Even in this digital age, manual processes still constrain healthcare providers. Documenting clinical encounters, reviewing complex patient records (e.g., labs and imaging), tracking patient care gaps, and reviewing and refilling medications require significant time and effort. These cumbersome processes present an opportunity for technology to play a bigger role in reducing clinician burden and improving patient care.
To take the next step in using data-driven technology to improve care, health systems need to increase analytics and machine learning (ML) adoption, then leverage those capabilities to provide the most relevant information to clinical decision makers.
The digitization of healthcare via the EHR has provided clinicians with more information than ever. However, that information can quickly become more than a clinician can process, especially in the brief timeframe they have with a patient. For technology to improve care, solutions and tools need to make data meaningful to the clinician and impactful to patient care.
EHRs increase providers’ access to a patient’s basic health data, but they continue to fall short of making that data actionable. Organizations can bridge this gap by using advanced analytics and ML to deliver more valuable information at the point of care. Instead of the clinician reviewing a few blood pressure readings as they sit with the patient, an advanced ML algorithm could be reviewing the last five years of blood pressure readings, BMI, cholesterol, race, family history, and socioeconomic information.
With greater speed and precision, the algorithm could give the clinician critical information, such as the patient’s risk for stroke, myocardial infarction, or kidney disease. The algorithm could also suggest possible treatment options based on the latest relevant literature.
Machine learning can process large amounts of patient data beyond the scope of human capability, then reliably convert that analysis into medical insights that help clinicians plan and deliver care. Clinicians can use these data-driven insights—based on the patient’s data and data from other patients with similar conditions—to understand diagnoses and treatment options, achieve better outcomes, and lower costs of care.
Establishing trust in ML requires transparency in the process. Clinicians are used to reviewing medical science from the perspective of clinical trials, designed and implemented by experts. Once complete, the trial is peer reviewed and published with information about the processes, data, and outcomes. This transparent process lends itself to scrutiny, evaluation, and explanation.
Widespread adoption of ML in medicine will likely require similar processes. Subject matter experts will play an integral role in helping build and evaluate algorithms. Algorithms will also require transparency, explaining the reasoning behind recommendations, and how recommendations can improve clinical outcomes.
When providers understand how ML can support data-informed decision making, they will view the capability as a real-world tool to augment patient care. Analytics organizations can also use real-world scenarios to earn team member ML buy-in.
Radiology, for example, has been on the forefront of adopting ML in clinical practice. Instead of predictions that ML would replace radiologists, it is now seen as a tool that can augment workflow, efficiency, and diagnostic capabilities. Technology companies including the American College of Radiology and the U.S. Food Drug Administration are collaborating with large academic institutions with large data sets. This collaborative effort between medical subject matter experts, data scientists, and governing bodies is key to future ML adoption. As experts develop, test, and demonstrate algorithm success with clinical outcomes and publish results for evaluation, radiologists will incorporate these algorithms into their workflow to benefit patients.
As team members learn to trust and adopt ML, they can apply the data science capability to standardized processes with large image data sets. Certain areas of medicine that involve pattern recognition, such as radiology, dermatology, and pathology, have seen increasing ML development. Data scientists can train ML models to look at images, identify abnormalities, and augment clinician interpretation, with the potential to improve the diagnostic accuracy and ultimately improve patient care.
For example, a high-risk skin cancer patient comes in for a routine mole check to screen for changes to the size, shape, and color of moles worrisome for melanoma. The dermatologist of the future has access to skin imaging, which takes a picture of the patient’s skin and then kicks off an ML algorithm that can document each mole in minute detail, compare moles to past images, and directs the dermatologist to specific moles that may need additional evaluation.
As health systems gain access to more data and invest in integration and interoperability infrastructure, ML will potentially reach all aspects of medicine. The most accurate ML models will typically come from organizations with big data sets and the supporting infrastructure, including a data platform and ML technology (e.g., Healthcare.AI by Health Catalyst). These tools are necessary to aggregate data from enterprise data warehouses, data platforms, and third-party data sources, then identify the relevant data inputs for ML. This process enables comparative effectiveness and research, resulting in unique, accurate ML algorithms.
The increase in healthcare data and accurate algorithms means ML-generated insights can reach more areas of medicine. For example, a primary care provider (PCP) treating a patient with hypertension could review ML-generated information during the clinic visit. The process could begin with an analysis of home-generated data from connected devices, including blood pressure readings, weight, fitness data, sleep data, compliance with medications, and nutritional data. Future wearable sensors may help collect additional data. An ML algorithm could combine that data with the health system’s data, such as additional diagnosis, symptoms, lab tests, imaging, and genomic data. By integrating all available patient data in real time, ML will augment the PCP’s ability to better understand the patient’s current state and future health risks and enhance medical decision making to improve that patient’s long-term outcomes.
This potential to augment care exists across all specialties of medicine as more data is available. For example, oncology will see advances in diagnosis with ML-augmented imaging and pathology, and ML’s analysis of complex genetic data will improve clinical care and inform treatment. ML algorithms will be like an additional expert consultation, aggregating and informing oncologists with the latest clinical trial results across a broad spectrum of cancers, allowing easier access to newer treatment options and even helping refer patients to clinical trials with promising investigational drugs.
Machine learning-powered tools can sift through more data, including libraries of similar patients, diagnoses, and genetics, than one person can process. As such, ML opens data resources that include treatment options and predictions for each treatment’s effectiveness, mortality rates, side effects, and cost. In this way, ML can put in infinitely more work behind the scenes, delivering real-time, accurate information to the point of care.
Many health systems already leverage ML in everyday clinical practice, advancing medicine into a new realm. While the benefits of ML to augment provider decision making seem endless, health systems first need to understand ML’s role in healthcare and then invest in supporting tools and infrastructure. With this process, ML is becoming commonplace in healthcare.
Analyzing patients’ real-time status along with that of similar patients at other health systems will enable providers to make the most informed decisions and more deeply understand the best course of action for each patient. The analytics engine and the ML algorithm can analyze millions of data sets and present relevant information to the clinician at the point of decision making, enabling care teams to spend less time with data and more time with patients.
Would you like to learn more about this topic? Here are some articles we suggest: