Developing Healthcare Analytics In An Agile World
By Stuart Jackson
Big Data is everywhere. However, as the list of Big Data tools grows and grows, organizations of all sizes often struggle to build and maintain analytics infrastructures that are future-proof. Nowhere is this clearer than in the development of healthcare analytics.
Thankfully, new ideas are emerging in the software development world that seek to better accommodate the unique nature of analytics development - so called agile data science principles. This is evidenced by the growing recognition that data science development methods differ in key ways from more traditional software engineering, with the first textbooks now starting to appear on this topic.
For example, one recent agile data science textbook emphasizes how software developers must be able to adapt to the more dynamic nature of data science, which incorporates elements of both engineering and research science [1]. This is essential to avoid the many typical pitfalls in healthcare analytics development and to deliver consistent value-creation.
Yet, the maintenance of repeatable healthcare IT development processes proves particularly burdensome. For example, clinical IT projects often suffer from poor up-front planning and ‘scope creep’, and face numerous challenges arising from issues such as customer technical abilities and the communication loops between developers, stakeholders, and customers [2]. Presumably, these challenges result most often in under-utilization of the value hidden in the EHR.
This is where agile methods applied to analytics development become so powerful, supporting the whole development process, from product definition, to stakeholder feedback, right through to successful continuous delivery. The benefits of agile methods over other development approaches (e.g., waterfall methods) are numerous, facilitating the continuous refinement of the technical details of an analytics product, alongside the simultaneous evolution of higher-level team goals.
Consider the typical maintenance issues that face any new healthcare analytic. Industry reference codes change sporadically, a client requests new software functionalities, or some important piece of implementation hardware becomes obsolete [3]. These and other related issues can negatively impact the operation of both administrative and clinical healthcare software over time. To counter this possibility, agile data science approaches encourage continual feedback in the development process, such that during the iterative refinement from an initial prototype to full-fledged product, many of these potential future bottlenecks have already been accounted for.
In addition, an important principle in doing agile data science in healthcare is to aim to increase the complexity or formality of the machine learning experiments performed by the software developers across each iteration cycle. This research-driven agile approach acknowledges the importance of scientific principles in healthcare IT development, while housing the endeavor in the more streamlined and accountable framework provided by traditional agile, keeping stakeholders happy.
While the challenges facing any new healthcare analytic are often substantial, one must remember that the opportunities for machine learning application in healthcare are also great [4]. Thus, understanding the best approach to take in product development is as imperative as ever.
About The Author
Stuart Jackson is a Ph.D.-trained consultant data scientist with more than 10 years of experience working with advanced analytics in both high-end research and industry domains. He is a former Fellow of Insight Data Science.
References (last accessed online May 12, 2019)
[1] https://www.oreilly.com/library/view/agile-data-science/9781491960103/
[2] https://365.himss.org/sites/himss365/files/365/handouts/550231707/handout-147.pdf
[3] https://www.frontiersin.org/articles/10.3389/fdata.2018.00007/full