R for HTA Annual Workshop 2023

Survival Analysis

Survival analysis was a key theme across several presentations at this year’s R for HTA workshop. Examples of more user-friendly approaches to working with survival data were presented, such as the use of R Shiny apps to readily obtain parametric extrapolations for survival data, and automating the digitisation of Kaplan-Meier curves to generate pseudo-IPD. Various presentations also focused on methodological advances in survival analysis, notably on how to incorporate externally obtained data, such as elicited expert opinion on long-term survival rates, when generating survival parameters.

Details of the R packages and tools for survival analysis that were presented at the workshop are provided below:

• survdigitizeR:1 An R package under development to digitise published KM curves and extract survival times and events to be used as input to a digitization algorithm to generate pseudo individual participant data (IPD)
• survextrap:2 An R package to model survival from a combination of standard survival IPD and external data sources, such as population data, registry data or elicited judgements
• SurvInt:3 An R shiny application which allows the user to specify population survival at key points and obtain parametric extrapolations that are consistent with those specified by the user. The application is available here
• expertsurv:? An R package for direct incorporation of expert opinion into survival models

Other relevant R packages and tools for statistical analysis and economic evaluation are detailed in the R for HTA website here .?

At Costello Medical, we share the ambition to generate the most precise and clinically plausible estimates of survival for evidence development, and will be incorporating the methods presented in our Statistics and Health Economics projects.?

Applications of R-Based Models

The current health economic modelling landscape is dominated by the use of Microsoft Excel-based cost-effectiveness and budget impact models. Interest in non-Excel-based models, especially those developed in open source languages such as R, has been increasing in recent years, with the establishment of the R for HTA workshops in 2018 and discussions at recent Professional Society for Health Economics and Outcomes Reseach (ISPOR) conferences, previously discussed here .

A few presentations discussed the benefits of developing models in R, such as the potential for vertical integration of clinical trial data analyses, the accessibility and efficiency of R (compared to Microsoft Excel), and the potential for developing user friendly interfaces for the R models themselves. However, given the advantages and disadvantages of R-based modelling have been discussed at length previously, the workshop was more focused around advancing methodologies for R-based models and how trust and transparency can be improved for such models. There was particular emphasis on the development of tested, well-documented, and transparent packages to form a reliable code base, and the development of tools to be able to validate the outcomes of these models.

Perhaps the most interesting session on R-based models was a panel discussion between members of the Finnish, Swedish, and Irish HTA bodies and the Norwegian Directorate of Health. Each member of the panel was enthusiastic about HTA modelling in R, with increased efficiency and potential for standardisation touted as key opportunities. Despite this, it was still acknowledged that barriers remain to mainstream adoption. While the potential for standardisation is there, with the opportunity to have specific ‘recommended’ packages to more efficiently develop economic models, there is a lack of guidance and precedence to establish appropriate methods for developing models in R. This is especially apparent when considering that only Zorginstituut Nederland (ZIN) have published guidance in this area, located here , and the Irish and Finnish representatives were not aware of receiving any R-based models to date.?

An important question following on from the workshop becomes who has responsibility for driving the development of this shift in HTA modelling? While there is a lot of ongoing discussion at workshops like R for HTA, these are often quite academic in nature and therefore can lack the capacity to drive large-scale policy shift. One suggestion during the panel discussion was that government bodies could provide incentives for companies to develop R-based analyses, however given increased pressure globally on keeping health spending down, it is unlikely HTA bodies can provide any financial incentives. Other incentives could include increased priority for HTA submissions supported by R-based models, however this has potential equity implications.?

Certainly, one key way in which HTA bodies could encourage more R-based modelling would be through the development of specific guidelines for developing economic evaluations in software beyond Excel, however, ultimately, it will not be until companies start to submit R-based models to HTA agencies that a precedence can be established which other companies and HTA bodies can refer to.

Given the interest expressed in R-based modelling from workshops like this to international conferences like ISPOR, it seems to be only a matter of time until R-based modelling becomes more mainstream for HTA submissions. The question then becomes will HTA agencies be reactionary to this change, allowing companies to have a larger impact on driving the precedence for R-based modelling, or will HTA bodies become more proactive, like ZIN, in shaping the development of robust guidance for new types of economic evaluation?

Thomas Kloska and Andrei Karlsson are employees at Costello Medical. The views/opinions expressed are their own and do not necessarily reflect those of Costello Medical’s clients/affiliated partners.

References

1. Zhang JZ, Rios JD, Pechlivanoglou P, et al. SurvdigitizeR: Digitizing Survival Curves. R package version 0.0.0.9000.?2023.
2. Jackson C. survextrap: Survival Extrapolation with a Flexible Parametric Model and External Data.?2023.
3. Gallacher D. SurvInt: A simple tool to obtain precise parametric survival extrapolations. Research Square, PREPRINT (Version 1) 2023.
4. Cooney P, White A. expertsurv: Incorporate Expert Opinion with Parametric Survival Models. R package version 1.0.0.?2023.
5. R-HTA Consortium. 2023-2025.
6. Zorginstituut Nederland. Richtlijn kosteneffectiviteitsmodellen in R.?2022.