In the United States, approximately 60% of the adult population is living with one or more chronic condition, from heart disease and asthma through to Alzheimer’s, kidney disease, and diabetes.1 This is putting a significant burden on healthcare systems in terms of their ability to deliver care and the cost of managing these illnesses. In the U.S. alone, almost three quarters of healthcare spend is linked to chronic conditions or associated complications.2 It is no surprise therefore that efforts are underway to look at how to best harness cutting-edge technologies such as artificial intelligence (AI) and machine learning (ML) to more effectively prevent, diagnose, treat, and manage these chronic conditions.
The increased use of smart phones, consumer-grade wearables, and home-networked diagnostic devices is what is enabling this transformation. Today there are healthcare apps to manage and monitor many chronic conditions, which seamlessly gather health information from individuals. These create the broader platform and ecosystem that may strengthen AI tools and in turn increasingly benefit people’s health.
At Ascensia, our sole purpose is to improve the lives of people with diabetes (PwD), a chronic condition that has particularly benefited from healthcare’s digital transformation. PwD may only see their healthcare provider (HCP) once every three months, meaning there are almost 2,200 hours between visits where they may not be monitored. Thankfully, with technological developments such as integration between glucose-monitoring devices and electronic health records (EHRs), it is now common for an app to facilitate the capture and analysis of real-time data to support the development of treatment plans and ongoing management. Also, the ability to amass and process large data sets such as this through cloud computing puts us at the tipping point where it is possible to build AI models to monitor people in between appointments and provide personalized plans, coaching, and feedback.
Whilst still relatively new, AI is beginning to become more broadly integrated into the healthcare ecosystem. An example of where it can be applied is in predicting early warning signs of a disease. In 2018, Google created a new AI algorithm to predict heart disease by using the data from analyzing retina scans of a patient’s eye. The company’s software can accurately deduce data, including an individual’s age, blood pressure, and whether or not they smoke. This can then be used to predict their risk of suffering a major cardiac event—such as a heart attack—with roughly the same accuracy as current leading methods. The possibilities of AI are limitless, and we are still scratching the surface of what can be achieved.
Soon, we are likely to see AI being seamlessly integrated into many healthcare solutions through education, lifestyle guidance, gamification, predictive analysis, and personalization. Nonetheless, while the benefits realized could be a game changer, in some circumstances, the path to truly embedded AI, where its full potential is utilized, is not without its challenges.
Artificial Intelligence Needs Quality Data
One of the prerequisites for building an AI platform is the availability of quality data. In short, the cleaner the data, the better the analysis and the better the care. It is therefore important to look at ways to collate, organize, and activate data to make it meaningful so that it can positively impact healthcare outcomes. The good news is that with increased use of devices and digital tools, the data pools are growing, and as many of these tools monitor more than a single vital statistic, they have the potential to provide a more holistic view of a person’s health profile. This is an important first step. If you then add in connectivity with EHRs, this further broadens the utility of the collective data, creating a real opportunity for AI to make a significant difference in patient care.
Regulations Need to Evolve
However, even with comprehensive data, to adopt and optimize AI in healthcare there are many hurdles to overcome. As advances in AI technology progress, governments and authorities are coming together to create legal and regulatory frameworks to ensure its use in a healthcare setting is safe and effective.
From a regulatory perspective, the approval of AI or ML software as a medical device creates new challenges, as AI software typically adapts its output based on the vast amount of new data generated on an ongoing basis. This is completely different from a drug or medical device approval, as it means the product itself is continuously evolving.
Putting this into perspective, data interpretation is of great relevance and can be split into two areas. For example, an algorithm is a predictable process that will always provide the same output when given the same inputs. This could stop an insulin pump administering insulin when a CGM detects low glucose levels. What’s important here is the recognition that an algorithm is predictable. AI, on the other hand, is when an algorithm may change through ML, and the outputs may vary based on the same inputs. Here, an example could be risk profiling where, based on a current database, the system identifies a person at risk of a medical event. However, as that database grows and becomes more refined and learns from previous successes and failures, that same person presenting with the same symptoms may be associated with a different level of risk. This in turn could lead to a different therapy decision and subsequently a different health outcome.
That said, FDA has approved a few AI-based medical technologies, such as solutions for the detection of retinopathy in PwDs, for ECG analysis, and for cancer diagnosis using CT or MRI scans. However, it is important to note that all those approved to date use locked algorithms, which cannot be changed or improved without further regulatory approvals. Such locked algorithms ultimately undermine the true value of AI, where it would otherwise constantly evolve its advice or information, based on regular updates of the data it consumes.
It is therefore likely to be some time before the full potential of AI and ML is imbedded in medical technologies as regulators continue to weigh the risks and benefits of products that are continuously changing. In the interim, perhaps a hybrid AI model where some features are restricted from change but others continuously adapt will help regulators to understand the clinical impact of certain technologies. For example, when a decision is deemed high risk, such as insulin dosing, the device will continue to be handled through more-traditional fixed algorithms or require human confirmation of the decision. Whereas those functions deemed lower risk, such as prompts to do a fingerstick, could integrate AI and ML for greater personalization and may still potentially optimize health outcomes if adherence to therapy and monitoring can be improved.
Data Must Be Protected
The next challenge to tackle involves cybersecurity and data protection. From a patient perspective, while there is increasing appetite to use digital tools to monitor and manage chronic conditions, there is some reluctance to share data with healthcare’s wider digital ecosystem. This caution is understandable. Cyber-attacks are becoming increasingly common, and as more and more personal and often sensitive data is being recorded, the healthcare industry has become a particular target.
Patient confidentiality is not a new concept and has always been at the core of healthcare delivery. The difference is that this information is no longer simply shared between patient and physician but also can be shared with medical devices, mobile phone, and social media platforms.
One way the sector is addressing the privacy issue is with the use of anonymized data sets, but that creates potential challenges and risks. First, to truly remove all identifiable information from large data sets is a huge task, if possible. Secondly, since AI relies on large, diverse data sets to build a holistic view, we need to consider how to share data between institutions, pharmaceutical companies, or healthcare providers safely and ethically. It will therefore be important that healthcare providers, AI developers, and policy makers alike consider, commit to, and appropriately address these fundamental cornerstones in relation to AI. Encouragingly, as data becomes increasingly important in global health, this is what we are beginning to see.
If handled with care, AI will dramatically improve the economics of healthcare provision globally and, more importantly, improve treatment outcomes for people living with chronic conditions.
AI has phenomenal potential and the technology that is emerging in healthcare is truly outstanding—moving from management to prevention can truly be a near possibility. The future looks promising!
- National Center for Chronic Disease Prevention and Health Promotion (NCCDPHP), Centers for Disease Control and Prevention, https://www.cdc.gov/chronicdisease/about/index.htm : last visited 5 July, 2021
- "An Empirical Study of Chronic Diseases in the United States: A Visual Analytics Approach to Public Health," Wullianallur Raghupathi and Viju Raghupathi, Int J Environ Res Public Health. 2018 Mar; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5876976/ last visited 22 June, 2021
For More Reading
- "Revisiting health information technology ethical, legal, and social issues and evaluation: telehealth/telemedicine and COVID-19," Kaplan B., Int J Med Inform. 2020;143:104239. doi:10.1016/j.ijmedinf.2020.104239
- "Digital Connectivity: The Sixth Vital Sign," Klonoff et al. Journal of Diabetes Science and Technology 1–6 2021, https://doi.org/10.1177/19322968211015241