to
. The practice of medicine, though incomplete, has undergone a remarkable transformation. Over the past 50 years,we have seen a steady shift from a field based primarily on expert opinion and anecdotal experiences of individual clinicians to a formal scientific discipline.
With the advent of evidence-based medical care clinicians were able to identify the most effective treatments for their patients, based on the quality evaluations of recent research. Precision medicine allows providers to use the genetic, environmental, and clinical information of a patient to further personalize care.
Precision medicine’s potential benefits come with new challenges. The amount and complexity data available for every patient is increasing rapidly. How will clinicians determine which data is most useful for a specific patient? What is the most efficient way to interpret data to select the best treatment option?
Computer scientists like me work to solve these challenges. My colleagues and I collaborate with experts in genetics and medicine to develop computer-based systems that use artificial intelligence to help clinicians integrate complex patient data and make the best decisions.
Evidence-based medicine is on the rise
Clinical decisions were made in the 1970smostly based on expert opinions, anecdotal experiences and theories of disease mechanisms, which were not supported by empirical research. Around that time, some pioneering researchers argued for the use of best available evidence to guide clinical decision-making. In the 1990s, evidence-based medicine was introduced as a discipline that integrates research with clinical expertise to make decisions about patient care.
Evidence-based medicine is built on a hierarchy of quality evidence which determines the types of information that clinicians should most heavily rely on when making treatment decisions.
CFCF/Mikimedia CC by-sa.
In randomized controlled trials, participants are randomly placed in different groups to receive either a treatment that is experimental or a placebo. These studies are also known as clinical trialsand are considered to be the best sources of individual evidence because they allow researchers the opportunity to compare treatment effectiveness without bias by ensuring that the groups are similar.
In observational studies such as cohort or case-control studies that do not involve researchers, the focus is on the health outcomes for a group of people. These studies are used in evidence-based medical practice, but they are weaker than clinical trial because they do not control for confounding factors or biases.
In general, systematic reviewswhich synthesize findings from multiple research studies provide the highest quality evidence. Single-case reports that detail the experience of a single individual are weak evidence, as they may not be applicable to a larger population. Personal testimonials and expert opinion alone are not supported with empirical data.
In the real world, clinicians can use evidence-based medicine framework to formulate a clinical question about a patient that can be answered clearly by reviewing the most recent research. A clinician might, for example, ask whether statins are more effective than diet or exercise in lowering LDL cholesterol in a 50-year-old man with no other risk factors. They can develop diagnosis and treatment plans by integrating evidence, patient preferences, and their own expertise.
As one might expect, gathering and putting together all the evidence can be a difficult process. Clinical guidelinesare often relied upon by clinicians and patients alike. These guidelines were developed by third parties, such as the American Medical Association and the National Institutes of Health. These guidelines provide recommendations for care and standards based on a systematic and thorough evaluation of the available research.
Dawning of precision medicine
Around this time, two other transformative developments were taking place in science and healthcare. Precision medicine is a new field that uses patient-specific data to tailor health care decisions for each individual.
First was the Human Genome Project, which began in 1990 and ended in 2003. It was designed to create a map of the human DNA or genetic information that cells use to function.
The map of the genome of humans enabled scientists to discover thousands of genes linked to rare diseases, understand how people respond differently to a drug, and identify tumor mutations that can be treated with specific treatments. Clinicians are increasingly analyzing DNA of patients to identify genetic variations. This helps them improve their care.
National Human Genome Research Institute/Flickr.
Second, the development of electronic recordsfor storing patient medical histories. The development of industry standards for digital records began in the late 1980s, even though researchers had been conducting studies on digital records for many years. The adoption of electronic medical records did not spread widely until the 2009 American Recovery and Reinvestment Act ().
Electronic records allow scientists to conduct large scale studies on the associations between genetic variations and observable characteristics that inform precision medicine. Researchers can use patient records in a digital format to train AI models that will be used in medical practice.
More Data, More AI, More Precision
On the surface, it is not a new idea to use patient health information for personalizing care. The ongoing Framingham Heart Study (19459139), which began in 1948 has produced a mathematical model that estimates a patient’s risk of coronary artery diseases based on individual health information rather than the average risk for the population.
The Human Genome Project, electronic medical records and other efforts to personalize healthcare today are fundamentally different from those made before. The mental capacity needed to analyze the vast amount of data and complexity available to each patient is far greater than that of the average human brain. Each person has hundreds to thousands genetic variants as well as hundreds to thousands environmental exposures. They also have a clinical history which may include a variety of physiological measurements, lab results and imaging results. In the ongoing work of my team, the AI models that we’re developing to diagnose sepsis in infants use dozens input variables, some of which are updated each hour.
Researchers, like myself, are using AI to create tools that help clinicians analyze this data and tailor diagnoses to each individual. Some genes can influence how certain medications work in different patients. Cost makes it difficult to screen all patients. Genetic tests can reveal certain traits. AI systems can analyze the medical history of a patient and predict if genetic testing is beneficial based on their likelihood to be prescribed a drug that is known to be affected by genetic factors.
A second example is diagnosing rare diseaseor conditions that affect less than 200,000 Americans. The diagnosis is difficult because there are thousands of rare diseases with overlapping symptoms and that the same disease can manifest differently in different people. AI toolsare able to help by examining the unique genetic traits of a patient and their clinical characteristics in order to determine which one is likely to cause disease. These AI systems could include components that predict if a patient’s genetic variation negatively impacts protein functionor if the patient’s symptoms are similar to specific rare diseases.
Future clinical decision-making
In the near future, new technologies will enable routine measurement of biomolecular information other than genetics. Wearable health devices can continuously monitor heart rate and blood pressure, as well as other physiological features. This data can be used by AI tools to diagnose diseases and personalize treatments.
Related research is already showing promising results in precision canceras well as personalized preventive health. Researchers are developing a wearable ultrasonic scanner for breast cancer detection, while engineers are developing skinlike sensor that detects changes in tumor size.
The research will continue to expand the knowledge of genetics, environmental exposures’ effects on health and how AI functions. These developments will have a significant impact on how clinicians make care decisions and provide it over the next fifty years.
Aaron J. Masino, Associate Professor of Computing at Clemson University.
The article is republished under a Creative Commons License from The Conversation. Read the original article.
