How can AI applications in healthcare reduce costs and administrative burdens?

Artificial intelligence applications in health care can provide tremendous value to clinical and administrative workflows in healthcare provider settings. Administrative burden adds hundreds of billions of dollars to healthcare costs annually. A National Academy of Medicine report estimated the US spends about twice the necessary amount on billing and insurance-related (BIR) costs. This administrative burden amounts to roughly $248 billion in excess spending annually. Many of these costs are unnecessary and are associated with fraud, waste, no- or low-value-added work, and lack of collaboration between stakeholders. In addition to reducing costs, AI tools can speed up healthcare administrative operations, like the transcription of medical notes, by eliminating manual data entry, correcting human error, and automatic processing of electronic health records (EHR).

 

What are the causes of high costs and administrative burden?  

Spending on healthcare is growing at an unsustainable rate in the United States. The McKinsey Global Institute found that the United States spent approximately $650 billion more than any other developed country on healthcare in 2006, and not due to an growing sick population. A Research Gate study also showed that supply expenses in the US per patient, per admission, were estimated to be $4,470. There are a number of institutional procedures and practices that result in this massive excess of spending. 

Lack of care delivery & coordination causes higher hospital readmissions-and higher costs.

Failed care delivery is the result of poor execution or the lack of widespread adoption of best practices – like effective preventive care practices or patient safety best practices – in care delivery. Delivery failures can result in patient injuries, negative clinical outcomes, and worse, resulting in higher costs. Failure of care coordination occurs when patient care is fragmented, disjointed, and/or lacks coordination. This can lead to unnecessary hospital admission, avoidable complications due to human error, and poor management of patient transition from one care setting to another. Nearly a fifth of fee-for-service recipients discharged from hospitals are readmitted within 30 days, resulting in $12 billion worth of potentially avoidable costs and readmissions annually. Three-quarters of those readmissions are in categories of diagnoses that could have been avoided, had an adequate treatment regimen been prescribed during the patient’s first admission. 

Physicians over-treat patients with costly or ineffective services.

Overtreatment involves outmoded operations of care, driven by providers’ preferences, rather than those of informed patients. For example, a reliance on defensive medicine and intensive care at the end of someone’s life, which overlook scientific findings and can be motivated more by financial incentive than the provision of optimal care for a patient. A study found that this category of overtreatment in health care added between $158 billion and $226 billion in wasteful spending in 2011. Burdens of overtreatment can also result from overdiagnosis. This occurs when physicians attempt to identify and treat a health issue in its earliest stages, despite the possibility that the disease may not progress. 

Another form of overtreatment is the application of higher-priced services when there are cheaper options available. The justification for directing patients and providers to the less costly alternative is clear when more expensive services provide negligible or no health benefits compared to less-expensive alternatives. The use of generic instead of name-brand drugs is an obvious example. For example, Budesonide, a popular drug used to treat asthma symptoms, has a generic price of around $40, yet the average retail price of its brand name, Pulmicort, can reach as high as $294. 

Administrative complexity in billing procedures. 

Administrative complexity leads to excess spending due to inefficient or flawed guidelines and overly bureaucratic procedures developed by private health insurance companies, the government, and/or accreditation agencies. For example, the US has a very complex and time-consuming billing process due to the lack of standardized forms and procedures.

  • Cost of Services and Pricing failures — These issues occur when the cost of a service exceeds that of its proper market value which should be equal to the specific cost of production and some profit. Healthcare providers in the US tend to utilize more expensive, but not necessarily more efficient modes of treatment in comparison to other countries. This can be largely attributed to a lack of transparency and competitive markets. For example, the costs of magnetic resonance imaging (MRI) and computed tomography (CT) scans in the US are several times more expensive than in other countries. Moreover, the US spent approximately 17.1 percent of the country’s GDP in 2017 on health, while Turkey spent a mere 4.2 percent of its GDP in the same year. The Trump Administration has announced new price transparency requirements, as of November 2019, with the goal of increasing competition and lowering healthcare costs for all Americans.
  • Fraud and abuse — fraud and abuse are also significant contributors to the high cost of healthcare. In addition to fake medical bills and scams, this category includes the cost of additional inspections and regulations to catch wrongdoing. A study conducted by Berwick and Hackbarth estimated that fraud and abuse added $82 billion to $272 billion to US health care spending in 2011.

Causes pulled from “Reducing Waste in Health Care.”

 

How are AI tools being used to reduce costs and administrative burden? 

Artificial intelligence (AI) research is growing rapidly among healthcare entities. In 2016, healthcare AI projects attracted more investment than AI projects within any other sector of the global economy. Much of the investment in AI for healthcare is focused on reducing costs and administrative burden. Here are some of the ways this investment in AI is being put to use in healthcare.

 

Natural Language Processing to Make Use of Large, Unstructured Datasets

Natural language processing (NLP) supports full speech recognition and interaction, helping to process large amounts of natural language data. Health providers have limited access to socioeconomic, behavioral, and environmental data about their patients. This data is needed to create actionable analytics, and lack the resources to analyze big patient data. NLP can be used to extract and interpret handwritten medical notes and text records, and other unstructured data sources. NLP can also support systems like answering unique free-text queries that require synthesizing numerous data sources.  

EXAMPLE USE CASE: Acusis Medical Transcription Service 

Acusis provides multiple options for dictation-capture for use with their medical transcription services. The first is AcuVoice, which is a cloud-based system by which you can dial in to a toll-free number and use a PIN to record your dictation. You can then use a web interface to access, replay, edit, and otherwise manage your recorded dictations. 

Alternatively, Acusis can provide an option to integrate with your existing EHR using AcuSuite, which is also a cloud-based system which can record audio from a number of sources available in most clinical facilities. 

 

AI Automation

Automation can also be used to automate processes that support EHR digitalization and deliver the same or better results more quickly than human processing. Robot Process Automation (RPA), a form of automation, is a very promising emerging technology. For example, clinicians typically manually type notes into a patient’s medical health record. With the use of RPA, clinicians will have the ability to auto-record notes and conversations, then combine RPA bots with NLP capabilities to automatically translate information from the conversation into a readable health narrative.

EXAMPLE USE CASE: “More Effective Management of Supply Processes: claims and billing.”

Stonehill’s Automation Platform can be set up to capture new patient appointments and automatically transition them into the various systems needed to initiate invoicing. Robotic process automation (RPA) not only starts the billing process earlier, but reduces the decline rate of claims made as well.  By automatically matching ICD-10 codes with covered treatments and coverage amounts, medical practices can be sure to submit bills properly the first time, reducing error rates and payment timing. The RPA platform can also expedite the recovery of co-pays or deductible fees. 

RPA software interacts with data sources and software platforms the same way staff typically would. Tasks can be designed, recorded, automated, and extended to provide better user experiences at an affordable cost. Stonehill helps healthcare providers implement RPA software, navigate potential implementation risks, and seize the significant opportunities offered by robotics and automation.

 

Machine learning  

A subset of AI, machine learning utilizes algorithms and statistical models to perform certain tasks without the use of explicit instruction. ML is being used in a variety of medical contexts. For example, Google recently developed a machine learning algorithm to help providers better identify cancerous tumors on mammograms. Stanford is also in the process of developing a deep learning algorithm to identify skin cancer.

Machine learning technology can also enable payers to review compliance on a greater number of claims in a more cost-effective, and less burdensome manner. Under the current manual process, reviews can only analyze about 1 percent of claims, leaving plenty of room for overlooked instances of medicare fraud, waste, and abuse. The Centers for Medicaid and Medicare Services are working to implement such technology to lower costs.      

EXAMPLE USE CASE: “KenSci“

“KenSci” uses machine learning to predict illness and treatment to help physicians and payers intervene earlier, predict population health risk by identifying patterns, surface high risk markers, and model disease progression. KenSci’s Risk Prediction Platform for Healthcare is engineered to ingest, transform, and integrate disparate sources of healthcare data, including EHRs, claims, administrative and financial, and other. KenSci was built by doctors and data scientists to help providers and payers intervene earlier by identifying future patterns of clinical and cost outcomes. The Risk Protection Platform  identifies what drives healthcare costs by modeling the complex interplay of disease progression and utilization to anticipate chronic and critical illness.

KenSci recently partnered with healthcare consulting firm T3K Health to focus on helping caregivers harness AI and machine learning for health records and workflow. The platform has pre-built models, applications, and workflow integration to enable its users to identify problems, and gather solutions more rapidly. Kensci provides a health analytics platform with data sources including US Annual Mortality by Cause,  National Health Expenditure % of GDP from the World Bank. This brochure and  demo provide details on the platform and world examples, highlighting the power of its analytics.