Leveraging artificial intelligence to improve healthcare organization operation

Or, it can be described as “a system’s ability to correctly interpret external data, to learn from such data, and to use that learning to achieve specific goals and tasks through flexible adaptation.” Unlike traditional analytics, AI shifts from reactive to proactive analytics that offer more real-time and actionable insights.

AI has particularly powerful potential in heavy data-generating industries such as healthcare. It offers enormous computational power that can reveal patterns and trends to facilitate predictions that enable optimal decision-making as well as empower business professionals to make better informed decisions.

Limitations on data use in healthcare analytics are numerous – in particular, different data is often siloed in repositories which do not communicate easily with one another. Not only does a large amount of data pose a challenge in terms of the resources required for analysis, gaining access and extracting from these different sources can consume at least as much time and resources as the data analysis itself. The healthcare industry currently relies, largely, on traditional analysis that yield merely descriptive analytics.

Descriptive analytics summarize raw data to make them easier to understand but require humans to interpret this past behavior and consider how it may influence future outcomes. With AI, healthcare organizations can reap myriad benefits by leveraging predictive and prescriptive algorithms to improve operations, reduce fraud, manage patient health, and improve patient outcomes.

Improved operational performance

While accurate, real-time information is crucial for effective business decision-making, healthcare organizations face numerous data obstacles. Data is housed across departments and exists in a variety of formats, from a range of sources and with different purposes.

For example, within a healthcare organization a wide range of data exists, including: clinical data in claims and utilization management, grievance and inquiry data in member services and other departments, case management and initial health assessment data, cost data from providers, network adequacy data, regulatory survey data, and self-reported data from patients through healthcare organization-owned applications. Formats vary, with certain data fields limited to specified coding standards, such as those for diagnoses and treatments, and claims management data must be coded to be “processable.”

Additionally, healthcare data comes from a variety of sources including the providers and the payees. The patients themselves can even be the source for a range of information from clinical data, such as self-monitored blood pressure readings, to contextual data such as customer satisfaction comments in surveys or shared on social network platforms. Finally, to further complicate matters, healthcare data is uniquely subject to confidentiality and privacy constraints that make its manipulation highly monitored and sometimes requires extra steps, prior to any analytics processing, to anonymize data. As in other highly regulated industries, various federal and state-specific policies are imposed to provide safeguards for the public, reinforcing the requirement to accurately generate, track, analyze, and report on more data.

By addressing these data obstacles, healthcare organizations have the opportunity to make great strides in their operational performance. At a minimum, it is critical for healthcare organizations to establish and utilize clear definitions of terms and data fields to ensure the accuracy of subsequent analytics and mitigate data discrepancies. Once the gap in the disparate data sources is bridged, a healthcare organization can more effectively monitor operational data and develop predictive analytics to improve operational performance. Some examples of this performance include the following:

• Improve Network Information. A health plan can combine data from member service inquiries and grievances, utilization management, network adequacy and regulatory survey data inform networking decisions. For example, this data can help predict the plan’s ability to serve an area prior to expansion, or it can help to identify gaps within the plan’s network prior to their causing any impact on members or resulting in regulatory non-compliance.
• Better Resource Allocation. With AI fed by multiple data sources, such as claims data, utilization management data, network adequacy data, initial health assessment data, enrollment data, and regulatory rule change data, a health plan can better predict changes in future claims, appeals, and provider dispute volumes to most efficiently allocate its processing resources and anticipated workloads on related issues, such as capitation deduction and recovery.
• Improve Financial Decisions. Healthcare organizations can utilize machine learning algorithms to make better-informed investment decisions. AI can recognize patterns, develop strategies, and show different scenarios while considering multiple variables.
• Harness AI and Deep Learning to Enhance Online Security. In particular, AI can be used to detect malware and thwart attacks.
• Administrative Efficiencies. Any healthcare organization can utilize AI to optimize electronic health record (EHR) work flow by enhancing natural language processing tools and automation.
• Improved Fraud Prevention and Detection. By using AI with multiple data sources, healthcare organizations can better detect and prevent fraud. For example, geographic information can be matched to claims data to identify outlying pharmacies or providers.

Case and utilization management

The inherent complexity of case and utilization management make it well-suited for AI. The high number of variables present in relevant patient data drive the complexity of this type of analysis and require powerful, high-performance computing environments to output trained models with a superior predictive capability. By relying on descriptive analytics, even the most sophisticated ones, organizations are limited to reactive, rather than proactive, solutions. The predictive analytics resulting from AI add “intelligence” to many mainstream healthcare applications and, when combined with human interpretation, can drive better decision-making.

Using data to establish risk profiling and predictive modeling, a healthcare organization can establish early and timely interventions that reduce unnecessary action and ultimately improve the quality of care. Similarly, when combinations of risk factors evident within the data indicate a high probability of utilization of specific services, a healthcare organization can establish case or care management programs to more effectively address member needs and potentially avoid utilization of unnecessary services. For example, by combining medical and pharmacy data, healthcare organizations can develop interventional programs to reduce emergency and hospital services for members with chronic diagnoses, such as severe congestive health failure.

Additional applications for AI within case and utilization management include:

• Population Health. Populations can be scored against the risk of developing chronic disease based on social determinants. The power of AI and advanced analytics can be fully unleashed when the data used extends to external or contextual data – such as social-demographic surveys or social networks.
• Genetic Makeup and Possible Disease Progression. When predictive analysis is applied to patients’ genetic makeup and the normal progression of diseases, invaluable benefits are created for patients, providers and payers.
• Avoidable Procedures. Technology can help identify ineffective patient paths or service patterns and help reduce the number of healthcare procedures and services performed that could have been avoided without negatively impacting patient health.

Claims Management Over the years, payers have built a relatively robust knowledge of common anomalies when processing patient claims. These anomalies are typically organized by typology, in “libraries” that constitute the basis for screening data using “rules.” Errors range from simple data entry issues to inconsistent or incomplete information to misinterpretation of existing data. Significant opportunities exist to improve the coding and definition of known errors and to automate the processing of large volumes of data in near-real-time. By harnessing the power of AI, real-time oversight of claims processing may become possible.

Claims data itself can be used as a management tool to ensure the ongoing monitoring of all claims and their processing. For example, reports generated from the claims data can assist leadership in monitoring myriad operational and compliance issues related to claims, including managing timeliness, duplicate claims, misdirected claims, claims in pending status, medical record requests, aging and check runs, and accuracy of claims payment, including interest and penalties, and managing provider disputes and recovery.

Claims data, or encounter data, holds a wealth of valuable information for any healthcare organization. Claims data can be used to assess the overall cost of individual and collective service and can be the basis of focused programs and strategies to reduce cost and improve the quality of care.

For example, if an organization is experiencing high cardiology costs, it could assess claims with cardiac-related diagnoses and those with diabetes-related diagnoses, as well as cross-validating this with other information, such as pharmacy data. This analysis may help the organization identify a pattern, which could result in a focused strategy to ultimately address this issue.

Claims data is an excellent source of information to monitor provider performance regarding coding accuracy, over- or under-utilization, and the identification of general patterns which can be used to compare to the rest of the network, benchmarks, and industry standards. Moreover, the provider information derived from the claims data can be used to validate the appropriateness of capitation payments, as well as support associated contracting decisions.

While healthcare organizations are aware that fraud is rampant within the industry, detecting fraud remains challenging. Fraud occurrences are difficult to catch because they are “hidden” in a large mass of transactions and get lost in statistical averages. They often straddle data silos and take advantage of organizations’ inability to merge disparate data sources.

Even more challenging is trying to identify previously unknown fraud schemes. Fraud schemes are not static phenomena. They shift, patterns morph, and new schemes emerge. While known phenomena are better addressed by predictive and prescriptive technologies, such as supervised machine learning, new patterns discovery is better done using unsupervised methods, such as self-organizing maps.

Upon identifying these anomalies, deeper probing can be done by asking more precise and pointed questions using supervised techniques. By utilizing AI to bridge multiple data sources, healthcare organizations can better prevent and detect fraud. For example, when organizations can combine provider data, member data, and vendor data with historical claims data, as well as other data sources, organizations will be in much better positions to detect and ultimately prevent fraud. While the development of systems to prevent and detect fraud using these types of sophisticated analytics technologies, such as both supervised and unsupervised learning, are still underway, the potential impact of such systems is immense and provides an attractive return on investment.

Conclusion

While the application of AI presents significant opportunities, there are limitations. In particular, AI can never replace human intelligence, but can complement it. Without good source data, AI cannot create valuable insights. An organization’s investment in AI should be an ongoing process rather than a one-time event. To grow into a mature data-driven company, this investment will require far-reaching changes throughout the organization:

• Operations. Evolving from “data from various sources produced and stored in siloes” to “consolidated company-wide data with basic reporting capabilities” to “analytics on top of automated consolidation of various data to gain understanding of business performance.”
• Patient data. Going from “manually sharing patient data between entities with limited manual integration across systems” to “automatically integrating and centralizing all available data using standardized data models and basic analytics.”
• Institutional. Moving from “collaboration between business units and/or departments to transfer knowledge” to “leveraging synergies between stakeholders driven by strong collaboration and knowledge sharing across business units and/or departments.”

Enablers of these changes include mirroring current data warehousing infrastructure with the creation of modern, flexible and evolutionary new ones, such as data lakes, and reinforcing technical teams’ expertise by gaining access to new skills in data science and data architecture – internally or through trusted partners.

Healthcare organizations are well-positioned to enjoy countless benefits by employing the use of AI. In particular, the rising costs associated with healthcare delivery, the highly regulated environment and ever-changing regulatory demands, the ongoing prevalence of fraud, and the numerous varied data sources facing healthcare organizations make AI an invaluable tool going forward.

Ready to get started?

Contact us

Published on March 13, 2019

The information provided here is for general guidance only, and does not constitute the provision of tax advice, accounting services, investment advice, legal advice, or professional consulting of any kind. The information provided herein should not be used as a substitute for consultation with professional tax, accounting, legal or other competent advisers.