Individualised Utility Based Decision Modelling for Personalised Clinical Decisions

About the Project

Traditional cost‑effectiveness models underpin healthcare resource allocation by estimating population‑average outcomes and utilities associated with treatments. From this, health service funders can decide which treatments to provide. However, individual patients do not necessarily reflect the population averate. They often hold very different valuations of outcomes, different attitutes towards risk (risk averse or risk seeking), and have different time preferences to those of the population average. They may also face personal costs, both lost income and out‑of‑pocket expenses, which are not captured in typical cost-utility (QALY) analysis.

We would like to develop a cost-effectiveness model of a medical intervention from the perspective of the individual patient. Conventional cost-effectiveness analysis weighs up the costs and benefits based on average valuations of health outcomes, average resource costs, using an average discount rate and assuming a neutral attitude towards risk (neither risk averse nor risk seeking). However, individuals vary in the weights they give to health states, their personal time preferences and their attitude towards risk. They also incur different costs (out of pocket and productivity losses). The optimum decision at a population level is therefore not necessarily the optimum decision for any individual.

The idea is to construct an economic model where the probabilities of different outcomes are provided by the best evidence. But the valuations of health outcomes and costs is elicited from the individual making the decision. This includes the weights given to outcomes, personal time-preferences (i.e. personal discount rate). The model then combines these with the probabilities from best evidence and indicates their personally optimum decision. Individuals can conduct their own sensitivity analyses by altering inputs to reflect their own uncertainties. This was done once many years ago for a decision relating to hysterectomy with or without oophorectomy (Pell 2002), but it was not taken forward.

This is experimental, which is why it is a PhD project. We are fairly flexible about the clinical decision to be modelled.

The doctoral thesis includes

The technical challenge of building a model:

• identifying epidemiological data to populate the model
• building a cost-effectiveness model (e.g. discrete event modelling; Markov modelling)
• incorporating uncertainty into the model

Eliciting data to populate the model:

• how to elicit personalised health state valuations, time-preferences, attitudes towards risk
• how to elicit effects on income/productivity

Usability of the model

• developing an interface for users
• communicating the model's personal recommendation and conveying uncertainty
• investigating the acceptability to patients and clinicians

Skills Developed

• Epidemiological modelling
• Decision analysis & health economics
• Utility elicitation techniques
• Bayesian and probabilistic modelling
• Qualitative research and mixed-methods integration
• Communication of risk and uncertainty
• Ethical considerations in personalised decision support

Requirements

Applicants should be able to demonstrate existing skills in the some of the following areas:

• Statistical modelling (regression, survival analysis)
• Working with large datasets (e.g., cohort studies, RCT datasets)
• Using statistical software such as R, Stata, or Python
• Understanding of basic causal inference methods
• Ability to interpret epidemiological results (effect measures, risk models)

The project is suitable for someone who is has strong quantitative skills and an interest in health or social sciences. e.g. a 1st class or upper 2nd class degree or a MSc in:

• Public Health
• Epidemiology
• Health Data Science
• Biostatistics
• Health Economics
• Medical Statistics
• Operational Research
• Psychology (especially cognitive or behavioural decision science)
• Economics
• Behavioural Economics
• Decision Sciences
• Statistics or Applied Mathematics
• Computing science

Funding Notes

Self funded only.

No funding is available for this project.