10 lessons I’d like to share from my experiences in working with healthcare data and software systems…

A presentation for @LetsDoDigitalCI last week… 1. The same data can be used for direct care, service management, continuous improvement, audit governance, and research if we do our job right. Different levels of granularity, and governance wrap, but we artificially separate these.

I gave a talk on “Unbundling the electronic health record” at #reclojure22 - . I think we need to turn what we think of as the “electronic health record” inside out. The conventional approach means thinking in terms of applications, and not data, and is centred on organisations / providers, not the patient, or the wider ‘service’. Architecturally, it is a monolith, at least to outsides, and proprietary and organisationally-centric.

What even is a ‘current medication list’?

Who coined the term shared medication ‘repository’ and does that concept even make sense? In certain constrained clinical environments, such as inpatient care, it might be a reasonable assumption that the inpatient electronic prescribing medicines administration (ePMA) system is the 'golden record' representing the ‘current medication list’

Graph queries across distributed heterogeneous patient data make run-time analytics and inference magically easily. Here I issue a query - a bit like SQL - for some patient data for patient 17490 - I’m asking for attributes but something special is happening. What is this doing? If you look carefully, I’m asking for attributes - but I’m drilling into those and having my queries resolved by multiple microservices; this is a federated distributed electronic health record. We start at an edge and navigate seamlessly. Notice how I can resolve ethnicity and

So I built an electronic patient record system that's been running since 2007. It was, as far as I am aware, the first implementation of SNOMED CT in a live clinical system. Most users have never known that SNOMED is its *lingua franca*, and most think it is a bespoke 'database’. They think it’s bespoke because built multiple subject and specialty-specific "prisms" through which to see, what is essentially, single record. I just take some basic building blocks and stitch them together.

Hi Pritesh. I’ve answered like this so I can do in a thread.

I guess I’d start by some looking at some basic principles and then weighing up how both openEHR and FHIR meet those principles.

What’s important to us?

https://twitter.com/mustbemistry/status/1373204999254200322

1. We want the right information at the right time, for patient, for professional, for direct care, for management, for quality improvement and for research. We use that information to make the best decisions we can.

So graph-based APIs / queries are a natural fit in health and care - building a semantic graph across health data using first-class properties. Think RDF for health.

https://twitter.com/amidgley/status/1359180844133416963

For instance, I'm now able to resolve e.g. an index of deprivation score against a patient by walking the graph of properties from current address to LSOA to the index. At the client level, I don't care how those data are fulfilled.

We need to take open-source mainstream for health and care, with public sector investing in in-house software development capability. We need to move from trying to integrate full-stack applications to thinking about data and software services in a platform. There will be a flywheel effect from the availability of open-source, standards-based foundational services and components that can be composed in interesting ways to solve user problems in health and care. Big wins from composing small, independent little wins that stack up!

My new blog post: how to use "appointments and scheduling" to help us learn how to build a seamless, "one system" health and care service, made up, paradoxically, of small, modular internet-era software and data services. wardle.org/platform/2019/… Most us of recognise that health and care need to provide a suite of computing & data services forming a platform on which a range of user-centred solutions can be built, underpinned by robust open standards?

Let's move from command & control to empowerment & collaboration.

So I've been practicing my @swardley mapping to try to look at one facet of our health and care architecture and ecosystem: patient demographics. Wardley maps use a value chain to show user need and their dependencies running from top to bottom. Understanding patient identity is pretty much essential to all that we do in health and care. One of major problems in health is that legacy applications have tended to be "full stack", containing different components that are user-facing, do business logic and that do data.

A new blog post: Let me tell you why we, in health and care, need to copy how the highest performing technology companies build & deploy software. You might accuse me of trying to create a public sector /NHS “Google” or “Amazon” and you might be right.
wardle.org/strategy/2019/… Firstly, I’m sure that we all agree that we want to create a seamless patient-centred service that blurs the boundaries between disciplines and organisations across health and care. Likewise, we surely recognise that technology is an enabler of those ambitions.

An important message for healthcare. The future is not running your own data centres. It is meaningful data operated upon by stateless cloud-based logic, created by humans and machines, stitched together declaratively, with analytics pipelines & not just single data repositories.

https://twitter.com/swardley/status/1071427090812162048

But not clear to me that policy-makers or industry have the foresight to embrace this new future, the next generation of electronic health systems, in which data drives decisions for direct care, service design, quality improvement, research & design of the systems themselves.

This is a brilliant article. There are some key points: realising “digital” isn’t about digitisation but deeply understanding our work and how we manage change, and manage our organisations.

https://twitter.com/Atul_Gawande/status/1059431597546594305

He gives an example of a real-life/model mismatch, thinking a single problem list can be shared within an organisation of scale. The description of the clinician’s frustration is one reason I have designed multiple context-bound problem lists based on pathways and services.

@bengoldacre @swardley Currently limited. My own EPR publishes structured data but everything else is application-bound or administrative and not usually clinically meaningful. It is, therefore, aspirational. Bedrock is meaningful data in open structured formats. @bengoldacre @swardley Thinking about it, it depends what you mean by API. If you are asking whether they exist as a specification, product or commodity, then yes - either running as a repository or a facade via HL7 FHIR.

How can we become data-driven in healthcare? It means making the logical patient record the core of our thinking and design and building from that solid foundation. But where to start? Can we use situational awareness and mapping (from @swardley ) to help us? We start with a simple one dimensional value chain with users at the top, their needs and then how we currently serve those needs.