Optimising Cancer follow-up using mathematical models and AI

Erasmus University, TU Delft and Erasmus Medical Centre have established a partnership through interdisciplinary research and education. As part of the work in the alliance, we are looking for a PhD candidate to develop and evaluate a mathematical model and algorithm to better tailor (risk-based) follow-up after cancer treatment. The role is highly impactful for ensuring sustainable cancer care in the future, and we are looking for an exceptional candidate with strong quantitative, and modelling (e.g., POMDP) expertise combined with the interest to work with patients and clinicians on improving cancer care. In the role you will combine data analysis using clinical registries, evidence synthesis, participatory research with patients and clinicians, applied modelling, and development of AI decision-making algorithms to answer a question that health services across Europe are struggling with: how much surveillance is enough, for whom, and delivered by whom?

Background
Follow-up after primary cancer treatment is, in several tumour streams, more intensive than the evidence currently supports. Where the aim is recurrence detection, frequent and intensive surveillance shows limited benefit for clinical outcomes while imposing psychological burden and out-of-pocket costs on patients and consuming scarce professional capacity.

These pressures are intensifying. Survivor numbers are rising, early-onset cancers extend the survivorship horizon, treatment is shifting from surgery towards systemic therapy, and complex care is concentrating in fewer specialised centres. Deciding where surveillance capacity is best spent has therefore become a question of workforce sustainability as much as one of clinical effectiveness.

Aim and objectives
The project aims to use modelling to propose risk-based follow-up strategies that match surveillance intensity to individual recurrence risk. It pursues three objectives.

• Optimising follow-up. Determining when, for how long and in what form patients should be seen, and identifying the factors — age, tumour stage, molecular profile, treatment received — on which stratification can be based. Given clinical and molecular characteristics of many tumours, a Partially Observable Markov Decision Process (POMDP) is a preferred approach.
• Defining successful implementation. Engaging patients to understand their concerns and preferences, and assessing the clinical feasibility of alternative follow-up arrangements, including nurse-led, physician-led and online delivery, with attention to professional roles, responsibilities and workload.
• Data and model requirements. Identifying suitable data sources such as the cancer registry, specifying the model along clinical, operational and organisational dimensions through participatory design, and building an extrapolation framework that transfers findings from large datasets to low-volume tumour streams.

The work combines several strands. A systematic review with LLM-assisted extraction of policy documents updates the international consensus on follow-up and risk stratification. Second, secondary data sources (e.g., cancer registry) will be used to map variation in current practice, estimate (yearly) recurrence incidence, and define the risk factors on which cancer follow-up can be stratified. Focus groups with patients establish what implementation would require of them and focus groups with professionals address feasibility and their own role. These feed a model whose outputs cover capacity and resource use alongside clinical outcomes, with risk factors and organisational constraints — hub-and-spoke versus centralised care, nurse-led delivery — entering as explicit variables.

Colorectal, lung and bladder cancer are the primary candidate tumour streams, with prostate and breast under consideration. Ideally, the project approach and results should be generalisable to other tumour streams.