___________________

¹ This workshop was organized by an independent planning committee whose role was limited to identification of topics and speakers. This Proceedings of a Workshop was prepared by the rapporteurs as a factual summary of the presentations and discussions that took place at the workshop. Statements, recommendations, and opinions expressed are those of individual presenters and participants and are not endorsed or verified by the National Academies of Sciences, Engineering, and Medicine, and they should not be construed as reflecting any group consensus.

Through Enhanced Registries and Beyond on July 29 and 30, 2024. Lisa Richardson, director of the Division of Cancer Prevention and Control at the Centers for Disease Control and Prevention (CDC) and co-chair of the workshop planning committee, introduced the workshop as an opportunity to explore opportunities to enhance and modernize cancer surveillance to improve cancer research, care, and outcomes for all patients. Robin Yabroff, scientific vice president of Health Services Research at the American Cancer Society and workshop co-chair, said discussions would focus on opportunities to improve the timeliness, quality, and completeness of cancer registry data.

This Proceedings of a Workshop summarizes the workshop presentations and discussions. Observations and suggestions from individual participants are discussed throughout the proceedings, and highlights are presented in Boxes 1 and 2. Appendixes A and B include the Statement of Task and workshop agenda, respectively. Speaker presentations and the workshop webcast have been archived online.²

PERSPECTIVES ON THE CURRENT STATE OF CANCER SURVEILLANCE

Patient Perspectives

Kelly Shanahan, director of research at Metavivor, a physician, and a person living with metastatic cancer, highlighted the importance of cancer surveillance for improving patient care and the need for cancer registry databases to be representative of all patients in the United States. When Shanahan was diagnosed with Stage 2 breast cancer in 2008, she chose bilateral mastectomy over lumpectomy and weeks of daily radiation therapy because the closest oncology services were a 45-minute drive from her home. She also continued working at her obstetrics/gynecology practice during chemotherapy. In 2013, she experienced severe back pain but dismissed it as a pulled muscle or herniated disk. However, several months later, scans revealed widespread bone metastases and several fractures.

Shanahan said people living with metastatic cancer often feel they are not captured in national databases that provide information on cancer statistics that drive efforts to reduce the cancer burden. For example, she explained that the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute (NCI) collects cancer incidence and survival data

___________________

² https://www.nationalacademies.org/event/41759_01-2024_enabling-21st-century-applications-for-cancer-surveillance-through-enhanced-registries-and-beyond-a-workshop (accessed October 8, 2024).

___________________

³ See https://www.who.int/standards/classifications/classification-of-diseases (accessed October 10, 2024).

___________________

⁴ See https://seer.cancer.gov (accessed October 11, 2024).

registries covering 48 percent of the U.S. population (see Figure 1) (Penberthy and Friedman, 2024). Penberthy noted that registries submit de-identified data to the SEER Data Management System,⁵ a central database that can help with facilitating data linkages. In addition to collecting data from the patient’s electronic health record (EHR), SEER taps up to 35 other data sources to inform the patient’s case record. These include medical claims data; pharmacy data; clinical trial information; pathology, imaging, genomic, and genetic results reports from specialty services; and nonmedical data sources, such as motor vehicle administration, veteran, and Social Security information.

Penberthy described some challenges experienced by population-based registries, such as SEER and the CDC National Program of Cancer Registries (NPCR). These challenges include the rapid advancement and increasing complexity of how cancer is diagnosed and treated, the lack of data outside of clinical trials on the dissemination and use of new cancer diagnostics and treatments and their impact on patient outcomes, and the lack of access to the full patient health record across multiple potential cancer data sources. She noted that treatment guidelines are frequently based on data from clinical trials, which include only about 5 percent of the population of adult patients with cancer (Unger et al., 2016). She added that these limited, often non-representative, datasets impact the generalizability of treatment guidelines to real-world use. She also emphasized the need for rapid and efficient collection of population-level data.

Penberthy said that a strength of cancer registries is their general ability to adapt as clinical care evolves. For example, SEER is expanding to include genomic and genetic data and information on treatments and social determinants of health. However, the manual collection of data remains a limitation of registries in many cases, and accessioning and reporting can be slow, with complete and finalized registry data running 2–3 years behind diagnoses, she said.

Penberthy agreed with Shanahan that population-level data on recurrent metastatic cancer are lacking. She explained that surveillance of metastatic disease is challenging because diagnoses of recurrence are made across a wide range of care specialties, clinical settings, and diagnostic modalities. Overcoming these challenges will require a multimodal approach, Penberthy said. She suggested that one potential near-term practical solution is a SEER Program collaboration with the Department of Energy (DOE) to develop the Modeling Outcomes using Surveillance Data and Scalable Artificial Intelligence for Cancer (MOSSAIC) project,⁶ an application programming interface (API) that can capture data on metastatic recurrence from pathology reports (Hsu et al., 2024). Penberthy added that one potential policy approach to improving

___________________

⁵ See https://seer.cancer.gov/seerdms (accessed October 11, 2024).

⁶ See https://computational.cancer.gov/about/mossaic (accessed October 20, 2024).

___________________

⁷ See https://www.congress.gov/bill/102nd-congress/senate-bill/3312 (accessed October 13, 2024).

⁸ See https://www.congress.gov/bill/102nd-congress/house-bill/4206 (accessed October 13, 2024).

⁹ HL7 is a set of global standards for the electronic exchange of health information. See https://www.hl7.org (accessed October 8, 2024).

¹⁰ See https://www.aphl.org/programs/informatics/pages/aims_platform.aspx (accessed October 11, 2024).

for cancer data reporting. Challenges for real-time data reporting via a cloud platform, Benard said, are that pathology data are often unstructured and in narrative text. Because additional patient information will also be needed, she said, CDC is pilot testing an infrastructure for leveraging EHR data.

CDC’s vision for cancer surveillance is a cloud-based infrastructure¹¹ to streamline the reporting of registry data to a single portal, Benard said. To this end, she said CDC has developed tools to automate processes and a testing site for cloud infrastructure. She added that CDC has worked with the state- and territory-level central cancer registries on data governance and is continuing to work with partners to develop a minimum dataset, additional tools, and pilot projects to ensure the availability of high-quality, timely cancer surveillance data.

The National Cancer Database

The National Cancer Database (NCDB)¹² is a nonfederal joint program of the Commission on Cancer (CoC)¹³ and the American Cancer Society that was established in 1989. Bryan Palis, senior manager and senior statistician at the American College of Surgeons, said that the NCDB drives the development and implementation of evidence-based quality-of-care metrics that are established standards of best practice based on demonstration of survival benefit. Palis described the NCDB as comprehensive, accruing 1.5 million case records per year from more than 1,400 hospitals; detailed, capturing a wide range of data for more than 70 disease sites and covering about 74 percent of U.S. cases; high quality, as accreditation standards for cancer programs from the CoC drive data accuracy; and contemporary, as cancer cases are submitted at least monthly to the Rapid Cancer Reporting System (RCRS).¹⁴

“High-quality cancer registry data are essential to accurately assess treatment and outcomes,” Palis said. He explained that standardized abstraction

___________________

¹¹ See https://www.cdc.gov/national-program-cancer-registries/data-modernization/cloud-based-computing.html (accessed October 13, 2024).

¹² See https://www.facs.org/quality-programs/cancer-programs/national-cancer-database/ (accessed October 13, 2024).

¹³ The American College of Surgeons Commission on Cancer (CoC) is a group of professional organizations dedicated to improving outcomes for persons with cancer through the monitoring of comprehensive quality care as well as promoting cancer prevention, research, and education. See https://www.facs.org/quality-programs/cancer-programs/commission-on-cancer (accessed October 15, 2024).

¹⁴ See https://www.facs.org/quality-programs/cancer-programs/national-cancer-database/rapid-cancer-reporting-system (accessed October 15, 2024).

processes outlined in the Standards for Oncology Registry Entry manual¹⁵ and staging system developed and compiled by the American Joint Committee on Cancer (AJCC)¹⁶ help ensure the quality and consistency of the data submitted to NCDB and used in CoC reports. He noted that more than 90 percent of CoC-accredited programs are compliant with CoC requirements for cancer registry data quality control and timeliness of data submission. A recent evaluation of NCDB quality control procedures found NCDB to have “a high level of case completeness, comparability with uniform standards for data collection, timely data submission, and high rates of compliance with validity standards for registry and data quality evaluation,” Palis said (Palis et al., 2024). NCDB is working to facilitate real-time quality reporting, including its 2020 implementation of the RCRS.¹⁷ One of the challenges NCDB is working to address is the limited data on systemic therapy, recurrence, and progression. Although these data might be captured, he said, they are not reported by NCDB due to data missingness and coding variability. Palis suggested data collection might be improved by using synoptic checklists that incorporate a continuum of cancer outcomes (e.g., no evidence of disease, recurrence, and progression).

Lawrence Shulman, professor of medicine at the University of Pennsylvania Perelman School of Medicine, provided examples of how NCDB data are used, including for benchmarking. He described the use of NCDB data on Stage 3 breast cancer survival as a metric of quality of care. The analysis found that survival was better at NCI-designated comprehensive cancer centers versus large or small community hospitals (Shulman et al., 2018). Shulman said, however, that the data are not sufficiently detailed to determine why this is the case and therefore are of limited use for quality improvement efforts.

Shulman also shared examples of the Cancer Quality Improvement Report¹⁸ for the Hospital of the University of Pennsylvania. The first example compared 30- and 90-day mortality for a particular complex cancer surgery at the University of Pennsylvania versus all CoC programs and high-volume CoC programs, identifying areas for improvement. The second example benchmarked hospital compliance with the quality measure requiring pathology analysis of at least 12 regional lymph nodes for primary colon cancer surgery;

___________________

¹⁵ See https://www.facs.org/media/vssjur3j/store_manual_2022.pdf (accessed October 8, 2024).

¹⁶ See https://www.facs.org/quality-programs/cancer-programs/american-joint-committee-on-cancer/cancer-staging-systems/cancer-staging-system-products (accessed October 8, 2024).

¹⁷ See https://www.facs.org/quality-programs/cancer-programs/national-cancer-database/rapid-cancer-reporting-system (accessed October 13, 2024).

¹⁸ See https://www.facs.org/quality-programs/cancer-programs/national-cancer-database/qualitytools (accessed October 15, 2024).

Shulman shared data showing that increased compliance improved survival outcomes across hospital types, and that sharing data with cancer programs was associated with improved compliance over time (Shulman et al., 2019).

CoC-accredited programs can also apply to use NCDB for research, and Shulman noted about 1,300 applications per year. Research use of NCDB has resulted in more than 4,500 publications, and Shulman shared an example that described differences in care for early-onset colorectal cancer that can inform cancer surveillance (Nogueira et al., 2024).

North American Association of Central Cancer Registries

NCI began publishing cancer mortality atlases in the 1970s, which Betsy Kohler, executive director of the NAACCR,^19,20 said inspired an increase in cancer registries across the country as states sought to understand their cancer death rates. The mission of NAACCR, a nonprofit, volunteer organization representing the United States and Canada, includes education and training for registries, registry certification, data publication, and promotion of data use for research and public health purposes (NAACCR, 2024). Its certification involves independent assessment of central registry data quality based on objective and measurable criteria for data completeness, timeliness, and quality of reporting (White et al., 2017).

Kohler said that NAACCR also works collaboratively with partners including NCI, CDC, CoC, and the National Cancer Registrars Association (NCRA) to develop cancer surveillance standards. She explained further that within the collaboration, a mid-level tactical group (MLTG)²¹ focuses on change management and implementation of data standards, while a high-level strategic group (HLSG)²² focuses on collaborative decision making and interagency strategic planning. Standards are updated annually. Kohler said the process of introducing a new data item takes about 18 months and is “incredibly complex” (see Figure 2).

Kohler said population-based cancer registries experience ongoing challenges, including the complexity of collating individual patient data from

___________________

¹⁹ See https://www.naaccr.org (accessed October 13, 2024).

²⁰ Betsy Kohler was the executive director of the NAACCR at the time of the workshop. At the time of publication of the proceedings the director is Karen Knight. See https://www.naaccr.org/board-resolutions/#1727105112085-6447aff5-03f1 (accessed December 16, 2024) and https://www.naaccr.org/naaccr-board (accessed December 16, 2024).

²¹ See https://my.naaccr.org/committee-details?cid=a0O5e000002k6msEAA (accessed October 15, 2024).

²² See https://my.naaccr.org/committee-details?cid=a0O5e000002k6mqEAA (accessed October 15, 2024).

___________________

²³ See https://www.who.int/standards/classifications/other-classifications/international-classification-of-diseases-for-oncology (accessed October 24, 2024).

²⁴ https://datascience.cancer.gov/collaborations/nci-department-energy-collaboration (accessed October 24, 2024).

that the current workforce of registrars is likely not sufficient to do this. Benard agreed that such an approach would require funding and resources that are not currently available. Penberthy and Shulman also added that not all recurrences are diagnosed in a hospital setting and that the data come from a wide range of sources. Thus, a holistic approach to data collection is needed. Kohler noted that data captured from multiple sources for a single patient could be conflicting because not all data are captured at the same time and would need to be resolved. Penberthy said this is an example of why some level of manual review will always be needed, such as to issue judgment on such conflicts. Furthermore, Eric Durbin, director of the Kentucky Cancer Registry and president-elect of NAACCR, suggested that a national patient identifier would help to facilitate data linkages and solve an array of cancer surveillance challenges, including the capture of recurrence data. Richardson agreed and said this has long been discussed, but concerns about privacy and confidentiality are often raised.

Building on Kohler’s point on state policies that restrict data sharing from central registries, Roy Jensen, William R. Jewell, M.D. Distinguished Kansas Masonic Professor of Cancer Research at the University of Kansas, noted that it would be essential to have model legislation that balances patient privacy protections with the need for health care professionals and researchers to access registry data. Kohler agreed and said that it is difficult to balance population size, privacy, and data utility, and challenges vary by state. Shulman said that some of the programs that report data to NCDB are very small and noted that there are approaches to develop “cohorts” for analyses that ensure data privacy.

LESSONS LEARNED FROM U.S. AND INTERNATIONAL CANCER SURVEILLANCE EFFORTS

Several speakers representing local, state, and international surveillance efforts and end users of registry data shared their perspectives on potential strategies for improving cancer surveillance, including opportunities for multidisciplinary collaboration.

U.S.-Based Perspectives

A Regional Registry Perspective: Greater Bay Area Cancer Registry

Scarlett Lin Gomez, professor of epidemiology and biostatistics at the University of California, San Francisco (UCSF) and director of the Greater Bay Area Cancer Registry (GBACR) in California, said that the GBACR²⁵ is one of

___________________

²⁵ See https://cancerregistry.ucsf.edu (accessed October 15, 2024).

the nine original registries in the SEER Program (Kohler et al., 2011). After the statewide California Cancer Registry²⁶ was established in 1988, the GBACR was expanded from five to nine counties and is based at UCSF. Gomez shared her perspective on opportunities and challenges for local registries in four areas: data dissemination, data access, data availability and quality, and linkages.

First, data dissemination has evolved from hardcopy reports to interactive online dashboards and maps, which Gomez said provides opportunities for broad dissemination to a wide range of end users. She pointed out that the tools in the California Health Maps²⁷ allow for reporting of data from “more geographically meaningful units,” such as zones comprising similar census tracts. Los Angeles County, for example, is divided into several hundred zones, she said. One challenge, Gomez said, is that “numerator data” are available from the registry, but there is often a lack of timely “denominator data” (e.g., census tract data, detailed population subgroup data such as race/ethnicity) for analysis of cancer trends. Gomez noted that another challenge at the local level is balancing data granularity with data suppression rules for small populations.

Second, there are increasing challenges for registries trying to release data and researchers trying to access data (Iyer et al., 2023). Gomez said that limiting data use slows research progress, and it is unclear whether patients are better protected by restrictive use policies. Gomez said that data access policies vary by state, and she advocated for changes in policies that limit data access.

The third point Gomez raised is data availability and quality challenges. Examples mentioned by Gomez included the high percentage of missing registry data for key elements (e.g., birthplace and Hispanic origin); lack of data on cancer progression and recurrence, health behaviors, and patient sociodemographic information; and variability in data quality and completeness across registries, even within a state (In et al., 2019; Lee and Singh, 2022; Mendez et al., 2024; Montealegre et al., 2014).

Last, Gomez said that data linkages can provide opportunities to fill gaps in data availability. However, linkages are “administratively difficult given [the] need to share patient identifying information,” she added. She suggested there are opportunities to use methods such as privacy protecting record linkage; approaching institutional privacy offices to develop data use agreements with indemnity language as required by states; and engaging patient advocates in conversations about the use of their data.

Despite the challenges, Gomez said that there have been many successes in gleaning new knowledge from cancer registry data (Abrahão et al., 2024; Giannakeas et al., 2024).

___________________

²⁶ See https://www.cdph.ca.gov/Programs/CCDPHP/DCDIC/CDSRB/Pages/About-the-CCR.aspx (accessed October 15, 2024).

²⁷ See https://www.californiahealthmaps.org (accessed October 15, 2024).

___________________

²⁸ See https://www.legis.state.pa.us/WU01/LI/LI/US/PDF/1980/0/0224..PDF (accessed October 15, 2024).

²⁹ See https://www.cdc.gov/national-program-cancer-registries/registry-plus/index.html (accessed October 13, 2024).

³⁰ See https://www.pa.gov/en/agencies/health/diseases-conditions/cancer/pa-bccedp.html (accessed October 15, 2024).

ticularly important for cancers that are late stage at diagnosis and rapidly fatal. Schwartz explained that presumptive cases for the CSS were first identified through screening of daily, weekly, and monthly hospital electronic pathology feeds. He said that trained staff then used a case-finding software system to determine if a patient was a likely registry candidate based on site, histology, age, sex, and residence. A request to complete a NAACCR abstract was then sent to the hospital. After the abstract was received, a new incidence record was accessioned into the CSS registry. While case finding and accessioning each took only a couple of days, receiving the abstract from the hospital could take weeks to months, he said. To address this, it was determined that the core data needed to confirm an incident case were available after case finding, and it could then be accessioned into CSS for use in surveillance and research, while the NAACCR abstract was being completed. As a result, Schwartz said, cases from the institutional electronic pathology feeds can now be available in CSS within a week.

Schwartz also described the effort to access state administrative health data, particularly All-Payer Claims Databases³¹ and hospital discharge databases, to expand the scope of the registry content. Relevant information in these databases include data on diagnostic modalities and treatments, the health of cancer survivors, and payments (which could be used to estimate costs of care). This effort has not been successful, as access to the Washington State All-Payer Claims Database is “substantially limited by state regulation,”³² he said. Access is available only for “circumscribed research projects.” Use for public health surveillance is not permitted, and data may not be rereleased (and therefore could not be reported to or made available via external linkages to NCI or CDC databases). Schwartz added that the state cancer registry also cannot access these data, and lack of access impacts other disease surveillance efforts.

A Registry End-User Perspective: Kaiser Permanente Washington Health Research Institute

The Kaiser Permanente Washington Health Research Institute (KPWHRI)³³ is part of the Health Care Systems Research Network (HCSRN),³⁴ a consortium of research centers at 20 U.S. health care systems, said Jessica Chubak, senior investigator at KPWHRI. “Each research center has a similarly structured data

___________________

³¹ See https://www.ahrq.gov/data/apcd/index.html (accessed October 13, 2024).

³² See https://app.leg.wa.gov/RCW/default.aspx?cite=43.371.020 (accessed December 16, 2024).

³³ See https://www.kpwashingtonresearch.org (accessed October 15, 2024).

³⁴ See https://hcsrn.org (accessed October 15, 2024).

warehouse with many different types of comprehensive health and health care data,” she said. Although the sources of cancer data might be different across health care systems (e.g., internal cancer registry versus external data sources), data in each research center’s HCSRN virtual data warehouse are saved in a common structure. Chubak said Kaiser Permanente Washington collaborates with the Seattle-Puget Sound Registry,³⁵ the SEER registry at the Fred Hutchinson Cancer Center.

Chubak offered two proposals to advance registry-based cancer research from her perspective as an end user of registry data. First, Chubak proposed easing data sharing restrictions, as other speakers had already suggested. She said that linkages between registries and health care systems data are critical for cancer research and that broader access to registry data by health systems could improve research efforts.

Chubak’s second proposal was to require and fund the inclusion of recurrence in cancer registries. She listed several overarching challenges for collecting such data, including distinguishing recurrence from progression; the ability to follow a patient over time and across health systems, payers, and geographies; the accuracy of data collection methods; and cost. She suggested that a multipronged approach to recurrence data collection is needed, with likely no “one-size-fits-all” solution. Funding and regulatory barriers will need to be addressed for any solution, she concluded.

International Perspectives

Cancer Data, Registries, and Surveillance in Canada

Craig Earle, chief executive officer of the Canadian Partnership Against Cancer (CPAC),³⁶ shared a perspective on cancer surveillance in Canada. He noted that health care is under provincial jurisdiction in Canada, and CPAC is a federally funded organization tasked with advancing a pan-Canadian cancer strategy in collaboration with provinces and other partners (CPAC, 2024). One area of focus for CPAC is improving cancer data.

Each of the 13 provincial jurisdictions has a cancer registry that reports data to a central Canadian registry (Statistics Canada, 2024). Earle said cancer staging was not routinely entered into most cancer registries in Canada about two decades ago, so one of CPAC’s initiatives was to introduce staging, through first the Collaborative Stage system and later the AJCC staging sys-

___________________

³⁵ See https://seer.cancer.gov/registries/sps.html (accessed October 15, 2024).

³⁶ See https://www.partnershipagainstcancer.ca (accessed October 16, 2024).

tem.³⁷ Cancer stage is now captured by registries across Canada, but challenges persist. The AJCC standards are frequently updated. The process is “time, training, and resource intensive,” Earle said, and as a result, “many jurisdictions only stage select cancer sites.” To improve capture of cancer stage, CPAC supported two synoptic reporting initiatives, one for collection of pathology and surgical data and another in which communities of practice would use synoptic data for quality improvement (Arnaout et al., 2024). Although synoptic reporting of pathology data has been adopted across Canada, Earle said neither synoptic surgical reporting nor the communities of practice for quality improvement were sustainable.

Cancer data are also needed for cancer system performance reporting, and from 2008 through 2018, CPAC issued annual reports comparing cancer system performance across jurisdictions (CPAC, 2018). However, there were concerns about the resources required within the jurisdictions for this process and varying opinions about the benefit of making such comparisons, Earle said. In 2018, the approach was changed to “monitoring jurisdiction-specific progress towards the priorities of the Canadian Strategy for Cancer Control,”³⁸ but he added that there has been recent interest in again reporting on cancer system performance.

Furthermore, Earle said that the ability to link cancer registries to other data sources in Canada varies across provinces. In Ontario, for example, ICES (formerly the Institute for Clinical Evaluative Sciences)³⁹ has an advanced virtual data warehouse. Similarly, through Statistics Canada,⁴⁰ the Canadian Cancer Registry has been linked with sources of social determinants of health data (e.g., individual tax files, immigration data, and census data).

In 2023, CPAC and the Canadian Cancer Society⁴¹ released the Pan-Canadian Cancer Data Strategy.⁴² Earle said interested parties, including health administrators, researchers and academic institutions, as well as federal, provincial, and territorial policy makers, were engaged in an iterative process

___________________

³⁷ The American Joint Committee on Cancer (AJCC) staging system uses TNM, which is a global system that evaluates cancer through Tumor (e.g., size, spread), lymph Node spread, and Metastases. The Collaborative Stage system was a combination of TNM, SEER Extent of Disease, and SEER Summary Stage data that is no longer in use. See https://www.cancer.gov/publications/dictionaries/cancer-terms/def/ajcc-staging-system and https://seer.cancer.gov/tools/collabstaging (accessed October 13, 2024).

³⁸ See https://www.partnershipagainstcancer.ca/wp-content/uploads/2019/06/Canadian-Strategy-Cancer-Control-2019-2029-EN.pdf (accessed October 13, 2024).

³⁹ See https://www.ices.on.ca/our-organization (accessed October 16, 2024).

⁴⁰ See https://www.statcan.gc.ca/en/start (accessed October 13, 2024).

⁴¹ See https://cancer.ca/en (accessed October 16, 2024).

⁴² See https://dev.partnershipagainstcancer.ca/about-us/corporate-resources-publications/pan-canadian-cancer-data-strategy (accessed October 8, 2024).

to identify data gaps and establish funding priorities, such as improving the efficiency, timeliness, and quality of data capture and access; enhancing linkages to current data; and filling gaps in current data collection.

Cancer Surveillance and Registries in Europe

Europe has 180 population-based cancer registries across 41 countries, covering about two thirds of the population of Europe (Arndt and Siesling, 2024), said Volker Arndt, director of the Baden-Württemberg Cancer Registry and head of the Unit of Cancer Survivorship at the German Cancer Research Center or Deutsches Krebsforschungszentrum (DKFZ).⁴³ “Registration practices differ, and harmonization is a key task of the European Network of Cancer Registries,”⁴⁴ he said, adding that the European Commission has recently committed €13 million to support quality improvement of European cancer registry data. Arndt shared his perspective on cancer surveillance and registries based on his experience with cancer registration in Germany.

Arndt said that prior to the Federal Cancer Registry Act in 1995,⁴⁵ Germany had only a few cancer registries (Robert Koch Institute, 2016). By 2009, all states had registries (Robert Koch Institute, 2019). Additional legislation followed in 2009, addressing the collection and sharing of harmonized data at the federal state-level;⁴⁶ in 2013, enabling quality assurance through registries;⁴⁷ in 2021, addressing access to cancer registry data;⁴⁸ in 2022 a call was launched to leverage AI to enhance the use of registry data;⁴⁹ and, in 2024, addressing linkage of cancer registry and other data for research use.⁵⁰ Arndt said that Germany now has 15 federal state-level cancer registries and one nationwide childhood cancer registry. “All data are pooled at the national level,” Arndt added (Katalinic et al., 2023).

___________________

⁴³ See https://www.dkfz.de/en/krebsregister/index.html (accessed October 16, 2024).

⁴⁴ See https://encr.eu (accessed December 16, 2024).

⁴⁵ The Federal Cancer Registry Act was passed in Germany parliament in November 1994 and was in effect from 1994 to 1999. See http://www.bgbl.de/xaver/bgbl/start.xav?startbk=Bundesanzeiger_BGBl&jumpTo=bgbl194s3351.pdf (accessed November 23, 2024).

⁴⁶ See https://www.gesetze-im-internet.de/bkrg (accessed November 23, 2024).

⁴⁷ See http://www.bgbl.de/xaver/bgbl/start.xav?startbk=Bundesanzeiger_BGBl&jumpTo=bgbl113s0617.pdf (accessed November 23, 2024).

⁴⁸ See http://www.bgbl.de/xaver/bgbl/start.xav?startbk=Bundesanzeiger_BGBl&jumpTo=bgbl121s3890.pdf (accessed November 23, 2024).

⁴⁹ See https://www.bundesgesundheitsministerium.de/ministerium/ressortforschung/handlungsfelder/forschungsschwerpunkte/krebsregisterdaten.html (accessed November 23, 2024).

⁵⁰ See https://www.gesetze-im-internet.de/gdng (accessed November 5, 2024).

___________________

⁵¹ See https://www.basisdatensatz.de/basisdatensatz (accessed October 16, 2024).

⁵² See https://www.recht.bund.de/bgbl/1/2024/102/VO.html (accessed October 16, 2024).

USCDI+⁵³ for cancer, a domain-specific list to accompany the U.S. Core Data for Interoperability (USCDI) dataset⁵⁴ maintained by the Office of the National Coordinator for Health IT (ONC). Attention is also turning to the need for data elements to describe information derived from omics-based tests.⁵⁵ The challenges, Yu said, include data capture in both human and machine-readable formats (e.g., structured data entry in EHRs and synoptic pathology reports), harmonization of data dictionaries, and adoption of transmission standards (e.g., state cancer registry capabilities and bidirectional data sharing). There are emerging methods, such as common data models, but he noted that new approaches can be complex, and they could create additional burdens for data entry. There are potential roles for new technologies, such as generative AI, and knowledge graph⁵⁶ extensions, which can capture relevant data from existing sources (e.g., social determinants of health information from census data). The ultimate goal of capturing data is to inform a learning health care system (see Figure 3) and improve care quality and value, which Yu said involves applying data science and machine learning to high quality digital data to answer key questions.

Many panelists representing a range of organizations involved in data collection and curation for cancer surveillance shared their perspectives on current approaches and emerging tools for automation and analytics.

Curating Real-World Electronic Health Record Data with the Flatiron Health Database

Emily Castellanos, senior medical director at Flatiron Health, said that Flatiron Health curates real-world data from more than 3 million patient records across a network of oncology practices (Castellanos et al., 2024). About three quarters of the network is community oncology practices that

___________________

⁵³ See https://www.healthit.gov/topic/interoperability/uscdi-plus (accessed October 14, 2024).

⁵⁴ See https://www.healthit.gov/isp/united-states-core-data-interoperability-uscdi (accessed October 16, 2024).

⁵⁵ Omics refers to any of several areas of biological study defined by investigating the entire complement of a specific type of biomolecule or all of a molecular process within an organism, such as proteomics, metabolomics, genomics, and transcriptomics. See https://www.britannica.com/science/omics (accessed October 14, 2024).

⁵⁶ Knowledge graphs are models used to organize interlinked concepts to connect data and context, thereby empowering analytics, integration, and human–system interfaces. See https://www.turing.ac.uk/research/interest-groups/knowledge-graphs (accessed October 16, 2024).

___________________

⁵⁷ See https://flatiron.com/oncology/oncology-ehr (accessed October 16, 2024).

⁵⁸ See https://www.ssa.gov/dataexchange/request_dmf.html (accessed October 16, 2024).

⁵⁹ Regular expressions are specific syntax used to search for terms. See https://www.csfieldguide.org.nz/en/chapters/formal-languages/regular-expressions (accessed December 16, 2024).

___________________

⁶⁰ See https://www.facs.org/quality-programs/cancer-programs/commission-on-cancer/standards-and-resources/2020/implementation (accessed October 23, 2024).

⁶¹ First course treatment includes “all types of treatment documented in the treatment plan and given to the patient prior to progression of disease or recurrence of disease.” See https://training.seer.cancer.gov/treatment/first-course (accessed December 17, 2024).

has made particular treatment choices. To enhance the content in EHRs, the University of Pennsylvania has developed and implemented an EHR-based documentation tool that prompts the entry of key data in a structured format. Shulman said that “the smart form,” together with cancer staging and routinely collected patient-reported outcomes and other structured data in the EHR (drugs, doses, lab results, etc.) will give a more comprehensive picture of persons with cancer across their care continuum. He emphasized that the form facilitates documentation, does not add work, and need not be reiterated in a free text entry.

Shulman also described a recent study that looked at the impact of payment policies on how clinical information is documented. Clinicians focus on information required for reimbursement, he said, and the result is “poor-quality, low-value documentation that is not structured or computable and contributes to clinician burnout” (Gabriel et al., 2023). The authors proposed a payment approach to reimburse for documenting key data in a structured format.

Lessons From the COVID-19 and Cancer Consortium

Jeremy Warner, founding director of the Center for Clinical Cancer Informatics and Data Science at Brown University, said informatics requires people, process, and technology. During the COVID-19 pandemic, the ability to gather timely data on persons with cancer who were diagnosed with COVID-19 was lacking. The people and technology were there, Warner said, but the processes were missing. The COVID-19 and Cancer Consortium (CCC19)⁶² was formed to fill the gap. The consortium now includes 123 institutions and more than 600 investigators.

Warner provided information on some essential elements that needed to be rapidly assembled to launch CCC19, from overall governance of the consortium to committee structures, the data model, terminology, quality metrics, analytics, and outreach and patient engagement. Warner noted that processes were copied from other organizations when possible, and persons with cancer who were experiencing COVID-19 were engaged to inform the development of the database.

Echoing a number of other speakers, Warner said that critical elements needed to populate the CCC19 registry, such as cancer treatment regimens, treatment toxicity, cancer status, and performance status, were not found in structured EHR data and often not in unstructured data either (Kim et al., 2019).

___________________

⁶² See https://ccc19.org (accessed October 16, 2024).

___________________

⁶³ The ECOG Performance Status Scale, developed by the Eastern Cooperative Oncology Group (ECOG), now known as the ECOG-ACRIN Cancer Research Group, was published in 1982. This scale assesses a patient’s level of functioning, with and without restriction, using a variety of factors, including the ability to care for oneself, daily activity, and physical ability (e.g., walking, working). This scale provides a grade from 0 (fully active, no restrictions) to 5 (deceased). See https://ecog-acrin.org/resources/ecog-performance-status (accessed October 23, 2024).

⁶⁴ See https://www.cdc.gov/national-program-cancer-registries/registry-plus/emarc-plus.html (accessed October 16, 2024).

⁶⁵ See https://info.cap.org/future-of-cancer-data-summit-insights/downloads/NPCR-ePath-FactSheet.pdf (accessed October 30, 2024).

___________________

⁶⁶ See https://www.cdc.gov/national-program-cancer-registries/data-modernization/advancing-electronic-reporting.html (accessed October 23, 2024).

⁶⁷ See https://www.cdc.gov/national-program-cancer-registries/about/pediatric-young-adult-cancer.html (accessed October 16, 2024).

⁶⁸ See https://www.healthit.gov/topic/interoperability/policy/trusted-exchange-framework-and-common-agreement-tefca (accessed October 19, 2024).

⁶⁹ A Quality Health Information Network (QHIN) is a network that directly connects participating organizations to achieve interoperability. The qualification process includes completing an application, onboarding, and designation. See https://www.healthit.gov/topic/interoperability/policy/trusted-exchange-framework-and-common-agreement-tefca (accessed October 19, 2024).

⁷⁰ See https://www.hl7.org/codex (accessed October 23, 2024).

to improve patient experience and outcomes in real care settings. A broad range of CodeX use cases are underway, including a pilot focused on cancer registry reporting.

mCODE⁷¹ is a core set of common cancer data based on the FHIR common health data exchange standard. Chen explained that mCODE is an open data model that is standardized and affords API capabilities, which can help to implement digital health solutions at scale. She said that standards are only valuable if they are used widely, effectively, and in actual health care settings. Chen noted significant uptake by vendors, with about half of U.S. patient records in vendor systems that use mCODE. A recent study at the Mayo Clinic found that mCODE used in the context of routine patient care increased data capture with minimal to no clinician burden, she continued (Tevaarwerk et al., 2024). In addition, mCODE is now being used for data capture beyond academic hospitals in community cancer clinics.

Now is the time to reshape the cancer data landscape, Chen said, emphasizing that industry, community, and federal agency efforts are aligning, with momentum building from federal agency cancer initiatives, such as the Cancer Moonshot.⁷²

The OneFlorida+ Clinical Data Trust Model for Facilitating Research

Elizabeth Shenkman, professor and chair of the Department of Health Outcomes and Biomedical Informatics at the University of Florida, discussed the OneFlorida+⁷³ Clinical Data Trust⁷⁴ as a model of a central data repository to facilitate research. OneFlorida+ partners with academic, regional, and statewide health systems across Florida and select partners in Alabama, Arkansas, California, Georgia, and Minnesota and is a PCORnet⁷⁵ network partner. At the core of OneFlorida+ is the central data trust, which contains EHR data for over 26 million patients spanning 2012 to present, she said.⁷⁶

Data in the OneFlorida+ Data Trust are derived from EHRs and other sources (see Figure 4) and stored in the formats of the Observational Medical Outcomes Partnership (OMOP) Common Data Model⁷⁷ and PCORnet

___________________

⁷¹ See https://hl7.org/fhir/us/mcode (accessed October 23, 2024).

⁷² See https://www.cancer.gov/research/key-initiatives/moonshot-cancer-initiative (accessed November 1, 2024).

⁷³ See https://onefloridaconsortium.org/about (accessed October 17, 2024).

⁷⁴ See https://oneflorida.sites.medinfo.ufl.edu/data-2 (accessed October 17, 2024).

⁷⁵ See https://pcornet.org (accessed November 2, 2024).

⁷⁶ See https://oneflorida.sites.medinfo.ufl.edu/data-2 (accessed December 4, 2024).

⁷⁷ See https://www.ohdsi.org/data-standardization (accessed October 17, 2024).

___________________

⁷⁸ See https://pcornet.org/data/common-data-model (accessed October 17, 2024).

⁷⁹ IRBs are committees that oversee research involving human subjects to protect human rights and welfare. See https://www.fda.gov/about-fda/cder-offices-and-divisions/institutional-review-boards-irbs-and-protection-human-subjects-clinical-trials (accessed October 17, 2024).

⁸⁰ See https://blogs.nvidia.com/blog/gatortrongpt (accessed October 17, 2024).

⁸¹ See https://flshotsusers.com (accessed October 17, 2024).

⁸² See https://clinicaltrials.gov/study/NCT05006833?term=Text (accessed November 26, 2024).

mCODE, and others. He said that MSK is deep phenotyping⁸³ a cohort of patients for whom deep tumor genotyping data are available from the MSK-Integrated Mutation Profiling of Actionable Cancer Targets (IMPACT®) genotype panel.⁸⁴ He explained that by using CCDE, MSK has treatment and outcomes data for a cohort of 24,000 deeply phenotyped patients. These data are highly curated and structured, have greater than 95 percent accuracy, and are used by MSK as a reference standard for AI model development and validation of vendor models, Stetson added.

Stetson said that the benefits of the CCDE real-world data model are that it is “relieving the data quality constraint for AI,” offers a “proving ground” for AI model development, and enables data sharing outside of MSK. He added that AI-enabled curation also increases data density more quickly.

Stetson offered his perspective on a road map for moving forward, including creating an automated cancer registry on a national scale; “federated tuning of AI-enabled curation,” establishing academic–industry partnerships to facilitate a “computable phenotype ecosystem,” and enabling “cross-institutional computable journeys” to be able to predict treatment response based on full patient history across time and place of care.

Emerging Directions in Multimodal Health Data Analytics

Peter Clardy, lead of the Clinical Enterprise Team at Google Health, discussed opportunities to integrate AI tools to manage multimodal patient data. Traditional programming systems involve structured inputs, algorithms, and outputs that are highly deterministic, he said. The original AI and machine learning tools involved supervised learning, which Clardy said “would allow datasets that are highly curated with a ground source of truth to perform certain kinds of task-specific data transformations.” Emerging generative language models, such as GPT,⁸⁵ learn from consuming huge volumes of data and then use an inference process for tasks, such as answering questions or data abstraction.

Clardy described three waves of AI, beginning in the 1980s with limited capabilities and evolving in the 2010s to neural networks that could successfully complete narrow tasks but could be fragile, required expensive training data, were poor at multimodal and sequential data, and could be difficult to interpret (Sharma et al., 2020). Now, in the third wave, the cost of training is lower,

___________________

⁸³ In precision medicine, deep phenotyping is an emerging process that gathers more specific, individualized details on disease manifestations and presentations to better understand potential connections between genetic variations and disease subtypes (Delude, 2015).

⁸⁴ See https://www.mskcc.org/msk-impact (accessed October 18, 2024).

⁸⁵ See https://www.ibm.com/think/topics/gpt (accessed October 24, 2024).

and generative AI is more capable of handling multimodal and sequential data, resulting in “rich, expressive outputs,” he said (Howell et al., 2024).

The use of AI in health care was initially focused on what Clardy described as “mid-generation supervised learning tasks,” such as lung or breast cancer detection, pathology predictors, and genomics (Howell et al., 2024). Today, the focus is on multimodal models that can integrate data from a range of sources and formats, learn, and create a range of outputs for different applications (Acosta et al., 2022). As mentioned, a challenge for AI is that “longitudinal comprehensive patient records are multimodal, heterogeneous, and laden with unstructured data,” Clardy said. Other challenges are that health-related questions are “complex and nuanced,” and the more preprocessing of the questions and the data that are done, the more rigid and costly the systems become. And, while tools such as FHIR have increased data interoperability, capturing information from unstructured data is a persistent challenge.

Clardy noted many opportunities to integrate AI tools into health care (e.g., knowledge graphs, optical character recognition tools, and multimodal patient data management), and each will have advantages and disadvantages. Current systems are still relatively deterministic, but he said there is an increasing number of human-in-the-loop hybrid AI tool pilots. Clardy added that although it is still in the proof-of-concept stage, a generative future is emerging, where large models can extract useful individual- and population-level data from large volumes of patient records. A consideration, however, is that clarity regarding how these systems work decreases as the use of generative tools increases, Clardy concluded.

Getting Data into the Record

Several session participants commented on a range of topics related to populating the EHR with the data needed by cancer registries.

Data Elements

Shulman reiterated that essential data elements include cancer recurrence and factors that provide information about why a patient’s clinician has made particular care decisions. The challenge, he said, is balancing “all the things we might want to know with the things that we absolutely have to know.” Chen highlighted the need to consider what other data elements might be needed for clinical actionability. Another area where data elements are lacking across data models is survivorship. She raised a concern that creating “different siloed versions of these data elements” could exacerbate interoperability barriers. Chen called for cross-sector collaboration to establish a cohesive platform that does not add burden, and Yu noted the need for public–private collaboration

to develop a shared vision. Warner said that cancer is many different diseases and suggested there is a need to consider when highly detailed data are needed versus when a subset is sufficient. He added that, whether machine or human, effort is finite, and one cannot have all the elements on every cancer.

Large Language Models

Gabriel observed that large language models work well with natural language and perhaps less so with medical language and lingo. He mentioned ambient listening as an emerging technology that might be used to extract information from clinical conversations (e.g., patient–clinician conversations and tumor board meetings) and provide prose for AI models to transform into structured data. Clardy said the patient record is often a poor reflection of a patient’s situation due to problems such as data missingness, copy-and-paste text, coding limitations, and lack of record interoperability (Turbow et al., 2021; Wells et al., 2013). “The language of health is multimodal and expanding, and an incremental approach with a human in the loop is needed,” Clardy added.

Incentives

Shulman said that the quality of clinic notes varies widely and reiterated that the data need to be available to be extracted. He noted that inadequate data entry is not malicious but rather the result of the numerous priorities competing for the clinician’s time. Current payment systems do not reward quality data entry, and he said incentives are needed to ensure clinicians document critical data elements at every patient encounter. For example, incremental reimbursement for a pathology report could be contingent on the report being in a synoptic format with structured data. Shulman also highlighted the need to eliminate unnecessary data elements in the EHR and include what matters most.

Warner pointed out the disconnect between what the clinical community needs to care for its individual patients and what the registry community needs at the population level for public health. He added that clinicians do not generally receive any training on cancer registries, and he suggested education is needed so that clinicians understand why registries are important for their work.

Yu said continuing to add more data elements contributes to clinician workload and burnout. Although smart forms can facilitate data entry, he said, many clinicians are resistant to clicking through a form and still prefer narrative, even as a redundant task that adds time. Jones noted the software tools that enable the clinician to speak their responses to the form questions. She said one issue is that clinicians are overwhelmed and should not spend their time searching for codes.

with an “incrementalist approach” of applying technology to improve operational efficiencies, such as automating tasks to reduce the drain on staff resources. Clardy said that improved long-term outcomes, such as lives saved, are very difficult to demonstrate, but hours saved is generally a metric that all parties can support. He also cautioned that “there is a tendency to use generative AI to solve for everything,” which he said has “the potential for very non-deterministic outcomes and is also computationally extremely expensive and time-consuming.” Clardy recommended “working up the level of complexity and risk … and applying deterministic or machine learning systems where you can, and then only leveraging these more generative models when nothing else will do” and making sure to include “human-in-the-loop and process supervision.”

THE INFRASTRUCTURE AND WORKFORCE SUPPORTING CANCER SURVEILLANCE

Several speakers discussed issues related to the evolving infrastructure for cancer surveillance and the shifting role of the workforce, with a focus on ensuring the capture of high-quality clinical data and reducing abstraction time.

Infrastructure

College of American Pathologists Cancer Protocols: Harnessing Structured Data to Optimize Cancer Surveillance

The vision for the College of American Pathologists (CAP) Cancer Protocols is to “improve patient outcomes through structured cancer pathology reporting solutions,” said Samantha Spencer, senior director of Standards and Guidelines at CAP. Protocols are based on current standards and include the core elements required by accreditation organizations (e.g., the American College of Surgeons, CoC, and CAP Laboratory Accreditation Program).⁸⁶ There are 101 CAP Cancer Protocols,⁸⁷ and templates for reporting by pathologists are available in MS Word and PDF formats and as electronic forms integrated into laboratory information systems (LIS) that can generate synoptic reports.⁸⁸

To create and edit the electronic Cancer Protocols, CAP developed a robust, web-based content generation tool based on the structured data

___________________

⁸⁶ See https://www.cap.org/laboratory-improvement/accreditation/laboratory-accreditation-program (accessed October 19, 2024).

⁸⁷ See https://www.cap.org/protocols-and-guidelines/cancer-reporting-tools/cancer-protocols (accessed October 19, 2024).

⁸⁸ See https://www.cap.org/protocols-and-guidelines/cancer-reporting-tools/cancer-protocol-templates (accessed October 8, 2024).

capture object model,⁸⁹ Spencer explained. This approach facilitates capture of structured data within the medical record; mapping of data elements to standardized terminologies, including SNOMED CT⁹⁰ and ICD-O; and transmission of structured, standardized data. Spencer noted that vendor implementation of the electronic Cancer Protocols is a key to success in clinical practice. To ensure successful vendor implementation, CAP works collaboratively with vendors, promotes bidirectional communication and feedback, and conducts implementation validation sessions to ensure fit into the workflow. The next step, which is still in development, is facilitating data transmission to downstream users, including registries (Torous et al., 2021).

Spencer emphasized that multidisciplinary collaboration is essential for realizing CAP’s vision for structured pathology reporting, and CAP is engaging with public health partners and standards-developing organizations on a range of topics. CAP is also engaged in commenting on federal health IT initiatives, such as USCDI, USCDI+, and HTI-2.⁹¹

A challenge, as discussed throughout the workshop, is that data in the clinical record are frequently not structured or standardized, Spencer said, and “human interpretation and translation is still needed.” There are also challenges with standardized data coding, and she said version control also needs to be addressed.

The common vision, Spencer summarized, is to improve patient care through interoperable cancer data reporting and downstream data use (Torous et al., 2021). Education and training are critical elements of success, and she noted the need to provide training about registries early in medical education. She also referred participants to the CAP Cancer Data Champions webinar series.⁹² Focus on implementation, work with vendors and end users to understand their needs and ensure functionality in clinical workflows, and support adoption at a national level through coordination, collaboration, and change management, she concluded.

Use of EHRs and Health Information Exchanges to Streamline Registry Processes

Mary Charlton, professor of epidemiology and director of the Iowa Cancer Registry⁹³ at the University of Iowa College of Public Health, dis-

___________________

⁸⁹ See https://www.healthit.gov/topic/scientific-initiatives/pcor/research-evaluation/structured-data-capture-sdc (accessed October 19, 2024).

⁹⁰ See https://www.snomed.org/what-is-snomed-ct (accessed October 24, 2024).

⁹¹ See https://www.healthit.gov/sites/default/files/page/2024-07/HTI-2_ProposedRule_Overview_Factsheet_508.pdf (accessed October 24, 2024).

⁹² See https://www.cap.org/calendar/webinars/cancer-data-champions (accessed October 8, 2024).

⁹³ See https://shri.public-health.uiowa.edu (accessed October 24, 2024).

cussed the role of EHRs and health information exchanges (HIEs) in cancer registration. HIEs facilitate the electronic transmission of patient medical information among health care-related professionals and facilities,⁹⁴ and she said that most states have one or more (ONC, 2019).

It is essential to collect data from all patients, not just those receiving care at academic medical centers, Charlton said. In Iowa, many patients in rural areas receive cancer care at one of 81 critical access hospitals, and not all of these hospitals have an oncologist (Schroeder et al., 2022). She said that HIEs can be a mechanism for registries to access EHR data for their entire catchment area, rather than collecting data from each hospital. The use of this approach is “highly variable across cancer registries and generally in early stages,” she said, and most registries are doing manual abstraction. Although technology has a role in helping registries collect an increasing number of data elements in the face of declining resources, Charlton said that technology cannot fully replace people, and human relationships can be essential to gathering data from rural hospitals.

Charlton described several lessons learned from linking the Iowa Cancer Registry to EHR data in CancerLinQ.⁹⁵ As mentioned by several other speakers, successful linkages with EHRs require good identifiers (e.g., Social Security number, full name, birth date, sex, race, and address), and this can be very challenging in states with strict policies regarding protected health information, she said. Another challenge for registries is linking EHR data for persons with multiple cancers. As an example, she said, if a patient had one tumor in each breast, data were insufficient to “accurately link EHR data to the correct tumor.” Linkage of the registry to EHR data in CancerLinQ was beneficial for improving the capture of treatment data, she said, particularly hormone and immunotherapy data (Charlton et al., 2022). There are also efforts to automate data extraction from EHRs, and she mentioned ExtractEHR⁹⁶ as an example, which is “an R⁹⁷ software package to automate laboratory adverse event ascertainment and grading” in pediatric cancer (Miller et al., 2022). She described the progress as promising and said that “in the future, these activities,

___________________

⁹⁴ See https://www.healthit.gov/topic/health-it-and-health-information-exchange-basics/what-hie (accessed October 24, 2024).

⁹⁵ For information on CancerLinQ, see https://www.cancerlinq.org/. In December of 2023, CancerLinQ ownership was transferred from the American Society of Clinical Oncology to ConcertAI. See https://connection.asco.org/magazine/features/building-cancerlinq-future-next-step-ambitious-asco-project (accessed October 19, 2024).

⁹⁶ See https://www.cancer.gov/research/areas/childhood/childhood-cancer-data-initiative/progress/projects (accessed October 24, 2024).

⁹⁷ R is a programming language for statistical computing and data visualization. See https://www.r-project.org/ (accessed November 5, 2024).

coupled with more advanced interoperability standards, could allow for more efficient flow and use of EHR data through [HIEs] to registries.”

There are also opportunities to leverage HIEs to streamline registry processes beyond EHR data, Charlton said. One example is for case finding, which she said, “can be a tedious and resource intensive process for hospitals, as well as for registries,” especially in rural states. Charlton said HL7 notifications to admit, discharge, and transfer (ADT)⁹⁸ likely contain variables useful for linkage and could be used for case finding, as well as for auditing the completeness of hospital case ascertainment. She noted that the Iowa Cancer Registry has partnered with CyncHealth (the HIE in Iowa)⁹⁹ to use ADTs for case finding because nearly all of Iowa’s hospitals share data with this HIE. As another example, she said, the Maryland Cancer Registry is partnering with the state HIE, the Chesapeake Regional Information System for Our Patients,¹⁰⁰ to find information on persons who are found to have cancer only at the time of death. Charlton pointed out that patients with insufficient registry data are generally excluded from research, and HIEs can be used to find the needed information.

Jackie Gerhart, chief medical officer at Epic, discussed how AI in EHRs can empower the cancer care workforce. For example, rule-based logic provides clinical decision support, such as flagging cancer screenings the patient is due for. At the other end of the spectrum is generative AI, and she emphasized that such tools to support the workforce should be fit for purpose (i.e., using AI only when needed).

Castellanos also discussed EHR technologies used to increase efficiency for abstractors. For example, Flatiron Health has a technology to identify and highlight documents that are missing needed information. There are also “form validators to help prevent errors at time of entry,” tools to monitor performance and changes in documentation patterns so root causes can be addressed, and protocols for abstractors to escalate complex cases for clinical quality review.

The Cancer Surveillance Workforce

The Influence of Cancer Registry Professionals on Cancer Surveillance

Cancer registrars are data information specialists who capture and maintain a complete history, diagnosis, treatment, and health status for every person with cancer in the United States,¹⁰¹ said Nadine Walker, senior director of professional practice at NCRA.

___________________

⁹⁸ See https://www.interfaceware.com/hl7-adt (accessed October 19, 2024).

⁹⁹ See https://cynchealth.org (accessed October 19, 2024).

¹⁰⁰ See https://www.crisphealth.org/about-crisp (accessed October 24, 2024).

¹⁰¹ See https://www.ncra-usa.org/About/Become-a-Cancer-Registrar/What-Do-Cancer-Registrars-Do# (accessed November 13, 2024).

___________________

¹⁰² See https://www.ncra-usa.org/ODS-Credential (accessed October 24, 2024).

¹⁰³ See https://www.cancerregistryeducation.org (accessed October 24, 2024).

¹⁰⁴ See https://www.ncra-usa.org/Portals/68/PDFs/Workforce/WorkloadExecutiveSummary_2024.pdf (accessed October 24, 2024).

¹⁰⁵ See https://sph.emory.edu/GCCS/gcr/intro.html (accessed October 24, 2024).

and territories, cancer is a reportable disease. In Georgia it is reported to the Georgia Department of Public Health, and the law¹⁰⁶ allows the health department to make policies accordingly. Ward explained that this means that implementing new data collection practices, such as collecting recurrence data, can be done through policy change and not necessarily legislative action.

Ward said pathology data elements are critical for population-based registries, second only to the data in the NAACCR record layout.¹⁰⁷ Ninety-four percent of all cancer cases in the United States have microscopic confirmation. However, he noted that pathology data are often lacking for brain tumors (which are frequently confirmed by radiology) and highly fatal cancers. He concurred with Charlton that the timely capture of every case of cancer in the state is necessary for surveillance and to support research. Fourteen federally funded R01¹⁰⁸ research studies in Georgia are using registry data because, he said, the law allows access to the data and the ability to invite patients to participate in studies. Having timely pathology data facilitates rapid case ascertainment and identification of patients who might be suitable for a particular study (Petkov et al., 2024; Tucker et al., 2019). He added that “the majority of people want to participate in cancer research studies.”

The current pathology workflow, Ward summarized, spans real-time data generation by laboratories; standardized messaging; screening for reportability using tools such as eMaRC Lite and transmission to registries; autocoding, which can enable more rapid identification of patients; and production of reports by registries.

Data collected by facilities that provide care are reported to registries in electronic format, Ward said, and central registries merge multiple reports for the same patient into a single record. At the hospital level, the model is for an ODS-Certified cancer registrar to abstract EHR and pathology data for entry into the registry software. As the field of cancer registration transitions to new data models, Ward said, a shift in workforce roles will also be needed. The ODS-Certified cancer registrar, as the expert on cancer data, will work collaboratively with IT staff and vendors to integrate EHR and pathology data into registry software, he said. With less manual abstraction needed, the ODS-Certified cancer registrar’s work will shift to quality control of the data. Importantly, they will continue to be involved in the abstraction of complex data elements and in decision making (e.g., abstracting multiple primary cancers).

___________________

¹⁰⁶ See https://dph.georgia.gov/chronic-disease-prevention/georgia-comprehensive-cancer-registry/reporting-cancer (accessed October 24, 2024).

¹⁰⁷ See https://apps.naaccr.org/data-dictionary/data-dictionary/version=25/chapter-view/record-layout-table (accessed October 24, 2024).

¹⁰⁸ See https://grants.nih.gov/funding/activity-codes/R01 (accessed October 19, 2024).

___________________

¹⁰⁹ See https://www.ncra-usa.org/Education/Accredited-Education-Programs-Old/Associate-and-Certificate-Programs (accessed October 24, 2024).

¹¹⁰ See https://www.ncra-usa.org/About/Store/4th-Edition-Textbook (accessed October 24, 2024).

¹¹¹ See https://www.ncra-usa.org/About/Become-a-Cancer-Registrar (accessed October 8, 2024).

¹¹² See https://shri.public-health.uiowa.edu/cancer-data/reports/iowa-cancer-reports/ (accessed October 19, 2024).

NCRA and Rutgers University to institute a certificate program for undergraduate students in public health.¹¹³ She noted the significant interest in the program, with 60 students enrolling in the first session. Ward and Charlton discussed the need to start early and raise awareness of the profession among high school students. Walker agreed and reiterated that NCRA does outreach to high schools, and she clarified that an associate’s degree is a requirement for pursuing ODS certification.¹¹⁴

Ward said that Georgia offers opportunities for some ODS-Certified staff to engage in research activities. Castellanos said appropriate compensation is also important, and Ward added that NCRA conducts salary surveys and noted that salaries for ODS staff are increasing.

Spencer highlighted the need to prepare registrars for their future roles as the workflow evolves. Winn observed that technology is evolving at a rapid pace and said it is important to consider how a variety of educational pathways might enhance the cancer surveillance workforce. Palis suggested that the role of the registrar could evolve beyond data entry and data quality to being “at the center of patient quality of care.” He said that registrars at CoC-accredited hospitals are involved in “monitoring compliance with key metrics” and discuss reports at cancer committee meetings “down to the case level.” One workshop participant suggested that the registry profession needs to be marketed to those with interest and expertise in data analytics. An ODS-Certified cancer registrar not only collects data but can understand and present them to the hospital administration in the context of monitoring patient quality of care.

Several session panelists also discussed promoting ODS-Certified careers to nurses. Ward said some of the most successful ODS-Certified staff are nurses because of their medical and health care background. Walker confirmed that NCRA does see many nurses transition to the profession and agreed with the need to raise awareness about the profession broadly among health care professionals.

Lori Swain, executive director of NCRA, said NCRA is working with the Bureau of Labor and Statistics to develop a definition of medical registrar in anticipation of the 2028 revisions to the Standard Occupational Classification.¹¹⁵ She noted that registrars are currently classed with health information technicians, but before 2017, registrars were classed with medical records specialists.

___________________

¹¹³ To learn more about the Cancer Surveillance and Epidemiology Certificate for undergraduate students at Rutgers University, see https://bloustein.rutgers.edu/undergraduate/#certificates (accessed November 25, 2024).

¹¹⁴ See https://www.ncra-usa.org/ODS-Credential/Potential-ODS/Eligibility (accessed October 24, 2024).

¹¹⁵ See https://www.bls.gov/soc (accessed October 19, 2024).

___________________

¹¹⁶ See https://www.cdc.gov/national-program-cancer-registries/about/pediatric-young-adult-cancer.html (accessed November 13, 2024).

¹¹⁷ This included a pediatrician, radiologist, orthopedic surgeon, neurologist, emergency room physicians, hospitalist, pathologists, medical oncologists, neurosurgeons, radiation oncologists, anesthesiologists, neuro-ophthalmologist, pediatric physiatrist, occupational speech therapist, robotics therapist, pharmacist, clinical trial investigator and coordinator, palliative care oncologist, social worker, and hospice nurses.

___________________

¹¹⁸ See https://cancercontrol.cancer.gov/research-emphasis/supplement/childhood-cancer-registry (accessed October 19, 2024).

¹¹⁹ See https://www.cancer.gov/research/areas/childhood/childhood-cancer-data-initiative (accessed October 19, 2024).

¹²⁰ See https://www.childrensoncologygroup.org/about (accessed October 19, 2024).

___________________

¹²¹ See https://surveillance.cancer.gov/delay (accessed October 24, 2024).

¹²² See https://news.va.gov/press-room/va-and-national-cancer-institute-announce-historic-data-sharing-partnership-to-better-understand-and-treat-cancer-among-veterans (accessed October 24, 2024).

¹²³ See https://myriad.com/about-myriad-genetics (accessed October 19, 2024).

¹²⁴ See https://allofus.nih.govs (accessed October 19, 2024).

¹²⁵ BReast CAncer gene, better known as BRCA, is a group of genes that code for DNA repair proteins. Mutations convey a higher risk for cancer, especially breast and ovarian cancers. See https://www.cancer.gov/about-cancer/causes-prevention/genetics/brca-fact-sheet (accessed October 19, 2024).

ing in every state” would facilitate real-time reporting and the assessment of patients for potential enrollment in clinical trials. “Mandated authorization for registries to access all payer claims data” would enable capture of data on longitudinal treatment, recurrence, and comorbid conditions, she continued. Penberthy concluded by urging CMS to share claims data for surveillance and incorporation into the registry and mandate recurrence data reporting.

Resource Needs for High-Quality Cancer Surveillance

Micky Tripathi, assistant secretary for technology policy at ONC, provided an overview of current and future ONC activities related to cancer surveillance. ONC has been working with the National Institutes of Health (NIH), NCI, the White House Cancer Moonshot Team, and other partners to improve the availability of data, both bringing data into the EHR and sharing it for authorized uses.

Tripathi said 97 percent of hospitals and 80 percent of clinician offices in the United States use ONC-certified EHR systems, which support standardization and data availability.¹²⁶ ONC maintains USCDI (the minimum dataset required for EHR systems) and has been working with public and private partners to develop domain-specific extensions, including USCDI+ for Cancer, which will support the data requirements of the CMS Enhancing Oncology Model¹²⁷ and clinical trial matching.

As directed by the 21st Century Cures Act,¹²⁸ ONC implemented TEFCA in late 2023 to support “nationwide network-to-network interoperability,” Tripathi said, and there are currently seven TEFCA QHINs¹²⁹ that are sharing clinical data for treatment purposes. The next step in TEFCA implementation is to support the nationwide exchange of clinical information for research purposes, which Tripathi explained has additional complexities. Challenges to be addressed include standardized consent (in a computable format usable across EHR systems) and standards for data de-identification. ONC has also been working with payers to facilitate clinician access to claims data via TEFCA, in line with the requirements of the CMS interoperability rule.

Tripathi said there will be continuing need for intermediary data curation services, such as registries, so that data generated for health care delivery can

___________________

¹²⁶ See https://www.healthit.gov/buzz-blog/health-it/oncs-next-chapter (accessed October 19, 2024).

¹²⁷ See https://www.cms.gov/priorities/innovation/innovation-models/enhancing-oncology-model (accessed October 19, 2024).

¹²⁸ See https://www.fda.gov/regulatory-information/selected-amendments-fdc-act/21st-century-cures-act (accessed October 19, 2024).

¹²⁹ See https://www.healthit.gov/sites/default/files/page/2023-11/TEFCA_2-Pager_Digital_508.pdf (accessed October 24, 2024).

be leveraged for other purposes, including research and public health needs. The goal is to use emerging tools to make the curation process “better, faster, and cheaper than it is today.”

The Need for Data to Assess the Quality Impact of Payment Models

“Value-based care ties the amount that health care providers earn for their services to the results they deliver for their patients,” said Nancy Keating, professor of health care policy and medicine at Harvard University (Lewis et al., 2023). She discussed the CMS Oncology Care Model (OCM)¹³⁰ for chemotherapy administration as an example of how data are used in alternative payment models.

While OCM was in effect from 2016 through June 2022, more than 1.5 million patient episodes of care were completed across more than 200 participating practices, Keating said.¹³¹ She explained that practices could bill $160 per patient per month during a 6-month chemotherapy episode. Practices were also eligible for performance-based payments based on cost and quality of care. Keating described the OCM as “modestly successful from a financial standpoint, saving about $600 per episode on average across the whole performance period” (OCM Evaluation Team, 2024).

CMS evaluated the impact of the OCM on the quality of care across 12 measures in areas of performance-based payment, timeliness of chemotherapy, guideline recommended treatment, chemotherapy associated use, and end-of-life care. Keating reported that there was no evidence that OCM improved the quality of cancer care for any of the measures.¹³²

Keating listed several areas of opportunity to advance progress. First, payment models focus on patients who receive treatment but do not provide insight into who is able to access treatment or whether patients are receiving appropriate guideline-based treatment. Second, these are payment models led by payers, and the data at their disposal are limited to submitted claims and data reports from practices in the model. She also noted the critical importance of having appropriate comparison groups to understand the quality impact of payment models. “Quality measurement in oncology care is early in its development,” Keating continued, and “claims-based measures identify a

___________________

¹³⁰ See https://www.cms.gov/priorities/innovation/innovation-models/oncology-care (accessed November 3, 2024).

¹³¹ See https://www.cms.gov/priorities/innovation/data-and-reports/2023/ocm-evaluation-pp1-9 (accessed November 20, 2024).

¹³² See https://www.cms.gov/priorities/innovation/data-and-reports/2024/ocm-final-eval-report-2024-exec-sum (accessed October 8, 2024).

very limited aspect of quality.” Last, she said, “Real-time, high-quality registry data are needed to fully assess the impacts of payment models on quality.”

Advancing Data Linkages to Biospecimens to Enable Population-Based Studies

Michaela Dinan, associate professor of epidemiology at Yale University, said that to conduct population sciences research that explores the intersection between social and biological factors, it is essential to have population-level data with linkages to biological samples, but few if any such linkages exist. Dinan observed the “disconnect between population sciences and basic biology,” including how investigators identify themselves, the types of funding for research in each field, the research infrastructure needed, and their journal audiences. Furthermore, genomic datasets are often siloed, and privacy concerns impede sharing patient genetic data.

To address this challenge, Dinan and colleagues conducted a pilot feasibility study linking SEER data, Medicare claims data, and physical tumor samples to analyze socioeconomic, clinical, and genomic factors associated with mortality in women with newly diagnosed breast cancer (Robinson et al., 2021). She said they are now linking “SEER–Medicare data to select cohorts of patients at the population level … to detect rare phenotypes or outcomes of interest for further study.”

Dinan highlighted several opportunities to address policy barriers to population-level genomic analyses, particularly current restrictions on data sharing and use. To protect patient privacy, publication of individual-level data is prohibited per the data use agreement for the NCI SEER–Medicare data linkage.¹³³ Derived data cannot be shared in a public data repository and must be destroyed after studies are completed, she explained. This is counter to the general practice of submitting generated genomic data to public repositories to enable “validation and replication of published research and re-use by other researchers for further discovery-based analysis of these unique and often costly cohorts,” she said.

The requirements of the NIH Genomic Data Sharing (GDS) policy¹³⁴ also present challenges, Dinan said. It mandates that data from large-scale genomic research conducted with NIH funds must be shared, and consent for sharing is required from all patients (living and deceased). Dinan said, “obtaining consent at the population level makes these types of studies logistically impossible,” which essentially rules out using NIH funding for the study

___________________

¹³³ See https://healthcaredelivery.cancer.gov/seermedicare/obtain/seerdua.pdf (accessed October 19, 2024).

¹³⁴ See https://sharing.nih.gov/genomic-data-sharing-policy (accessed October 19, 2024).

of larger cohorts. She also pointed out that the GDS policy and the SEER–Medicare data use agreement are “in direct conflict” regarding the sharing of genomic data.

Another challenge, Dinan said, is that “not all SEER registries participate in these linkage programs,” and she noted that she was aware of seven SEER registries participating in a Virtual Tissue Repository program.¹³⁵ In addition, she said, “logistical workflows to maintain de-identification of these data linkages are complex,” and there are significant costs associated with obtaining formalin-fixed, paraffin-embedded tissue samples.

Aligning Incentives to Improve Data Capture

Gerhart offered actions key collaborators can take now to drive improvement in data capture and enable improvements in cancer care, research, and patient experience.

Reiterating that data standards are essential for the exchange of health care data, Gerhart encouraged those in government to learn about TEFCA and advocate for organizations they work with to participate in TEFCA. She urged payers, and CMS in particular, to “measure and pay for what matters.” Metrics are often quantity based (e.g., number of cancer screens ordered), but there are other metrics of value to oncology clinicians and researchers that could be identified and implemented, and less relevant metrics could be eliminated.

Gerhart also reiterated the need for clinicians and health systems to embrace the use of structured data and highlighted the importance of taking an approach that incentivizes and rewards the uptake of structured data entry rather than a penalty-based approach. She said another potential action that can be taken now is to automate and simplify electronic synoptic pathology reporting. The technology exists, as do existing systems that can be leveraged, and vendors are ready to work with health systems and clinicians to improve these processes. Gerhart said technology companies could focus on ease of use in product development, such as developing simple, smart forms for data capture by the patient’s care team and cancer registrars. Vendors could also design products to “make the right thing to do the easy thing to do,” she said.

Advocates for improved cancer surveillance can bring interested parties together in interdisciplinary working sessions to share experiences and lessons learned and collaborate on solutions, Gerhart said. Advocacy efforts are also needed to show the value of cancer registries and registrars to policy makers and clinicians and market the ODS profession to grow the workforce.

___________________

¹³⁵ See Sanchez et al., 2024.

___________________

¹³⁶ Multicancer detection tests are blood tests that screen for multiple cancer types through the presence of certain biomarkers. See https://www.cancer.gov/publications/dictionaries/cancer-terms/def/mcd-test (accessed December 17, 2024).

the existing infrastructures that use novel technologies (e.g., surveys, bioassays, and wearables) to capture other types of population data, such as the All of Us program, Hantel said. He noted that “opt-in datasets” can exacerbate representation biases but added that recent opt-in programs have made strides to minimize this issue through community engagement and stratified sampling, such as in the All of Us program.

Hantel also discussed the need to develop communications policies that more effectively disseminate registry-related information to target audiences, including the public, local legislators, and funders. Registry data are routinely shared via publication in medical journals and public health initiatives, but even public reports on cancer registry websites can be very technical. Communications policies are needed to leverage technology for multimodal storytelling through social media and interactive registry website content for lay audiences. Hantel suggested that a public educated about the benefits of cancer surveillance programs can be advocates or “cancer registry influencers” for funding allocation and local legislation.

Policy Opportunities to Improve Sharing of Patient Data

Deven McGraw, chief regulatory and privacy officer at Ciitizen,¹³⁷ discussed policy opportunities to improve data sharing from a privacy and data governance perspective. She expanded on the concept raised earlier of a national patient identifier to better facilitate data linkages. Creating such an identifier was authorized under HIPAA in 1996. However, every year since, the Health and Human Services (HHS) Appropriations Bill has included language prohibiting HHS from funding or acting on this authorization. Policy advocacy is needed to demonstrate the importance of a national patient identifier for patient care, registry and surveillance efforts, and research, she said, and show that data sharing can occur while preserving patient privacy.

The impact of state policies on data sharing by registries was also raised earlier, and McGraw pointed out that state legislators meet infrequently and often do not receive sufficient education about measures under consideration because of the time constraints of their sessions. She supported the idea of developing model legislation for states that would allow data sharing by registries while protecting patient privacy.

McGraw said that patient organizations are starting to proactively develop registries and fund research directly.¹³⁸ She noted that provisions of the 21st Century Cures Act, particularly the prohibitions on information blocking, and

___________________

¹³⁷ See https://www.citizen.health/about (accessed October 24, 2024).

¹³⁸ See https://www.iqvia.com/newsroom/2023/10/patient-organizations-gave-22-billion-in-grants-for-research-patient-services-and-access-to-care-ove (accessed October 8, 2024).

the tools being developed by ONC are helping to enable these efforts. As an example, McGraw mentioned how Ciitizen “worked with the Leukemia and Lymphoma Society¹³⁹ to quickly power a study on the efficacy of COVID-19 vaccines in their patient community” (Greenberger et al., 2021). Ciitizen has also worked with the Cholangiocarcinoma Foundation¹⁴⁰ to enable its research registry, which matches patients with studies. Patients want to be “part of the solution, and this is one way to do that,” she said.

Ethics, Privacy, and Cancer Registries

Aaron Goldenberg, professor and vice chair in the Department of Bioethics at Case Western Reserve University, discussed contextual and social distinctions that influence the development of ethical policy and data governance for cancer registries. Contextual distinctions include the needs of individuals versus populations; whether the data in question are identified or identifiable versus de-identified or anonymized; and whether goals of data access are for clinical, quality improvement, or quality assurance versus research use. Social distinctions include the potential for physical and practical harms or benefits versus dignitary harms or benefits; consent versus data security; and being trusted by communities versus being trustworthy in the eyes of the community. However, “these are all false dichotomies,” Goldenberg said. He suggested that the distinction between patient privacy and data access discussed throughout the workshop is also a false dichotomy.

Goldenberg proposed an approach to developing cancer registry policies that address the challenges facing different populations and focus on the foundational principles of engagement and ethical stewardship. He explained that bioethics centers around patient and community values. Cancer registry policies need to reflect those values “in much more holistic ways than they currently do,” and he emphasized the need to be proactive in engaging legislators and ensuring they take patient values into account in policy development. Approaches could include providing model legislation to states or conducting legislative workshops. It is possible to develop policies that both maximize patient privacy and facilitate data access, he said.

As discussed, Goldenberg said patients are generally supportive of their data being used. However, it is important to acknowledge the concerns some communities have regarding the use of their data and ensure that their values and concerns are reflected in the policies developed. Based on his research, Goldenberg said that patient willingness to participate in registries and biorepositories is associated with the extent to which the research done using their

___________________

¹³⁹ See https://www.lls.org (accessed October 19, 2024).

¹⁴⁰ See https://cholangiocarcinoma.org (accessed October 19, 2024).

data will be relevant to and bring benefit to their community (Goldenberg et al., 2011). Recent interviews with biobank leaders highlighted the importance of ethical stewardship of patient samples and data. Many registries emphasized their responsibility to serve as honest brokers of patient information and to engage patients as partners in the research use of their data. Goldenberg added that, for many biobank leaders, concerns about ethical stewardship also included ensuring the preservation and protection of data if a registry shuts down (Cadigan et al., 2024). He noted a policy opportunity to address “the lack of policies and governance structures that allow for sustainability if grant funding ends.”

REFLECTIONS

In the final session, moderators from the prior sessions highlighted observations made throughout the workshop and summarized opportunities discussed to advance progress in cancer surveillance based on speakers’ remarks (see Boxes 1 and 2). The data collection challenges experienced by registries were a recurring theme, Nadia Howlader, program director in the Statistical Research & Applications Branch of the Surveillance Research Program at NCI, said. These included difficulties in capturing comprehensive, real-world cancer data due to rapid advancements in treatment, variability in data sources, and limitations in current reporting systems. Many participants discussed motivating the submission of complete and accurate cancer data by linking reimbursement to cancer reporting.

Other approaches to improving data collection that were discussed included developing and promoting the use of standardized reporting protocols to improve consistency and accuracy of reporting across health care facilities and providing the training and resources needed for clinicians and others to comply with new reporting requirements, Howlader reported. Several participants suggested using registries to gather select data (e.g., recurrence and treatment) from a population-representative sample to support inferences about the total population.

Another main topic across the workshop was the need for improved data integration and interoperability of data sources, Richardson said. Many speakers discussed establishing national standards for data exchange; partnering across government, health care organizations, and technology vendors to foster widespread adoption of standards and platforms; and providing the funding and technical support needed to support upgrading of data systems and compliance with new standards, Richardson summarized.

Nekhlyudov highlighted the significant impact that state and governmental regulatory policies have on data access, linkages, and sharing. Multiple participants discussed revising state policies to facilitate data access, linkages, and sharing while still protecting patient privacy, Gomez said. Approaches

suggested by several speakers included leveraging privacy-preserving record linkages, federated data query models, the NAACCR Virtual Pooled Registry Cancer Linkage System,¹⁴¹ and partnering with patient advocates to raise awareness of the importance of registry data, she said.

Many speakers representing registries and registry data users described other challenges, such as level or reduced funding in the face of rising costs for cancer surveillance; delays in data availability for reporting and research; data gaps, particularly for recurrence and disease progression; challenges linking to other data sources; a lack of denominator data for analysis of trends; and balancing the need for data granularity with preserving patient privacy in small populations, Nekhlyudov summarized. Gomez said that, in considering how to capture more timely and complete surveillance data, many participants noted the need for a multipronged approach that incorporated human review. Suggestions included creating efficient data linkages to fill gaps (including patient surveys) and increasing funding to support the implementation of improved registry infrastructure and advanced technologies.

Yu summarized the range of methodological and technological challenges to be addressed, for example, data are frequently not in a structured format and require human curation; critical data are often missing; registry data are not available as timely as is needed; harmonized standards to enable data interoperability and sharing are lacking; use cases are needed to establish what data are really needed; the ideal balance across human curation, structured data capture, and generative AI is still to be defined; clinicians already are overburdened by documentation; and current payment models do not incentivize quality data capture. The need to develop the business case for organizational investment in infrastructure to support data entry, extraction, and registry population was also discussed, Yu said.

Shulman said that increased incentives will be needed to ensure structured entry of critical data. The challenge is to increase data timeliness and granularity without increasing the burden on clinicians and cancer registrars, he said. Several speakers discussed the need for federal agencies to align around data standards and data sharing policies and the need to harmonize state-level policies that impact data sharing and use and take a fresh look at data privacy laws that were passed decades ago, Shulman summarized. Many speakers also noted the importance of establishing public and private partnerships in addressing these challenges.

Regarding infrastructure and the evolving cancer surveillance workforce, several panelists discussed improving the pathology workflow and patient outcomes by using the CAP Cancer Protocols and leveraging opportunities for data automation, Palis said. Richardson highlighted several of the opportuni-

___________________

¹⁴¹ See https://apps.naaccr.org/vpr-cls (accessed December 18, 2024).

ties to advance progress noted by many speakers, including providing continuing education regarding the evolving landscape of data capture; recruitment and retention of ODS-Certified registry staff, marketing the profession to highlight the role of the ODS-Certified professional in data analytics; developing tools to assist registries with case finding to enable timelier reporting; and leveraging linkages to EHR data to provide a more complete picture of the patient’s cancer experience. A recurring theme mentioned by many participants was the need to ensure that every person with cancer is counted, Richardson said.

Bradley recapped other key issues identified by speakers. Multiple types and sources of cancer data exist, data capture and curation are labor intensive, and efforts are often duplicative, she said. Multiple panelists highlighted the need for real-time, high-quality registry data to support quality improvement and evaluation of care and payment models and inform timely treatment decisions by patients and families. Additional aspects of data linkages discussed included the costs of maintaining and updating them and the need for patient identifiers that could more readily facilitate linking patient data. Several participants raised the issue of sustainability of registries and legacy biospecimens, given the funding challenges facing some registries and tissue banks. The topic of false dichotomies that influence the development of ethical registry policies and data governance was also discussed.

Yabroff highlighted other policy opportunities discussed throughout the workshop. These included suggestions by some participants for mandated reporting of electronic cancer pathology data; funding of collaborations between registries, EHR vendors, and other partners to develop multipronged approaches for addressing data gaps (e.g., synoptic reporting and payor mandates); developing model legislation for states that enables data linkages and sharing and addresses NIH policy barriers to data sharing; and developing workforce education and training that incorporates “both high tech and high touch.”

There was also discussion of partnering with advocates to raise awareness of the need to allocate federal funding for a national patient identifier and educate the public and policy makers about the value of cancer registries, cancer surveillance, and sharing of cancer data in general. Nekhlyudov suggested the need to engage professional organizations to better inform clinicians.

In closing, Richardson said, “population-based cancer registries are a public good” and reiterated the need to remember that every data point in a registry represents a patient’s journey.

This page intentionally left blank.