Logistical optimization of radiotherapy treatments
Due to the COVID-19 crisis measures the PhD defence of Bruno Vieira will take place online without the presence of an audience.
The PhD defence can be followed by a live stream.
Bruno Vieira is an external PhD student affiliated with the Netherlands Cancer Institute and the University of Twente. His supervisors are prof. dr. Wim van Harten and prof. dr. ir. Erwin Hans, both from the faculty of Behavioural, Management and Social Sciences (BMS), and his co-supervisor is dr. Jeroen van de Kamer from the Netherlands Cancer Institute.
The delivery of radiotherapy (RT) involves the use of rather expensive resources and multi-disciplinary staff. In RT, timeliness is crucial, and literature shows that delays in the start of treatment have shown to increase the risk of tumor progression in various cancer types, and patients experience greater psychological distress when subject to longer waiting times. As the number of cancer patients receiving RT increases, timely delivery becomes increasingly difficult due to the complexities related to, among others, variable patient inflow, specialized patient routing, and the joint planning of multiple resources. In this thesis, we studied, designed, and developed several Operations Research (OR) models for the logistical optimization of RT processes to support decision-makers use their resources more efficiently. We propose innovative approaches for solving logistical problems encountered in the whole RT chain of operations, from pre-treatment to treatment, while optimizing for the most important KPIs related to timeliness and patient-centeredness. The developed research work is practice-oriented, with models being built, validated, and tested using real-world information and data provided by six collaborating RT centers.
Chapter 2 presents a literature search of OR methods for resource planning in RT in six databases covering journal papers in the medical and mathematical domains. Data extraction included, amongst others, the subject of research and methods applied, the extent of implementation according to a six-stage model, and the (potential) impact of the results in practice. The review shows that most models addressed the problem of scheduling patients on treatment machines, while little focus has been given to the pre-treatment phase of the process, despite its high impact on timeliness. We also found that the development of OR methods for RT macro-planning (strategic level) is rather low and can be further extended. Furthermore, we verified that none of the 33 analyzed papers reported a full implementation of the model indicating that, unless there is lack of reporting, implementation rates of OR models in RT are rather low.
In Chapter 3 we introduce a mathematical approach for the optimal allocation of radiation therapy technologists (RTTs) to the several operations they perform throughout the (pre-)treatment chain of operations. Besides performing multiple activities, RTTs have specific rotation needs to maintain specialized skills and partial availability, which makes the allocation of RTTs a complex task especially due to the highly variable patient arrivals and care content. In our approach, we use a novel stochastic mixed integer linear programming (MILP) model to optimize the allocation of RTTs over a set of scenarios of patient inflow generated from historical patient data. The goal is to maximize the (expected) number of patients completing the pre-treatment phase within the waiting time target. Results for a case study in the RT department of the Netherlands Cancer Institute (NKI) showed that, on average, the number of patients able to start treatment within the maximum waiting time standards may increase from 91.3% to 97.9% for subacute patients, and from 96.3% to 100.0% for regular patients.
Radiotherapy pre-treatment workflow is either driven by the scheduling of the first irradiation session, which can be set right after consultation (pull strategy) or after the pre-treatment operations have been completed (push strategy). In Chapter 4 we assess the impact of using pull and push strategies and explore alternative interventions for improving timeliness in radiotherapy using discrete-event simulation modeling. Staff surveys, interviews with managers, and historical data from the NKI (2017) were used to generate model inputs, in which fluctuations in patient inflow are considered. Results showed that a pull strategy allows for 41% fewer patients breaching the waiting time target than a hybrid strategy (40% pull / 60% push), in spite of slightly longer waiting times (12%).
Chapter 5 focuses on the problem of automatically scheduling RT sessions while satisfying patient preferences regarding the time of their appointments. While most literature focuses on the timeliness of treatments, collaborating RT centers have expressed their need to include patient preferences when scheduling appointments for irradiation sessions. Therefore, we propose a MILP model to solve the problem to optimality, scheduling all sessions within the desired window in reasonable computation time for small size instances up to 66 patients and 2 linacs per week. For larger centers, we propose a heuristic method that pre-assigns patients to linacs to decompose the problem in subproblems (clusters of linacs) before using the MILP model to solve the subproblems to optimality in a sequential manner. We tested our methodology using real-world data from the NKI and found that, with our combined approach, the problem can be solved in reasonable computation time (3.3 hours) with as few as 2.8% of the sessions being scheduled outside the desired 90-min time window.
In Chapter 6 we adapted the MILP model of Chapter 5 to generate weekly schedules for the RT treatment scheduling problem for two Dutch RT centers. The model was iteratively adjusted to fulfill the technical and medical constraints of each center until a valid solution was attained. Patient data was collected from the internal databases, and the feasibility of the obtained schedules was verified by staff members of each center. We verified that the practical application of the OR model has helped RT planners produce a feasible, high-quality schedule in an automated (faster) way than currently done in practice. The weekly schedule improved in both centers, with the average standard deviation between sessions’ starting times decreasing from 103.0 to 50.4 minutes (51%) in one center, and the number of gaps in the schedule being reduced from 18 to 5 (72%) in the other center. The process required 5 minutes respectively 1.5 hours of computation time for the two centers at test, as opposed to 1.5 days when performed manually by both centers.
The OR models proposed in this thesis have demonstrated considerable potential benefits when results are extrapolated on a national scale. The algorithm of Chapter 3, for instance, showed that our model could potentially lead to cost savings of 69 FTEs nationwide. Similarly, by obtaining more compact schedules for one of the test cases (Chapter 6), our model showed that savings of around 188 FTEs could be achieved overall. However, the implementation of OR tools in RT practice is not straightforward. We found that matching the mathematical value with actual willingness and practical options for change in practice is an ultimate uncertainty that needs to be confronted when aiming to actually implement the theoretical results. This requires input and organizing commitment from all relevant stakeholders and researchers from the start of the project in order to deal with sociodynamic aspects of work processes from an early stage. Implementation efforts performed within this project also revealed that gradual model adaptations performed in small steps and constant communication are needed to ensure the translation OR models into clinical practice.
Concluding, the research models proposed in this thesis proved to be able to provide decision-making support for RT centers aiming to organize their processes more efficiently for the benefit of patients and staff. Although several hurdles hamper the application of those models in practice, OR has proven effective in solving highly constrained logistics problems in RT where feasible (near-optimal) solutions are difficult to achieve manually. Overall, advanced analytical methods based on OR have proven capable of optimizing the logistics of radiotherapy treatments for a better quality of care.
