By Rahul Mehta, senior vice president and head of data management proficiency, CitiusTech.
The sheer volume and variety of data, such as claims, EMRs, lab systems, and IoT now available to healthcare organizations is mind-boggling. The potential to pull data from these myriad sources to work for real-time care intervention, clinical quality improvement, and value-based payment models is unfolding fast.
Yet, as organizations seek to aggregate, normalize and draw insights from large and diverse data sets, the importance of data quality becomes apparent. Consider an activity as fundamental as identifying the correct patient. According to Black Book Research, roughly 33 percent of denied claims can be attributed to inaccurate patient identification, costing the average hospital $1.5 million in 2017.
For example, the average cost of repeated medical care due to inaccurate patient identification with a duplicate record is roughly $1,950 per inpatient stay and more than $800 per emergency department visit.
As data quality become more important, healthcare organizations need to understand the key characteristics that affect quality: accuracy, completeness, consistency, uniqueness and timeliness. However, data reliability and integrity also depend on other key factors, including data governance, de-duplication, metadata management, auditability and data quality rules.
With a strategic approach, healthcare organizations can employ a unified data strategy with strong governance for data quality across all data types, sources and use cases, giving them the ability to scale and extend to new platforms, systems and healthcare standards. The result is an approach that uses a combination of industry best-practices and technology tools to overcome common challenges and assure data quality for the long term.
Understanding Data Quality Challenges
Historically, providers and payers alike treated data quality as a peripheral issue, but that is no longer viable in today’s complex data ecosystems. First, there are a diversity and multiplicity of data sources and formats: EHRs, clinical systems, claims, consumer applications, and medical devices. Add to that, challenges associated with legacy applications, automation needs, interoperability, data standards and scalability.
Lastly, there are increasing numbers of use cases for clinical quality, utilization, risk management, regulatory submission, population health, and claims management that need to be supported.
Considering the current data environment, the downstream effects of data quality issues can be significant and costly. For example, in the case of patient matching as referenced above, something as common as two hospitals merging into the same health system, but following different data-entry protocols, can lead to duplicate and mis-matched patient records. It can also lead to critical patient data elements, such as date of birth, being documented differently by different facilities and then made available across multiple systems, in varying formats.
By Jeff Springer, senior vice president of healthcare solutions, CitiusTech; and Fernando Schwartz, vice president of data science and consulting, CitiusTech.
As consumers, we experience artificial intelligence
(AI) every day. In fact, we’ve come to expect the consumer experience it
enables, such as Amazon’s highly personalized suggestions for additional items
we might like. Now is the time to gain AI’s rewards in healthcare as well. Dr.
Eric Topol, founder and director of Scripps Research Translational Institute,
put it this way: “If properly and humanely deployed, AI has the potential to
restore efficiency to a wide array of burdensome healthcare processes, freeing
up physicians to treat their patients in the way they deserve. The path won’t
be easy, and the end is a long way off. But with the right guard rails,
medicine can get there.” 
While there are significant challenges with
interoperability, data sharing, and data access, organizations are taking
several approaches to overcome them. With an incremental approach to data
management and analytics, healthcare organizations can reap the benefits of AI
– specifically machine learning (ML) and natural language processing (NLP) faster,
enabling them to overcome challenges and achieve success in value-based care.
and Its Subsets
it’s important to have a shared understanding of AI concepts. AI is a term used
to describe the ability of machine intelligence to imitate human intelligence
through cognitive functions and behavior. Both ML and NLP are subsets of that broader concept. ML applies
algorithms and statistical models that effectively perform a specific task or
make predictions using patterns and inference. NLP enables computers to
understand, interpret and manipulate human language using computational
linguistics. At the highest level, these tools can be applied in healthcare to
help find the answers to questions and identify root causes—leading to workflow
improvements at a massive scale.
Advances Speed AI Adoption
Advances in data and analytics technology are making
it possible to significantly reduce implementation time for AI projects. From a
data management perspective, five years ago, it would have taken two years to
implement a strategy and infrastructure for data management needed for AI.
Hence, many organizations applied a project-by-project approach and didn’t take
advantage of the opportunity to reuse data for multiple projects. The results
were data silos and missed opportunities to leverage data across the
Today, new technology and new data approaches
expedite projects from the data, analytics and learning perspectives.
Specifically, today’s late-binding architecture enables organizations to take
only the data needed for a question, metric or pattern, and then curate
additional data as needed. This opens the door for going beyond the standard
healthcare sources, such as claims, HL7, and CCDA, to also include historically
cost-prohibitive data, such as social media, benefit information and
unstructured data. These data sets can be leveraged to determine the most
effective engagement models, risk patterns and communications. Today, if organizations
start by asking the right questions before implementing a new data set, they
could have answers in as little as three months, rather than years. By
leveraging existing infrastructure and data in conjunction with new technology,
organizations can begin to see success more quickly, while building out an
incremental, long-term strategy.
AI to Work Answering Healthcare Questions
organizations take advantage of the new technologies that enable sophisticated
data analytics, they can more easily apply ML and NLP to specific data sets.
Start with a question and determine what data can best provide answers. The
examples to follow illustrate questions organizations might ask to improve
clinical outcomes, revenue assurance, or operational efficiency.
Clinical Best Practices: Using
ML, healthcare organizations can more effectively analyze treatment patterns by
asking questions, such as: What interventions during a clinical encounter help
avoid sepsis? Or how can an ED visit be prevented? Once there is an
understanding of what contributed to positive outcomes, organizations can embed
care protocol improvements within clinical workflows.
Population Health: Again,
using ML, organizations can thoroughly analyze populations, treatment patterns
and results by asking questions, such as: Which segments are having trouble and
how can they be addressed? Are issues arising that are related to geography,
benefit structure, or disease pattern? ML can be applied to sift through many
dimensions and identify root causes. For example, a member of a diabetes
population segment lives within a certain zip code, sees a certain provider,
and is more likely to be readmitted. By asking the right question, variations
in care can be identified, and protocols and workflows can be adjusted to
Management: Healthcare organizations, especially
ACOs, strive to improve quality while tightly controlling costs. Using ML,
organizations can look for variations in practice patterns across providers,
patients and conditions. From there, organizations can implement protocols that
operationalize utilization management. For example, duplicate exams, such as
expensive MRIs, can be avoided by making recent results available to providers
at the point of care. And providers can be guided to the most cost-effective
location for a given procedure or exam.
Data Aggregation: Even with
standard EMR implementations, there can be tremendous variation in how certain
types of data are captured. For example, data points needed for population
health management may be captured in data fields and as unstructured text.
Using NLP, healthcare organizations can now parse many different types of data
from many sources. This enables access to data for risk-based contracts, such
as social determinants of health like transportation, weight loss, food
insecurity, and electricity.
AI to Gain Quick Wins in Quality Improvement
With today’s technology, healthcare organizations
can move more quickly to take advantage of AI, especially ML and NLP— seeing
results in months rather than years. With modern data and analytics approaches,
organizations can proceed incrementally to identify questions, metrics or
patterns that deliver quick wins in clinical, financial and administrative
outcomes. At the same time, these achievements contribute to a successful
long-term strategy to continuously improve quality and outcomes while assuring appropriate
payment in today’s value-based care environment.