Beyond Initial Treatment: Navigating Recurrence Risk and Clinical Decisions in Cancer Survivors
By Léon Van Wouwe, Clinical Innovation Director, Volv Global Emerging artificial intelligence solutions offer promising strategies to address critical challenges in recurrence risk assessment and decision-making post-curative cancer treatment. Clinical management of cancer survivors following curative treatment presents a complex set of challenges, particularly in accurately assessing recurrence risk and making informed decisions about the necessity of additional therapies. These clinical complexities are compounded by evolving surveillance strategies, substantial patient heterogeneity, and inherent limitations in current evidence bases and conventional risk assessment tools. Key Clinical and Diagnostic Challenges After curative treatment, pinpointing the precise moment when a patient transitions into being “at risk” for recurrence is often unclear. Clinicians face difficult decisions regarding the optimal timing to initiate surveillance and risk assessment, whether immediately upon confirmation of disease-free status, upon completion of treatment, or at another clinical milestone altogether. The chosen time scale for follow-up and survival analysis, be it from initial diagnosis, the start of treatment, or the confirmation of being disease-free, can significantly influence clinical interpretations and research outcomes. (1) Surveillance regimens currently vary widely depending on cancer type, individual patient risk factors, and continually evolving international guidelines. Although the overarching goal of surveillance is to detect recurrence at a stage where intervention could be most effective, the evidence supporting intensive surveillance remains mixed. Additionally, the benefits of earlier detection regarding survival rates and overall quality of life remain uncertain for many cancer types. Notably, recurrence may be discovered either through routine surveillance or symptomatic presentation. Often, there exists a critical window – characterised by increased healthcare consultations in the months before recurrence diagnosis – where earlier identification may indeed be feasible. (2) The assessment of recurrence risk is further complicated by several critical factors: Lack of population-level data: Most data on recurrence and disease-free survival originate from clinical trials, typically involving younger, healthier patients who might not reflect the broader community population. (4) Variability in guideline adherence: Differences in interpretation and application of risk assessment guidelines by healthcare professionals can lead to significant underestimations or misclassifications of recurrence risk. For instance, guideline-consistent and clinician-assessed risk levels often diverge significantly in breast cancer cohorts.(3) Patient heterogeneity: Factors including age, tumour stage, molecular subtype, treatment history, and comorbidities profoundly influence recurrence risk, as well as the appropriateness of further therapeutic interventions. (5) Limitations and Opportunities in Risk Assessment Tools Traditional risk assessments primarily depend on clinicopathologic features, biomarkers, and imaging technologies. While indispensable, these tools often lack the sensitivity required to detect minimal residual disease effectively and might not fully capture recurrence risk complexity, particularly within diverse patient groups. (2, 4) Emerging techniques, such as circulating tumour DNA (ctDNA), advances in PET imaging, and new genomic markers, among other methods, are increasingly incorporated into surveillance and risk stratification. However, their optimal application and overall impact on clinical outcomes remain under active investigation. (2) The integration of artificial intelligence (AI) and machine learning (ML) presents a substantial opportunity to advance clinical practice. Recent studies in other fields, such as ML-based models predicting progression from community-acquired pneumonia (CAP) to acute respiratory distress syndrome (ARDS), demonstrate that outcome risk prediction models can achieve remarkable accuracy and clinical utility. These innovative models exploit large, heterogeneous datasets and advanced analytics, offering the potential to identify high-risk patients significantly earlier and more precisely than conventional methods. Translating such sophisticated approaches into oncology could fundamentally enhance recurrence risk assessment by: Seamlessly integrating diverse clinical, genomic, and imaging data to deliver individualised predictions. Identifying phenotypic subgroups and previously unknown risk factors. Supporting more precise stratification of patients for decisions regarding adjuvant therapies. Decision-Making for Adjuvant Therapy Determining the initiation of adjuvant or additional therapies following curative treatment necessitates a careful balancing act. Clinicians must weigh the recurrence risk against potential adverse effects of overtreatment, such as treatment-related toxicity, financial burden, and impacts on patient quality of life. (4, 6, 8) Decision-making complexities are further magnified by: Patient values and preferences: Integrating patient perspectives is essential, given the broad variability in individual risk tolerance and treatment goals. (2, 9) Guideline variability: Divergences in clinical practice guidelines, driven by evolving evidence and methodologies, complicate the standardisation of care and the assurance of optimal outcomes. (5) Communication gaps: Many patients remain unaware of their personal recurrence risk or lack detailed discussions with healthcare providers, underscoring the necessity of improved risk communication and shared decision-making. (9) Summary Table: Clinical Management Challenges Challenge/Factor Current Limitation/Status Emerging Solution/Trend Example Scenario Defining “at risk” period Unclear clinical milestones for risk onset Standardised time points, AI-driven analytics Post-treatment surveillance timing Risk assessment tools Limited sensitivity/specificity, population bias Genomic profiling, ctDNA, ML/AI models Breast cancer recurrence risk Surveillance strategies Mixed evidence for intensive follow-up Individualised, risk-adapted surveillance Lung cancer regular scans Guideline adherence Divergence between clinician and guideline assessments Education, decision-support tools Breast cancer risk underestimation Patient heterogeneity Diverse risk profiles, comorbidities Personalised medicine, AI stratification Elderly or multimorbid patients Communication and shared decisions Gaps in risk discussion and preference integration Patient-centered risk communication Survivorship care planning Conclusion: The Potential of AI-Driven Outcome Prediction Assessing recurrence risk and making informed adjuvant therapy decisions post-curative cancer treatment remains fraught with clinical and operational difficulties. Traditional tools and guidelines, while vital, often fail to capture the full complexity of recurrence risk across diverse patient populations. AI and ML technologies, exemplified by successful outcome prediction models in conditions like CAP to ARDS, represent a transformative advancement for oncology. These methods can synthesise vast, heterogeneous datasets, delivering precise individualised risk predictions. Such innovation will support more accurate therapeutic decisions, reduce overtreatment, and enhance patient quality of life, ultimately fostering more confident, patient-centric care for cancer survivors. About the author Léon van Wouwe has 20+ years’ global experience in clinical development and operations, uniting data science with pharma and research. He drives cross-functional collaboration to advance innovative treatments. References American Cancer Society. (n.d.). Follow-up care after cancer treatment. Retrieved from https://www.cancer.org/cancer/survivorship/long-term-health-concerns/importance-of-follow-up-care.html American Society of Clinical Oncology (ASCO). (2024). Guideline concordance