A Quantitative Paradigm for Decision-Making in Precision Oncology

Intricacies of Human-AI Interaction in Dynamic Decision-Making for Precision Oncology: A Case Study in Response-Adaptive Radiotherapy

Background

Adaptive treatment strategies that can dynamically react to individual cancer progression can provide effective personalized care. Longitudinal multi-omics information, paired with an artificially intelligent clinical decision support system (AI-CDSS) can assist clinicians in determining optimal therapeutic options and treatment adaptations. However, AI-CDSS is not perfectly accurate, as such, clinicians' over/under reliance on AI may lead to unintended consequences, ultimately failing to develop optimal strategies. To investigate such collaborative decision-making process, we conducted a Human-AI interaction case study on response-adaptive radiotherapy (RT).

Methods

We designed and conducted a two-phase study for two disease sites and two treatment modalities-adaptive RT for non-small cell lung cancer (NSCLC) and adaptive stereotactic body RT for hepatocellular carcinoma (HCC)-in which clinicians were asked to consider mid-treatment modification of the dose per fraction for a number of retrospective cancer patients without AI-support (Unassisted Phase) and with AI-assistance (AI-assisted Phase). The AI-CDSS graphically presented trade-offs in tumor control and the likelihood of toxicity to organs at risk, provided an optimal recommendation, and associated model uncertainties. In addition, we asked for clinicians' decision confidence level and trust level in individual AI recommendations and encouraged them to provide written remarks. We enrolled 13 evaluators (radiation oncology physicians and residents) from two medical institutions located in two different states, out of which, 4 evaluators volunteered in both NSCLC and HCC studies, resulting in a total of 17 completed evaluations (9 NSCLC, and 8 HCC). To limit the evaluation time to under an hour, we selected 8 treated patients for NSCLC and 9 for HCC, resulting in a total of 144 sets of evaluations (72 from NSCLC and 72 from HCC). Evaluation for each patient consisted of 8 required inputs and 2 optional remarks, resulting in up to a total of 1440 data points.

Results

AI-assistance did not homogeneously influence all experts and clinical decisions. From NSCLC cohort, 41 (57%) decisions and from HCC cohort, 34 (47%) decisions were adjusted after AI assistance. Two evaluations (12%) from the NSCLC cohort had zero decision adjustments, while the remaining 15 (88%) evaluations resulted in at least two decision adjustments. Decision adjustment level positively correlated with dissimilarity in decision-making with AI [NSCLC: ρ = 0.53 ( p < 0.001); HCC: ρ = 0.60 ( p < 0.001)] indicating that evaluators adjusted their decision closer towards AI recommendation. Agreement with AI-recommendation positively correlated with AI Trust Level [NSCLC: ρ = 0.59 ( p < 0.001); HCC: ρ = 0.7 ( p < 0.001)] indicating that evaluators followed AI's recommendation if they agreed with that recommendation. The correlation between decision confidence changes and decision adjustment level showed an opposite trend [NSCLC: ρ = -0.24 ( p = 0.045), HCC: ρ = 0.28 ( p = 0.017)] reflecting the difference in behavior due to underlying differences in disease type and treatment modality. Decision confidence positively correlated with the closeness of decisions to the standard of care (NSCLC: 2 Gy/fx; HCC: 10 Gy/fx) indicating that evaluators were generally more confident in prescribing dose fractionations more similar to those used in standard clinical practice. Inter-evaluator agreement increased with AI-assistance indicating that AI-assistance can decrease inter-physician variability. The majority of decisions were adjusted to achieve higher tumor control in NSCLC and lower normal tissue complications in HCC. Analysis of evaluators' remarks indicated concerns for organs at risk and RT outcome estimates as important decision-making factors.

Conclusions

Human-AI interaction depends on the complex interrelationship between expert's prior knowledge and preferences, patient's state, disease site, treatment modality, model transparency, and AI's learned behavior and biases. The collaborative decision-making process can be summarized as follows: (i) some clinicians may not believe in an AI system, completely disregarding its recommendation, (ii) some clinicians may believe in the AI system but will critically analyze its recommendations on a case-by-case basis; (iii) when a clinician finds that the AI recommendation indicates the possibility for better outcomes they will adjust their decisions accordingly; and (iv) When a clinician finds that the AI recommendation indicate a worse possible outcome they will disregard it and seek their own alternative approach.

Deep reinforcement learning identifies personalized intermittent androgen deprivation therapy for prostate cancer

The evolution of drug resistance leads to treatment failure and tumor progression. Intermittent androgen deprivation therapy (IADT) helps responsive cancer cells compete with resistant cancer cells in intratumoral competition. However, conventional IADT is population-based, ignoring the heterogeneity of patients and cancer. Additionally, existing IADT relies on pre-determined thresholds of prostate-specific antigen to pause and resume treatment, which is not optimized for individual patients. To address these challenges, we framed a data-driven method in two steps. First, we developed a time-varied, mixed-effect and generative Lotka-Volterra (tM-GLV) model to account for the heterogeneity of the evolution mechanism and the pharmacokinetics of two ADT drugs Cyproterone acetate and Leuprolide acetate for individual patients. Then, we proposed a reinforcement-learning-enabled individualized IADT framework, namely, I$^{2}$ADT, to learn the patient-specific tumor dynamics and derive the optimal drug administration policy. Experiments with clinical trial data demonstrated that the proposed I$^{2}$ADT can significantly prolong the time to progression of prostate cancer patients with reduced cumulative drug dosage. We further validated the efficacy of the proposed methods with a recent pilot clinical trial data. Moreover, the adaptability of I$^{2}$ADT makes it a promising tool for other cancers with the availability of clinical data, where treatment regimens might need to be individualized based on patient characteristics and disease dynamics. Our research elucidates the application of deep reinforcement learning to identify personalized adaptive cancer therapy.

Treatment of evolving cancers will require dynamic decision support

Cancer research has traditionally focused on developing new agents, but an underexplored question is that of the dose and frequency of existing drugs. Based on the modus operandi established in the early days of chemotherapies, most drugs are administered according to predetermined schedules that seek to deliver the maximum tolerated dose and are only adjusted for toxicity. However, we believe that the complex, evolving nature of cancer requires a more dynamic and personalized approach. Chronicling the milestones of the field, we show that the impact of schedule choice crucially depends on processes driving treatment response and failure. As such, cancer heterogeneity and evolution dictate that a one-size-fits-all solution is unlikely-instead, each patient should be mapped to the strategy that best matches their current disease characteristics and treatment objectives (i.e. their 'tumorscape'). To achieve this level of personalization, we need mathematical modeling. In this perspective, we propose a five-step 'Adaptive Dosing Adjusted for Personalized Tumorscapes (ADAPT)' paradigm to integrate data and understanding across scales and derive dynamic and personalized schedules. We conclude with promising examples of model-guided schedule personalization and a call to action to address key outstanding challenges surrounding data collection, model development, and integration.

MuTATE-an R package for comprehensive multi-objective molecular modeling

MOTIVATION

Comprehensive multi-omics studies have driven advances in disease modeling for effective precision medicine but pose a challenge for existing machine-learning approaches, which have limited interpretability across clinical endpoints. Automated, comprehensive disease modeling requires a machine-learning approach that can simultaneously identify disease subgroups and their defining molecular biomarkers by explaining multiple clinical endpoints. Current tools are restricted to individual endpoints or limited variable types, necessitate advanced computation skills, and require resource-intensive manual expert interpretation.

RESULTS

We developed Multi-Target Automated Tree Engine (MuTATE) for automated and comprehensive molecular modeling, which enables user-friendly multi-objective decision tree construction and visualization of relationships between molecular biomarkers and patient subgroups characterized by multiple clinical endpoints. MuTATE incorporates multiple targets throughout model construction and allows for target weights, enabling construction of interpretable decision trees that provide insights into disease heterogeneity and molecular signatures. MuTATE eliminates the need for manual synthesis of multiple non-explainable models, making it highly efficient and accessible for bioinformaticians and clinicians. The flexibility and versatility of MuTATE make it applicable to a wide range of complex diseases, including cancer, where it can improve therapeutic decisions by providing comprehensive molecular insights for precision medicine. MuTATE has the potential to transform biomarker discovery and subtype identification, leading to more effective and personalized treatment strategies in precision medicine, and advancing our understanding of disease mechanisms at the molecular level.

AVAILABILITY AND IMPLEMENTATION

MuTATE is freely available at GitHub (https://github.com/SarahAyton/MuTATE) under the GPLv3 license.

Predictive nonlinear modeling of malignant myelopoiesis and tyrosine kinase inhibitor therapy

Chronic myeloid leukemia (CML) is a blood cancer characterized by dysregulated production of maturing myeloid cells driven by the product of the Philadelphia chromosome, the BCR-ABL1 tyrosine kinase. Tyrosine kinase inhibitors (TKIs) have proved effective in treating CML, but there is still a cohort of patients who do not respond to TKI therapy even in the absence of mutations in the BCR-ABL1 kinase domain that mediate drug resistance. To discover novel strategies to improve TKI therapy in CML, we developed a nonlinear mathematical model of CML hematopoiesis that incorporates feedback control and lineage branching. Cell-cell interactions were constrained using an automated model selection method together with previous observations and new in vivo data from a chimeric BCR-ABL1 transgenic mouse model of CML. The resulting quantitative model captures the dynamics of normal and CML cells at various stages of the disease and exhibits variable responses to TKI treatment, consistent with those of CML patients. The model predicts that an increase in the proportion of CML stem cells in the bone marrow would decrease the tendency of the disease to respond to TKI therapy, in concordance with clinical data and confirmed experimentally in mice. The model further suggests that, under our assumed similarities between normal and leukemic cells, a key predictor of refractory response to TKI treatment is an increased maximum probability of self-renewal of normal hematopoietic stem cells. We use these insights to develop a clinical prognostic criterion to predict the efficacy of TKI treatment and design strategies to improve treatment response. The model predicts that stimulating the differentiation of leukemic stem cells while applying TKI therapy can significantly improve treatment outcomes.

Feasibility Study of a Network Meta-Analysis and Unanchored Population-Adjusted Indirect Treatment Comparison of Niraparib, Olaparib, and Bevacizumab as Maintenance Therapies in Patients with Newly Diagnosed Advanced Ovarian Cancer

Selecting a first-line (1L) maintenance option for ovarian cancer is challenging given the variety of therapies, differing trials, and the lack of head-to-head data for angiogenesis and poly(ADP-ribose) polymerase (PARP) inhibitors. Thus, indirect treatment comparisons (ITCs) can aid treatment decision making. This study assessed the feasibility of two ITCs, a network meta-analysis (NMA) and a population-adjusted ITC (PAIC), comparing the efficacy of the PARP inhibitor niraparib in the PRIMA trial (NCT02655016) with other 1L maintenance treatments. A systematic literature review was conducted to identify trials using the Cochrane Handbook for Systematic Reviews of Interventions to assess differences in trial design, population characteristics, treatment arms, and outcome measures. All 12 trials identified were excluded from the NMA due to the absence of a common comparator and differences in survival measures and/or inclusion criteria. The PAIC comparing PRIMA and PAOLA-1 trials was also not feasible due to differences in inclusion criteria, survival measures, and the previous receipt of chemotherapy/bevacizumab. Neither ITC met recommended guidelines for analysis; the results of such comparisons would not be considered appropriate evidence when selecting 1L maintenance options in ovarian cancer. ITCs in this setting should be performed cautiously, as many factors can preclude objective trial comparisons.

Patient participation in treatment decision-making of prostate cancer: a qualitative study

PURPOSE

Despite increasing development in decision-making strategies for patients with prostate cancer, little is known about patients' individual experience and perception throughout the decision-making process. The objective of this study was to explore patients' experiences and perceptions towards treatment decision-making.

METHODS

We conducted a qualitative interview study with 30 patients diagnosed with prostate cancer. We transcribed interviews verbatim and inductively identified codes. Thematic analysis was used to develop and refine a codebook that aided in the identification of themes.

RESULTS

Three key themes and nine subthemes emerged, which were as follows: I. less involved in treatment decision-making, (i) passive decisional control, (ii) lack of medical knowledge, and (iii) domination by family members; II. the right to be informed of the disease condition and to choose treatment options, (i) sociocultural influences, (ii) patients believe that they should know the true facts of the disease, and (iii) patient autonomy during treatment; and III. future consideration and advance care planning, (i) fewer future concerns, (ii) advance care planning is poorly understood, and (iii) acceptance of advance care planning.

CONCLUSION

The study results show that patients with prostate cancer have a diversity of needs to cultivate their ability to make treatment decisions, and healthcare professionals should empower patients, as well as provide decision aids or decision support for patients.

On the Controllability of a System Modeling Cell Dynamics Related to Leukemia

In this paper, two control problems for a symmetric model of cell dynamics related to leukemia are considered. The first one, in connection with classical chemotherapy, is that the evolution of the disease under treatment should follow a prescribed trajectory assuming that the drug works by increasing the cell death rates of both malignant and normal cells. In the case of the second control problem, as for targeted therapies, the drug is assumed to work by decreasing the multiplication rate of leukemic cells only, and the control objective is that the disease state reaches a desired endpoint. The solvability of the two problems as well as their stability are proved by using a general method of analysis. Some numerical simulations are included to illustrate the theoretical results and prove their applicability. The results can possibly be used to design therapeutic scenarios such that an expected clinical evolution can be achieved.