Good practice guidance for the prediction of future outcomes in health technology assessment (HTA).
NIHR Doctoral Research Fellowship awarded to Benjamin Kearns.
Duration of fellowship: January 2017 to December 2019.
This fellowship has two main aims:
- Identify / develop methods that can be used when generating predictions of future outcomes (extrapolating observed data).
- Develop good practice guidance that may be used by (a) Analysts when extrapolating data, and (b) Reviewers when critiquing produced extrapolations.
My original plans for the fellowship can be found in the plain English summary from my application. As is common with PhDs, the work performed hasn't fully followed these plans... The below describes work performed so far (last update: August 2018). A key hypothesis is that time-series techniques will be very useful when extrapolating survival outcomes. When used in this setting, resulting models are known as dynamic survival models.
Reviews of the methodological literature.
I have performed two literature reviews. The aim of the first review was to identify a long-list of statistical methods (models) that may be used to analyse and extrapolate survival (time-to-event) data. I will be investigating a subset of these in more detail during the fellowship. I am currently writing-up the results of this review - a broad overview of the methods identified is provided here.
The second review was on estimation methods for dynamic survival models. The results of this have been presented as a poster at the 2018 RSC. I also gave a talk based on the results of the first review. Both the talk and poster may be found here.
There are currently two on-going studies. The first is comparing dynamic survival models against commonly used parametric survival models (such as the Weibull and Gompertz). It is looking at the question "how well do these models describe the observed data?"
The second is comparing dynamic survival models against a range of other survival models (tbc). This seeks to answer the question: how well do they predict (forecast) the future?
I am using simulation studies for both of the above, as they allow me to know what the 'truth' (or 'correct answer') is, and so make it easier to answer the above questions.
To date I have undertaken the following public involvement activities:
- Held workshops to discuss the proposed work, and how best to involve members of the public. There have been three so far: one prior to submitting my application, and one each in years one and two.
- Registered the fellowship to be a Public Involvement Standards Freestyle Project.
- Entered the Three Minute Thesis Talk.
More details on all of the above can be found here.
I am very interested in how best to involve members of the public within methodological research (such as this fellowship), so would like to talk to others that are also working on this.
I originally planned to perform case-studies are in the areas of cancer and mental health. These would re-use trial data requested from data-sharing websites. I am still awaiting these data so these case-studies are currently on hold.
The main output of this fellowship will be the development of good practice guidelines. These will be developed in combination with the public advisory group and other stakeholders. They are likely to include: a summary of the statistical methods that I have covered; a discussion of their strengths and weaknesses (including the assumptions required); guidance on both when the methods should be considered and how to choose between them; and details on what information should be reported. The guidelines may also include: flowcharts to assist with conducting analysis; reporting guidelines; and position statements.
Results shall be disseminated through a variety of channels, including academic publications, conference presentations, Plain English summaries, and social media.
If you would like to know more details, please contact myself at: email@example.com
This website reports work funded by the NIHR (www.nihr.ac.uk)