The digital–physical divide for pathology research

Abstract

In 2018 we launched the Network of Enigmatic Exceptional Responders, a patient-directed network to study individuals who had an exceptional response to cancer treatment.1Perakslis ED Kohane IS Treating the enigmatic “exceptional responders” as patients with undiagnosed diseases.Sci Transl Med. 2016; 8340ed8 Crossref Scopus (2) Google Scholar Recruitment, which began in late 2018 and is ongoing, was carried out though a national public radio story and social media posts, after which participants underwent genomic, T-cell-receptor, and microbiome analysis and completed questionnaires, and their health records were obtained from their respective institutions. However, when we attempted to obtain images of microscopic and routinely stained tissue, tissue sampled from excised tumours, or biopsy samples from tumours, many of our requests were unfulfilled—including more than 90% of our requests for tumour samples. The stark difference between obtaining tissue images or samples from hospitals and obtaining other types of patient data will delay advances in multimodal digital medicine. For example, the performance of large language models in making diagnoses on the basis of text contained in electronic health records could be improved if a sufficient quantity of pathology images, characterised with the assistance of artificial intelligence, was routinely available for the same patients. Pathology samples are not inherently digital, and most health-care institutions have not invested in the digitisation of these samples on the same scale as for radiology images. A purely digital approach to pathology will continue to be unlikely, if only because the preservation of tissue samples—which has proven to be valuable for investigating mechanisms of disease—remains a substantial obligation. Further, the large sums of money invested in integrating electronic health records with department information systems have not been widely extended to pathology, and current pathology practices benefit from a reduced need to make extensive, time-consuming electronic annotations to the health record. However, the opportunity cost of not digitising images of available tissue and integrating them with other clinical data—as is increasingly done in radiology—has slowed the use of such images for discovery research. In an era in which we are redefining disease taxonomies using multimodal techniques, this digital divide could slow advances in precision medicine. The standard and ubiquitous staining of tissue with haematoxylin and eosin already provides rich information on cellular and tissue architecture and component cell types. With work over the past 5–10 years in diagnostic pathology, histology images can also be used to predict mutational status in various cancers, including colorectal and breast cancers.2Van der Laak J Litjens G Ciompi F Deep learning in histopathology: the path to the clinic.Nat Med. 2021; 27: 775-784Crossref PubMed Scopus (258) Google Scholar Immunohistochemistry stains using antibodies against specific proteins, depending on the tissue type or the diagnosis, are now also used routinely. Sharing these stained images, linked to clinical records at a health-care-system scale, will enable a major acceleration in discovery research. Moreover, the ability of machine learning to predict the result of immunostaining or mutational status from standard haematoxylin and eosin slides—without needing to conduct the relevant staining or sequencing—is rapidly increasing.3Ektefaie Y Yuan W Dillon DA et al.Integrative multiomics—histopathology analysis for breast cancer classification.NPJ Breast Cancer. 2021; 7: 147Crossref PubMed Scopus (15) Google Scholar, 4Liechty B Xu Z Zhang Z et al.Machine learning can aid in prediction of IDH mutation from H&E-stained histology slides in infiltrating gliomas.Sci Rep. 2022; 1222623 Crossref Scopus (3) Google Scholar Obtaining these pathology images in digital format requires the use of specialised scanning equipment, but the major costs lie in changes to the current workflow, staffing, computational infrastructure, and regulatory validation—covering all stages of the process from obtaining a specimen to clinical reporting and archiving. Just as the Human Genome Project benefited from international sequence sharing, it is at least plausible that sharing these image resources at scale will accelerate our understanding of human pathology. A novel channel for the sharing of pathology images at the national scale has emerged in the USA as a result of the 21st Century Cures Act, which called for a roadmap to sharing data with patients. This roadmap is the United States Core Data for Interoperability (USCDI),5Gordon WJ Gottlieb D Kreda D Mandel JC Mandl KD Kohane IS Patient-led data sharing for clinical bioinformatics research: USCDI and beyond.J Am Med Inform Assoc. 2021; 28: 2298-2300Crossref PubMed Scopus (6) Google Scholar which, since October, 2021, has included an extensive and growing list of data elements beyond the simple medications, laboratory studies, procedures, and diagnoses that were initially required. The USCDI requires electronic data pertaining to individual patients to be made available to them. Were pathology images to be created, they would be shareable under USCDI, with patient consent, for uses including multi-institutional studies.6Pallua JD Brunner A Zelger B Schirmer M Haybaeck J The future of pathology is digital.Pathol Res Pract. 2020; 216153040 Crossref Scopus (54) Google Scholar Sharing tissue is a far more challenging proposition than sharing images. In the short term, tissue is required for non-routine staining analyses and for destructive cellular analyses such as DNA single-cell sequencing and full-tissue mass spectrometry analysis. Requirement for physical storage and curation is therefore necessary. However, the regulatory requirement for tissue storage is time-limited (eg, only 10 years in the USA), so the information contained in samples for which images are not digitised is lost forever for millions of patients every year. A patient's legal and ethical right to access their own data has long been recognised if not operationalised.7Institute of Medicine of the National AcademiesBeyond the HIPAA Privacy Rule: enhancing privacy, improving health through research. National Academies Press, 2009Google Scholar The equivalent understanding and agreement for pathology samples is absent. Health-care professionals, particularly pathologists, are designated as stewards of these samples. A patient does not have a right to donate tissue specimens. The ability to donate such specimens depends on the willingness of the stewards to share them, and requires stewards to take time out of their clinical schedule to go through an extensive process. Even with a nominal fee, compensating these already-burdened clinicians for lost revenue is challenging. Moreover, these requests to pathologists to share specimens between institutions are in addition to requests from researchers within the same health-care system. The absence of standard policy and legislation means that suitable consent processes are not consistently in place, adding to the uncertainty and administrative costs of sharing these samples for research. Aside from policy, merely finding the correct samples—preserved in the appropriate manner—for specific tissues among different institutions, and then linking these samples to associated clinical and genomic annotations, can take months of inquiries and add a major efficiency barrier to such tissue studies. Perhaps most frustrating are the challenges of maintaining a culture of sharing tissue samples for research purposes. Even within institutions, investigators who have been funded to create biobanks often treat them as their own personal libraries. Such actions go against the wishes and expectations of patients, who have consented to the use of their tissue samples for research. Finally, incomplete consent and a lack of acknowledgement of patients’ wishes in terms of tissue sharing and autonomy often harm vulnerable and disenfranchised communities and serve to undermine confidence in research institutions and the research agenda, most visibly in the case of Henrietta Lacks.8Wolinetz CD Collins FS Recognition of research participants' need for autonomy: remembering the legacy of Henrietta Lacks.JAMA. 2020; 324: 1027-1028Crossref PubMed Scopus (41) Google Scholar, 9Javitt G Why not take all of me: reflections on the immortal life of Henrietta Lacks and the status of participants in research using human specimens.Minn J Law Sci Technol. 2010; 11: 713Google Scholar In conclusion, although the challenges of sharing images and tissues in pathology are large and systemic, several steps can be taken in the short term (table). Solutions require realigning incentives—both academic and financial. We have used some of these short-term solutions in our own work on the Network of Enigmatic Exceptional Responders study, when obtaining tissue samples from patients who had an exceptional response to cancer treatment, and are optimistic that we can make progress in obtaining access to samples in the coming year.1Perakslis ED Kohane IS Treating the enigmatic “exceptional responders” as patients with undiagnosed diseases.Sci Transl Med. 2016; 8340ed8 Crossref Scopus (2) Google Scholar Nevertheless, longer-term solutions are required to achieve broader gains for biomedical research.TableChallenges and potential solutions for the digitisation and sharing of pathology images and tissue samplesShort-term solutionsLonger-term solutionsSharing pathology slide imagesEnsure pathology images obtained for clinical care are included in the USCDI dataset for sharing. Fund imaging technicians or pathologists for their additional work to digitise sample images of high value.Require full digital imaging of all slides obtained during clinical care as conditions of reimbursement and accreditation. Make all clinical and research pathology images available digitally for distribution under appropriate governance and patient consent. Provide academic credit to those who generate and annotate those images.Sharing tissue mechanicsCreate cross-institution, tissue-finding information tools and online templates for patients to consent to the use of their samples for research.Update standardised consent agreements to include broader sharing of samples. Store samples in a central biobank (community-based, regional, or national) with suitable process automation and scale.Sharing tissue cultureEducate patients about the value of sharing their samples for research. Engage pathology societies and departmental leadership to promote sharing culture.Develop data authorships and include efforts made towards sample sharing in faculty promotion criteria.10Lo B DeMets DL Incentives for clinical trialists to share data.N Engl J Med. 2016; 375: 1112-1115Crossref PubMed Scopus (36) Google Scholar Increase reimbursement fees for academic centres and departments who participate most in the sharing of samples.USCDI=United States Core Data for Interoperability. Open table in a new tab USCDI=United States Core Data for Interoperability. We declare no competing interests.

Ähnliche Arbeiten