We are all aware that technology is a change agent. When compared to 20, 10, or even five years ago, the healthcare sector has undergone significant growth and advancements, much of which can be directly attributed to technological innovation. The possibilities appear limitless as technology advances in intelligence, speed, and dependability.
For patients, doctors, and medical researchers, artificial intelligence (AI), and particularly machine learning (ML), is anticipated to have a significant impact on the future of the healthcare sector. For instance, a 2016 study from Johns Hopkins identified medical errors as the third-leading cause of death in the United States. These errors were the result of individual and/or system-level blunders. This percentage can be decreased by implementing AI technology in healthcare settings that are centred on patient treatment and care. However, getting there will require some effort. Before AI technologies are widely adopted in the healthcare sector, there needs to be more education on their advantages and how to use them effectively.
What can AI do?
How these technologies may be used most effectively is one of the main topics being posed about AI and ML right now. What benefits does using AI in my practise or hospital bring to me as a doctor or healthcare organisation? Which use cases are crucial? How can this technology help my patients in the long run?
At a fundamental level, artificial intelligence can serve as a prescreening tool for physicians, giving them important context details about the patient before they meet with them. As a second opinion on diagnosis and treatment, these technologies can then offer augmentation.
Providing triage and diagnostic data
We know there is a significant correlation between those who smoke cigarettes and those diagnosed with lung cancer. However, many other factors (occupation, chemical exposures, medications, supplements) may contribute to or increase an individual’s risk of developing lung cancer.
An intelligent virtual assistant, also known as a chatbot, may be used to navigate an initial conversation with a patient to collect health history, symptoms or other important information – ideally before they even set foot in their physician’s office. This data can then be used as input to an ML model that provides the patient with initial triage/diagnostic data – while offering the doctors an overall risk score or probability that a patient may develop lung cancer. Research has shown the results of AI-based triaging to be comparable to the accuracy of human doctors, so there is a great deal of potential here.
Not only can the ML model provide doctors with a patient risk score, but it can also identify the patient’s specific current and historical features that contributed the most to their score – such as the duration or occurrences of exposure to other carcinogens in their work environment. Sometimes these correlations are evident to the trained professional, but other times they are not. AI and ML can help uncover these complex patterns in patient data that are not always easily identifiable by humans. This is valuable prescreening information that can help better prepare doctors for examinations, streamline the patient’s experience in the hospital and reduce human error in diagnosis and/or treatment plans.
AI to the rescue for healthcare workers, too
Another increasingly important use case for AI and ML is combating burnout and even risk of suicide among healthcare workers. Suicide rates among physicians were already one of the highest in the country and have only worsened due to the strain occurring from the COVID-19 pandemic. While this is a highly sensitive issue that requires thoughtful human intervention, technology will increasingly have a key role to play, too.
Hospitals are now looking to AI and ML to help identify healthcare workers at greater risk of negative mental health impacts (fatigue, burnout, depression) so they, in turn, can offer support. Advancements in natural language processing (NLP) techniques – which can search and analyze unstructured data, such as physician notes – open opportunities within healthcare organizations to analyze electronic health record (EHR)-based activity logs and workload metrics to identify trends in negative sentiment and overwork.
Enhancing diagnostic data
Comparable to a system of checks and balances, ML can also do a deep dive into a patient’s diagnostic imaging. Using radiographs, CT scans and other types of imaging, ML methods can link patterns and trends in images that may not be easily detected by physicians, including early signs of diseases and levels of degradation in tissue or bone. For example, research has determined that COVID-19 can cause the presence of ground-glass opacities (GGO) in the lungs. While many different diseases are also known to cause this, the presence of GGO specific to COVID-19 demonstrates a unique pattern and location. ML models have been trained to identify the presence of COVID-19 using previous patient radiographs as a baseline, with current sensitivity scores of .90 and accuracies of around 91%.
Another prominent example involves patients and the wearable devices they may issued by their doctors to help monitor their diagnostics over time. These devices can be particularly critical for those managing chronic, degenerative conditions such as heart disease or diabetes. The data collected from these wearable technologies can be incorporated into ML algorithms to provide even more detailed insights into habits and trends, as well as a patient’s future state if a given trend (such as persistently higher blood pressure levels) continues. Additionally, doctors can use this data to identify high-risk patients struggling to maintain healthy levels and/or share guidance to help the patient self-manage between checkup visits.
Having insight into these types of data points earlier in the process can help doctors provide more reliable treatment responses sooner, while also helping to reduce the risk of a misdiagnosis. This is especially important when studies have shown that there are an estimated 40,000 to 80,000 deaths each year in U.S. hospitals related to misdiagnoses.
Using AI in healthcare safely and effectively
As is the case with most new technologies, there is some hesitancy and unknowns relative to AI. First and foremost, AI is not meant to be a replacement for doctors. Rather the technology should be used as an aid to help doctors more effectively assess and treat their patients. It’s also important to note that the machines do have limitations and won’t be 100% accurate all the time – there will need to be continued human involvement throughout the diagnosis and treatment process.
When used in this context, AI and ML can offer the healthcare industry a myriad of benefits. Yet, the organizations using it also need to know not just how to develop AI models, but also how to do so effectively, safely and ethically.
While Gartner reports have found that an estimated 85% of AI projects will fail through the year 2022, this estimate does not have to be the reality for a given organization. The reason that so many AI projects fail is not necessarily due to the AI processes themselves, but rather the lack of strong data governance, collaboration and problem definition. Organizations that successfully leverage AI typically start their design with a clear end goal in mind relative to the role that AI will ideally play or the problem it will solve. This approach allows organizations to work backward, guiding them through the foundational components that will help contribute to the end goal.
An organisation must also have the right data governance and security in place to start leveraging AI and ML technologies successfully and maximise return on investment. The architecture of a healthcare institution depends on the data used in AI projects being accurate and trustworthy because machine learning (ML) is mainly focused on probability and statistics. For instance, analysing patterns in patient photos collected over a specified period of time using AI and ML can be quite helpful. Where and how the imaging is processed can be a problem. The model may produce inaccurate findings as a result of adjustments to the angle, location, or imaging processing. To have the highest success rate with the utilization of AI and ML, organizations must ensure that they have quality data and defined processes. This starts with building a strong foundation of data governance and security.