Table of Contents
- Key Points
- Background: Why This Research Matters
- How the Research Was Conducted
- What Studies Were Included
- AI for Treatment Recommendations and Decision Support
- AI for Drug Dosing and Precision Medicine
- AI for Treatment Adherence and Patient Management
- Challenges and Concerns in AI-Driven Decision-Making
- Ethical and Legal Issues
- What This Means for Patients
- Limitations of the Research
- Recommendations
- Frequently Asked Questions
- Source Information
Key Points
- In reviewed studies, AI matched expert cancer treatment recommendations in many cases, but performed better in straightforward cases and struggled with complex ones, so human oversight remains essential.
- Patients have the right to ask whether AI is used in their care, how their data is used, and what AI's limitations are; informed consent should cover these points.
- The review identified concerns including unclear accountability for AI-guided decisions, data privacy risks, algorithmic bias, and AI systems that are hard to explain.
- AI showed promise in personalizing drug dosing, improving adherence to oral anticancer treatments, and predicting symptoms like pain and fatigue with varying accuracy.
- Researchers recommend robust informed consent models, clear legal accountability, diverse training data, and equitable access as AI becomes more integrated into oncology.
Background: Why This Research Matters
Artificial intelligence (AI) is a fast-growing family of technologies. These technologies are used in many fields, including economics, environmental science, and manufacturing. In healthcare, AI has the potential to change both research and clinical practice. AI can automate tasks, improve efficiency, and make more accurate predictions based on advanced data analysis.
The data used by AI can include genetic information, proteomic data (information about proteins in the body), clinical records, imaging results from radiology, pathology, and endoscopy, and demographic information. In other words, AI can use virtually every part of patient care that we are able to gather, measure, and repeat.
Machine learning (ML) models accumulate information, detect patterns, and provide predictions or recommendations based on input data. However, these models are only as effective as the data used to train them.
Healthcare professionals must stay informed about technological advancements. They must also engage with the new and evolving ethical and legal implications of applying AI to medicine. This includes identifying, understanding, and regulating issues related to:
- Data privacy
- Patient consent
- Algorithmic transparency (how AI systems make their decisions)
- Accountability for decisions made by AI systems
Automated decision-making and profiling have the potential to compromise individual rights and freedoms. This underscores the need for an appropriate and robust legal framework. Knowledge of these areas will enable clinicians to minimize the risk of harm, ensure equitable access, and protect patient rights.
Similarly, patients should be made aware of their rights regarding data security and informed consent. Patients need to understand how their health data is used in treatment decisions.
The concept of "big data" continues to evolve. Yet the fundamental questions it seeks to answer remain consistent. For oncologists, the question is: what are the most effective treatments to recommend? For patients, the question is: what is the best therapeutic strategy for my individual needs?
A shared awareness of the increasing role of data-driven decisions is essential. This awareness helps ensure that big data is used responsibly, maintaining patient autonomy, privacy, and equality in care.
This systematic review specifically examined the ethical, legal, and informed consent challenges associated with clinical trials involving AI-driven therapeutic decision-making in oncology. The review focused on how these considerations affect clinicians and patients as AI becomes integral to cancer treatment.
How the Research Was Conducted
The researchers reviewed available literature on AI models used for aiding therapeutic decisions in oncology. They highlighted the ethical and legal challenges of integrating these technologies into patient care. The goal was to identify areas lacking transparency and explainability, and to discuss the barriers hindering ethical and patient-centered implementation of AI in oncology decision-making.
The systematic review followed the 2020 PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist and guidelines. The researchers adhered fully to the methodology of the Cochrane Handbook. They followed a study protocol that was initially registered on PROSPERO with the registration number CRD420251046482.
Eligibility criteria: The researchers conducted a systematic review of literature published between January 2015 and May 2025. The review explored how AI technologies are applied in therapeutic decision-making for solid tumors. The researchers paid particular attention to ethical, legal, and consent-related implications.
The search was intentionally broad and inclusive in terms of methodology. This allowed for a comprehensive understanding of the diversity of AI interventions and their clinical and ethical ramifications. The researchers included studies employing qualitative, quantitative, or mixed-method approaches. This encompassed experimental, observational, cross-sectional, and qualitative designs.
To be eligible, studies had to meet several criteria:
- They were required to implement at least one AI intervention in the context of systemic treatment decision-making in oncology. This could be a specific subset such as machine learning, deep learning, or natural language processing.
- The included studies needed to focus exclusively on cancer patients.
- The studies needed to address at least one dimension related to the ethical, legal, or informed consent aspects of using AI in therapeutic settings.
- Only articles written in English and available in full-text form were considered.
The researchers excluded studies where AI was applied solely in non-therapeutic domains. These excluded domains included cancer diagnosis, histopathology, imaging, surgery, or radiotherapy without reference to systemic treatment planning. They also excluded review articles, case reports, and publications limited to conference abstracts or posters. These were excluded because of their limited methodological transparency and peer-review status.
The overall search strategy was conducted following the PICO framework:
- Population (P): Cancer patients undergoing AI-supported therapeutic decision-making
- Intervention (I): The use of AI-based tools in medical oncology to aid treatment selection
- Comparison (C): Conventional care relying on clinician-led decisions without AI input
- Outcome (O): The clinical adoption of AI decision-support systems during oncology consultations, whereby physicians actively incorporated algorithm-generated insights into patient management
Search strategy: The systematic search was conducted on May 5 in three databases: CENTRAL Cochrane Library, PubMed, and EMBASE. The search strategy used a combination of MeSH terms arranged in a specific search string. The search string combined terms for oncology, cancer, or oncologists with terms for artificial intelligence, machine learning, AI, or deep learning. It also combined terms for decision, treatment, therapy, treatment planning, personalized oncology, or clinical trial with terms for ethics, informed consent, legal, or data privacy.
Selection process: The selection of studies was performed independently by three reviewers. All identified references were initially imported into EndNote 20 to eliminate duplicates. Titles and abstracts were then screened using Rayyan, followed by a full-text review. At each selection stage, inter-rater agreement was assessed by calculating Cohen's kappa coefficient. Any discrepancies between the three reviewers were resolved through consultation with a fourth independent reviewer. After relevant articles were selected for inclusion, the reference list and citations of each article were inspected for additional eligible publications.
Data collection process: Data extraction was conducted independently by three reviewers, with disagreements adjudicated by a fourth reviewer. A standardized data extraction sheet was employed to systematically collect key information from eligible studies. This information included article title, first author, year of publication, and Digital Object Identifier (DOI). It also included population demographics and particularities, AI application type, and key findings on therapeutic decision-making impact. Ethical, legal, and consent challenges were also included.
What Studies Were Included
A total of 429 records were identified through the systematic literature search. Following screening and eligibility assessment, 15 studies met the inclusion criteria and were included in the final analysis.
Out of 15 studies, both medical oncologists and patients with various types of cancer were included in the study populations. The AI intervention consisted of:
- Generative artificial intelligence such as large language models (in four studies)
- Supervised machine learning, for instance random survival forest (in one study)
- Explainable AI techniques (in one study)
- A machine learning-based web application (in two studies)
The primary impact of therapeutic decision-making was assessed for each study. The studies were subsequently classified into thematic groups for the analysis of ethical, legal, and informed consent challenges.
Of the chosen papers, all of them addressed AI-related ethical issues. However, only eleven covered legal concerns, and ten addressed the informed consent process.
AI for Treatment Recommendations and Decision Support
AI models have exhibited strong performance in aligning with expert recommendations for cancer treatment. In work done by Lazris and colleagues in 2024, ChatGPT-3.5 provided case-specific treatment recommendations in 81% of cases. The overall treatment strategy concordance was 83%. However, exact treatment plan concordance was lower at 65%. The AI system had challenges in chemotherapy regimen recommendations and follow-up protocols.
Similarly, AI-based models demonstrated high sensitivity and accuracy in classifying cancer risk. This supported clinicians in making informed treatment decisions.
Ng and colleagues in 2023 described how they designed ADBoard to improve decision-making in multidisciplinary meetings (MDMs). Their tool ensured the completeness of patient information and enhanced the explainability of decision protocols. The adoption of AI in MDM is expected to streamline workflow, reduce administrative burdens, and allow clinicians to focus on complex cases.
Action point: Future practices should focus on integrating AI precision in chemotherapy recommendations, enhancing multidisciplinary decision transparency, and streamlining clinical workflows for oncology treatment support.
AI for Drug Dosing and Precision Medicine
AI-driven platforms have shown promise in personalized dose optimization. The CURATE.AI platform successfully provided individualized dosing recommendations. The platform aimed to enhance therapeutic decision-making by optimizing drug administration.
Tan and colleagues in 2021 conducted a pilot study to evaluate the viability of CURATE.AI. The study monitored the timeliness of dose recommendations and the frequency of clinically significant dose modifications. The results suggested feasibility for broader implementation in precision oncology. Secondary outcomes included physician adherence to AI-recommended doses.
Furthermore, AI-assisted molecular tumor boards (MTBs) demonstrated close alignment with ideal treatment plans, which are also called consensus annotations. The AI model often outperformed conventional MTB recommendations. The AI model facilitated standardized MTB discussions, reduced physician workload, and supported precision oncology frameworks that could be expanded globally.
Action point: To effectively formulate precision medicine treatment recommendations, MTBs must navigate the vast array of molecular alterations and immune markers unique to each patient's tumor. Integrating AI and ML into MTB workflows could significantly enhance accessibility, facilitate standardized implementation, and enable routine adoption even in smaller clinical centers.
AI for Treatment Adherence and Patient Management
Masiero and colleagues in 2024 found that AI-driven clinical decision support systems (DSS) could improve adherence to oral anticancer treatments by predicting patient behavior. A machine learning-integrated DSS was evaluated for its effectiveness in promoting medication adherence. Secondary objectives included identifying new predictive variables to refine adherence behavior models. The study anticipates that improved adherence will positively influence clinical outcomes and reduce the economic burden of nonadherence.
Similarly, AI was utilized to support personalized patient follow-up strategies. This enhanced patient engagement and optimized clinician decision-making regarding supportive care interventions. These AI-assisted approaches are expected to improve quality of life by reducing treatment-related toxicity and unnecessary interventions.
Beyond direct therapeutic decision-making, AI has shown potential in predicting nonvisible symptoms relevant to palliative care. AI models achieved varying levels of accuracy, ranging from 55.5% to 88.0%, in predicting symptoms such as:
- Pain
- Dyspnea (difficulty breathing)
- Fatigue
- Delirium (sudden confusion)
- Anxiety
These predictive capabilities could help optimize symptom assessment and management. They could assist clinicians in delivering early interventions for patients requiring supportive and palliative care.
Action point: AI models could help shape multiple different patient profiles to assist the medical team in better identifying individual issues, needs, and expectations. Once addressed, these might improve treatment results beyond the generally predicted values obtained from clinical trials and real-world registries.
Challenges and Concerns in AI-Driven Decision-Making
While AI has demonstrated potential in optimizing therapeutic decisions, several concerns remain. Clinicians acknowledged AI's role in validating their prognostic judgments and guiding treatment discussions. However, they expressed reservations regarding AI accuracy, variability in patient responses, and the risk of over-reliance on algorithmic predictions.
According to Stalp and colleagues in 2024, ChatGPT's treatment recommendations were generally sufficient. High accuracy was observed in HER2-positive breast cancer cases. The AI performed better in primary, non-complicated scenarios, particularly for chemotherapy. However, its limitations became evident in complex and postoperative cases. In these cases, the AI struggled to provide proper chronological treatment sequences and precise recommendations. This underscores the need for human oversight in intricate therapeutic decisions.
Li and colleagues in 2024 conducted a cross-survey study to explore Chinese oncologists' perspectives on integrating AI into clinical practice. Cancer care providers expressed concerns about AI integration, particularly regarding:
- Its potential to mislead diagnoses and treatments
- Over-reliance on its recommendations
- Risks related to data security
- Algorithm bias
- Patient privacy
Additionally, many highlighted the slow adaptation of regulations to AI advancements. Views on AI's impact on the doctor-patient relationship were mixed. Some feared increased disputes. Opinions on whether AI might replace physicians remained divided, with no clear consensus.
In their qualitative study, Hesjedal and colleagues in 2024 emphasized various opinions regarding AI's role in therapeutic decision-making. Scientists pointed out the need for robust validation and high-quality training datasets. Clinicians underscored the necessity of human oversight in AI-generated recommendations. Patients exhibited caution, expressing trust in physicians to ensure AI's reliability and safety in clinical practice. Ethical and regulatory concerns, including data security, algorithm bias, and AI's potential impact on the doctor-patient relationship, were also highlighted.
Evaluations are most often carried out retrospectively on small datasets. These datasets may not accurately represent the language, demographic, and epidemiological characteristics of patients encountered in clinical practice.
Ethical and Legal Issues
The review identified several ethical and legal challenges that must be addressed for safe and equitable implementation of AI in oncology.
Algorithmic transparency and explainability: Many AI systems function as "black boxes," meaning their decision-making processes are not easily understood. This lack of transparency makes it difficult for clinicians to explain treatment recommendations to patients. It also complicates the informed consent process.
Accountability: When AI-guided decisions lead to harm, it is often unclear who is legally responsible. The responsibility could fall on the physician, the hospital, the AI developer, or a combination of parties. This ambiguity creates challenges for both patients seeking redress and clinicians using these tools.
Data privacy: AI systems require large amounts of patient data to function effectively. Protecting this data from unauthorized access and ensuring it is used ethically remains a significant concern. Patients need to understand how their health data is used in treatment decisions.
Informed consent: Traditional informed consent models may not adequately address the complexities of AI-assisted decision-making. Patients need to understand:
- What role AI plays in their treatment recommendations
- What data is being used and how
- What the limitations of AI recommendations are
- That they have the right to ask questions about AI involvement in their care
Algorithmic bias: AI models are only as good as the data used to train them. If training data does not represent diverse patient populations, the AI may produce biased recommendations that disadvantage certain groups. This raises concerns about equity in cancer care.
Regulatory gaps: Many clinicians highlighted the slow adaptation of regulations to AI advancements. Current legal frameworks may not adequately address the unique challenges posed by AI in healthcare.
What This Means for Patients
AI is increasingly being integrated into cancer care. This means that patients may encounter AI-assisted tools during their treatment journey. Understanding what this means can help patients participate more actively in their care.
AI can help doctors in several ways:
- Providing treatment recommendations that align with expert opinions in many cases
- Personalizing drug dosing to optimize effectiveness
- Predicting symptoms that may not be immediately visible
- Supporting follow-up strategies and patient engagement
However, AI is not a replacement for human judgment. The review found that AI performs better in straightforward cases and struggles with complex scenarios. This means that human oversight remains essential in cancer treatment decisions.
Patients should be aware that:
- AI recommendations are tools to assist doctors, not replace them
- They have the right to ask whether AI is being used in their care
- They have the right to understand how their data is being used
- They should feel empowered to ask questions about their treatment plan
The review emphasizes that patient autonomy, privacy, and equality in care must be maintained as AI becomes more integrated into oncology.
Limitations of the Research
This systematic review has several limitations that patients and clinicians should consider.
First, the review included only 15 studies. This relatively small number reflects the emerging nature of AI applications in oncology decision-making. More research is needed to draw definitive conclusions.
Second, evaluations of AI systems are most often carried out retrospectively on small datasets. These datasets may not accurately represent the language, demographic, and epidemiological characteristics of patients encountered in clinical practice. This limits the generalizability of findings.
Third, the studies included in this review used various types of AI interventions. This heterogeneity makes it difficult to compare results across studies directly.
Fourth, the review focused specifically on therapeutic decision-making for solid tumors. Findings may not apply to other areas of oncology or to non-solid tumors.
Fifth, the rapid pace of AI development means that some findings may become outdated quickly. New AI tools and applications are continuously being developed.
Recommendations
Based on this review, the researchers offer several recommendations for the responsible integration of AI into oncology.
For healthcare systems and policymakers:
- Develop robust informed consent models that adequately address AI involvement in treatment decisions
- Establish clear legal accountability frameworks for AI-guided decisions
- Create ethical frameworks and regulatory mechanisms to protect patient autonomy
- Address algorithmic bias through diverse training datasets and ongoing monitoring
- Ensure equitable access to AI-enhanced cancer care
For clinicians:
- Maintain human oversight of AI-generated recommendations
- Communicate clearly with patients about the role of AI in their care
- Stay informed about the ethical and legal implications of AI in medicine
- Use AI as a tool to support, not replace, clinical judgment
For patients:
- Ask your healthcare team whether AI is being used in your treatment planning
- Request information about how your data is being used
- Understand that AI recommendations are tools to assist your doctors
- Feel empowered to ask questions about your treatment plan
- Report any concerns about privacy or data security to your healthcare provider
The researchers emphasize that future research must establish ethical frameworks and regulatory mechanisms. These are needed to protect patient autonomy and responsibly integrate AI into oncology.
Frequently Asked Questions
Can artificial intelligence really help decide my cancer treatment?
In the reviewed studies, AI matched expert treatment recommendations in many cases. For example, one 2024 study found ChatGPT-3.5 gave case-specific recommendations in 81% of cases, with overall strategy agreement of 83%. AI also helped personalize drug dosing and predict symptoms. However, AI is a tool to assist doctors, not replace them, and human oversight remains essential.
What are the risks of using AI in cancer treatment decisions?
The review identified several concerns: AI systems can be 'black boxes' with unclear decision-making, making it hard for doctors to explain recommendations. Legal accountability for harm is often unclear. Data privacy is a significant worry. AI may also produce biased recommendations if training data lacks diversity. Additionally, AI struggles with complex cases, so human oversight is critical.
How accurate is AI at recommending cancer treatment?
Accuracy varies. In one 2024 study, ChatGPT-3.5 provided case-specific recommendations in 81% of cases, with 83% overall strategy agreement but only 65% exact treatment plan agreement. It performed better in primary, non-complicated cases, especially for chemotherapy, but struggled with complex and postoperative cases. AI models also predicted symptoms like pain and fatigue with 55.5% to 88.0% accuracy.
Can AI help with my drug dosing and follow-up?
Yes, in some studies. The CURATE.AI platform provided individualized dosing recommendations in a 2021 pilot study, showing feasibility for broader use. AI-driven decision support systems also improved adherence to oral anticancer treatments by predicting patient behavior. AI-assisted molecular tumor boards aligned closely with ideal treatment plans and sometimes outperformed conventional recommendations, potentially reducing physician workload.
What should I ask my doctor about AI in my treatment?
You can ask whether AI is being used in your treatment planning, what role it plays, what data is used and how, and what the limitations are. You have the right to understand how your data is used and to ask questions about your treatment plan. Report any privacy or data security concerns to your healthcare provider. These questions help you participate actively in your care.
Is AI regulated for use in cancer care?
The review found that many clinicians highlighted the slow adaptation of regulations to AI advancements. Current legal frameworks may not adequately address unique challenges like algorithmic transparency, accountability for AI-guided decisions, and data privacy. The researchers recommend developing robust informed consent models, clear legal accountability frameworks, and ethical and regulatory mechanisms to protect patient autonomy and ensure equitable access.
If AI is helping decide my cancer treatment, when should I get a second opinion?
Ask for a second opinion when AI is part of your treatment planning, especially if your case is complex or postoperative. AI recommendations align with expert opinions in many cases, but accuracy is higher in straightforward, primary scenarios and lower in complex cases, where chronological treatment sequences and precise recommendations become difficult. AI is a tool to assist doctors, not replace them, and human oversight remains essential. You have the right to ask whether AI is used in your care and how your data is used. Diagnostic Detectives Network provides independent expert second opinions.
Source Information
Original article title: Artificial Intelligence and Decision-Making in Oncology: A Review of Ethical, Legal, and Informed Consent Challenges.
Authors: Froicu EM, Creangă-Murariu I, Afrăsânie VA, Gafton B, Alexa-Stratulat T, Miron L, Pușcașu DM, Poroch V, Bacoanu G, Radu I, Marinca MV.
Publication: Current Oncology Reports (2025) 27:1002–1012
DOI: https://doi.org/10.1007/s11912-025-01698-8
Accepted: May 29, 2025
Published online: June 17, 2025
Registration: PROSPERO CRD420251046482
This patient-friendly article is based on peer-reviewed research.
