Multistate modelling for cost-effectiveness analysis

: The impact of common simplifying assumptions explored in the context of implants for total knee replacement

Abstract

Traditional multistate models in cost-effectiveness analysis of total knee replacement
(TKR) surgery are cohort-based, discrete-time models, which assume patients
transition between discrete health states, independent of patient history (e.g., revision
surgeries). This dissertation develops continuous-time, individual-level models that
incorporate patient history to assess the cost-effectiveness of TKR implants, addressing
these limitations.
In my dissertation, I first reviewed all types of economic models used in TKR. I then
implemented a discrete-time Markov model (DTMM) for implants in TKR using estimates
derived from a Network meta-analysis that I performed, and analyses of observational
cohorts, costs, and utilities derived from the UK-based National Joint Registry, NHS, and
Patient Reported Outcome Measures in England-for Hip and Knee Replacement
Procedures. Next, I expanded the modelling framework to include continuous-time
Markov model (CTMM) and multistate microsimulation. Finally, I conclude by discussing
the pros and cons of the different modelling techniques and their impact on results and
conclusion. All models were implemented in R.
Under constant log hazard rates, DTMMs and CTMMs yielded similar results, with
“Cemented Cruciate-retaining Fixed Modify” being the most cost-effective implant.
However, with spline function-based hazard rates, CTMMs lead to different implants
being most cost-effective. Including patient history through multistate microsimulation
further altered findings, identifying “Oxininum-coated Cemented Cruciate-retaining
Fixed Modify” as most cost-effective for women aged 65-74. The increased model
complexity required High Performance Computing facilities for CTMMs and multistate
microsimulation.
This dissertation highlights how more detailed models, like multistate microsimulation,
offer nuanced insights but require significant computational resources. The higher
complexity may also reduce transparency, increase the chance of errors, and make
reporting more challenging. In contrast, DTMMs provide quicker, efficient results for
constant-hazard scenarios, making them suitable for rapid decision-making when the
assumptions hold.

Date of Award	18 Mar 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Howard H Z Thom (Supervisor), Elsa M R Marques (Supervisor) & Nicky J Welton (Supervisor)