top of page

FDA & Adaptive AI, Systems Plan for New Digital Pay Rates, VNAs for Imaging-The HSB Blog 10/27/20

FDA Patient Engagement Advisory Committee (PEAC) Meeting on Adaptive AI and ML

Event: (10/22) On October 22nd, the FDA Patient Engagement Advisory Committee (PEAC) held a meeting to discuss complex issues surrounding Artificial Intelligence (AI) and Machine Learning in Medical Devices. Attendees included several patients, companies, clinicians, and other high-level leaders. The meeting was centered on issues around the medical devices, its regulation, and usage by patients, in addition to their concerns on data privacy and transparency. The committee made recommendations on improving AI and machine learning in medical devices by giving careful consideration to datasets and how it can be used to train and improve algorithms. Adaptive AI presents new challenges for patients, companies, and regulators regarding informed consent, considering where the device would learn algorithms, use algorithms to improve decision-making, and make advanced improvements moving forward.

Description: AI technology will be used to improve diagnosis and make recommendations for a course of treatment. Using adaptive AI will allow machines to learn more freely and improve what it learns from each sequence it is running. For future AI models to be successful, there must be a diverse representation in the intended patient population's data. Diverse and unbiased patient cohorts will allow for less biased data and remove clinical variables that lead to AI devices' mistakes. When data is not present for certain groups in a population, AI researchers cannot study them and find ways to combat societal issues. Incorporating and collaborating with patients in the creation of AI technologies will provide transparency and confidence in the products; however, many barriers exist even after a product makes it to market because criteria can change over the device's life. Adaptive AI makes decisions dependent on the life it sees out there – an apparent reason why it must be regulated because devices should differentiate between good and bad decisions to ensure safe outcomes. The committee struggled with a decision on how to explain the challenges of how explainability and transparency in AI models can be presented to patients.

Implications: Large clinical datasets can be used to train and improve AI algorithms, leading to immense improvements in AI-based systems that diagnose as well as those that determine treatment protocols and actually treat patients. AI technologies will change clinician workflow and ensure continuous safety and efficacy in delivering ongoing quality care to patients.However, because ML algorithms can “learn” and “change” over time, medical diagnoses and treatment can change as adaptive AI models change as well, raising ethical concerns around transparency and informed consent. In addition, adaptive AI models also suffer from the shortcoming of biased training sets and under representation of underserved communities. As a result, patients and regulators need the ability to understand and examine models too so that the risks and consequences of any changes to treatment protocols due to improvements in the models can be assessed. In addition, the committee appeared to strongly imply that new informed consent releases would be required when changes to AI algorithms make meaningful changes in treatment or diagnosis protocols.

After COVID-19 Spurred a Boom in Telehealth, Systems Mull How to Sustain Momentum

Event: (10/20) Healthcare Dive recently published an article highlighting the Center for Connected Medicine/Klas Research’s 4th Annual “Top of Mind for Top Health Systems Report”. Among other things the report looked at trends in telehealth reimbursement post-COVID and health system’s potential responses.

Description: When the pandemic began both commercial and government insurers waived numerous rules and regulations to increase access to telehealth including waiving co-pays for telehealth visits and equalizing reimbursement for in-office and telehealth visits. While the Federal and State regulators continue to explore which rules to make permanent (for Medicare and Medicaid beneficiaries), some private insurers have recently changed some rules (including the waiver of patient co-pays) for telehealth visits. This survey of 117 executives at 112 provider organizations gave some insight into how provider organizations were planning to respond depending on how these changes settle out.

Implications: As noted by Healthcare Dive “a lot of how much [providers] are going to invest in technology is going to have to have to do with what regulations end up looking like.” Since the regulations will to a large extent dictate reimbursement (which were approximately 30% lower for telehealth prior to COVID), despite the recent uptick in volumes, profitability will drive continued usage. For example, over 30% of respondents to the study noted that they were unsure of what they would do if reimbursement returned to prior levels, while 20% said they would continue providing telehealth services and 16% said they would analyze the continued viability of continued use. Moreover, it looks like payment and regulation will be top of mind for the foreseeable future with 25% or respondents noting that payment models and regulation were the top areas for future telehealth improvement.

How [Vendor Neutral Archives] VNAs Can Address the Challenge of Data Accessibility and Analysis

Event: (10/19) A recent article in Mobihealthnews noted that if healthcare providers had more accessibility to aggregated data, it would be easier to combat difficulties such as reducing clinician burnout, improving efficiency, and increasing patient satisfaction. Vendor Neutral Archives (VNAs) can drive better outcomes by enabling data to be provided as a source of AI, business intelligence, or to deliver clinical information and insights.

Description: A recent survey conducted by McKinsey & Company found that data integration and analytics is one of the major trends reshaping imaging services and influencing the future of healthcare. VNAs which allow access to data from any vendor system in multiple formats, improving workflow and reducing the time taken to complete different stages of an imaging exam will enable data to use artificial intelligence to provide decision-making support. This is an essential component in connecting existing systems to open platforms and specialists with patients. VNAs will eliminate the need to switch between systems or wait for data to arrive from other sites because prompt access will be given to correct data from any location in the network.

Implications: The world and healthcare in particular is in the middle of a data explosion. The amount of healthcare data collected is currently projected to double every 73 days compared to 2010 when it was said to be doubling every three and a half years. As a result, healthcare costs have been growing faster than expected, and payers are looking to more sophisticated technology to improve the management of costs. Data integration and analytics to realize the value of data have become more crucial for healthcare delivery. Using VNAs will improve clinician workflow, financial, and operational performance of departments and organizations, reduce clinician demand, lower costs, and provide better patient outcomes.

CDC's 'Virtual Human' Relays Prostate Cancer Info Through Candid Conversations

Event: (10/19) Mobihealth News recently published an article that the CDC launched a prostate cancer information program with a uniquely digital face: a "virtual human" named Nathan who speaks with men about screening and treatment. The CDC noted that the tool can help increase screening since “the decision to be screened and treated for prostate cancer can be overwhelming and complicated” … “some of this is driven by discomfort and long-identified stigma” with a prior examination protocol.

Description: "Talk to Someone About Prostate Cancer" takes the form of an in-browser interactive video developed by health simulation company Kognito. In the video, Nathan introduces himself and prompts the viewer by asking how confident they are talking to their provider about prostate cancer screening. Users indicate their response by clicking one of several text responses, prompting a relevant reply, and kicking off a series of question and answer conversation trees.

Implications: Prostate cancer is among the most common cancers in American men. Thirteen in 100 men will develop prostate cancer during their lifetime, and two or three will die due to it, according to the CDC. As symptoms vary, there is a high need for men to speak to doctors about whether or not they should undergo screening. Virtual health engagement tools that employ digital avatars can help some patients open up about difficult-to-discuss topics, providing a new way for individuals to become more engaged in their care. The CDC's virtual conversation effort comes after this summer's "Access Initiative for Quitting Tobacco," which combines free nicotine replacement therapies and conversations with a similar digital human named Florence that used a microphone and artificial intelligence to interpret spoken questions.

COVID-19 Impacts on Cancer Care

Event: (10/21) On October 21st, the Journal of Clinical Oncology Clinical Cancer Informatics published an article highlighting the significant drop in cancer screening, diagnosis, and treatment for Medicare beneficiaries. The changes in the utilization of cancer care services is attributed to the decrease in routine healthcare visits brought on by the COVID pandemic.The research was conducted by Avalere, a leading healthcare consulting firm, in collaboration with the Community Oncology Alliance, a non-profit advocating for community oncology practices.

Description: The study found that for the period March-July 2010 compared to 2019, there was a significant decrease in cancer screening, biopsies, surgery, office visits, and therapy with variation by cancer type and site service. At the peak of the pandemic in April, screening for breast (-85%), colon (-75%), prostate (-74%) and lung cancer (-56%) were all all significantly lower as indicated. The decrease in diagnosis and delay in care is attributed to the stay at home order to reduce COVID-19 transmission, especially amongst the elderly and immunocompromised. As a result, many healthcare providers accommodated short term adjustments to cancer care delivery, such as temporarily discontinuing non-emergent care screenings, shifting delivery of care to telehealth, and delaying surgeries and other in-office cancer services to reduce transmission risk. Although the stay at home orders were lifted in May and June, utilization of certain oncology services continue to lag, fewer patients are undergoing screening, with many providers and patients choosing to reschedule or completely forego screening, leading to fewer cancer diagnoses.

Implications: With the decrease in screenings, diagnosis, and cancer care delivery, researchers found that cancer is likely to present itself at a later stage and require more complex care thus lowering the likelihood that patients will respond and ultimately be cured. To combat this, stakeholders are encouraged to increase awareness of the dangers of medical distancing and regain seniors’ confidence in their ability to safely receive care. Policies and technology to promote access to cancer care have the potential to reduce the projected morbidity and mortality amongst the population. However, further studies need to be conducted to understand the overall impact on different patient populations and potentially take corrective action.

Researchers Combine AI with EHRs to Improve [COVID] Hospitalization Prediction

Event: (10/21) A recent study by researchers at the NYU Grossman School of Medicine has applied AI to EHR (electronic health records) data to predict more accurately good outcomes from COVID-19 treatment. In particular researchers wanted to provide clinicians with actionable information that could be easily integrated into their exister workflows

Description: Researchers undertook the study to apply AI to EHR data to help hospitals and doctors more accurately manage patients who had tested positive for COVID. The model analyzed 3000 retrospective cases and developed a model that could identify hospitalized patients likely to have good outcomes with 90% accuracy. In addition, researchers deliberately sought to reduce the number of features and variables used in the model, opting to go with the minimum data requirement needed to make a prediction.

Implications: The study predicts outcomes in hospitalized COVID patients with a high degree of accuracy while integrating to current data infrastructure. However, researchers noted the need to remain vigilant when designing AI related tools as there are instances where the physician’s knowledge exceeds that of the machine and must be integrated with the AI output. The researchers also found that to be effective AI learning solutions require two distinct components; 1) addressing a clearly defined use case that clinical leaders will champion and 2) [developing a model that] motivates changes in clinical management, based on model predictions.


Search By Tags
Recent Posts
Follow Us
  • Facebook Basic Square
  • Twitter Basic Square
  • Google+ Social Icon
bottom of page