Despite decades of medical research that have informed therapies and healthy behaviors shown to improve patient outcomes and lower costs, patients’ refusal to adhere to known interventions continues to be a substantial worldwide health problem.
Artificial intelligence has the potential to be a game changer by offering innovative solutions in monitoring, motivating, and sustaining positive habits related to medical adherence.
Medical adherence describes the degree to which a patient correctly follows medical advice, whether it be taking medication as instructed, showing up for scheduled appointments, or generally following dietary, exercise, and mental health plans to completion [1]. Of these, the most impactful and well studied is medication nonadherence, which has historical roots dating all the way back to Hippocrates, who was the first to note the behavior in 400 BC [4]. Today, researchers estimate that up to 60% of patients are medication nonadherent [5-8]. The personal and societal cost is substantial.
Medication nonadherence has a dramatic negative impact on healthcare, costing the U.S. hundreds of billions of dollars and resulting in tens of thousands of premature deaths annually [3].
Historically, the statistics for remediation have been bleak. For example, a 2014 review studied 182 randomized controlled trials for adherence intervention and found only five to report a modest improvement in both adherence behavior and clinical outcomes [9]. Not an encouraging determination.
For healthcare to do a better job at promoting adherence, holistic and innovative solutions are needed at multiple levels, e.g., regimen, patient, provider, and system levels, with an eye toward increased personal engagement and monitoring [2]. Our working hypothesis for this article is that medical nonadherence remediation efforts require innovative solutions centered around personalized care, moving away from conventional scenarios involving intermittent office visits and periodic clinician-patient touchpoints to a more modern and interactive 24/7 healthcare approach. Clinicians don’t have the time nor do healthcare systems have the resources to offer at-home or remote care services at the level that they can achieve with in-office visits. However, technological advancements in AI can help fill the gap in issues related to monitoring, habit formation and motivation surrounding medical adherence.
We explore each of these facets and how AI is currently used or can be used to engage patients and provide relevant solutions.
MONITORING
For patients who are prescribed a multitude of drugs, to be taken at different times of the day and for long periods of time, keeping up with their medication can be challenging. For those who may need assistance in remembering to take their pills, there exists a lot of innovative technology to help remind and passively monitor them. Some of the more innovative technologies are described below:
ingestible sensors: Miniaturized radio frequency identification (RFID) tagged gelatin capsules are now a reality. These pills, once swallowed, transmit a unique signal to a relay device, which in turn transmits a time-stamped message to a cloud-based server that functions as a direct measure of medication adherence [10].
smart pill dispensers and bottles: Pill dispensing technology via interactive hardware or bottles with embedded weight sensors to measure remaining pill count, blister sleeves, and text messaging for pills not being taken on time: HERO, PRIA, TinyLogics, CYCO, Pillsy, Nomi, MEMS Technology.
Rapid point-of-care drug assays: Rather than focusing on the taking of individual pills, this technology involves the use of quick “bedside” or in-clinic testing apparatus (such as mass spectrometers) to assess drug adherence via urine or serum samples [11, 12].
AI plays a central role in complementing many of these technologies. For example, computer vision applications in mobile phone apps have the ability to detect that a pill is in the scene and is subsequently swallowed, and the event is registered with the app as a means of verifying medication adherence [13]. Researchers have also combined RFID technology, embedded into pill bottles, with computer vision based face and object recognition to passively observe elderly patients in their homes and infer their level of adherence to taking prescribed medication [15]. AI also plays an important role in security for some systems equipped with a video camera, by using deep learning models to recognize and confirm a person’s face prior to dispensing medication.
MOTIVATION AND HABIT FORMATION
Medication adherence is a highly complex behavior, influenced by a wide variety of factors. However, according to the American Medical Society,
Most nonadherence is intentional—patients make a rational decision not to take their medicine based on their knowledge, experience and beliefs.
Willful nonadherence is very difficult to overcome as it involves changing a patient's perceptions, attitudes, and behaviors regarding medication and/or how they may feel about their care provider. Advanced monitoring solutions, such as fancy pill dispensers and RFID-tagged pills, are impractical solutions if a patient’s rational intent is to be purposefully nonadherent. So, what exactly are a typical patient's concerns that prohibit them from taking their medicine? The AMA suggests that there are eight primary factors related to intentional medication nonadherence, which naturally fall into emotional, financial, and educational categories [14]:
emotional
fear: patients are fearful of potential side effects
overwhelmed: patients are more likely to be nonadherent the more medications that are prescribed and the higher the dosing frequency
worry: concerns about becoming dependent on medication
depression: patients who are depressed are less likely to adhere to a prescribed medication plan
mistrust: given the influence of the pharmaceutical companies over doctors, patients may mistrust their doctor’s motivation in prescribing a particular drug
financial
cost: patients may not be able to afford the medication
educational
lack of symptoms. patients who don’t feel any differently when they start or stop their medicine might see no reason to take it.
misunderstanding: patients may not understand the need for the medicine, the nature of the side effects or the time it will take to see results. This is particularly true for patients with chronic illness, because taking medication every day to reduce the risk of something bad happening can be confusing. Failure to see immediate improvement may lead to premature discontinuation.
In each of these categories, AI can be of help. Arguably, the most difficult category for AI to tackle is that of emotion. When people feel rooted in a cause, be it political, sports, or health related, it is often very hard to uproot them and move their way of thinking towards more neutral ground. This is particularly true when their beliefs are engrossed in fear, mistrust, or worry, as may be the case for patients who don’t adhere to their medication plan. When patients’ fears are not justified, changing their minds means changing how they view the problem. Part of the solution then is to attenuate cognitive distortions, which is one component of cognitive behavioral therapy (CBT) [16]. Recently, AI has been making significant inroads in CBT via AI-driven, CBT chatbots such as Woebot, initially developed at Stanford University. These therapeutic chatbots could be reengineered to address patients’ concerns of medication usage, allowing them the opportunity to explore their fears in an anonymous and private environment. The chatbot could respond to recognized fears with informative videos or facts, which may help assuage their fears and potentially convince patients to continue taking their medication and/or encourage contact with their clinician to continue the discussion.
Emotional nudges, done respectfully and carefully, may play a powerful and persuasive role in keeping patients on their medication.
Artificial intelligence is also highly leveraged in advertising, using behavioral models to target a person with “just the right messaging at just the right time” in order to maximize profits [17]. The very same technology can be used to tailor helpful, healthy and personalized messaging to patients, for example, to check whether patients are doing okay with daily check ins and to ensure that they are staying on track with their medications. Combined with electronic medical records (EMR) and patient-level historical adherence trends, AI can help better understand a patient as an individual rather than a number in the system and thereby facilitates more effective and influential communication. Ideally, tailored and personalized messaging can be adaptive, sensitive to the current point of a patient’s treatment trajectory. Such messaging is likely to keep the patient in-the-loop and more involved with their own care and possibly promote stronger feelings towards adherence. These AI-driven models may conceivably alter how a doctor and other care providers interact with a particular patient, giving them a better sense of how to engage with patients on a personal level and, in doing so, foster trust and encourage a better relationship. These ideas are not just thought experiments. Many companies are already in the game of developing personalized adherence models to influence behavior and forecast adherence tendencies, e.g., McKinsey & Company, AllazoHealth, Neura, AiCure, and Nomi to name a few.
Daily check-ins via a mobile application may also be used to address directly patients’ concerns revolving around education. For example, imagine a scenario where a patient was prompted by a mobile app with the question: “How are you feeling today?” to which the patient may respond, “I feel the same...I don’t think my medication is working” or “I’ve developed a rash and am worried I’m allergic to my medication”. Natural language processing can be used to intelligently parse the patient’s response and pull up relevant information for them to read, or queue up a relevant video for them to watch, addressing their concerns. The app may also alert their clinician that the patient isn’t doing well on their medication, which could lead to a phone call or intervention by their clinician. Again, this simple daily check-in keeps the patient in the loop and gives them an avenue to express their concerns, and the catered response to these concerns may lead to the patient staying on their medication or receiving refined guidance from their doctor.
That leaves only the real issue of finance to tackle. For many, particularly in these trying times of a COVID-19 pandemic, the economic future may be very uncertain. At a critical time when we need our health care to support people, their jobs and health may be in jeopardy. How can AI help in this situation? The answer may be rather straightforward.
As the financial and health costs of medication nonadherence are significant, one approach may be to give patients direct monetary compensation when they have demonstrated sufficient adherence to their care plan.
AI may come into play by predicting what form of compensation is most appropriate, based on a patient’s learned behavioral model. Is the patient a golfer? Maybe offer a discount at a local golf course or free session at the driving range. They seem to love movies... perhaps a movie pass to a local theater? Or perhaps, cold hard cash is most relevant? Whatever the case, reciprocity in the form of financial gain might be well received and provide a strong behavioral improvement adherence incentive. Such compensation might be baked into the health plan for an individual undergoing treatment. With the high cost of nonadherence, small and reasonable compensation measures for adherence might make economic sense. Such offers may seem unrealistic to some but similar concepts are already part of our current reality. For example, in 2017 HealthCare Finance News reported on a test study where specialized contracts were formed with big pharma companies requiring drugmakers to take on some of the outcomes-based risk in drug performance, which involves medication adherence, in such a way that the cost of certain drugs for certain disease types are reduced [18]. The reduction in drug prices provides at-risk populations with access to cheaper medication, which in turn may promote adherence.
SUMMARY
Medication nonadherence is a major health problem, costing the U.S. healthcare system hundreds of billions of dollars per year, and is associated with over a hundred thousand premature deaths annually. Historically, proposed solutions have fallen short, calling for innovation and a change in standard healthcare practices to address the problem. With the promise of better understanding a patient as an individual, rather than a number in the system, AI may serve a central role in helping to combat medication nonadherence by providing tailored and timely messaging, keeping patients engaged and informed. In addition, AI-based chatbots may be used to address emotional and educational concerns patients may have regarding prescribed medication. Finally, AI may be used to help identify at-risk populations, so that agreements made between insurers and drug companies better serve these populations by offering lower drug prices.
REFERENCES
https://en.wikipedia.org/wiki/Adherence_(medicine)
Stirratt MJ, Curtis JR, Danila MI, Hansen R, Miller MJ, Gakumo CA. Advancing the Science and Practice of Medication Adherence. J Gen Intern Med. 2018;33(2):216‐222. doi:10.1007/s11606-017-4198-4
Meera Viswanathan, PhD, Carol E. Golin, MD, Christine D. Jones, MD, MS, Mahima Ashok, PhD, Susan J. Blalock, MPH, PhD, Roberta C.M. Wines, MPH, Emmanuel J.L. Coker-Schwimmer, MPH, David L. Rosen, MD, PhD, Priyanka Sista, BA, Kathleen N. Lohr, PhD. Interventions to Improve Adherence to Self-administered Medications for Chronic Diseases in the United States: A Systematic Review. December 2012. https://doi.org/10.7326/0003-4819-157-11-201212040-00538
Hugtenburg JG, Timmers L, Elders PJ, Vervloet M, van Dijk L. Definitions, variants, and causes of nonadherence with medication: a challenge for tailored interventions. Patient Prefer Adherence. 2013;7:675‐682. Published 2013 Jul 10. doi:10.2147/PPA.S29549
https://www.managedhealthcareexecutive.com/health-management/three-reasons-nonadherence
Briesacher BA, Andrade SE, Fouayzi H, Chan KA. Comparison of drug adherence rates among patients with seven different medical conditions. Pharmacotherapy. 2008;28(4):437–43. doi: 10.1592/phco.28.4.437.
Vink NM, Klungel OH, Stolk RP, Denig P. Comparison of various measures for assessing medication refill adherence using prescription data. Pharmacoepidemiol Drug Saf. 2009;18(2):159–65. doi: 10.1002/pds.1698.
Yeaw J, Benner JS, Walt JG, Sian S, Smith DB. Comparing adherence and persistence across 6 chronic medication classes. J Manag Care Pharm. 2009;15(9):728–40.
Nieuwlaat R, Wilczynski N, Navarro T, Hobson N, Jeffery R, Keepanasseril A, Agoritsas T, Mistry N, Iorio A, Jack S, Sivaramalingam B, Iserman E, Mustafa RA, Jedraszewski D, Cotoi C, Haynes RB. Interventions for enhancing medication adherence. Cochrane Database Syst Rev. 2014 Nov 20; (11):CD000011.
Hafezi H, Robertson TL, Moon GD, Au-Yeung KY, Zdeblick MJ, Savage GM. An ingestible sensor for measuring medication adherence. IEEE Trans Biomed Eng. 2015 Jan; 62(1):99-109.
Gupta P, Patel P, Štrauch B, Lai FY, Akbarov A, Marešová V, White CMJ, Petrák O, Gulsin GS, Patel V, Rosa J, Cole R, Zelinka T, Holaj R, Kinnell A, Smith PR, Thompson JR, Squire I, Widimský J Jr, Samani NJ, Williams B, Tomaszewski M. Risk Factors for Nonadherence to Antihypertensive Treatment.Hypertension. 2017 Jun; 69(6):1113-1120.
McNaughton CD, Brown NJ, Rothman RL, Liu D, Kabagambe EK, Levy PD, Self WH, Storrow AB, Collins SP, Roumie CL. Systolic Blood Pressure and Biochemical Assessment of Adherence: A Cross-Sectional Analysis in the Emergency Department. Hypertension. 2017 Aug; 70(2):307-314.
https://sbir.nih.gov/statistics/success-stories/aicure
https://www.ama-assn.org/delivering-care/patient-support-advocacy/8-reasons-patients-dont-take-their-medications
Hasanuzzaman FM, Yang X, Tian Y, Liu Q, Capezuti E. Monitoring Activity of Taking Medicine by Incorporating RFID and Video Analysis. Netw Model Anal Health Inform Bioinform. 2013;2(2):61‐70. doi:10.1007/s13721-013-0025-y
http://cogbtherapy.com/cbt-blog/cognitive-distortions-all-or-nothing-thinking
https://www.marketingaiinstitute.com/blog/ai-in-advertising
https://www.healthcarefinancenews.com/news/payers-test-risk-contracts-big-pharma-improve-adherence