The Devil's in the Digital Health Data

Data, data and more data. We're increasingly obsessed with it but is it actually clinically useful and do we even need it?

Whether it's IBM Watson or the myriad of analytics startups springing up all over the world, there's clearly an appetite to apply advanced systems such as machine learning and natural language processing to clinical data. Building the technology though is only half the battle so here are the other challenges we need to consider:

1. Workflow

There are so many intersecting clinical and administrative workflows in a single hospital, let alone between institutions, it can be difficult to map. Many analytics systems need to enter healthcare focusing on a single workflow, such as a cancer pathway to start, before expanding their scope. The first step is therefore to make sure you have a team that can spend time in hospital with staff to observe how data flows before planning how to analyse it.

2. People

Clinicians are notorious for the workarounds they deploy to get around sluggish systems and processes. Therefore many data flows can be unpredictable and volatile to map out and subsequent inferences sometimes nonsensical. The best way to understand the flow of data and how it impacts the clinical user experience is to have clinical expertise on your side as a core member of the product team.

3. IT Ecosystems

IT system interoperability is woeful in healthcare but is critical for any analytics solution to be successful. Analytics vendors need to have strong partnerships with a range of IT vendors, particularly in the EHR and EDM fields, to be successful. More importantly, a shift to open API based integration and even open source components may be required in the long term.

4. Outputs

A key question for any analytics platform is the type of of metrics it outputs. These need to be tangible and actionable and in a language that is relevant for both clinicians and management staff to work with. Many inferences and data trends are frankly useless and so contextualised outputs are key to a worthwhile solution.

5. Cost

Despite the perceived benefits of analytics solutions, cost is still a major barrier to adoption. This is especially the case when trying to sell directly to healthcare providers and without strong data source partnerships (e.g. EHR vendors etc.) can make a product a non-starter. Business models need to be considered closely and costs addressed through strong contextually relevant business cases.

These insights may sound relatively basic but you'd be surprised how often technical developments are hyped above the practical realities of implementation which are often far more important.