The information ecosystem is growing at unprecedented speed, and technologies with advanced capabilities to track and assess that information are multiplying. Smartphones. Wearables. Connected medical devices. All of these innovations harness the power to transform health outcomes; all require constant data collection and submission to do so.
Welcome to the world of big data.
In a report on big data in healthcare from Healthbox, experts shared their insights on how to break through the noise in a health ecosystem swelling with more data than ever before. The report noted that the term ‘big data’ was originally coined in the 1990s to describe data sets too large or complex for traditional databases to handle.
Webinar: Harnessing Big Data in Healthcare
According to James Gaston, the senior director of maturity models at HIMSS, “[Our cultural definition] is moving away from a brick-and-mortar centric event to a broader, patient-centric continuum encompassing lifestyle, geography, social determinants of health and fitness data in addition to traditional healthcare episodic data.” The industry is on the cusp of learning just how powerful big data in healthcare is, he noted.
“The sheer volume, velocity and variety of data being collected poses challenges for harnessing and ensuring its validity to benefit both the macro, population-level health and the micro, evidence-based precision medicine,” the report stated. In other words, finding meaning within mountains of data is a massive undertaking for any individual in any role across the health system.
This is where the power of innovations such as artificial intelligence (AI) come into play. Lily Peng, MD, PhD, product manager in the Google Brain AI Research Group, explained that while human intelligence is best suited for integrating small numbers of very large effect factors, AI is particularly adept at combing through and identifying patterns in vast numbers of very small effect, or obscure factors.
The report also emphasized an important point about AI: human and artificial intelligences each have their own unique variances which will inevitably influence how to best apply each and embed them into workflows.
In a world awash with data, people can rest assured that though AI and big data in healthcare hold immense potential, there are still limitations that prevent them from being a substitute to universal decision-making. No single innovation should exist as the single solution.
Incorporating a complementary approach to care with big data can help catalyze actionable health insights rather than adding a new layer of complexity to clinical workflows. This, however, will require careful consideration into evolving models of care provision and decision-making, the report notes. The result could very well be the evolution of enhanced clinical decision-making and more personalized care delivery than ever available before.
Pairing the power of machine learning and human intelligence to obtain valuable insights from large data sets will require focus on four different areas, as shared in the report.
“The lens that each investigator brings to big data creates inherent biases,” the report noted. This can include everything from categorization of evaluated data, how that data was collected, etc. “It is assumed that the power of high-dimensional data resides in the absence of hidden confounders that remain undisclosed in the data. Unfortunately, this assumption is far from being a forgone conclusion and poses a threat to the validity of conclusions derived from big data by AI techniques.”
“Due precaution must be taken to structure analyses such that reverse engineering of patient identities does not occur; however, it is worth noting that the benefit of shared open data exceeds the adverse potential for re-identification of the individual.”
“Society will have to come to grips with weighing the ethical trade-offs between the benefits of shared open access to data and the finite, but real, possibility for re-identification of individual people by reverse engineering of the segmented data. Human intelligence, not artificial intelligence, will be required to grapple with these questions.”
Use of big data in healthcare can pave the way to provide patients with more detailed, comprehensible guidance for on how to manage chronic diseases and other major health conditions. But will the increased access to this information directly lead to improved outcomes, satisfaction, and consumer experience as a whole?
“The integration of data, AI-derived knowledge, and informed clinical decisions must be adopted by and tightly interwoven into clinical processes and workflow to drive potential benefit in the care of patients. Appropriately structured clinical trials are needed to demonstrate that the incremental benefits of a data-driven care process justify any costs (and complications) incurred by these decisions.”
Healthbox emphasized the fact that in data analysis, it’s important to keep in mind the age-old rule that correlation does not imply causation. It’s also important to “ensure that the data subjected to analysis does not suffer from the omission of confounders that may be causally related to the measured outcomes.”
“Domain expertise and human intuition will always be required to work in tandem with artificial intelligence to confirm the absence of hidden confounders... The use of machines can help humans reveal these undiscovered or unanticipated variables.”
With these insights in mind, it is clear that through a collaborative approach, we can better strategize for success with big data in healthcare, which will get us further on our way to harnessing the ultimate powers of health innovation. The ongoing advent of AI technologies will amplify the value of big data, paving the way for a more collaborative, human-centered approach that is helped, not hindered, by new technologies in health and care.
Learn how one health system used analytics to shift clinical operations, patient and staff scheduling, revenue cycle activities and population health initiatives during COVID-19.
Originally published 6 March 2019; updated 14 July 2020