Artificial Intelligence and Customer Service in Health Care
This chapter explores the intersection of AI, natural language processing, and speech recognition technologies to create a trusted virtual character. These virtual custom assistants (VCA) can provide consistent customer service to health care consumers, and, more importantly, when they need it.
Traditional customer service in health care is marred by inaccurate or incomplete information, inconsistent behavior of call center representatives, difficulty in predicting volume of calls and resultant staffing issues, and a high cost of human-to-human interaction for routine questions. For some demographics, mainly under age 35, the lack of a self-directed customer journey is essentially a poor customer experience. As cited in his article “Improving Health Outcomes with Better Patient Understanding and Education,” Robert John Adams (2010) states: “for individuals to realize the benefits of health education also requires a high level of engagement … Interventions to improve self-care have shown improvements in self-efficacy, patient satisfaction, coping skills, and perceptions of social support.”
Enter an intelligent virtual character. Unlike chatbots, which are voice-only, task-oriented characters, an intelligent virtual character has a humanistic form, can be animated in 3D, understands the intent of the consumer, and can access a large library of relevant content. It can be voice activated, text, or both and is conversational rather than task-oriented. It is also contextual in its knowledge, and not suited for random interactions on a variety of topics, the way Siri and Cortana were designed.
Technology
Health care insurers have complex websites that attempt to answer frequently asked questions, allow customers to file claims, request replacement insurance cards, and perform a host of activities. Some are designed better than others, but in the end a large percentage of visitors to the website either use chatbots or call a human to achieve their desired result. Why? It’s not the answer to the first question that matters, it’s the ability to ask a follow-up question where the websites fail to satisfy the needs of the customer. It is the conversation that is ultimately missing.
Today, intelligent virtual characters have a combination of technologies that have reached a level of sophistication that allow people to have detailed conversations with them. It begins with automated speech recognition (“ASR”) and text-to-speech (“TTS”) engines, which are used to recognize the spoken words and reply with speech. The next technology is natural language processing (“NLP”), which provides context to the speech so that the intent of the customer is recognized. Some words sound the same, and others have the same spelling, but in each case their use in a sentence can have different meanings. NLP helps keep the conversation on track.
Next is emotion detection. While personality traits may be helpful in extended conversations and relationships, understanding the immediate mood of the customer is essential in customer service. Several software engines use micro-facial detection, tone of voice, choice of words or a combination of these clues to detect the mood of the customer, and the changes in mood as the conversation progresses. The mood then determines the response from the intelligent virtual character as well as the animation that characterizes an appropriate physical response. In an article, Milliard (2017) quoted Anthony Chambers, director in the life sciences practice at the Chicago-based consultancy West Monroe Partners, in an interview with Healthcare IT News:
If we could use an interactive bot, where the patient then has a point of conversation via smartphone or something, that could be a game changer because of the challenge of clinical trials being so stressful on the population, and the expense of running the trials. What gets really fascinating—we have yet to see it, but we’re seeing discussions of it—is potential uses around the quality of care. That remains an untapped potential where the promise of emotional intelligence, in combination with AI, could play out.
How does artificial intelligence (“AI”) fit in? AI, as a term of art, has lost its definition as companies try to market their products as having AI incorporated in them. In this chapter, AI is used to describe the software on a device, which allows the device to recognize unstructured data, such as a text, pictures, or videos, as input from a customer, and provide similar data to the customer in response to the input. For example, a table of frequently asked questions may try to recognize key words of text and match many answers from a database containing the same words. Some answers will be relevant, many may not. On the other hand, AI allows the customer to “speak” to the device in any number of ways, and the device “understands” the intent of the question and provides a highly contextual answer that it has been “trained” to provide. The device then verifies if the answer is satisfactory to the customer and, if not, goes to the next answer. Over time, AI determines the probability of the correct answer and refines the sequence of answers it provides over time. In essence, the process of understanding and responding mimics that of a human being answering the question, albeit in a more consistent and highly scalable manner. Virtual reality and AI are becoming an integral part of nursing. An article by Ferguson et al. (2016) in the Contemporary Nurse Journal stated: “These technologies can be utilized in many settings to assist in health behavior change interventions, optimize care, and improve health outcomes of individuals across all care settings.”
AI also offers the opportunity for a follow-up question on the answer presented by the virtual character, an essential element to ensuring a high quality of customer service. There are numerous AI engines available, and most utilize application programming interfaces (“APIs”) to enable developers to quickly create software programs using the AI capability. The bulk of the work these days is in training the AI to be conversational in its interaction, and an ecosystem has been built around companies that help with this process, like companies that assist in integrating medical records into an electronic medical record (“EMR”) system.
AI is often incorrectly used interchangeably with deep machine learning, which is an advanced form of analytic and predictive software with neural networks, which is the science of nonlinear machine processing also used in pattern prediction. This chapter is limited to technology that enables a device to listen to a person and respond by voice or text to answer a question.
Health care is one of the few industries where the customer is an indirect payer for services received. Pricing for the services is rarely transparent, and the quality of the provider is hard to determine before receiving the services. In health care, consumers often speak with call center representatives and, for the purposes of this chapter, the experience is separated by (i) customer acquisition and (ii) cost reduction.
Customer Acquisition/Retention
All large health care payers, providers, and physician practices have a website; some are elaborate and others are rudimentary. The purpose of the website may vary from merely creating a presence with information about services and locations, to allowing a user to perform transactions, obtain health care test results, and customize the experience. Once developed, it is rare for the design to be revised over several years, with content creation being the only activity devoted to its maintenance. The initial cost of development may be a few thousand or several hundred thousand dollars, but very few are developed with user experience at the center of the design effort. The technology described above enables the implementation of a digital character (agent/advisor/avatar/assistant or some variant is used by numerous vendors), that can guide a user through an engaging experience.
Open Enrollment for Insurance Plans
Selection of an insurance plan can be a daunting task. An avatar leads you to the plan selection page, where graphical sliders allow you to adjust the deductibles, network flexibility, and monthly premiums. Based on the user’s selection of these sliders, the avatar identifies plans that meet these criteria and displays the availability of Health Savings Account (HAS) and employer-match options for such plans. The user can ask questions about a specific plan, by voice or by text, and get answers from content approved by the insurers’ risk management group. The content may include text or video format and the user can leave the website and resume the interaction, just as one would with a Digital Video Recorder (DVR). Plans can be compared side-by-side as the user explores various options.
Once a plan is selected, the avatar pre-populates any required forms (if the user is already a member who selects a different plan or changes participant coverage) or assists the user in entering the required information for a new member. At the end of the guided experience, the user is enrolled as a member without having any direct interaction with a call center associate.
The metrics for use of an avatar-based open enrollment experience are compelling:
- 87 percent pull through of users to completion of enrollment
- 90 percent increase in accuracy of completed forms
- 33 percent reduction in labor costs of call center operations
The next generation of interaction will allow an existing member to select a plan based on the member’s historical claims data, use of specific providers, customer satisfaction scores for providers, pharmacy usage, and other content from the insurer that is made available during the guided experience. This level of personalization will allow the member to select plans that are outcome-based, providing the insurer with a “customer-for-life” opportunity in customer service.
Cost Reduction
According to the Center for Disease Control and Prevention (2017), over 75 percent of the cost of health care in the United States is driven by chronic diseases. There are 22 chronic diseases, of which the top five account for 80 percent of the cost of chronic diseases. In general, the treatment of chronic diseases is episodic, similar to the treatment of acute health issues. However, management of chronic diseases requires care coordination, symptom management, and behavior modification to reduce the cost associated with managing the disease. Ward et al. (2014) indicated in a report titled Multiple Chronic Conditions among US Adults: A 2012 Update that
Other factors, such as medication compliance and use of urgent care, also pay a large role in these costs. Some of the primary goals of the national CDC initiative focused on addressing multiple chronic conditions in the United States include strengthening health care and public health systems, improving self-care management of [them], and providing better tools and information to health care providers.
Care coordination is currently a highly manual process, conducted by skilled labor such as nurses, physician assistants, and social workers. On average, a case manager is assigned over a thousand patients and spends a considerable amount of time measuring body vitals on patients, entering the data, tracking it, and responding to alerts generated when the data shows a patient to be outside normal ranges. Most patients end up in urgent care or emergency departments anyway, since the process is not scalable and the employee turnover tends to remain high among case managers. Collier, Fu, and Yin (2017) cited an Accenture study that stated:
Growth opportunities are hard to come by without significant investment, but artificial intelligence (AI) is a self-running engine for growth in health care. When combined, key clinical health AI applications can potentially create $150 billion in annual savings for the US health care economy by 2026.
Chronic Disease Management
AI is particularly suited for this use case. A virtual assistant can be implemented to interact with the patient, either on demand or on a scheduled basis, to inquire about the patient’s status on:
- Activities of daily living to determine ability to live independently
- Eating, bathing, dressing, toileting, transferring (walking), and continence
- Mood
- Medication adherence
- Exercise levels
- Body vitals
This information can be collected without human intervention, and can be used to identify trends in the progression of the disease. The trending and daily information can be triaged based on medical best practices, and patients that need attention can be brought to the attention of the case manager. Now, instead of conducting data collection and analysis, the case manager performs triage and intervention, allowing the case manager to handle a much higher volume (2× to 5× in most cases) of patients, with better outcomes and lower cost of intervention. The patient feels empowered and engaged in management of their condition and over time, and the frequency and intensity of symptoms decrease through this approach, which cannot be effectively implemented without the use of AI-driven technology.
The development of medical-grade sensors by companies that traditionally focus on consumer products, such as the recently released Apple Watch, will allow machine-driven collection of body vitals in real time, further improving the quality of data and the frequency of collection to improve the analytics necessary to monitor the condition of patients with chronic diseases. Use of the virtual assistant will ensure patient engagement levels that in the past have eluded manufacturers of exercise devices/monitors, since the data collected now has relevance to the patient’s health. The data will be more accurate than patient-reported data, since devices will communicate directly via Bluetooth or other emerging technologies.
Conclusions
Implementing AI in a clinical setting requires a thorough understanding of the workflow and processes currently in place in the departments likely to be affected by the technology, and a willingness to modify these processes to leverage the technology rather than using AI to incrementally improve existing processes. Here are four strategic areas that lend themselves to implementation of AI.
Case Management
The elderly population continues to increase as a percentage of the overall U.S. population, and adults over 50 years of age already have multiple chronic diseases. Congestive heart failure (CHF) and chronic obstructive pulmonary disease (COPD) are two of the top five chronic diseases in terms of prevalence and cost. The occurrence of symptoms often leads to emergency care and care for these patients is usually managed through intervention by nurses and case managers.
Studies show that intervention during early signs of symptoms saves a considerable amount of cost for these diseases. In the case of CHF, a weight gain of 2 pounds indicates a patient is likely to need urgent care in about 24 to 48 hours. For COPD, increased frequency of rescue inhalers is a sign that intervention in a clinical setting will be necessary in short order. Hospital systems employ thousands of case managers to reach out to patients and monitor these trends; communication usually is by telephone and may include texts and e-mails. Response from patients varies greatly and case managers have a high rate of burnout in their jobs.
Enter a virtual assistant powered by AI. The hospital provides the patient with a dedicated device, generally a tablet, preloaded with the assistant, along with sensors to measure body vitals. Each day, at a time convenient for the patient, the assistant engages the patient by asking a few questions and then measuring body vitals that are related to the disease. The sensors communicate to the virtual assistant in real time through Bluetooth. Once the measurements are completed, the information is transmitted to the case manager in a report format. Each patient has metrics assigned, and trend data as well as variations from prescribed values alerts the case manager when a patient needs intervention. By focusing only on patients that need attention, the case manager can manage over 2,500 patients, a 5× increase from current standards, while providing far better care to those patients who need it. In his article titled “Market Insight: How Service Providers’ Strategic Planners Should Target the Remote Patient Monitoring Market,” Gupta (2016) argues that
Remote patient monitoring offers the potential to deliver hospital-grade monitoring outside of the care setting. Strategic planners need to understand why RPM is being discussed now, where the main growth segments are and what to know to effectively position their offerings in this market.
Patient Portal
Most health systems and their physician practices have patient portals, where secure access is given to the patient for lab results, appointments, and physician notes from the patient’s records. Its purpose is to provide a self-directed experience for the patient to obtain information at their own pace and on their own time. The portals are poorly designed without the user experience in mind, and the patient ends up calling the doctor’s office to speak to a nurse. If the patient has questions about the meaning of the test results, the nurse generally schedules a conversation between the doctor and the patient.
A virtual assistant with AI can substantially reduce the burden for doctors and their staff. It can notify the patient when the results are posted on the patient portal, identify any abnormalities and provide context around the variations, identify the need and type of follow-up necessary, and schedule the subsequent appointments, all without interaction with the nurse or doctor. It does this conversationally, just like the patient would experience with a person, and has content that allows it to answer questions consistently and repeatedly, without judgment and without irritation. Specialists within the health care system get automatic referrals and can even prioritize calendaring based on the patient’s condition.
Primary Care
Access to primary care and subsequent treatment is another major cost factor in health care. Pharmacies now have clinics, the Internet provides millions of resources to self-diagnose, and call centers staffed with doctors are available to prescribe medications. Ride-sharing services are available to take patients to clinics and concierge medicine is on the rise.
In most cases, the symptoms that require primary care are straightforward and the diagnosis is fairly consistent. AI allows a patient to describe symptoms to a virtual assistant, and the virtual assistant can clarify the symptoms by asking the patient questions, much like a nurse would. Using specific algorithms to perform a differential analysis, the virtual assistant sends the information to a nurse or doctor at a remote location along with suggestions of the likely diagnosis; the doctor or nurse reviews the information, verifies the diagnosis, and prescribes medication or suggests an office visit, handling substantially higher volumes of patients than possible if the patient was having a direct conversation with the doctor or nurse.
OTC Drug Interaction
A substantial number of drugs dispensed in the United States are nonprescribed. This includes drugs that formerly required prescriptions but are now available over-the-counter (OTC). Pain relievers and allergy medication are two of the more prevalent types of drugs in this category. When pharmacists dispense drugs, they are likely to watch for drug interactions, proper dosage, and other conditions that may affect the patient. They have access to databases that provide them with necessary information. Once the drug is an OTC drug, however, the consumer is on their own. Labels on the medication can be confusing and often similar in appearance between the different strengths of the drug. Taking 10 mg of a drug instead of 1 mg can have disastrous effect on the consumer.
Using a device such as a smartphone, an app can access the same information available to the pharmacist and alert the consumer on possible contraindications. The camera on the smartphone can scan the code on the product and compare it to medical information previously entered by the consumer. The interaction can be conversational right at the point of sale, and can even commence with a recommended dosage and list active ingredients before the consumer goes to the pharmacy.
The current political debate has focused on how to pay for health care in the United States; the real issue is the high cost of care, the inconsistency among providers in the outcomes, and disintermediation of the patient—physician relationship. Iuga and McGuire (2014) of Johns Hopkins Bloomberg School of Public Health state in Adherence and Heath Care Costs that
In 2010 the costs of health care in the US exceeded $2.7 trillion and accounted for 17.9 percent of the gross domestic product. Projections indicate health care will account for 20 percent of the US gross domestic product by 2020. Twenty percent to 30 percent of dollars spent in the US health care system have been identified as wasteful. Providers and administrators have been challenged to contain costs by reducing waste and by improving the effectiveness of care delivered.
The use of AI technologies will result in a revised model of health care in the United States and many parts of the developing world, by increasing patient satisfaction and levels of engagement in their own health, early symptom tracking, outcome-based care models, and continuous monitoring of chronic diseases. However, the real advantage will be improved quality of life for patients and satisfaction among health care professionals on a job well done!
References
Adams, R.J. 2010. “Improving Health Outcomes with Better Patient Understanding and Education.” . Viewable at https://.ncbi.nlm.nih.gov/pmc/articles/PMC3270921/ (accessed October 14, 2010).
Center for Disease Control and Prevention. 2017. “Preventive Health Care, what’s the problem?” . Viewable at https://.cdc.gov/healthcommunication/toolstemplates/entertainmented/tips/PreventiveHealth.html (accessed September 15, 2017).
Collier, M., R. Fu, and L. Yin. 2017. “Artificial Intelligence: Healthcare’s New Nervous System.”,.Viewable at https://.accenture.com/t20170418T023006Z__w__/us-en/_acnmedia/PDF-49/Accenture-Health-Artificial-Intelligence.pdf
Ferguson, C., P.M. Davidson, P.J. Scott, D. Jackson, and L.D. Hickman. 2016. “Augmented Reality, Virtual Reality and Gaming: an Integral Part of Nursing.” Viewable at http://.tandfonline.com/doi/abs/10.1080/10376178.2015.1130360?journalCode=rcnj20 (accessed January 14, 2016).
Gupta A. 2016. “Market Insight: How Service Providers’ Strategic Planners Should Target the Remote Patient Monitoring Market.” , Viewable at https://.gartner.com/doc/3383517/market-insight-service-providers-strategic (accessed July 19, 2016).
Iuga, A.O., and M.J. McGuire. 2014. “Adherence and Health Care Costs.” , Viewable at https://.ncbi.nlm.nih.gov/pmc/articles/PMC3934668/ (accessed February 14, 2014).
Miliard, M. 2017. “The Next Big Thing in AI, Emotional Intelligence, Could Give Hospitals a Competitive Edge.” Viewable at http://.healthcareitnews.com/news/next-big-thing-ai-emotional-intelligence-could-give-hospitals-competitive-edge (accessed August 17, 2017).
Ward, B.W., J.S. Schiller, and R.A. Goodman, 2014. “Multiple Chronic Conditions Among US Adults: A 2012 Update.” , Viewable at https://.cdc.gov/pcd/issues/2014/13_0389.htm (accessed April 17, 2014).