Interview Questions with PMWC 2025 Luminary Honoree: Ida Sim, UCSF

1.     Your work with JupyterHealth is pioneering the secure collection and computation of health data from various sources. Could you share how JupyterHealth is currently being implemented in real-world healthcare settings, and what impact you anticipate it having on patient outcomes?

Answer:

JupyterHealth is currently in early development with use cases drawn from real-world primary and subspecialty care from UCSF and Weill Cornell. Our participatory design process has included frontline clinicians and lay users of wearable sensors.

JupyterHealth is a technology platform that will accelerate the process of sensor data acquisition, digital biomarker development, and integration of digital health solutions into clinical workflows. JupyterHealth is disease agnostic. As such, we anticipate broad impact on patient outcomes across multiple disease and prevention domains. For example, we anticipate JupyterHealth accelerating companies that draw on multiple sensor streams and electronic health record data to generate individualized metabolic profiles and personalized decision support for patients to optimize their own metabolic health. While such companies currently exist, lowering the technical barrier to entry and the backend operating costs while creating an open community for sharing and co-creating new computational approaches should result in more robust digital health markets that will then impact patient outcomes.

2.     As Co-Founder of Vivli, you’ve revolutionized data sharing for clinical trials. How do you see the role of large-scale data sharing evolving in the next five years, particularly in the context of integrating data from diverse sources like sensors and medical records?

Answer:

Vivli has been an inspiring success story for how understanding of technology, research culture, and industry priorities can be harnessed together to create a new way forward towards open science. With the NIH’s data sharing mandate (in which funded studies must make their data available for reuse), scientific data sharing will increase in scale and scope in the coming years. Major expansions to sharing of imaging and genomics data are likely.

Medical records– in the form of the US Core Data for Interoperability (USCDI) dataset – can already be accessed for individual patients and in bulk through federally mandated APIs. JupyterHealth will further unleash the ability to securely access sensor data from individuals, integrated with USCDI access. The value of data sharing comes only if the shared data are used, however. It is not enough to bring standardized data together. Fostering large-scale data reuse requires scalable analytic platforms that not only bring sensor and medical record data together but will connect them to reusable analytic pipelines that can easily be packaged into implementable solutions. Building on the proven success of Project Jupyter, JupyterHealth uniquely addresses both large-scale data acquisition and data access and large-scale data use and reuse in digital health. Like Vivli, JupyterHealth reflects a deep understanding of enabling technology, digital health scientific needs, and industry dynamics to build a new pathway for digital health impact.

3.     With your dual role in academic research and clinical practice, how do you envision the integration of advanced computational tools into everyday clinical workflows, and what are the biggest challenges you foresee in achieving this integration?

Answer:

Currently, clinical workflows are anchored in the electronic health record (EHR) system. Integration of computational tools into clinical workflows are therefore constrained by EHRs, which often limit the ability of 3^rd party computational tools to access the care context (what task is the clinician trying to complete? What are the latest relevant test results?) or to write back into the official medical record (which limits the utility and potentially billing of services provided).

FHIR and SMART-on-FHIR standards are important mechanisms for enabling greater workflow integration but are not by themselves sufficient to ensure large-scale effective integration into everyday care.

Moreover, the heavy lift for integrating advanced computational tools is far more than a technical problem. Workflow re-engineering requires organizational change capabilities that ideally include the ability to rapidly prototype and adapt tools to local practice styles. JupyterHealth’s support for iterative “dev-ops” aims to facilitate such rapid prototyping.

Lastly, with recent AI advances, health systems need to establish governance processes to ensure appropriate AI usage and to conduct ongoing monitoring for performance bias and drift. These processes require clinical AI expertise, which is not yet widely available.

Interview Questions with Victor E. Velculescu

 

1. Integration of Liquid Biopsies:

“Given the rapid advancements in liquid biopsy technologies, including your development of DELFI, how do you envision their integration into routine clinical practice? Considering the current limitations, what specific innovations do you believe are necessary to overcome these hurdles, and how do you see the role of interdisciplinary collaboration in achieving these breakthroughs?”

Until recently, liquid biopsies were considered expensive, logistically complex, and used a needle-in-a-haystack approach to search for specific genetic sequence changes or methylation patterns.  We have developed a new AI liquid biopsy approach based on genome-wide fragmentomics, the discovery that cancer cells are more chaotic than normal cells and, when they die, leave behind tell-tale patterns and characteristics of cell-free DNA (cfDNA) fragments in the blood.  This new machine-learning fragmentome approach called DNA EvaLuation of Fragments for early Interception (DELFI) provides a sensitive and accessible next-generation liquid biopsy approach for early cancer detection.   As described in our recent study (Cancer Discovery, 2024) this approach has been clinically validated as a liquid biopsy test for lung cancer, a cancer type that we know benefits from early detection and yet less than six percent of eligible individuals are currently getting screened.  To be successful clinically, any new liquid biopsy test providers will have to work closely with all members of health system teams, including primary care, pulmonology, oncology, radiology, phlebotomy, and nursing staff.  I am optimistic that a new generation of AI liquid biopsy approaches that are affordable and accessible will revolutionize early cancer detection in the US and around the world. 

2. Translating Genomic Discoveries into Therapies:

“Your research has unveiled novel genes involved in neoplasia, such as PIK3CA. What are the most significant obstacles in translating these genomic insights into actionable therapeutic targets? How do you propose we address the biological complexity and variability among patients to ensure these therapies are both effective and personalized?”

In the mid 2000’s we performed the first genome-wide sequence analysis of human cancers.  These analyses revealed a variety of driver genes and pathways not previously known to be involved in tumorigenesis.  These discoveries included PIK3CA, one of the most commonly altered genes in cancer, among others, and revealed the complex landscape of genomic changes in individual tumors.  These observations paved the way for personalized approaches to cancer patients, including new FDA approved therapies against PI3K and IDH1, and FDA approved diagnostic tests for comprehensive tumor profiling. Although challenges remain in developing effective therapies against cancer, understanding the genomic landscape of these diseases for every patient marks an important step in matching each patient with the right therapy.  I expect over time, a combination of new targeted and immune therapies, tailored to an individual’s genomic alterations will be essential to providing effective personalized therapies. 

 

3. Cost-Effectiveness and Overdiagnosis in Liquid Biopsies:

“Precision medicine based on liquid biopsies holds immense promise, but there are concerns about cost-effectiveness and potential for overdiagnosis. In the context of global healthcare disparities, how do you propose making liquid biopsies accessible and reliable across different economic settings? What measures can be taken to mitigate the risk of overdiagnosis while maximizing early detection benefits?”

We set out some time ago to develop a liquid biopsy approach that would be high performing and widely accessible.  The effort led to developing an AI fragmentome platform which is inherently lower cost than first-generation liquid biopsy approaches.  This method uses simple lab processes and low-coverage whole genome sequencing and has lower costs than conventional liquid biopsy methods which historically have required complex chemistries and ultra-deep sequencing.  The fragmentome approach results in millions of genome-wide data points that are captured in a cfDNA fragmentation profile which when combined with AI analyses provide highly sensitive algorithms for distinguishing those with cancer from those without.  These low laboratory costs combined with high performance make this approach well suited for broad public health efforts like cancer screening in the US and globally. 

One way we can avoid overdiagnosis in cancer screening is to focus on detecting cancer types where we know screening saves lives.  In these settings, the diagnostic pathways are well established and the problem is rather that not enough people are getting screened.  As an example, we chose to focus on early detection of lung cancer as this is the greatest cancer killer in the US.  Unfortunately less than 6% of eligible individuals at risk of this disease are screened using low-dose computed tomography (CT) imaging.  The clinically validated fragmentome blood test for lung cancer will help more at risk individuals get screened.  As the next step after a positive blood test is a low-dose CT scan that is recommended in this population, and there is no risk for overdiagnosis.  We estimate that even a modest uptake of this approach will save thousands of lives in the US over the next five years. 

4. Leveraging Transcriptomics for Personalized Treatment:

“As a pioneer in transcriptomics, you’ve shaped our understanding of cancer at the molecular level. Given the rapid pace of advancements in this field, what are the next big leaps you anticipate in transcriptomic research? How can we leverage emerging technologies, such as single-cell sequencing and spatial transcriptomics, to further refine personalized treatment regimens and improve patient outcomes?”

One of the completely unanticipated connections that have emerged recently has been the intersection between the “transcriptome”, or the compendium of transcripts representing the expressed genes in a cell, and the detection of these changes through alterations in cell-free DNA (cfDNA).  It turns out that the genome-wide cfDNA “fragmentome”, which represents the compendium of changes in cfDNA fragments, can capture changes that are occurring as a consequence of gene expression.  These include alterations in cfDNA fragmentation at sites of transcription factors as well as at regions of other epigenetic changes such as methylation, histone modification, or larger scale chromatin changes.  I believe that in the future, this connection between the transcriptome and fragmentome may provide new noninvasive ways to identify personalized therapies and mechanisms of resistance for cancer patients and improve outcomes. 

 

5. Broader Applicability of DELFI Technology:

“While DELFI shows promise for lung cancer screening, its potential application for other cancers, especially those lacking established screening methods, remains a critical area of interest. What are the unique challenges you face in adapting DELFI for these cancers, and how do you foresee integrating AI and machine learning to enhance its diagnostic accuracy and broaden its applicability?”

The DELFI approach was built to solve population health problems.  The application of AI to genome-wide cell-free DNA (cfDNA) fragmentation profiles has provided a cost-efficient platform that is ideally suited to cancer detection in various applications.  Our platform uses millions of data points to sensitively identify individuals who may have cancer, including early-stage disease, and the cancer’s tissue of origin. Beyond our clinically validated test for detection of lung cancer, other applications of the approach include early detection of liver cancer (published in Cancer Discovery in 2023), detection of other cancer types (as described in our original Nature study in 2019 and in other studies currently underway), and monitoring therapeutic response or resistance of patients with cancer to targeted and immunotherapy.   Most recently, at the AACR 2024 meeting this year, I presented findings from a study using fragmentomics for early detection of ovarian cancer, the fifth leading cause of cancer death in the U.S.  This is an incredibly deadly disease with no great biomarkers for screening and early intervention.  Our approach showed high performance for the detection of ovarian cancer and demonstrates that the DELFI platform has broad opportunities for the detection of a variety of human cancers.