncology AI:

TipASCO Principles for AI in Oncology

Before Adopting Oncology AI:

  1. Demand high-quality evidence:
    • Prospective validation studies
    • External validation in diverse populations
    • Clinical outcomes (not just prediction accuracy)
  2. Ensure transparency:
    • Explainable AI (especially for treatment decisions)
    • Clear description of training data
    • Known failure modes disclosed
  3. Maintain human oversight:
    • AI assists, never replaces, oncologist judgment
    • Multidisciplinary tumor board review remains standard
  4. Assess equity:
    • Performance in underrepresented populations
    • Access considerations (cost, technology requirements)
  5. Consider patient preferences:
    • Some patients prefer human-only decision-making
    • Informed consent when AI significantly influences care

Safe Implementation:

  • Pilot testing in low-stakes applications first
  • Parallel validation (AI + standard approach)
  • Clear escalation pathways for AI-human disagreement
  • Systematic monitoring for bias and errors
  • Patient feedback mechanisms

Red Flags:

❌ Claims of autonomous treatment decision-making ❌ No validation in diverse populations ❌ Black-box recommendations without rationale ❌ Vendor resistance to independent evaluation ❌ Replacing rather than augmenting tumor boards

12.1 Conclusion

AI in oncology holds immense promise—from earlier cancer detection to personalized treatment selection to accelerating drug discovery. But IBM Watson’s failure demonstrates the perils of premature deployment. Oncologists must demand rigorous evidence, transparent algorithms, and proof of clinical benefit before integrating AI into high-stakes cancer care decisions.

The goal is not just better predictions, but better outcomes for patients—especially those from communities bearing disproportionate cancer burdens.

12.2 References