Client: Care Quality Commission
Timeframe: February 2018 – May 2018
The CQC has a duty to ensure all relevant healthcare services are providing safe, effective and high-quality care, and to take appropriate action when services require improvement so that patients and service users are safeguarded. It operates as a number of specialised area teams who inspect different healthcare sectors, such as Adult Social Care (ASC) and Hospitals. These teams cover a geographic area, and hold responsibility for the appropriate inspection and follow-on actions within their portfolio.
The client wished to explore options for the implementation of an ‘Expert System’ to provide several core functionalities that allow the organisation to operate more effectively and efficiently, and to develop new methods of organisational learning and decision support with novel applications of data science. The requirement was to develop a set of functionalities that match the attributes of what an ‘Expert System’ needs to provide, and to focus on areas where there is the largest potential for gain or value. This implicates a range of system features, and a range of matched technologies based on need and term – hence the term ‘Expert’ is relative to the ability to meet client needs and urgencies of those needs.
This project focused on ASC as an example of a provider set with fewer and less organised available datasets. There can be a need for frequent unscheduled inspections where intelligence indicates a concern. In addition, the sector is especially reliant on structured on-site observational inspection.
Following a series of meetings with inspectors, managers and programme leads, we produce a series of traditional, enabling, AI and machine learning (ML) components to meet the needs and priorities of the organisation, its workforce, and patients under the care of providers. Our solutions are specifically designed around a phased delivery to manage past, present and future information streams, as well as maximising benefits from contemporary technological advances. Plans have been structured to account for development delays and attempt to guarantee value by ensuring that essential components and the prerequisite functions for an AI processor are introduced initially, with more advanced functionality arriving at a time the client is ready.
Due to a complex set of provider-inspectorate factors, there are justified reasons as to why the inspection process currently operates as it does, as well as reasonable concerns about the challenges involved in transitioning to a system that employs AI.
The aims of the project are:
- Evaluate the hardships faced by inspection teams
- Discover what the most valuable functionality is
- Gauge organisational appetite and preparedness at several hierarchical tiers
- Review past efforts and existing systems and processes
- Develop a set of system attributes with the client
- Propose a technical solution for implementation, with justification for technologies suggested.
- Assemble these in the most appropriate order.
To deliver the following benefits for the CQCs ‘business plan’:
- Efficiency (capacity)
- Choice and control
- Risk management
Several opportunities were identified from our discussions. We found universal optimism for an ‘Expert System’ that makes information easier to find and faster to analyse, with an emphasis on increasing the time inspectors spend on high value activities. The key findings were:
- Pre-processing of information inflows to a single central system. As intelligence currently stems from many origins, having relevant information assembled in a structured way is seen as a valuable way to assist inspector review, workload and intelligence processes.
- Intelligent scheduling. Scheduling priorities change with information and time. Adaptive scheduling through the Expert System, based on a series of information streams can provide inspection teams with options to mitigate time-criticality, and promote improved balance to workflows.
- Inclusion of the information and intelligence back catalogue. Making relevant information and intelligence rapidly available, indexed appropriately and tractable for the purposes of analysis.
- Algorithm visibility. All machine outputs need to be visible and verifiable to increase trust and improve accuracy through guided learning. A ‘black box’ will not provide the necessary clarity, and the system must return information that matches human responses, and is trained by staff to improve accuracy over time.
- Risk management. Early, predictable, repeatable, responses to risk are essential. There is a potential to develop ‘swarm intelligence’ across a range of inspection teams and sectors.
- Automation and semi-automation of certain time intensive tasks
- Capturing tacit knowledge. Ensuring that intelligence is available for analysis.
- Decision support.
Some of these components are standalone, while others build as the system develops. A system that learns from human ‘training’ is preferred, so that predictions improve with multiple human adjustments.
A phased development was proposed to provide tangible value, and build system functionality gradually.
Phase 1 – Building blocks
- Content identification and extraction via document and textual analysis.
- Defining information relationships – creating the systems ontology.
- Making tacit knowledge explicit
- Testing and optimisation
Phase 1 introduces rapid and accurate information retrieval, sufficient and full evaluation, and supports automation in later phases. It also allows information to be used in justifications an explanation.
Phased 2 – Enhanced information capture
This phase introduces the main components necessary to build a guided expert system, and introduces interactivity and the necessary design to promote it.
- Metadata, and automated development of metadata
- Decision learning and validation
- Interactivity and visual aids
Phase 3 – Supported decision making
This phase introduces most of the AI components, and are independent functionalities built on the foundation of a validated Expert System developed to Phase 2.
- Evidence mapping to build knowledge models based around decision contexts. This makes use of ontologies, indexed information and metadata to assist decision making. This can also provide alerts when an observed difference differs more widely than is acceptable from the expected.
- Potential for developing automated information models that inform decisions. These can produce novel and non-obvious insights from pattern recognition.
- Managing temporal information. There is very significant potential for automated management of information using date stamps and more sophisticated attributes such as temporal clustering, identification of repeated patterns and synchronous events that may be linked and presented to users through the visualisation of such relationships for accessibility and clarity.
More Case Studies
Product: Cows Milks Allergy Formula Sector: Lifescience Summary Methods Analytic…
Case Study | Lifescience
Client: Sanandco Sector: Lifescience Background Sanandco developed a device that…
Case Study | Lifescience
Client: NHS Improvement Sector: Healthcare Timeframe: February 2018 – May …
Case Study | Analytics
Client: Hillingdon Hospital Sector: Healthcare Timeframe: February 2018 – …
Case Study | Analytics
Southern Derbyshire CCG sought an evaluation of 12 initiatives on the Urgent Car…
Case Study | Analytics
Client: Care Quality Commission Sector: Healthcare Timeframe: February 2018 R…
Case Study | Analytics,General