A 2019 Emerg survey called ‘Machine Learning in Healthcare’ asked over 50 healthcare executives where they believe AI in healthcare will be by 2025. Over 50% said AI would be ubiquitous in healthcare.
A ReportLinker forecast predicts that AI in healthcare will grow from $2.1 billion in 2018 to $36.1 billion by 2025. Clearly, the future is bright for AI in healthcare. In this article, we’re going to look at how that growth is benefiting hospitals and the healthcare industry in general. By understanding how machine learning is benefiting patient care and outcomes, we can get a feel for what is happening on the ground and how it is affecting real lives.
Understanding how machine learning is benefiting patient care and outcomes, we can get a feel for what is happening on the ground and how it is affecting real lives.
Predicting Patient Outcomes
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is trying to help doctors make more efficient decisions. With a barrage of data from multiple patients and hospitals and often in inconsistent formats, making real-time treatment decisions can a challenge, to say the least.
AI is great at taking voluminous amounts of data, finding patterns in it, and providing a digestible result. ICU Intervene is one such AI-based approach created by CSAIL. It takes large amounts of data from previous ICU cases and makes suggestions for treatment. Additional notes are present in the use cases fed to ICU Intervene, which makes ICU Intervene suggestions that much more valuable.
“The goal is to leverage data from medical records to improve health care and predict actionable interventions.”
“The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment,” says Ph.D. student Harini Suresh, lead author on a paper about ICU Intervene, to MIT News. “The goal is to leverage data from medical records to improve health care and predict actionable interventions.”
A 2018 study published in Nature Digital Medicine, showed that an advanced algorithm was able to predict readmissions, in-hospital deaths, and long hospital stays more accurately than previous approaches. The algorithm used de-identified health records from 216,000 adult patient hospitalizations. From this data, its model was able to determine patterns leading to its final results.
Computing power has greatly increased, allowing AI to be applied to real-world problems rather than theoretical ones
Applying AI to medical records is only possible because so many medical records have been digitized in recent years. By 2016, 96% of US hospitals were using certified EHR (electronic health records), according to HealthIT. Additionally, computing power has greatly increased, allowing AI to be applied to real-world problems rather than theoretical ones (i.e., AI 1.0).
Machine Learning (ML) Use Cases
To get a broader view of how machine learning is being used by real companies in healthcare, we’re going to look at several companies.
Supporting EMR systems can be costly. ML-based enterprise software products are able to trim costs by removing the tedious, laborious process of cross records analysis and replacing it with automated, real-time analysis. Such solutions also have the benefit of greatly reducing human work from hours to mere seconds.
Predicting Illnesses and Treatment
Enterprise-grade machine learning means taking a company’s unstructured data from multiple sources, formats, and infrastructures and transforming it into data that AI models can work with. The end result is new insights that were not previously available to the client.
Such software is ultimately a patient prediction platform. It helps doctors understand who will get sick, what they will come down with, and how much treatment will cost.
Enhancing health information management and exchange of health information
Greater health can be achieved through better health information management. By analyzing health information, companies can make operational improvements, optimize revenue, and create better patient outcomes.
Immuno-oncology research with help from IBM Watson
Pfizer and IBM’s Watson created a medical research collaboration in 2016. The collaboration utilizes Watson’s machine learning capabilities and Pfizer’s medical scientist to accelerate research in immuno-oncology.
The nursing field is facing a shortage of nurses, and the prognosis doesn’t look as though it will get better anytime soon. The Bureau of Labor Statistics’ Employment Projections 2016-2026, shows that nursing (RNs) is expected to be the fastest-growing field through 2026. The Bureau projects that 203,700 new RNs will be needed each year.
AI may provide some relief in the form of automated tasks. These are routine tasks performed by nurses that aren’t high value. Other tasks will be augmented by AI, freeing up some time for nurses to focus on more important areas.
Some routine tasks that nurses perform are taking patient vitals and asking various questions to see how a patient is doing. After the patient session, all information must then be recorded and stored. In some hospitals, notes can be recorded in real-time as they are being taken. This saves some time in having to separately transfer notes into another system.
Virtual avatars exist that are able to communicate empathically and naturally with customers/patients. Its clientele includes insurers, pharmacies, and various other companies that use the avatar to answer employee questions. These avatars are virtual nurses.
Patients at home can bring up an app on their smartphone. The avatar appears and helps patients take their blood pressure, weight, and also ask a few questions about how the patient is doing. Because the avatar is interactive and empathetic, patients feel comfortable using the technology. All of the information collected by the avatar is available to a healthcare provider. Depending on patient session results, the avatar may tell the patient to call a real nurse.
Utilizing avatars allows nurses to save time in collecting vitals and asking various questions. An avatar is not able to make any type of diagnosis. By analyzing patient data and question responses in real-time, it is able to make a determination if further assistance is needed (i.e., call a nurse).
Google AI provides an excellent example of how image analysis or computer vision is being used in healthcare. By working with doctors and clinicians, Google has focused its healthcare machine learning efforts on ophthalmology and digital pathology.
Diagnosing diabetic eye disease
Google developed a dataset of 128,000 eye images. These images were used to train a deep neural network to detect referable diabetic retinopathy. They then tested their algorithm’s performance on two separate clinical validation sets. Combined, these datasets totaled approximately 12,000 images.
The end result is that Google’s models performed as well as ophthalmologists. As more images are fed to the models, and they are able to improve their results through further training, this kind of machine learning will be able to augment the detection of referable diabetic retinopathy.
Assisting pathologists in detecting cancer — the next area that Google focused on was analyzing pathology slides for detecting breast cancer in lymph node biopsies. Analyzing pathology slides is extremely time consuming and requires years of training, expertise, and experience. Google’s machine learning pathology efforts are meant to augment the expertise of physicians rather than be used as a final diagnosis.
A recent Chil Mark Research report titled ‘Precision Medicine and Health IT: New Data, New Challenges’ concludes that the full potential of precision medicine won’t be possible without utilizing advanced machine learning and deep learning technologies.
Full potential of precision medicine won’t be possible without utilizing advanced machine learning and deep learning technologies.
The authors of the report recognize that the volume of medical data being produced will only increase as more computers, devices, and methods of gathering and accumulating data continue to grow, outpacing what clinicians and researchers are able to keep up with and extract value from. Machine learning has the ability to analyze such volumes of data and provide results in real-time.
The above use cases all show how machine learning is being utilized today in healthcare. The overarching theme that we can see is that AI is an augmentor and not a technology that is meant to replace healthcare professionals.
AI is an augmentor and not a technology that is meant to replace healthcare professionals.
Healthcare.ai — Bringing AI to Healthcare
Most big industry efforts are lead by some neutral organization. For AI in healthcare, healthcare.ai has taken on that role. It’s a great place for learning about advancements in the field of healthcare driven by AI.
Healthcare.ai does have some practical use for software developers. For those looking to apply AI in healthcare, the site might not be the best place for such solutions as you will need to know how to use programming languages such as Python or R.
Will The Government Slowdown ML in Healthcare?
Ensuring that AI software is safe to use in healthcare falls under the realm of the FDA. It might seem odd that the FDA is approving AI software but this is nothing new to the FDA. It has been approving AI software in the healthcare industry since 2017. In fact, it approved the Apple Watch ECG in September 2018, which became available in the Apple Watch Series 4.
AI algorithms are always evolving due to their learning nature. This presents a problem for the FDA. To approve an algorithm, the algorithm must stop evolving. Once approved, the software can be published.
AI algorithms are always evolving due to their learning nature.
The FDA is still designing a framework for approving AI software. It recognizes that an approval process will slow the design and evolution of AI. For now, it is looking at a middle ground. After initial approval, allow AI software to evolve to a certain extent with well-known safeguards.
Machine learning has been with us for decades. However, it sat in the shadows until advances in computing power and growth of digital data made possible the results we see today. AI is no longer a term followed by “what-if”’s.
Real-time analysis of data and precise identification of pathologies are what machine learning is best at. After initial training, machine learning models have proven that they are able to produce practical results.
Doctors, clinicians, and nurses aren’t going to be replaced by AI for the foreseeable future. For the time being, AI instead will help augment these professionals in their everyday work of saving lives.