“Good care starts with a good diagnosis,” said Hricak, who added that an estimated 60 to 70 percent of all decision making in health care is influenced by diagnostic testing (Dzau et al., 2016; Forsman, 1996). Stephen
4 See http://www.nationalacademies.org/hmd/Activities/Disease/NCPF/2018-FEB-12.aspx (accessed April 24, 2018).
Grubbs, vice president of clinical affairs at the American Society of Clinical Oncology (ASCO), agreed, adding: “When I order the right test, I [need to] know I am getting the right answer” to inform subsequent decision making in cancer care.
However, a number of workshop speakers described challenges to ensuring high-quality diagnosis in cancer care, such as
- Diagnostic uncertainty and the potential for diagnostic errors;
- The rapidly changing landscape of oncologic imaging, pathology, and precision oncology care;
- Disparities in patient access to oncologic imaging and pathology expertise and technologies; and
- Lack of adequate communication and collaboration among oncologists, pathologists, and radiologists in cancer care.
Despite the importance of an accurate diagnosis for subsequent care, several speakers noted that some patients with cancer experience diagnostic errors.5 Patty Spears, scientific research manager and patient advocate at the University of North Carolina at Chapel Hill Lineberger Comprehensive Cancer Center, stressed that patients often assume their diagnosis is accurate, and may not realize that their diagnosis could be incorrect, “making it especially important that there are systems in place to ensure such accuracy.” She added that reaching an accurate and complete diagnosis is critical so that the patients have “all the information that they need to make the correct decisions during their whole continuum of cancer care.”
Dana Siegal, director of patient safety services at Controlled Risk Insurance Company (CRICO) Strategies, reviewed findings from CRICO’s Comparative Benchmarking System, a national database of medical malpractice claims from captive and commercial insurers across the United States.6 Siegal reported that diagnostic-related claims are the third largest in volume and the first largest in dollars lost (based on indemnity and expenses) compared with other types of medical malpractice claims (e.g., surgical treatment or medical care). She added that the failure to diagnose cancer and delays in a cancer diagnosis represent approximately 30 percent of all diagnostic-related malpractice claims. Among all cancer-related malpractice cases, Siegal said that clinical judgment contributed to 72 percent of these cases, including the misinterpretation of diagnostic testing. For both the pathology and radiology specialties, misinterpretation of diagnostic testing was the primary cause of cancer-related malpractice claims. In the pathology specialty, additional contributing factors included insufficient communication to the ordering clinician and poor technique; in radiology,
5 According to Improving Diagnosis in Health Care, a diagnostic error is “the failure to (a) establish an accurate and timely explanation of the patient’s health problem(s) or (b) communicate that explanation to the patient” (NASEM, 2015, p. xiii).
6 See https://www.rmf.harvard.edu/Malpractice-Data/Annual-Benchmark-Reports#CF (accessed June 18, 2018).
Michael Cohen, interim chair and director of anatomic pathology at Wake Forest School of Medicine, reported that diagnostic error rates in anatomic pathology9 range from 1 percent to 6 percent, depending on factors such as anatomic site and the definition of diagnostic error that is used (Frable, 2006). He added that a study on mandatory review of surgical pathology reports found major disagreements (e.g., a change in diagnosis) in 2.3 percent of the cases; in 9 percent of cases, there were minor disagreements (e.g., a change in cancer staging, margin status, or other aspects that influence decision making in cancer care) (Manion et al., 2008).
Otis Brawley, chief medical officer of the American Cancer Society, added that a study of diagnostic concordance in breast pathology found that the overall agreement among individual pathologists and consensus panel reference diagnoses was approximately 75 percent (Elmore et al., 2015). Disagreement from the consensus panel reference diagnosis was greatest for pathologists who had lower weekly case volumes for interpretation, worked in smaller practices, or worked in non-academic settings. Brawley noted that these findings are likely because pathologists in high-volume settings or who practice in academia see more cases of breast cancer than pathologists who practice in community settings.
Hricak highlighted a number of studies that have found discrepancies in imaging interpretations between general radiologists and those with expertise in oncology (Coffey et al., 2017; Corrias et al., 2018; Hatzoglou et al., 2016; Horvat et al., 2018; Lakhman et al., 2016; Lorenzen et al., 2012; Lysack et al., 2013; Spivey et al., 2015). Depending on the study, second reviews by oncologic radiologists led to a change in treatment planning for a substantial portion of patients (13 to 53 percent). Hricak also cited a study showing that radiologists with a specialty in genitourinary systems who interpreted magnetic resonance imaging (MRI) added incremental value to the assessment of prostate cancer, while general radiologists did not (Mullerad et al., 2004). This finding “makes you realize the gaps in
7 Incidental findings are potentially abnormal results that are found unintentionally during diagnostic testing.
8 This paragraph was updated since the prepublication release.
9 Anatomic pathology addresses the microscopic examination of tissues, cells, or other solid specimens, sometimes with the aid of ancillary testing to detect specific genes or molecules, said Cohen.
knowledge and in the importance of special expertise in oncologic imaging,” Hricak said. Another study showed that when subspecialty radiologists read images, the turnaround time of radiology reports is significantly faster because they are more familiar with possible findings, and consequently, the time to diagnosis is quicker (Stern et al., 2018).
Radiology has been moving more toward subspecialization in the United States, Hricak noted. In 2012, a workforce survey found more than 35 percent of radiologists reported they were general radiologists; in 2017, fewer than 10 percent of radiologists reported they were generalists (Bluth et al., 2017). She pointed out that the majority of patients with cancer are treated in community settings, suggesting the need for strategies to extend oncologic imaging expertise from academic medical centers and cancer centers to community settings of care.
Brawley also noted that diagnostic testing has performance limitations (e.g., the potential of a diagnostic test to yield a false-positive or false-negative result), and interpretation of results can be subjective. “Uncertainty in medicine is just not appreciated and everybody thinks everything is binary,” he said, adding that many molecular tests for predicting cancer recurrence divide patient populations into low, middle, or high risk of recurrence, but the thresholds for these categories can be difficult to determine and may not always reflect clinical differences among the population subgroups. Brawley said there also can be differences in the performance of a diagnostic test within clinical trials that have highly selected populations compared to its performance in clinical practice settings.
Several speakers noted how rapidly the landscape of oncologic imaging and pathology is changing, as well as the growth in precision oncology care. “We are in the era of the convergence of life sciences, physical sciences, and engineering, and each is essential to proper diagnosis,” said Hricak. “There will not be precision oncology without precision diagnostics, and bioengineering and computational methods are essential for next-generation diagnostics,” she said, adding, “The time has never been better to understand the importance of diagnostics—which is pathology and imaging—in moving cancer care forward.”
Digitization is changing the nature of the fields of radiology and pathology, several speakers noted. Ronald Kline, medical officer at the Center for
Medicare & Medicaid Innovation at the Centers for Medicare & Medicaid Services (CMS), said the digitization of radiologic images has helped to enable patients to receive second opinions by oncologic radiologists: “The ease of moving digital images to different hospitals in different locations allows you to get better care,” Kline said. Although the field of pathology is not as far along as radiology, efforts to digitize pathology are also under way, said Richard Friedberg, professor and chair of the department of pathology at the University of Massachusetts Medical School at Baystate. In April 2017, the Food and Drug Administration (FDA) approved the first digital pathology system to review and interpret whole-slide images prepared from biopsied tissue.10
Jeremy Warner, associate professor of medicine and biomedical informatics at Vanderbilt University, said the goal of precision oncology care is to ensure the delivery of the right care to the right patient at the right time. Precision oncology therapies target specific abnormalities in a patient’s cancer, facilitated by diagnostic testing that characterizes that cancer. Friedberg noted that “our pathology world is changing dramatically by getting more molecular, [more] scientific, and more technical.” He added that morphology11—the physical characteristics of a patient’s biopsy tissue—was once the primary data element in pathology, but now it is just one of many data elements. He said there is much greater reliance on molecular testing in pathology, and with this testing, the number of distinct subtypes of various cancers has grown remarkably. Brawley pointed out that when he graduated from medical school 30 years ago, only two types of lung cancer had been discovered; now there are dozens of subtypes of lung cancer, including those defined by the genetic abnormalities that can be targeted by certain therapies. “Lung cancer has gotten incredibly complicated in a very short time, and the ability of the pathologists to do these tests with certainty is guiding how medical oncologists treat the disease,” Brawley said. He added that such testing is being used to help estimate the risk of recurrence of an individual’s cancer, such as with Oncotype DX12 or MammaPrint.13 “This understanding of the differences in genomics among cancers is giving us a
10 See https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm552742.htm (accessed April 24, 2018).
11 The science of the form and structure of organisms.
12 See http://www.genomichealth.com/en-US/oncotype_iq_products/oncotype_dx.aspx (accessed April 24, 2018).
13 See http://www.agendia.com/diagnostic-products/mammaprint.html (accessed April 24, 2018).
21st-century definition of cancer that depends not just on the biopsy but also on the genomics,” Brawley stressed.
Warner noted that the complexity of such testing is continuing to increase: In just the past few years, molecular testing has grown from the testing of individual genes or several genes to include multiplex omics panels with hundreds of genes. He added that the FDA approval of the cancer immunotherapy pembrolizumab14 for patients whose cancers harbor biomarkers indicating high microsatellite instability or mutations in mismatch repair genes, rather than on where a cancer originated in the body, signaled a new era of biomarker-based treatment assignment. Also in 2017, the FDA approved the first in vitro diagnostic test capable of detecting genetic mutations in 324 genes and 2 genomic signatures in any solid tumor type.15 Warner said this will prompt a “vastly increased uptake of these large gene panels in the coming year.”
Michael Becich, chair and distinguished university professor in the department of biomedical informatics at the University of Pittsburgh School of Medicine, noted that computational pathology—which he described as an approach to diagnosis that incorporates multiple sources of data (e.g., hematoxylin and eosin staining, immunohistochemistry, immunofluorescence, and genomic data), presents clinically actionable knowledge, and provides decision support for precision medicine—is also helping to redefine the field (Louis et al., 2014, 2016).
Oncologic imaging is also increasing in complexity, said Hricak and Fiona Fennessy, director of the Cancer Imaging Fellowship Program at Brigham and Women’s Hospital/Dana-Farber Cancer Institute. Imaging is used in oncology to detect tumors, determine their size, and determine whether lymph nodes have been affected. In addition, Hricak noted that expertise in oncologic imaging requires a thorough understanding of cancer biology, including the disease’s ability to affect multiple anatomic systems. Oncologic imaging also requires knowledge of treatment options, given the advent of molecular imaging methods that can be used to select specific treatments. “The oncologic report has to not only be accurate, but be clinically relevant and actionable,” Hricak said. She added that the need for oncology training and expertise is not widely recognized because “you
14 See https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm560040.htm (accessed April 24, 2018).
15 See https://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/ucm587273.htm (accessed April 24, 2018).
usually don’t realize what you are missing.” At the end of their fellowship training, Hricak said that fellows often remark that they did not realize how much knowledge oncologic imaging requires.
New technologies and techniques in imaging are enabling greater precision and additional insights from imaging. For example, some new imaging techniques can detect early, subtle tissue changes indicative of a response to treatment, Fennessy said. With the advent of functional imaging, also called physiologic imaging—such as positron emission tomography (PET) or various forms of MRIs—radiologists can also assess tumor biology based on uptake of glucose or blood flow, presence of specific receptors, or other molecular features of interest. “In the past, most X-ray or ultrasound imaging has been structural or anatomic, but some of nuclear medicine imaging is more and more becoming imaging that’s both anatomic as well as physiologic, which is a new frontier,” Brawley said. Hricak added that much of that physiologic imaging is still only used in clinical trials and is not yet approved for more general clinical use.
Hricak and Becich added that a number of efforts are under way to integrate machine learning and AI with imaging technologies. Becich said the field of radiomics attempts to extract and analyze large amounts of quantitative data from medical images, using algorithms that may identify features associated with disease characteristics that human interpretation may fail to appreciate (Kumar et al., 2012). Becich noted that the increased complexity associated with the introduction of computational pathology and radiomics contributes to cognitive overload among pathologists, radiologists, and oncologists. Hricak suggested that there is a need to facilitate the development of AI and machine learning to take over some relatively straightforward repetitive tasks, such as tumor measurement and summarizing pertinent history, in order to give radiologists the time to focus on more challenging diagnostic tasks.
Given the rapid growth in complexity in imaging and pathology, several speakers said that a major challenge is how to convey complex diagnostic testing results to clinicians who are unfamiliar with emerging technologies. William Sause, director of radiation oncology at Intermountain Healthcare, noted that with the “explosion of scientific knowledge, the clinician is overwhelmed with the amount of information that can be provided. Trying to sort through that [information] and provide a succinct meaningful interpretation of the data is truly a problem.” Spears noted that conveying this complex information to patients in an understandable way is also critically important.
Despite the promise of newer diagnostic technologies, Brawley also cautioned against premature adoption and dissemination before they have been adequately vetted. Brawley gave historic examples of premature adoption of technologies and their potential for harm. For example, when mammography was first introduced, Brawley said that women with non-invasive breast lesions unnecessarily received mastectomies; in addition, after introduction of the Pap smear,16 some women with mild cervical dysplasia received radical hysterectomies or radiation therapy because it was not well understood that the majority of these cervical abnormalities regress without treatment. “We need to be very careful in how we disseminate these new technologies into the community,” said Brawley. Cohen pointed out that the field of molecular diagnostics is a rapidly evolving area, and many companies are offering omics-based tests with proprietary algorithms whose results cannot be reproduced in the scientific literature. He added that there is an element of subjectivity in this testing, especially in how cut points are determined, but there is limited or no information about how these decisions were made because of the proprietary nature of the tests. This testing can add value, “but they need to be incorporated very carefully,” Cohen said.
Brawley said there are disparities in access to high-quality diagnostic services for people with cancer. To receive high-quality imaging results, the necessary components include well-trained radiologists and radiologic technologists, adequate time for them to do their job properly, up-to-date and well-maintained equipment, and access to necessary contrast agents, Brawley said. Similar components are needed for pathology, including high-quality reagents. But some of these components can be lacking in community hospitals, especially those considered safety-net hospitals that offer access to care regardless of ability to pay and thus have a substantial share of patients who have low incomes, are uninsured, or are otherwise vulnerable, Brawley stressed.
He noted that in a study of the quality of colon cancer surgery, ade-
16 Also called a Papanicolaou test, in which cells from the cervix are examined under a microscope for cervical cancer or cell changes that may lead to cervical cancer. See https://www.cancer.gov/publications/dictionaries/cancer-terms/def/pap-test (accessed June 18, 2018).
quate cancer staging (the evaluation of at least 12 lymph nodes) occurred in fewer than half of the patients, and people who were African American or Hispanic were more likely to have inadequate lymph node examination after resection for colon cancer (Rhoads et al., 2011). Brawley noted that this could have contributed to understaging in cancer diagnosis, and thus undertreatment, of some people if additional lymph nodes had been examined and found to contain cancer cells. Brawley added that inadequate examination of lymph nodes was associated with the hospital where care was received, suggesting that understaging was related to workload, or inadequate time for lymph node examination during surgery.
Both Brawley and Richard Schilsky, senior vice president and chief medical officer of ASCO, noted that the introduction of innovative technologies can exacerbate health disparities because of the lag in their adoption in cancer care facilities with limited finances. “Worsening health disparities might signal the onset of new technologies, because new technologies often have their most immediate uptake in populations most easily able to access them,” Schilsky said.
Spears added, “We want the outcome to be better for the patient, and as technologies move forward, we need to be intentional and include everybody, because we know that [all] patients are not treated the same. We need to even that playing field and make sure everything is accessible, because right now, where you live and who you are matters, and you don’t want to make [disparities] bigger by adding new technologies.”
Sause noted that the new technologies to improve cancer diagnosis hold a great deal of potential, but their costs may impede “our ability to do a lot of other things and we have to . . . come to grips with this.” Schilsky added that it will be important to consider the value of new technologies and not just their costs.
Several workshop participants said a major diagnostic challenge is the lack of collaboration and communication among pathologists, radiologists, and oncologists (NASEM, 2015). Even within a single specialty, the work culture and time pressures can discourage collaboration. For example, David Larson, vice chair for education and clinical operations in the department of radiology at Stanford University, said radiologists generally work in parallel, meaning they tend not to consult each other except within