The RSS Medical Section organised an invited session at the RSS Conference 2021 on 9 September 2021, titled: 'Novel advances in Bayesian health economics'.
The three speakers were Anna Heath, a scientist at the Hospital for Sick Children in Toronto and assistant professor in the Division of Biostatistics at the University of Toronto; Konstantina Chalkou from the University of Bern in Switzerland; and Nick Latimer, a reader in health economics from the University of Sheffield, UK.
Dr Heath’s talk, 'Statistical developments in the calculation of Value of Information', covered Value of Information (VoI), a decision-theoretic approach to estimating the expected benefits from collecting further information, eg through a future research study. A VoI analysis is based on a decision-theoretic model, which aims to support decision making under uncertainty. VoI methods assess the sensitivity of models to different sources of uncertainty and help to set priorities for further data collection. They have been widely discussed in healthcare policy making, but the ideas are general to a range of evidence synthesis and decision problems. Dr Heath’s talk gave a broad overview of VoI methods, explaining the principles behind them, the range of problems that they can tackle, how they can be implemented, and the ongoing statistical challenges in in the area.
Dr Latimer talked about 'Causal inference analyses to estimate comparative effectiveness of treatments'. Randomised controlled trials (RCTs) - the preferred approach for estimating the effectiveness of new treatments. However, due to limitations around generalisability, short-term follow-up, limited comparators and feasibility, it can be useful to supplement evidence from RCTs with analyses of observational data. These analyses are prone to issues such as confounding by indication and time-dependent confounding, which can only be addressed with advanced causal inference statistical methods, such as 'g-methods'. And these methods will only produce unbiased results if good quality data are available on prognostic variables, and if models are well specified. In the UK, large amounts of data are collected on cancer patients, held by the National Cancer Registration and Analysis Service. Dr Latimer and colleagues are investigating whether reliable estimates of comparative treatment effects can be obtained using these data, by emulating target trials in pancreatic cancer.
The final talk from Dr Chalkou was titled 'Decision curve analysis for personalised treatment choice between multiple options'. Dr Chalkou explained that the use of individualised prediction models would theoretically allow treatment to be targeted to individuals for whom the probability of benefit is sufficient to outweigh treatment-related harms. Decision curve analysis (DCA) can be to determine whether application of a model in practice would lead to better clinical decisions. The aim is to extend the DCA methodology, considering patient preferences, to the scenario where several treatment options exist and evidence about their effects comes from a set of trials. Patient preferences can be considered via a personalised threshold probability, ie the probability that renders each treatment worthwhile taking. She presented the steps needed to estimate a prediction model’s net benefit using evidence from studies synthesised in a network meta-analysis (NMA) to compare personalized model’s strategy versus a one-size-fit-all treatment decision making strategy, such as ‘treat none’ or ‘treat all patients with a specific treatment’ strategies. Dr Chalkou and colleagues applied their methodology on a NMA individualized prediction model for relapsing-remitting Multiple Sclerosis patients, which can be used to choose between Natalizumab, Dimethyl Fumarate, Glatiramer Acetate, and placebo.
All speakers presented virtually from their different locations in Canada, Switzerland and the UK. The session was well attended, with interesting questions from the audience.
Nathan Green, UCL and the RSS Medical Section.