Poor data can sabotage AI initiatives in healthcare

healthcare data

Artificial intelligence (AI) and machine learning (ML) already are having a major impact on healthcare in numerous ways. Here are some headlines on AI and healthcare from just one recent Google search:

As you can see, AI is being applied to a diverse set of clinical and operational healthcare challenges. But all uses of AI in healthcare have one thing in common: They are driven by the ability of a machine (or algorithm) to “learn” – and thus improve accuracy and efficiency – by analyzing large amounts of data.

That’s why quality data is critical to any successful healthcare AI initiative, and why low-quality data can make it difficult or even impossible for providers to improve patient care and operational efficiency – or even to correctly identify patients.

How data quality affects healthcare

In a 2019 survey of healthcare providers by eHealth Initiative on the challenges of patient matching – the ability to access a patient’s accurate data across electronic health records (EHR) systems – 66% of respondents said data entry errors were the greatest contributors to patient matching problems, while nearly four in ten (38%) said lack of data governance was a barrier to improving patient matching rates.

Data-quality issues not only can affect individual patient care, they can impede efforts to collect and analyze social determinants of health (SDOH) such as addiction, food insecurity, homelessness, and poverty to improve population health.

“Healthcare leaders can’t address SDOH without understanding the populations they serve and without the capacity to drill down and create a comprehensive view and database of the health of individual patients. That’s impossible to do with inaccurate, incomplete or duplicative patient data, “writes Andy Aroditis, CEO of patient indexing technology vendor NextGate.

Conversely, he says, healthcare data that is accurate and complete “is well prepared for analytics.”

AI can be used to analyze SDOH data across multiple populations to gain comparative insights on health trends that may be useful to providers as well as medical researchers, government agencies, and community-based organizations.

“By ingesting materials ranging from electronic health records to scanned images, AI can use machine learning to train the model to hone in on the key social determinants relevant to a given patient,” writes Healthcare Finance News associate editor Jeff Lagasse. “AI’s ability to tease patterns out of large data sets allows SDOH integration to become more widespread, more relevant, and more actionable.”

Only, however, if AI is working with quality data.

3 steps toward better data

So what can providers do to ensure their AI initiatives are getting the right data to “learn” the lessons that can improve patient care and population health?

  • The first step is to determine the sources of bad data. In many organizations the main source often is data entry at patient registration. This can be addressed by establishing clear steps (or best practices) for collecting data, communicating the importance of clean data for patient care and organizational efficiency, and implementing a consistent training program for current and new employees.
  • Another common problem is data that is unstructured or lacks elements that can be identified or categorized in a provider’s system. This data must be standardized and structured; otherwise it is of little to no value.
  • Healthcare organizations also should determine if their internal processes are preventing the collection and accessibility of relevant patient and population data. Changes in process may entail further staff training.

As AI and machine learning become more deeply integrated into our healthcare system, it is critical that providers and other healthcare stakeholders are committed to collecting and storing high-quality data. Failure to do so can endanger patients and undermine population health efforts and medical research.

Speak Your Mind


This site uses Akismet to reduce spam. Learn how your comment data is processed.