Diagnosis and Treatment Applications

Since MYCIN was developed at Stanford for diagnosing blood-borne bacterial infections," diagnosis and treatment of disease has been a focus of AI applications. The early rule-based systems showed promise for accurately diagnosing and treating disease, but were not adopted for clinical practice. Comparisons suggested that they were not substantially better than human diagnosticians and they were poorly integrated with clinician workflows.

Rule-based “order sets” within EHR systems are widely used—within the most widely used electronic medical record systems such as EPIC and Cerner, and at the U.K.’s National Health Service (NHS),¹² but they lack the precision of more algorithmic systems based on machine learning. These clinical decision support systems are difficult to maintain as medical knowledge changes, and are often not able to handle the explosion of data and knowledge based on genomic, proteomic, metabolic, and other “omic-based” approaches to care.

However, attempts to find effective Al-driven diagnosis and treatment continue unabated despite some setbacks. IBM’s Watson received considerable attention in the media for its focus on cancer diagnosis and treatment. Watson employs a combination of machine learning and NLP capabilities to review scientific literature for diagnoses and treatment options. However, early enthusiasm for this application of the technology has faded as customers realized the difficulty of teaching Watson how to address particular types of cancer, and of integrating Watson into care processes and systems.¹³ A prominent implementation of Watson at the MD Anderson Cancer Center in Houston spent $62 million but was not used to treat a single patient and was not integrated with the hospital’s EHR system. The project was put on hold in late 2016 and has not been resumed.¹⁴ There is little documented evidence of improved treatment of patients or cost savings in healthcare from Watson.

There are many other approaches to using AI for diagnosis and treatment. Perhaps, the most common approach in current usage employs deep learning to analyze radiology images of potentially cancerous cells in, for example, breast lesions or colon polyps. While these applications have demonstrated impressive results in laboratory analyses compared to human radiologists and oncologists, they have not yet been widely applied to clinical practice.¹⁵ Guidelines for their effective use by radiologists and other clinicians have not yet been developed, though the American College of Radiology’s Data Science Institute is developing a framework for the application of machine learning to radiology and provides an ongoing list of AI algorithms that have been approved by the US Food and Drug Administration.¹⁶

Other approaches to diagnosis and treatment employ statistical analyses of genetic or metabolic data. Various researchers have built prediction models from big data to warn clinicians of high-risk conditions such as sepsis and circulatory failure.¹⁷ Jvion offers a “clinical success machine” that identifies the patients most at risk as well as those most likely to respond to treatment protocols. Other diagnosis approaches use approaches based on other input types, such as retinal scanning¹⁸ or genomic-based precision medicine.¹⁹ Each of these could provide decision support to clinicians seeking to find the best diagnosis and treatment for patients, and they are in the early stages of clinical application.

There are also several firms that focus specifically on diagnosis and treatment recommendations for certain cancers based on their genetic profiles. Since many cancers have a genetic basis, human clinicians have found it increasingly complex to understand all genetic variants of cancer and their response to new drugs and protocols. Firms such as Foundation Medicine and Flatiron Health, both now owned by Roche, specialize in this approach.

Outside of cancer, one medical condition that has received some incorporation into clinical practice is sepsis. Because sepsis and septic shock can cause rapid death if undiagnosed and untreated, the primary value of machine learning models is more rapid identification of the condition based on routinely gathered patient data in EHRs. These types of systems are part of the clinical workflow in providers such as Penn Medicine and the Mayo Clinic. One review paper of AI for sepsis prediction suggested that the approach offers considerable promise, but also that many models in use included variables in the predictive factors that are also part of the definition of sepsis, such as low blood pressure.²⁰

AI-enabled diagnosis and treatment, despite these initiatives, remain mostly present in research labs and in tech firms, rather than in clinical practice. Scarcely, a week goes by without a research lab claiming that it has developed an approach to using AI or big data to diagnose and treat a disease with equal or greater accuracy than human clinicians. Since these types of findings are based on statistically based machine learning models, they are ushering in an era of evidence- and probability-based medicine, which is generally regarded as positive but brings with it many challenges in medical ethics and patient/clinician relationships.²¹ The rise of EHRs in many large provider organizations, which capture and store considerable routing data, has made the application of machine learning to patient data much more feasible in recent years.²²

Both providers and payers for care are also using “population health” machine learning models to predict populations at risk of particular diseases,²³ or accidents,²⁴ or to predict hospital readmission (discussed below in Section Administrative Applications).²⁵ These models can be effective at prediction, although they sometimes lack all the relevant data that might add predictive capability, such as patient socio-economic status.

But whether rules-based or algorithmic in nature, Al-based diagnosis and treatment recommendations are sometimes challenging to embed in clinical workflows and EHR systems. Such integration issues have probably been a greater barrier to broad implementation of AI than any inability to provide accurate and effective recommendations. And many Al-based capabilities for diagnosis and treatment from tech firms are standalone in nature or address only a single aspect of care. Some EHR vendors have begun to embed limited AI functions (beyond rule-based clinical decision support) into their offerings, but these are in the early stages.²⁶ Providers will have to either undertake substantial integration projects themselves or wait until EHR vendors add more AI capabilities.

As with the Al-driven hype for autonomous vehicles, immediate hopes for Al-driven diagnosis and treatment need to be dialed back, and patience has become a necessary virtue.