Grand Rounds March 22, 2024: Early Diagnosis and Assessment of Autism Via Objective Measurements of Social Visual Engagement (Warren Jones, PhD)

Speaker

Warren Jones, PhD
Director of Research, Marcus Autism Center
Children’s Healthcare of Atlanta
Norman Nien Distinguished Chair in Autism
Associate Professor, Dept. of Pediatrics
Emory University School of Medicine

Keywords

Autism Spectrum Disorder, Biomarkers, Social Visual Engagement

Key Points

  • Autism affects 1 in every 36, impacting more than 9.1 million individuals and their families in the U.S. When we think about conditions that affect young children and their families, autism is one of the most common.
  • Parents in the U.S. spend an average of 2-3 years between the time when they first begin to worry and the time when they finally receive a diagnosis. There are not enough expert clinicians or expert centers to meet public need. Disadvantaged families wait even longer.
  • Clinicians need more measures that are objective, quantitative, dimensional and fine-grained, performance-based, standardized, efficient and community-viable, able to capture core features of social disability, and mechanistically relevant.
  • Social visual engagement measures how children look at and learn from their surrounding social environment. Children look at and acquire information from what they are looking at. Researchers use eye tracking data to measure social visual engagement, which reflects early-emerging differences in autism spectrum disorder (ASD).
  • In 3 studies, researchers tested the performance of eye-tracking-based assays of social visual engagement in 16-30-month-old children to accurately assess presence of ASD and accurately assess severity of ASD.
  • The Discovery study included 719 participants, and the Replication study included 370 participants aged 16-30 months old. The initial Discovery study and first Replication study showed high sensitivity and specificity when comparing eye-tracking-based measures of social visual engagement with expert clinician diagnosis in children approximately 16-30 months old.
  • With the results from the initial studies, the trial team embarked on a multi-site clinical trial at 6 sites across the U.S. 475 participants completed eye-tracking measurement of social visual engagement, expert clinical diagnosis, and a rating of expert diagnostic certainty.
  • The study resulted in 335 participants with a reference standard certain diagnosis for ASD and 140 with a reference standard uncertain diagnosis for ASD. The children seen in the study who did not have autism did have other developmental diagnoses, which highlights the challenge of diagnosing children with developmental delays.
  • Study results show a high sensitivity and specificity when comparing eye-tracking-based measures of social visual engagement with expert clinician diagnosis in children as young as 16-30 months old. Results also show a strong correlation with standardized assessments given by experienced clinicians

Learn More

Read more in JAMA.

Discussion Themes

-Is there an eventual hope that a tool like this could be used without a referral to a specialty center? Absolutely. We started at a point to try to develop a tool with gold standard outcomes, and we are going on to test screening studies in other age groups. Our hope is to extend and develop clinical tools that could be easier to use.

What did you learn that informed the development of tools that would be more generalizable? This work has been conducted in conversation with FDA for many years, successfully moving something that was lab-based to increasingly more real-world. We are working toward making this more usable by general clinicians, asking what they need and want to make it more useful. For the screening studies, we looked at the SMART study framework for clinical trials to get to who is most likely to need a diagnostic evaluation.

 

Tags

#pctGR, @Collaboratory1

Grand Rounds March 15, 2024: Antibiotic Choice on Renal Outcomes – The ACORN Trial (Edward Qian, MD, MSc)

Speaker

Edward Qian, MD, MSc
Assistant Professor of Medicine
Vanderbilt University Medical Center
Division of Pulmonary and Critical Care Medicine

Keywords

ACORN, Antibiotics, Acute Kidney Injury, Critical Care Medicine, Randomized Clinical Trial, EHR

Key Points

  • Sepsis is a common cause of critical illness and death. A common treatment is treating with vancomycin plus either piperacillin-tazobactam or cefepime. Piperacillin-tazobactam and cefepime have unique risk/benefit profiles but comparative data are lacking.
  • There have been associations with piperacillin and acute kidney injury (AKI). Acute kidney injury is associated with a 6-to-8-fold increase in mortality in critically ill patients. There was concern that there is an association between concurrent vancomycin and piperacillin-tazobactam with creatinine elevations, but a retrospective, observational analysis was inconclusive.
  • The ACORN Trial was randomized clinical trial to understand the effect of empiric antibiotic choice for patients in the ED and ICU. The trial enrolled 2,511 patients, who were adults who were in the hospital ED or ICU for less than 12 hours and received at least one dose of empiric cefepime or piperacillin-tazobactam. The intervention was a choice of empiric gram-negative antibiotic (cefepime or piperacillin-tazobactam).
  • The screening process was done via an interruptive alert in the electronic health record (EHR) that was activated based on a provider placing an order for drugs. The alert would remind for exclusion criteria, and ask if the patient is eligible for ACORN trial.
  • The trial operated under a waiver of informed consent. The research involved no more than minimal risk to subjects, and the research could not be carried out practicably without the waiver or alteration.
  • The participant randomization happened within the EHR. The delivery of the intervention happened in the EHR, starting with an alert to enroll. Providers were sent an order set depending on the arm. The study team monitored for compliance within the EHR.
  • The primary outcome was AKI ordinate scale between enrollment and Day 14. Secondary outcomes were major adverse kidney events within 14 days and days alive and free of delirium and coma to day 14.
  • The trial found that, compared to cefepime, piperacillin-tazobactam does not increase the incidence of AKI. Compared to piperacillin-tazobactam, cefepime may decrease the number of days alive and free of delirium and coma.

Learn More

Read more in JAMA.

Discussion Themes

-How much pilot testing did you do? I was trained as an EPIC physician builder. In preparation for this trial, I underwent this training so all the pieces that you saw, I built and tested in a relatively short time period.

How did you engage clinicians and make sure the question you were answering was one they wanted the answer to? There was a huge amount of leg work to engage stakeholders. We were talking to the leadership of ICU, nephrology, ID, and others for buy-in, to answer questions, and to make modifications based on suggestions.

How portable are your EHR modifications? In particular, how big a lift would it have been to implement the same protocol in multiple EPIC installations? This whole trial was done while I was getting a master’s in applied informatics. The answer is no. The screening and randomization could be portable. Each installed EPIC has slight differences and quirks. There will be small differences in the application of the intervention.

 

Tags

#pctGR, @Collaboratory1

Grand Rounds March 8, 2024: Public-Private Partnerships in the Trustworthy Health AI Ecosystem (Michael Pencina, PhD; Brian Anderson, MD)

Speakers

Michael Pencina, PhD
Chief Data Scientist, Duke Health
Duke University School of Medicine

Brian Anderson, MD
CEO, Coalition for Health AI

Keywords

Artificial Intelligence, AI, Coalition for Health AI (CHAI), Duke ABCDS, Governance, Regulation

Key Points

  • The regulatory landscape for AI is changing rapidly. There are a number of government agencies trying to establish their position on AI, including FDA and The U.S. Department of Health and Human Services Office of Civil Rights. The Office of the National Coordinator for Health Information Technology (ONC) published its final rule on AI in December.
  • We are living in a “wild west” of algorithms. There’s so much focus on development and technological progress and not enough attention to its value, quality, ethical principles, or health equity implications.
  • There are principles we can apply for the responsible development and use of AI, such as ensuring that AI technology serves humans; defining the task we want the AI tool to accomplish; describing what the successful use of the AI tool looks like; and creating transparent systems for continuously testing and monitoring AI tools.
  • Three years ago, Duke created an algorithm-based clinical support (ABCDS) function which monitors application of algorithms, AI, in our health system. The governance is now changing as AI moves from predictive models to large-language models.
  • This is complex environment where not everyone speaks the same language, clinicians, developers, patients, informaticists, statisticians, and more. It is very important that development teams are connected with what is happening at the health system level and the clinical teams on everything they build.
  • Model development follows the FDA process starting with model development, silent evaluation, effectiveness evaluation, and general deployment, with evaluations at regular intervals.
  • The Coalition for Health AI (CHAI) is a nonprofit that brings together more than 1,400 private sector organizations, including health care systems, payors, device manufacturers, technology companies, and patient advocates, plus U.S. government partners, to provide a framework for the landscape of health AI tools to ensure high quality care, increase trust among users, and meet health care needs.
  • Generative AI will become more impactful and ubiquitous. We have an urgent need to define what is safe and effective in this space and rethink how we regulate and align generative AI

 

Discussion Themes

-The field is really exploding. How are you forecasting how the federation will come together to address the potential volume of algorithms? Part of the challenge is that it is really important that we build consensus quickly and include as many voices as we can. We are identifying as many use cases as we can, building consensus around a set of shared definitions and frameworks, and then fleshing those out in the nuances of use cases. The other challenge is, once we have that defined, we need to share freely what we learn for wide adoption. We can’t do this in silos.

How are you approaching ethics in AI? How do you envision including an ethical component in algorithm development? For our ABCDS team for Duke, we have an ethicist on the team to make sure our principles are applied. This is not enough; we will need more as the field develops

Tags

#pctGR, @Collaboratory1

Grand Rounds March 1, 2024: Effect of an Intensive Food-As-Medicine Program on Health and Health Care Use: Evidence from a Randomized Clinical Trial (Joseph Doyle, PhD)

Speaker

Joseph Doyle, PhD
Erwin H. Schell Professor of Management and Applied Economics
MIT Sloan School of Management

Keywords

Food-as-Medicine, Randomized Clinical Trial, Diabetes

Key Points

  • Diabetes is common and costly. 9% of the U.S. has the condition; it is responsible for 300,000 premature deaths/year and $330 billion in annual healthcare spending. Payers should have some incentive to improve outcomes because sustained reduction in HbA1c from poor to fair can result in cost savings. Food insecurity is associated with diabetes.
  • Food-as-Medicine observational studies have had a large correlation with improved health.
  • For the intervention in this Food-as-Medicine trial, participants with diabetes were prescribed 10 meals/week for participants and their families that they could fill at the program’s clinic. Clinic staff included a dietitian, a nurse and a community health worker. The dietician met with the patient to share education about nutrition, portion size, recipes to make food taste good, and information via an optional diabetes self-management program. They also screened for complications and close care gaps. The average duration of the intervention was 1 year, and the cost was about $2,000 per participant.
  • The trial was for adults with HbA1c greater than 8.0, who were food insecure and lived within a residential zip code within 25 miles of a clinic. Recruitment was by phone calls and physician referrals, and consent was over the phone. Randomization was stratified by HbA1c greater than 9.5 and site. The intervention group started the program now. The control group started in 6 months and received a brochure that lists addresses of area food banks.
  • The primary outcome was HbA1c after 6 months. Lab results for HbA1c, cholesterol, triglycerides, blood pressure, weight were taken at 0, 6 and 12 months. There were also surveys to assess program education, diet, a self-efficacy questionnaire, and a self-assessed physical and mental health questionnaire at 0, 6, and 12 months. Participants received a $50 gift card for completing the labs and surveys.
  • Additional data sources came from EHR data, health plan claims, and program participation data including food visits and education.
  • 500 patients were randomized to intervention or control groups, with a utilization of 465 and 349 participants completing the 6-month HbA1c lab sample. The treatment group started with a mean 10.3 HbA1c it dropped to a mean of 8.8 at 6 months, which plateaued in 12 months. The control group followed similarly. There were no significant drops in cholesterol, triglycerides, or fasting glucose and weight did not lower. There was not a statistically significant difference between the 2 groups at 12 months.
  • The study found a null effect on HbA1c. The study did find substantial effects on diet and healthcare engagement.

Learn more

Read more in JAMA.

Discussion Themes

-Why do you think it was a negative study? The idea was that they would have to go get the food. Medically tailored meals is the alternative. We don’t have the diaries of what they do with the food. We see people visiting the fresh food pharmacies for up to 12 months. When they get home do they give the food away, eat it themselves, throw it away, we don’t know. In terms of the types of food they had, they were not especially culturally diverse.

Can you say something about the dose of the meal, which was 10 meals a week? Were people eating less healthy meals for the non-program meals? How much food to give is a parameter that needs more research. The dietitians in the program thought that 10 meals a week was the right number and that providing all meals would have been too much and resulted in waste. From a researcher standpoint, we need more information

Tags

#pctGR, @Collaboratory1

Grand Rounds February 23, 2024: Virtual Vigilance: Monitoring of Decentralized Clinical Trials (Adrian Hernandez, MD; Christopher J. Lindsell, PhD)

Speakers

Adrian Hernandez, MD
Executive Director, Duke Clinical Research Institute
Vice Dean, Duke University School of Medicine

Christopher J. Lindsell, PhD
Professor and co-Chief of Biostatistics, Department of Biostatistics & Bioinformatics
Director, Data Science and Biostatistics, Duke Clinical Research Institute
Duke University School of Medicine
Editor in Chief, Journal of Clinical and translational Science

Keywords

Decentralized Clinical Trials, Virtual Vigilance, Data and Safety Monitoring

Key Points

  • There has been global growth of decentralized clinical trials (DCTs) and with that a growing need to develop best practices and to ensure quality results are generated from trials. The global decentralized clinical trial market is expected to grow at a compound annual growth rate of 30.1% from 2021 to 2026.
  • But there are concerns to developing DCTs including lack of standardization and validation, regulatory and ethical uncertainties, engagement vs. coercion, data security and privacy issues, technological literacy and access, resistance to change and adoption, and lack of “safe” sharing.
  • There is agreement that trials need to meet the people, at home and covering clinical trial deserts.
  • There are 5 guiding principles for defining quality that should inform DCTs: Have we enrolled the right participants according to the protocol with adequate consent? Did participants receive the assigned treatment and did they stay on the treatment? Was there complete ascertainment of primary and secondary efficacy data? Was there complete ascertainment of primary and secondary safety data? Were there any major GCP-related issues?
  • Regardless of the trial inclusion and exclusion is routine. What we often do not think about is verifying the identity of the participant. In a remote study, it would be possible for duplicate enrollment, falsified or fabricated eligibility source documents, or data completed by surrogates. Consider secure digital identification, two-fact authentication or virtual/video visits. Balance with not adding barriers for participation.
  • Getting study drug and other study materials into the hands of a participant requires distribution via mail or courier, breaking the traditional chain of custody. Under RBM, the process by which study materials get to participants should be considered high risk and monitored accordingly.
  • As roles for sites change, it remains critical that participants can be actively managed and that data about patient status can be acted upon, including mechanisms for participants to ask questions and get timely responses, participants to report worrisome events, participants to report healthcare encounters and other events, tracking adherence to study intervention, and tracking adherence to data collection procedures. Solutions include a bi-directional EDC (electronic data capture system), MyChart for research, and active notifications to study personnel based on entered data.
  • Baseline state, treatment, outcome, and safety data are critical to understanding treatment benefits and risks. Outcomes including patient reported outcomes, functional assessments including via digital technology, healthcare events or mortality may require identify verification. Supporting documentation may include recordings of functional assessments, EHR data, or other information that can be uploaded for remote review. Note that the release of medical records may be needed for health systems unrelated to sites.
  • New issues to consider for DCTs: Geographic distribution of participants; enrollment of two or more participants who share the same digital resource; enrollment of participants who do not have sufficient digital resources; and rogue digital and social media recruitment practices

 

Discussion Themes

-Do you anticipate that a new set of clinical practices will emerge from this work? I think we must. One of the things we are in the process of sharing in terms of what we are observing is putting new procedures for what we monitor in a study. It is based on good study design and focusing on the key principles we are trying to adhere to. There has to be a range of approaches that could be fit for use.

Can you comment on how this is perceived at NIH and some of the institutes? I am sure NIH is trying to accelerate our understanding of DCTs. The knowledge base isn’t there yet in terms of what you need to look for that would be missing in terms of monitoring efforts. From my perspective, if one key goal for NIH is to reach people in underserved communities having sound practices for decentralized methods will be important.

How can we prepare research teams that do not always have the training and compensation to do all of these things? We need to train the workforce for what we need them to do. We are building out what are the core competencies, the technological, legal, ethical, clinical research administration competencies that are needed in addition to data management for a CRC. I hope we see further development of education programs that support the workforce with the skills needed. The skills may also vary between trials and getting participants the care they need.

Tags

#pctGR, @Collaboratory1

Grand Rounds February 16, 2024: Clinical Implications of the MINT Trial: p=0.07 (Jeffrey Carson, MD, MACP)

Speaker

Jeffrey Carson, MD, MACP
Principal Investigator and Study Chair
MINT Trial
Provost-New Brunswick, Rutgers Biomedical Health Sciences
Distinguished Professor of Medicine
Richard C. Reynolds, M.D. Chair in General Internal Medicine
Rutgers, Robert Wood Johnson Medical School

Keywords

Transfusion; MI; MINT; Anemia; Clinical trials

Key Points

  • Anemia is common in patients with acute MI. Due to the lack of evidence, indications for red blood cell transfusion in patients with MI are controversial. Prior to the completion of the MINT trial, three trials had compared transfusion thresholds in a total of 820 patients, and the results were inconsistent. Trials in other clinical settings suggest the use of restrictive transfusion strategy, which is a lower hemoglobin level, is safe. Most trials conducted prior to the MINT trial suggested that a restrictive transfusion strategy was comparable to a liberal transfusion strategy.
  • The MINT trial investigators looked at 30-day mortality and several outcomes including MI, heart failure, stroke, bleeding, infection, and clot. None of these relative risks or confidence intervals are significant, suggesting that you could safely use a restrictive transfusion strategy for these patients. The previous trials suggest that when you randomize people to either more or less blood, those who receive less blood get about 40% less blood overall in these studies.
  • Prior to the MINT trial, investigators detected very little difference between liberal and restrictive transfusion strategies. However, they saw very wide confidence intervals, which made it difficult to determine the best way to manage patients in this context.
  • The objective of the MINT trial was to determine whether the risk of death or MI through 30 days differed with a restrictive transfusion strategy with a hemoglobin threshold of 7 to 8 g/dL as compared to a liberal transfusion strategy with a hemoglobin threshold of 10 g/dL among patients with an acute MI and a hemoglobin concentration of less than 10 g/dL.
  • The team enrolled patients across 144 sites who fit the following criteria: 18 years or older, had an STEMI or NSTEMI, had Types 1, 2, 4b, or 4c MI, and had a hemoglobin concentration of less than 10 g/dL within 24 hours of enrollment.
  • They utilized 2 transfusion strategies, restrictive and liberal. In the restrictive strategy, transfusion was permitted, but not required, when the hemoglobin concentration was less than 8 g/dL and strongly recommended when the concentration was less than 7 g/dL or when angina symptoms were not controlled with medications. For the liberal strategy, 1 unit of packed red blood cells were administered following randomization, and red blood cells were transfused to maintain a hemoglobin concentration greater than or equal to 10 g/dL through hospital discharge or 30 days.
  • There were 3 primary limitations of the MINT trial. The assigned strategy was not masked; trial results were not adjusted for multiple comparisons; and although pre-specified, cardiac death was not designated as primary, secondary, or tertiary outcome or adjudicated.
  • The results of the MINT trial did not demonstrate a statistically significant difference in the rate of 30-day death or recurrent MI in patients with acute MI and anemia assigned to a restrictive compared to a liberal transfusion strategy. Additionally, while not statistically significant, the point estimates for the primary outcome and secondary outcomes consistently favored a liberal transfusion strategy. Finally, heart failure and other safety outcomes were comparable in the two transfusion groups.
  • In contrast to other clinical settings, the MINT trial results suggest that a liberal transfusion strategy has the potential for clinical benefit with an acceptable risk of harm and may be the most prudent approach to transfusion in anemic patients with acute MI.

Learn more

Read more about the MINT trial.

Discussion Themes

-When did you consent the patients – before or after the cath lab? We consented them as early in the hospitalization as we could. Some patients were consented before cath and others were consented after. In general, we tried not to consent patients until their hemoglobin was less than 10 in most situations. Whenever it was less than 10, then we consented them when we could get to the patient, but that varied. There was a solid mix of pre- and post- cath lab.

-Could you discuss the journal review process? Was there substantial disagreement among the authors, reviewers, and editors regarding the trial conclusions or clinical implications? There were differing opinions. One of our reviewers was incredibly helpful in helping us frame the relative risk confidence interval concepts that I described, which helped us shape the language. I think that the journal bought into that and helped us do it. It wound up being a very collaborative, constructive process. In general, I think as these things go, it was quite positive. There was some negotiation about how we would describe some of our conclusions, but overall, it was a good experience.

Tags

#pctGR, @Collaboratory1

Grand Rounds February 9, 2024: Pragmatic Recruitment of Underrepresented Groups – Experience From the Diuretic Comparison Project (Cynthia Hau, MPH)

Speaker

Cynthia Hau, MPH
Statistician
Boston Cooperative Studies Program Coordinating Center
VA Boston Healthcare System

Keywords

Diuretic Comparison Project, VA Health System, Patient Recruitment, VA Point of Care Program

Key Points

  • Diuretic Comparison Project (DCP) was one of the first full-scale studies for the VA Point of Care Program. It had a pragmatic recruitment model with an embedded design, multicenter study without local study investigators and management teams, and broad recruitment that included patients from all 50 states and Puerto Rico.
  • DCP started to understand if chlorthalidone would be more effective at preventing cardiovascular (CV) outcomes compared to hydrochlorothiazide for thiazide-type diuretic. Both drugs have a well-established safety profile and would be a good fit for a pragmatic design. At the time, more than 95% of VA patients received hydrochlorothiazide for thiazide-type diuretic.
  • The VA IRB determined DCP was a minimal risk study, with less restrictive eligibility criteria and EHR-based safety and outcome monitoring. The study started in June 2016 and ended in June 2022 and ultimately found no difference between the 2 drugs in preventing major CV outcomes and non-cancer death among hypertensive VA patients.
  • DCP aimed to recruit 13,500 patients across the U.S. DCP developed an EHR-based workflows and integrated workflows within the local VA primary care setting.
  • DCP planned to launch 5 sites in the first year starting with the Boston Healthcare System. DCP anticipated other customizations when expanding to regional settings. The study randomized 13,523 patients across the U.S. from 72 VA healthcare systems. Among study locations were sites in all 50 states including Alaska, Hawaii and Puerto Rico. 45% participants were from rural areas. 55% from urban residential areas.
  • One key success factor was having a flexible EHR. DCP developed 3 electronic workflows with 6 major configurations including applications needed outside of the EHR system. DCP developed a patient tracking tool that nurses could use to make sure patients got through each step.
  • Other key success factors were the creative thinking across the study team and the excellent study coordination and partnership within the health system.
  • DCP learned that leveraging health systems for large-scale clinical trials is feasible. Key elements for successful implementation and execution were having a flexible EHR with systemwide collaboration, a supporting community and excellent study coordination, and aligned incentives.

Learn more

Read more in JAMA Network

Discussion Themes

-One real takeaway is the incredible amount of preparatory work that was required before a participant was randomized. How long did the start up for the trial take? The preparation work started with an idea in 2010, which led to selection of the pilot study in 2014 to test out the system for randomizing patients within the existing infrastructure. The actual startup was about 18 months. There was a lot of work that went into startup, making sure incentives were aligned across the VA system, meetings with primary care, cardiology, the VA ethics office, and ultimately with the central IRB. During that time the team was developing the best-case EHR modifications but quickly learned those were not the best modifications, so the first year was iterative where we were testing the forms as we were going. There was a lot of learning on our part for how to do this, and once we figured that out, we were more successful in our randomization rate.

Regarding the critical importance of the health system as one of the take homes from your talk, do you encounter heterogeneity within the VA health system for “buy in” to participate in the trial or do all VA health systems cooperate equally? When people who are external to the VA view it as one homogeneous system, that is not the case. We did the study at 72 medical centers and encountered 72 different themes within the EHR. Centers did not engage in the same way or even the same way as their nearest neighbor. There were different reasons for not participating and it was interesting to see what was a main driver of participating. When people understood what we were trying to do, embed research into clinical care, people really engaged and a few regional directors made participation one of the drivers of research in their region and helped us get initial groundwork.

Tags

#pctGR, @Collaboratory1

Grand Rounds February 2, 2024: Strategies for Improving Public Understanding of FDA and the Products It Regulates…Why Should We Care, and What Might We Do? (Susan C. Winckler, RPh, Esq)

Speaker

Susan C. Winckler, RPh, Esq
Chief Executive Officer
Reagan-Udall Foundation for the FDA

Keywords

U.S. Food and Drug Administration, FDA, Misinformation, Communication, Health Information

Key Points

  • U.S. Food and Drug Administration (FDA) Commissioner Robert Califf asked the Reagan-Udall Foundation to conduct research and consult with experts to better understand how consumers find, consume, and perceive health information, especially regarding FDA-regulated products.
  • From January to September 2023, the Foundation conducted in-depth research and held 5 listening sessions to learn how people consume and understand health information. They held roundtable conversations with experts to understand their thoughts on what the FDA might do to better communicate, and they conducted polls and held 25 individual interviews. The research yielded a report that provides 5 observations, 16 potential strategies, and more than 40 potential tactics for the FDA to consider.
  • This work is important because the digital health information environment and limited public trust in government institutions represent pressing challenges for FDA. The spread of misinformation has accelerated, in part because more people than ever are accessing health information on the internet and via social media. The growing problem of misinformation undermines confidence in science and public health institutions, and the misuse of products can cause harm and confusion.
  • The primary finding is that clear consistent communication both to consumers directly and via medical channels is critical to the FDA’s mission to protect and promote public health. Consumers trust policy – and the scientific evidence on which it is based – if communicated to them properly. Sound science, sound policy, and sound communication are each fundamental to the Agency’s success.
  • Observation 1: There is a lack of understanding, particularly among consumers, about the FDA’s mission, responsibilities, and authority. Potential strategies: Increase direct-to-consumer education, emphasizing the scientific rigor behind, and the reasons for, FDA’s regulatory processes; collect, analyze, and use information about consumer use of FDA’s website; increase interaction with the media.
  • Observation 2: Information vacuums breed misunderstanding. Potential strategies: “Prebunk” health misinformation; collaborate with other organizations (within and beyond government); monitor and respond to misunderstandings.
  • Observation 3: Communication clarity, and consistency, matters. Potential strategies: Adopt a consistent, fact-based tone; use concise, digestible, plain-language approaches; use storytelling and personal narratives.
  • Observation 4: Healthcare professionals are trusted messengers, but they need information. Potential strategy: Increase regular communication with health care professionals.
  • Observation 5: Consistent, multi-channel messages resonate with consumers. Potential strategies: Collaborate with many partners (across and outside government); leverage communications expertise across FDA; communicate creatively; align multi-channel source material before announcements.
  • The FDA has an opportunity to take a more proactive approach to anticipating, listening for, spotting and defusing health misconceptions before they escalate. The FDA can leverage more communications channels, messengers, and mediums to reach the public. To build trust, the FDA must approach communications with an emphasis on clarity and humility.

 

Discussion Themes

-How can the FDA communicate the fact that the best advice tomorrow may be different from the best advice today? How do people want to hear that science is a moving target? In all of our conversations, someone would say, in an effort to be confident, we did that wrong in the pandemic. Some tactics are routinely saying this is based on our best understanding of the information we have today, to convey that our understanding of science changes.

Were you able to examine responses from groups that can be particularly challenging such as vaccine deniers and climate change skeptics? Will the strategies pierce the veil of misinformation with these groups? The government is not the communicator to reach these group, so we asked if there are opportunities for trusted groups or messengers to reach these audiences.

Did you consider examples of organizations in government and the private sector that may do a particularly good job of this? A current project at the Foundation is how can we take FDA information and power places where consumers look for information, similar to how a Weather app is powered by government information from NOAA.

 

Tags

#pctGR, @Collaboratory1

Grand Rounds January 26, 2024: Advancing the Safe, Effective and Equitable Use of AI in Healthcare (Mark Sendak, MD, MPP; Suresh Balu, MD, MBA)

Speakers

Mark Sendak, MD, MPP
Population Health & Data Science lead
Duke Institute for Health Innovation (DIHI)

Suresh Balu, MD, MBA
Director, Duke Institute for Health Innovation (DIHI)
Associate Dean, Innovation and Partnership
Duke School of Medicine

Keywords

AI, ML, health equity

Key Points

  • The Duke Institute for Health Innovation’s (DIHI) mission is to catalyze transformative innovation in health and healthcare through high-impact research, leadership development and workforce training and the cultivation of a community of entrepreneurship.
  • DIHI approaches this work through four pillars of innovation: implementation and health delivery science, health technology innovation, leadership and workforce development, and best practices development and dissemination.
  • In 2021, DIHI started a Health AI Partnership to empower healthcare professionals to use AI effectively, safely, and equitably through community-informed up-to-date standards. We have continued to see the digital divide, where a small number of teams have expertise. Through this partnership, we are trying to build those skills in low resource settings and build a community of practice.
  • The Health AI Partnership started with 7 organization partners and has expanded to about 20 organizations. The main deliverables for the first phase of the project were developing standard key decision points for the AI product lifecycle and developing the Health Equity Across the AI Lifecycle (HEAAL) Framework.
  • There are 8 key decision points in the AI product lifecycle: identify and prioritize a problem; evaluate AI as a viable component of the solution; develop measures of outcomes and success of the AI product; design a new optimal workflow to facilitate integration; evaluate pre-integration safety and effectiveness of the AI product; execute change management, workflow integration, and scaling strategy; monitor and maintain the AI product; and update or decommission the AI product.
  • The first key decision point is procurement, which begins with identifying a problem and ends with allocation of resources to either build or buy an AI product or solution.
  • Key decision point 1 is procurement. Frontline staff have to have buy-in to submit a proposal so organizational resources can be used to sustain the innovation. Align frontline staff and organizational leaders – create alignment throughout project selection.
  • The second key decision point is development and adaptation, which is either building or adapting an external solution for internal clinical use. The first step is to develop measures of success, the second step is designing the workflow, and the third step is evaluating pre-integration safety and effectiveness before a solution is put into clinical use.
  • The next key decision point is clinical integration. DIHI built a modular infrastructure to support many projects via a flexible data pipeline technology infrastructure that started with one model and now includes dozens of models. The final key decision point is lifecycle management, which includes monitoring and maintaining the AI product and updating or decommissioning the product.
  • The Health AI Partnership held a workshop that examined how to assess the potential future impact of a new AI solution on health inequities. The discussion resulted in five assessment domains to be evaluated across the span of the AI adoption process: accountability, fairness, fitness for purpose, reliability and validity, and transparency.

Learn more

Health AI Partnership

Duke Institute for Health Innovation

Discussion Themes

-Can you talk about how DIHI/Health AI Partnership go about with the challenges in medical AI adoption outside of academic medical centers? There is a massive inequity and teams like DIHI are very rare. The next phase of work is building out a practice network in building AI capabilities. It will be a massive undertaking. We have a small philanthropic gift to get started. There is a need for support of infrastructure to bring AI partnerships to low resource settings.

How do you get reliable information and data for care that is received outside of Duke? How do you mitigate bias because of incomplete data? For the CKD example, we combined data from Duke and claims data.

 

Tags

#pctGR, @Collaboratory1

Grand Rounds January 19, 2024: Why Are Imaging RCTs Different? Lessons From Chest Pain Evaluation Trials (Pamela S. Douglas, MD, MACC, FASE, FAHA)

Speaker

Pamela S. Douglas, MD, MACC, FASE, FAHA
Ursula Geller Professor of Research in Cardiovascular Diseases
Duke Clinical Research Institute – Duke University
Past President, American College of Cardiology
Past President, American Society of Echocardiography

Keywords

Chest Pain, Cardiovascular Imaging, Coronary Artery Disease

Key Points

  • Imaging has transformed cardiology and many other fields. In 2024, despite several large randomized controlled trials (RCTs) comparing different evaluation approaches there is no universal consensus on initial imaging strategies, who to test and how; there are ongoing concerns about over imaging lowest risk patients but there is no consensus on testing deferral pathways; and new imaging technologies may offer value but are untested.
  • Similar to most types of trials, there are pros and cons to both pragmatic and explanatory designs in imaging trials. Feasibility and generalizability affect design choices including inclusion/exclusion criteria, flexibility of imaging intervention being tested, guidance/control of subsequent care after imaging, and endpoints and outcomes.
  • People with angina-like symptoms are often not patients with a disease. Most do not have obstructive coronary artery disease (CAD), but a few are very high risk. There is a potential for over testing with significant false positive rate and potential for missed diagnosis with stress imaging.
  • RCT design will vary depending on which CAD manifestation(s) are reflected in the information provided by the imaging test being studied, which affects the treatment target(s) being evaluated in a therapeutic trial.
  • There are many imaging findings that are important for optimal care. There is no single CAD phenotype that can be targeted diagnostically or therapeutically. Imaging strategies must be multidimensional or account for this heterogeneity.
  • There are considerations for the flexibility of the intervention. When we evaluate imaging strategies for chest pain, is “usual testing” the appropriate comparator? Coronary CT angiography (CTA) may be the preferred test. Researchers also need to ask what is the optimal CTA intervention and should downstream care be required?
  • What are appropriate endpoints for imaging RCTs and what events are we trying to avoid? Angina may not effectively discriminate between strategies. Intermediate endpoints such as diagnostic and therapeutic thinking are useful with impact on treatment being a major determinant of long-term value. Process of care/efficiency measures are important. Costs are rarely a significant factor in comparing different testing approaches.

 

Discussion Themes

-Why aren’t payors more willing to fund these trials? I have not seen a payor fund a trial in this space. There has only been a hand full of trials. They are trying to follow the guidelines for symptoms and disease. The tests are so common in aggregate it is a big expense.

What is the sensitivity of portable AI ECHO devices? They are pretty good. We think about it as a screening tool to see if a patient needs an intervention.

 

Tags

#pctGR, @Collaboratory1