NHS Digital
contact "HAJ-NAJAFI, Arran (NHS DIGITAL)" <a.haj-najafi@nhs.net>
Three suggestions for 2019:
Models deployment
Models deployment NHS Digital use routinely collected data to build models to support NHS. Some of these are predictive models identifying high risk patients, the models themselves do not contain any sensitive or identifiable information. We invite you to explore/build tools to support the implementation and deployment enabling NHS Digital to share these models with health bodies so they can run these models at their end with minimum development work required. Model Background – Did Not Attend (DNA) are estimated to cost NHS over £1B a year. NHS Digital has explored HES data, and used feature engineering to build random forest models to identify patients at high risk of not attending an appointment. The output of the above machine learning model is a list of variables with estimates (i.e. gender; male = 0.234, female = 0.145). These models are to be run on booking data which Trusts hold with predictors and additional information on appointment. This will produce a RAG rating for every attendee i.e. Red – high risk of DNA, Amber – medium risk and so on. Challenge - There are multiple IT suppliers and various booking systems that Trusts use. These booking systems will have different formats and data quality issues amongst others. We invite you to explore how this model (R/Python script) can be executed on booking data at Trusts end with minimal development required for the Trusts. The model does not have any identifiable/sensitive information. However, the booking systems may contain personal, identifiable and/or sensitive information.
PoC - Live A&E attendance app for public
Emergency Care Dataset (ECDS) contains hospital code, hospital postcode, Accident and Emergency attendance amongst other variables. Create an app that is updated in real time showing attendance at any Accident and Emergency site to enable patients to make informed choice on which A&E site to visit when required.
* Could you clarify whether some part of the ECDS data is available as a realtime feed, or would it simply be used to train a predictive model for triage estimation?
Some kind of back end functionality seems necessary. We did discuss this one with one of his team in a call I had earlier in the summer. I think the plan was that the static dataset would be used to build a predictive model for improved interactive triage.
Enabling Independent Living for an ageing population.
When the NHS was founded in 1948, 48 per cent of the population died before the age of 65; that figure has now fallen to 14 per cent. Life expectancy at 65 is now 21 years for women and 19 years for men and the number of people over 85 has doubled in the past three decades. By 2030, one in five people in England will be over 65. This success story for society and for modern medicine has utterly transformed our health and care needs. Many people stay healthy, happy and independent well into old age, and there is mounting evidence that tomorrow’s older people will be more active and independent than today’s.
The right supply of housing in terms of location, affordability, size, tenure and facilities is a crucial factor in enabling people to remain in their own homes as they age. It is essential that new housing stock reflects the needs of the local ageing population, with sufficient extra care, sheltered and age-friendly housing available . Existing housing stock can also be adapted with aids and technology to assist older people with daily living and maximise their independence and safety. Adaptations and care packages can aid older people’s recovery after a hospital stay and can help them to remain in their own homes at the end of life. Providing adaptations to support an older person to remain at home for just one year can save £28,000 on long term care costs.
We invite you to build IoT tools to support Independent living for older adults cognisant of the fact that people over 55 make up 94 per cent of non-users of the internet (Centre for Ageing Better).
Idea led to Grand Remote
