Dr Rupert Dunbar-Rees Interview
Dr Tom Foley, Dr Fergus Fairmichael
Dr Rupert Dunbar-Rees is a GP by background, and Founder of Outcomes Based Healthcare. He trained in Medicine at Imperial College, gaining a degree in Orthopaedics from University College London. He was a Partner in general practice for five years before joining the Department of Health, London for three years, as Clinical Lead on the Commercial Team. He led the clinical work stream on a £1.25 billion NHS procurement programme, resulting in 265 new GP surgeries across the UK. Outcomes Based Healthcare is an organisation supporting the adoption of outcomes based approaches to healthcare. Its methodology involves co-creating outcomes with patients, whilst providing academic and technical rigour to the process of measuring and contracting for outcomes.
Outcomes Based Healthcare
There is a lot of work underway to improve how outcomes are collected and analysed.
OBH is currently working on a project exploring the correlation between passive recordings from mobile applications and actively recorded Patient Reported Outcome Measurements (PROMs). For example there are two ways you can measure how mobile someone is; either asking them or by measuring their mobility. This project will explore both in app behaviour (how people use an app) and out of app behaviour, making use of the various sensors in smartphones, to see if they can give us more information. The idea being that continuous passive recording should give greater quality of information. This project is funded by The Technology Strategy Board (TSB).
Correlating PROMS with passive recordings is an area that has not been greatly explored yet. There is some similar work done in Poland by Harimata, (www.harimata.co), who have developed an app to for autism in which children play games on an iPad. Behaviour patterns can be examined from their interaction with the games. This also examines multiple variables outside of the game itself, such as the way the iPad is held with the position of the gyroscope, the acceleration and deceleration and touch movements interacting with the screen. These variables are distilled down to reflect features that are most predictive of where a child lies on the autism spectrum. This is improving the accuracy of diagnosis.
Another company doing exciting work in this area is Ginger.io who use passive mobile data to examine how people interact with social networks via their mobile phone. They also incorporate PROMs via condition specific surveys and combine the results using behavioural analytics. They then evaluate predictive features that reflect a depression score and can indicate earlier intervention when appropriate.
The greatest impact and innovation in this field is unlikely to come from within the health system itself, but will be inspired by innovations from elsewhere.
There is currently a need to spend a lot of time trying to improve the quality of data for outcome measurement. There is very little access to patient level data, normally they are aggregated datasets.
Hospital Episode Statistics (HES) and Secondary Uses Service (SUS) data, primary care read codes, etc. are usually extracted at the aggregate level via Commissioning Support Units (CSU), Clinical Commissioning Groups (CCG) or the Health and Social Care Information Centre (HSCIC). These data sources are also combined with Office of National Statistics (ONS) and census data.
There can be significant delays, sometimes up to 18 months, in obtaining and cleaning this data.. Extracting data manually can be quicker, but created additional governance issues. There are CCGs who are starting to put in place integrated digital records that can make this job easier.
Incomplete data such as under-coding or comorbidities is a big problem. For example you may have a record of an intervention for diabetes, such as an amputation procedure in secondary care, but the data generated on discharge is often lacking the diabetes flag. This can be quite significantly under coded, with the National Diabetes Audit estimating that approximately 30-40% of cases are not flagged. This can be compensated for by merging primary care datasets with the secondary care data sets to add back the diabetes flag. This is not perfect, but accuracy has been improved from around 60% to 98%. This does create additional work but some of it can be automated. There is work underway on ontologies and looking at clinical records to fill in these missing gaps.
There are both public and professional concerns about confidentiality and information governance when linking primary and secondary data.
Lessons from other industries
Medicine is at least a decade behind some other industries in the way we use data. For example if you look at amazon who enrich your data and use it so effectively to suggest other purchases. Another industry that was changed dramatically thanks to the use of data was the gas industry, who have developed a far better understanding of their supply and demand and subsequently we have seen the decline of large gas stores such as gasometers.
If you take this principle and apply to the medical setting you can start to pre-empt complications of disease.
The UK is still crawling in terms of outcome measurement. Collection of meaningful outcome measurements goes well beyond the bounds of hospital or provider silos and into community services, social services and GP, which should work together around the individual. This is the biggest lesson we still have to learn about outcomes. There are a lot of non-outcomes in current frameworks. There are a lot of process indicators and measures for an individual providerís performance but it often does not reflect what is important to individuals across the whole of their care. There is a need to ensure that what is being measured is appropriate and that it is representative of what we are trying to measure.
Data collection is a huge undertaking and the easier it is to collect outcome measures the more likely that they will be adopted. There are efforts to ease the collection of outcome measurements with automation, for example automatic phone systems and early examples of mobile applications seeking PROMs.
There are two schools of thought with regard to standardisation of outcome measurement:
1. To standardise and agree on one set for all
2. To have different outcome measures that reflect what matters locally
Both methods have their own merit, and regional or international variation is important, but in general, one standardised set of outcome measurements for each condition would be very useful.
In the next five years meaningful clinical outcomes will start to be collected at a health economy level. PROMS will likely gather some momentum and become more widely used in this time, however, it will likely more than five years for them to gain significant traction. Predictive analytics will continue to develop and improve in tandem with this.
The use of outcome measurement for predictive modelling could produce a transformational change. Incorporating data from electronic records and other social determinants of health and using machine learning methods and clustered analysis to identify specific sub clusters of disease could give a better understanding of diseases. This could allow earlier interventions and specific behaviour change models to be incorporated in care pathways and create bespoke medicine for a specific subset of disease.
There are a lot of variables that can be recorded that may give some insight into health, including biochemistry, genetics, health seeking behaviour, medication compliance etc. Social data could be linked with this to give further insight.
There will be an increasing reliance on patient generated and patient held data. The patient will hold their data and use it as they see fit. Mobile health technologies will continue to play an increasing role. Patients may hold their health record on their mobile devices and link this with their own activity data and social media data. This could lead to an abundance of data that was not previously available and the patient would hold much more medical information about themselves than the NHS would. Doctors may have to ask patients for permission to see their data. With these systems evolving with increasing scale the patient generated information will be key to understanding their health.
Change in Research
Recording of outcomes could be a game changing application in research. Simple comparative effectiveness research, using routine data, is possible at present and understanding of all of the newly available variables and outcomes will give further information to these types of studies. This may not always provide the gold standard research evidence, but it may be the best available and the knowledge generated from such studies could be used until a randomised control trial examines the area.
Big data can be viewed as hypothesis generating system which could then be tested in the appropriate randomised control trial (RCT). This would be a major change in the way we look at so called evidence based medicine.