Multistate models are an important statistical tool to understand disease trajectories and can provide clinically important measures of effect and effect differences(1). Multistate models can be used to describe patterns of multimorbidity using large-scale electronic health records and, through quantification of utilities associated with disease states (e.g. quality of life measures), the individual burden of multimorbidity can be estimated. Nevertheless, there are numerous computational and analytical challenges in this framework, with multiple comorbidities and disease pathways making multistate modelling demanding without model simplification.
This studentship will develop methods for multistate modelling in multimorbidity research, which will lead to:
- innovation in trajectory modelling and clustering of multimorbidity pathways.
- methodological advances that underpin new data science developments in large-scale longitudinal linked electronic health records, as being curated in the Health Data Research Hubs.
The project will consider two important aspects in the developments of multistate models in multimorbidity research. Firstly, the development of models for complex disease trajectories, including approaches to relax the Markov assumption, including multiple time-scales in derivation of rates and incorporation of recurrent events. Secondly, the project will investigate efficient post-estimation simulation-based techniques to estimate clinically useful measures of effects and effect differences, such as disability- or quality-adjusted life-years and their contrasts. These measures are commonly calculated in health technology assessments using Discrete Event Simulation (DES) of multistate models (2). Variance reduction techniques such as cloning and antithetic sampling will be investigated to allow efficient estimation of incremental effects between two or more groups of interest.
These techniques will be applied to the investigation of cardiovascular and non-cardiovascular outcomes following a cancer diagnosis, utilising national data from the Virtual Cardio-Oncology Research Initiative (VICORI) (3). The student will develop user-friendly open-source software to encourage reproducible research and enable the methods to be utilized across health data research.
1. Crowther MJ, Lambert PC. Parametric multistate survival models: Flexible modelling allowing transition-specific distributions with application to estimating clinically useful measures of effect differences. Stat Med. 2017 Dec 20;36(29):4719–42.
2. Davis S, Stevenson M, Tappenden P, Wailoo A. NICE DSU Technical Support Document 15: Cost-effectiveness modelling using patient-level simulation. 2014 Apr;62.
3. Sweeting MJ, Oliver-Williams C, Teece L, Welch CA, De Belder MA, Coles B, et al. Data Resource Profile: The Virtual Cardio-Oncology Research Initiative (VICORI) linking national English cancer registration and cardiovascular audits. International Journal of Epidemiology. Accepted for publication.