Evaluating effects of complex treatments in chronic disease using large observational datasets

Patients with chronic conditions and their carers face difficult questions about the effects of treatments: What are the long-term effects of multiple treatments used in combination? What are the expected health outcomes for a given patient under different treatment choices? Randomized controlled trials are the gold standard for estimating treatment effects, but their ability to answer complex questions is limited. They are typically restricted to a subset of the eventual treatment population, have short follow-up, usually do not consider several treatments in combination, and can be unethical. Large longitudinal observational datasets from disease registries and electronic health records provide the means to gain understanding of treatment effects that would not be feasible in a trial. However, to do this we have to successfully overcome the fundamental difficulty that those who received a given treatment were prescribed it for a reason, and hence the groups who do and do not receive treatment are not directly comparable. How to handle this is a highly active area of research.

This fellowship will develop and evaluate statistical methods needed to address key questions about the effects of treatments on health outcomes using large observational datasets. It will also tackle crucial questions about the effects of treatments on health outcomes for patients with cystic fibrosis (CF) using data from the UK CF Registry. CF is an inherited, chronic, progressive, and life-shortening condition affecting about 10,000 in the UK. People with CF need intensive treatment and support from health services and families. The UK CF Registry captures data on nearly all UK CF approximately annually and the data contain demographic information, information on treatments used, clinical measurements, and dates of birth, diagnosis and death.

Key statistical advancements will include methods enabling estimation of the effects of multiple treatments used in combination on health outcomes; methods for answering questions about the impact of a one-time major intervention (e.g. lung transplantation) on survival in a way that is most relevant to patients and clinicians; and methods for providing personalised information to patients about their expected outcomes under different treatment choices.

Patients with CF take many treatments daily, which is time consuming and unpleasant. Questions addressed in this fellowship will focus on those that could help reduce treatment burden and help patients and clinicians in making treatment decisions. We will investigate the impact of three commonly used treatments on lung function, and whether existing treatments are redundant in patients receiving Ivacaftor, which is a major new treatment in CF since 2012 and effective for patients with a certain genotype. Another key question will be about the impact of lung transplantation on survival in CF. This research will provide personalised information for patients about what their expected survival would be should they choose to be listed for lung transplant or not.

In the second part of the fellowship, I will study the impact of treatments on health outcomes in other chronic disease areas of major public health importance, including type 2 diabetes, using the Clinical Practice Research Database (CPRD). CPRD is a database of over 11 million patients in the UK, obtained from General Practitioner records. It will be important to react to new questions arising in CF and in other chronic diseases. Further statistical developments will be an ongoing focus.

This research will be led by the Future Leaders Fellow with assistance from a postdoctoral researcher. The host institution is the London School of Hygiene and Tropical Medicine. The Cystic Fibrosis Trust, a UK-wide charity, is a Project Partner. Several collaborators will provide expert input: they include statisticians, clinicians, and patient representatives.

This fellowship will develop novel statistical methodology for estimation of treatment effects from routinely collected observational data, enabling effective exploitation of the wealth of computerised clinical data that is increasingly available. It will provide answers to key questions about treatment effects in cystic fibrosis (CF) by making use of longitudinal observational data from the UK CF Registry.

Immediate beneficiaries will be statisticians, epidemiologists and data scientists working with observational data and those working on statistical methodology for causal inference. The proposed methodology will be developed with practical applicability in mind and publications will be aimed at both statistical and applied audiences. Computer code, using the open source R software, will be made freely available, facilitating uptake. Findings will be communicated via presentations at international conferences and smaller focused workshops. Invited conference sessions will be proposed in years 3 and beyond and a high profile symposium is planned for the end of year 3. A wide audience will be reached via the networks of the LSHTM Centre for Statistical Methodology and the International Biometric Society. Dissemination via local networks at the host institution and collaborators will be used to encourage uptake of the methods during the lifetime of the project.

The applied results from this fellowship will be of direct benefit to patients with chronic diseases and their doctors. The first beneficiaries will be in the CF community. Results will include new information about combined treatment effects and information for the first time about the impact of withdrawing existing treatments for patients who are using a new disease-modifying drug. This has the potential to reduce treatment burden and to impact on health care policy and spending. This project will result for the first time in answers to key questions about the impact of lung transplantation on survival for CF patients in the UK, with a focus on providing information in the most relevant way for patents and clinicians. Collaborators include CF clinicians who are best placed to influence putting findings into clinical practice. Input from patient advisors will ensure that results are communicated appropriately. The wider UK and international clinical audience will be reached via publications and presentations at the UK, European and US CF Conferences. The CF Trust, which is a Project Partner, is in an ideal position to communicate findings to a wide audience, via its website and social media, and international influence. In the longer term, our personalised information about treatment effects could be incorporated into clinical practice. In a current project in collaboration with the CF Trust I am investigating approaches to how personalised information on predicted survival can be presented to doctors and patients. Experiences from that will inform this project.

This project will positively impact the UK CF Registry and the UK Transplant Registry, by placing them at the forefront of research into learning about treatment effects from registry data. Our results will motivate continued support for the registries from patients and health carers and encourage use of the registries for further research. Uptake of the methods in other national CF registries is anticipated within the lifetime of the project. Linked data will create a future resource for new studies about transplant in CF.

Research outputs will also include novel findings about treatment effects in other chronic diseases, including type 2 diabetes, using the Clinical Practice Research Database. The results will be communicated through traditional routes of publications and conference presentations. Discoveries with the potential for major public health impacts can influence public trust in their data being used for research.