Contributed Commentary by Joshua Reicher, M.D. and Michael Muelly, M.D.
October 26, 2021 | As we’ve seen with the COVID-19 pandemic, diagnostics deliver vital information that enable health providers to properly triage patients and provide the best treatment according to their illness. There are 14 billion laboratory tests, 70 million CT scans, and 40 million MRIs ordered annually in the United States, with the vast majority of medical decisions relying on these test results. However, diagnostic challenges remain due to complexity of clinician assessments and integrating complex results together. Enhanced solutions are needed to improve patient outcomes.
Machine learning and digital biomarkers could potentially change the field of diagnostics with better accuracy, reliability and cost-effectiveness, not only addressing the limitations with conventional diagnostics but also for use in future healthcare initiatives and the biopharmaceutical industry.
Challenges with Current Diagnostics
There are several challenges within the existing diagnostics and prognostics landscape. The first is the use of subjective, qualitative measurements. This can lead to a wide range of variability in clinical assessments. Trying to better standardize assessments across clinical sites and laboratories is a major area for improvement.
The second is that computationally-driven and/or quantitative techniques, including many prognostic calculators, are generally either not targeted to specific diseases or so highly targeted to be limited in access and usability. These methods rarely capture the complexities and variance of disease phenotypes or take into account each patient’s unique characteristics.
Third is the morbidity and mortality associated with some forms of tissue biopsy. As an example, it has been reported that the mortality rate following surgical lung biopsy for interstitial lung disease diagnosis can range as high as 16%.
In the past several years, there has been a significant growth in utilizing digital data to improve patient outcomes. Leveraging diverse clinical and laboratory datasets to identify distinct disease signatures has the potential to address many of the challenges with current diagnostic methods.
Emergence of Machine Learning and Digital Biomarkers
Digital biomarkers, which are derived from existing digitally-stored behavioral or physiological data that have been reanalyzed with machine learning in more depth and detail, are becoming more accessible within the industry due to significant technological advances in collecting and analyzing large amounts of data. Some examples of digital biomarkers include speech-based biomarkers or activity tracking data from wearable sensors.
Compared to traditional diagnostics and prognostics, digital biomarkers offer objective, often quantifiable signatures that can be used to detect disease earlier before clinical symptoms arise or provide valuable information about treatment response. Another advantage to digital biomarkers is that they offer the potential for widespread accessibility such that disease assessment is not reliant on one small set of expert providers.
Outside of conventional diagnostics and prognostics, digital biomarkers can also be used in other areas of healthcare. One example is through early disease screening in the evaluation of large populations during emergency situations. This can be a cost-effective method for establishing public health initiatives and lowering downstream expenditures through early intervention.
In life sciences research, digital biomarkers can be used for uncovering the underlying biology of disease progression, from cancer to lung disease to heart failure. In clinical trials, they can enable better patient stratification for precision drug targeting and enrollment screening. Furthermore, digital biomarkers can optimize trial assessments across multiple sites where there is subjective variability between locations and practices.
Digital Biomarkers and the Clinician
Digital biomarkers and machine learning in diagnostics are valuable assets that can be integrated into the clinician’s toolkit. In general, physicians are best suited at making broad diagnostic assessments, considering many different options at the same time while thinking through different disease variables. However, doctors, and most humans, struggle with complex statistical and mathematical calculations.
Machine learning algorithms and models can be used to supplement the clinician in cases where identifying signals are not ideal for human interpretation. These focused evaluations of specific, and often subtle, patient characteristics are vital for clinical decision-making, particularly for rare diseases. With digital biomarkers, the goal is to extract the data to a level that makes it more tractable for doctors to make informed choices about care.
As the digital biomarker space continues to expand, stakeholders need to focus on better methods for collecting functional, high-quality digital data. In many cases, there is not enough data available or the data collected does not directly correspond to information about the disease, preventing the establishment of precise diagnostic assessments. This issue needs to be addressed as the field moves forward.
With the continued support of digital health initiatives from industry, regulatory agencies, payers and providers, there is a tremendous opportunity to leverage clinical, engineering and computational expertise to solve the limitations with conventional diagnostics.
Dr. Joshua Reicher is Co-founder and CEO of imvaria. Dr. Michael Muelly is Co-founder and CTO of imvaria. Dr. Reicher can be reached at firstname.lastname@example.org.