Secondary Logo

Journal Logo

EDITORIALS

The B Reader Program, Silicosis, and Physician Workload Management

A Niche for AI Technologies

Gowda, Vrushab BS; Cheng, Glen MD, JD, MPH; Saito, Kenji MD, JD

Author Information
Journal of Occupational and Environmental Medicine: July 2021 - Volume 63 - Issue 7 - p e471-e473
doi: 10.1097/JOM.0000000000002271
  • Free

Artificial intelligence (AI) technologies hold considerable potential as healthcare tools. Their pattern recognition features, in conjunction with their speed and increasingly high-fidelity analysis, promise to redefine conventional diagnostic paradigms. These characteristics readily lend themselves to labor-intensive pulmonary imaging; a number of AI products have already received FDA clearance for this purpose and are currently marketed in the United States.1 These new technologies may find particular application filling an old regulatory compliance need – the B Reader Program. Administered by the Centers for Disease Control and Prevention's (CDC) National Institute for Occupational Safety and Health (NIOSH) for over half a century, the B Reader Program is tasked with monitoring pneumoconiosis among exposed industrial workers. However, it faces long-declining physician participation and a recently broadened mandate – two factors which may render compliance increasingly difficult. Moreover, a number of emerging reports have drawn attention to potential financial conflicts of interest (COI) which may bias independent B Readers toward overreporting positive findings, and employer-retained physicians toward underreporting them.2 Emerging AI technologies can offer a solution to both COI and supply-demand mismatch. Leveraging their innate machine learning capabilities to reduce physician workload, efficiently process large volumes of imaging data, and lower rates of human error, AI-enabled radiology tools may breathe new life into an ailing program.

FROM PATHOLOGY TO POLICY

Pneumoconiosis comprises a series of interstitial lung diseases which ultimately result in lung parenchymal fibrosis, often after protracted occupational or environmental exposure.3 At the cellular level, it is brought on by the chronic inhalation of airborne particulates, prompting alveolar macrophages to engulf foreign material and induce fibrosis.4 The resultant changes to lung architecture are progressive, irreversible, and dose-dependent. Over time, pneumoconiosis may produce a slew of downstream pathologies, including lung cancer, increased susceptibility to tuberculosis, and collagen vascular disease.5–8 Pneumoconiosis is generally classified according to the type of particulate matter implicated; chief among them are silicosis, asbestosis, and coal workers’ pneumoconiosis (CWP), also known as “black lung disease.”9

To mitigate the untoward effects of CWP, Congress passed the Federal Coal Mine Health and Safety Act in 1969, which among other actions directed the Department of Health and Human Services (then the Department of Health, Education, and Welfare) to establish a routine screening radiography program for coal workers.10 The Coal Workers’ Health Surveillance Program, administered by NIOSH, was developed to this end.

The NIOSH B Reader system finds its origins in this program, which aimed to identify, certify, and engage qualified physicians in a national CWP screening endeavor.11 As its name implies, “B” Readers were initially called upon to confirm suspected features of pneumoconiosis in a second-opinion capacity.12 The program today dispenses with the preceding “A” Readers and instead relies upon a cadre of highly trained B Readers to independently establish findings of CWP.13 Since 1976, the program has sought to minimize interobserver variability through a rigorous certification process, evaluating candidates on the basis of their ability to categorize images according to the International Labour Organization (ILO) International Classification of Radiographs of Pneumoconioses.14

WINDS OF CHANGE

In 2018, OSHA (the Occupational Safety and Health Administration) issued new regulations aimed at protecting individuals at risk of silicosis, specifying that employers have a “duty to exercise reasonable care to prevent and detect violations of the silica standard.”15 Just as NIOSH does with CWP, OSHA requires that at-risk individuals receive triennial chest X-rays interpreted specifically by a NIOSH-certified B Reader for screening and classification.16 This well-intentioned policy bears profound implications.

An estimated 2.3 million workers face routine occupational exposure to respirable crystalline silica, a figure dwarfing the some 50,000 employed in coal mining.17 Crystalline silica is far more pervasive than coal dust and is implicated across a range of industries, including building construction, maritime manufacture, glass production, mining, and agriculture.18 Having issued these new guidelines, OSHA vastly expanded the population subject to screening, just as the pool of B Readers has continued to diminish over the years. The mean physician age of qualified B Readers has soared into the sixties while pass rates on the certification exam have consistently deteriorated since at least the mid-1980s.19 From a peak of 750 in 1993, it has exhibited secular decline to its current level of 209 according to latest NIOSH statistics.20 The reasons for this decline remain to be determined. No dedicated studies interrogating the origins of this trend have been conducted, although some researchers have postulated that changing educational patterns, a broader move from plain radiography to CT imaging, and economic shifts within the affected industries are at least partly to blame.19 In addition, a recent study has documented endemic subjectivity among existing B Readers’ reports, which in turn may be related to financial COI; those read by federally funded independent B Readers tended to overreport pneumoconiosis findings, while those interpreted by employer-hired physicians were more likely to underreport them.21

Twelve states altogether lack B Readers, including Arizona, Hawaii, Idaho, Iowa, Maine, Missouri, Nebraska, New Hampshire, North Dakota, South Dakota, Vermont, and Wyoming.19 This is significant; unlike coal mining, the occupations implicating respirable crystalline silica exposure are not geographically concentrated. The totality of these trends would imply a potentially overwhelming burden upon current B Readers, who are mostly board-certified radiologists, pulmonologists, internal medicine, or occupational medicine physicians often engaged in full-time practice. Absent a change in federal screening guidelines, an increase in the number of qualified B Readers, or marked reduction in the occupational exposure to inhaled silica dust, this strain can only be expected to worsen over time.

A TECHNOLOGY-ENABLED SOLUTION

Enter AI-enabled radiology tools. These products may find myriad applications in screening for silicosis and other pneumoconioses, from workflow management and triaging to patent computer-aided detection (CADe).22 A number of existing FDA-cleared products offering each of these functions may be appropriate to the task. Perhaps more intriguingly, Australian, Japanese, and Chinese researchers have independently developed CADe algorithms specifically for pneumoconiosis detection and monitoring.23–25 Each applied deep learning techniques to chest X-rays to classify cases according to ILO criteria. These systems demonstrated superior sensitivity and specificity to human physicians, while diminishing the high interobserver variability which has previously plagued the B Reader Program.26 This can moreover mitigate the effect of potential COI by offering a standardized readout across interpreting providers. Although the B Reader Program is currently limited to analysis of plain film radiography, ultra-low-dose CT may offer a more precise modality for pneumoconiosis surveillance. This would limit radiation exposure while retaining the benefits of high-resolution imaging, which facilitate application of CADe techniques.27 To this end, Fujifilm and Kyoto University have developed a CT-based AI tool for pneumoconiosis detection, which they are currently commercializing and plan to market in 2021.28

The federal government can further the domestic research, development, and early deployment of similar products. This can be accomplished through a pilot program, spearheaded by either by the Department of Labor (the parent organization of OSHA) or the CDC (which oversees NIOSH). This would not represent an unprecedented move for either. In particular, the CDC has been actively engaged in technology demonstration projects over a span of years, and NIOSH has operated the Center for Occupational Robotics Research (CORR) since 2017 to conduct feasibility studies on automation technologies’ potential to reduce workplace injuries.29 Together with the National Science Foundation (NSF), CORR routinely provides grant funding to academic partners in support of related research.30 Entities seeking to develop AI-enhanced silicosis screening technologies could obtain funding via a joint CORR/NSF grant, and then engage with the NIOSH Future of Work Initiative, which explicitly names artificial intelligence applications among its key priorities.31

Through this avenue, NIOSH may endow promising technologies with the necessary support for further testing, early deployment, and eventual scaling. A prospective pilot program could bear two foci: (1) supporting existing FDA-cleared CADe products as they expand their approved indications to include pneumoconiosis and (2) furthering the development of novel platforms, such as those in the studies referenced above, for clinical use in the United States. Through either pathway, programs demonstrating high-fidelity pneumoconiosis detection can find integration within the B Reader Program itself. To be clear, these AI tools would not replace human B Readers. As with other CADe products, they would flag potentially concerning features to the interpreting physician, who remains the ultimate arbiter of clinical decision making.

After suitable programs receive FDA clearance, NIOSH can amend its B Reader guidelines to consider use of AI tools to assist human physicians. It has updated its regulations to keep pace with changing technologies once before; in 2012, NIOSH moved to permit submission of digitally acquired X-rays, as opposed to film radiographs, and issued specifications to that effect.32 This much-needed update provides a template for future NIOSH regulatory sanction of AI tools in pneumoconiosis screening initiatives. NIOSH should moreover consider utilizing this avenue to extend the B Reader Program to ultra-low-dose CT, so as to enhance diagnostic precision and offer a modality more suitable for CADe.

Of course, AI is no panacea. These technologies often require DR as opposed to previous generations of analog or computed radiography (CR) techniques; all are permissible under the B Reader Program.33 Moreover, pneumoconiosis screening is performed in a variety of clinical resource settings using a range of technologies without a uniform set of protocols. This heterogeneity may result in inconsistent image quality, thereby posing limitations to the widespread implementation of AI tools. NIOSH would be well advised to begin its pilot at institutions leveraging DR and offer a series of image acquisition guidelines formulated in conjunction with the AI developer.

CONCLUSION

Trends in the declining pool of participating physician B Readers, together with an expanded patient population requiring screening, high degree of interpretive subjectivity, and potential physician COI, will significantly complicate employer compliance with the NIOSH B Reader Program in the near future. Utilized appropriately, AI-enabled radiology tools offer considerable promise as a remedy. They stand to fill the need for regulatory compliance in pneumoconiosis screening, while offering a labor-saving solution to physician workflow issues and enhancing patient safety. NIOSH should contemplate opportunities to facilitate the development of AI-enabled radiology applications, their deployment to augment human B Reader performance, and ultimately further the ends of OSHA's silicosis screening objectives. NIOSH can capitalize upon an existing technology development infrastructure to foster a pilot program, in cooperation with OSHA, the NSF, and university partners. Its lessons may be scaled through the B Reader Program, building upon technical advances in the field and informing future policy.

REFERENCES

1. Benjamens S, Dhunnoo P, Meskó B. The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database. NPJ Digit Med 2020; 3:118.
2. Friedman LS, De S, Almberg KS, Cohen RA. Association between financial conflicts of interest and ILO classifications for black lung disease. Ann Am Thorac Soc 2021; Published online March 29.
3. Wang XR, Christiani DC. Respiratory symptoms and functional status in workers exposed to silica, asbestos, and coal mine dusts. J Occup Environ Med 2000; 42:1076–1084.
4. Mossman BT, Churg A. Mechanisms in the pathogenesis of asbestosis and silicosis. Am J Respir Crit Care Med 1998; 157 (Pt 1):1666–1680.
5. Pollard KM. Silica, silicosis, and autoimmunity. Front Immunol 2016; 7:97.
6. Sato T, Shimosato T, Klinman DM. Silicosis and lung cancer: current perspectives. Lung Cancer (Auckl) 2018; 9:91–101.
7. Rees D, Murray J. Silica, silicosis and tuberculosis. Int J Tuberc Lung Dis 2007; 11:474–484.
8. Makol A, Reilly MJ, Rosenman KD. Prevalence of connective tissue disease in silicosis (1985–2006)—a report from the state of Michigan surveillance system for silicosis. Am J Ind Med 2011; 54:255–262.
9. The National Institute for Occupational Safety and Health. Pneumoconioses. 2011. Available at: https://www.cdc.gov/niosh/topics/pneumoconioses/. Accessed February 12, 2021.
10. Pub. L. 91–173 (91st Congress).
11. Cummings KJ, Johns DO, Mazurek JM, Hearl FJ, Weissman DN. NIOSH's respiratory health division: 50 years of science and service. Arch Environ Occup Health 2019; 74:15–29.
12. Halldin CN, Hale JM, Blackley DJ, Laney AS. Radiographic features of importance in the National Institute for Occupational Safety and Health-administered Coal Workers’ Health Surveillance Program: characterising the use of the ‘other symbols’. BMJ Open 2017; 7:e015876.
13. Halldin CN, Hale JM, Weissman DN, et al. The National Institute for Occupational Safety and Health B Reader Certification Program – an update report (1987 to 2018) and future directions. J Occup Environ Med 2019; 61:1045–1051.
14. Halldin CN, Petsonk EL, Laney AS. Validation of the international labour office digitized standard images for recognition and classification of radiographs of pneumoconiosis. Acad Radiol 2014; 21:305–311. DOI 10.1016/j.acra.2013.11.019.
15. 81 Fed. Reg. 16285.
16. 29 CFR §1926.1153.
17. Occupational Health and Safety Administration. Silica, Crystalline. 2021. Available at: https://www.osha.gov/silica-crystalline. Accessed February 12, 2021.
18. Yassin A, Yebesi F, Tingle R. Occupational exposure to crystalline silica dust in the United States, 1988–2003. Environ Health Perspect 2005; 113:255–260.
19. Cite to Halldin et al. 2019, supra.
20. The National Institute for Occupational Safety and Health. NIOSH Certified B Readers. 2020. Available at: https://wwwn.cdc.gov/niosh-rhd/cwhsp/ReaderList.aspx. Accessed February 12, 2021.
21. Cite to Friedman et al. 2021, supra.
22. Dikici E, Bigelow M, Prevedello LM, White RD, Erdal BS. Integrating AI into radiology workflow: levels of research, production, and feedback maturity. J Med Imaging (Bellingham) 2020; 7:016502.
23. Zhang L, Rong R, Li Q, et al. A deep learning-based model for screening and staging pneumoconiosis. Sci Rep 2021; 11:2201.
24. Wang X, Yu J, Zhu Q, et al. Potential of deep learning in assessing pneumoconiosis depicted on digital chest radiography. Occup Environ Med 2020; 77:597–602.
25. Okumura E, Kawashita I, Ishida T. Computerized classification of pneumoconiosis on digital chest radiography artificial neural network with three stages. J Digit Imaging 2017; 30:413–426.
26. Cite to Fed. Reg., supra.
27. Kim Y, Kim YK, Lee BE, et al. Ultra-low-dose ct of the thorax using iterative reconstruction: evaluation of image quality and radiation dose reduction. Am J Roentgenol 2015; 204:1197–1202.
28. Fujifilm. Fujifilm and Kyoto University Jointly Developed AI-Based Diagnostic Support Technology for Interstitial Pneumonia. 2019. Available at: https://www.fujifilm.com/news/n190409.html. Accessed February 12, 2021.
29. The National Institute for Occupational Safety and Health. Center for Occupational Robotics Research. DHHS (NIOSH) Publication 2019-162 (2019).
30. The National Institute for Occupational Safety and Health. Robotics. 2020. Available at: https://www.cdc.gov/niosh/topics/robotics/. Accessed February 12, 2021.
31. The National Institute for Occupational Safety and Health. Future of Work Initiative. 2020. Available at: https://www.cdc.gov/niosh/topics/future-of-work/default.html. Accessed February 12, 2021.
32. 77 Fed. Reg. 56717.
33. Andriole KP. Productivity and cost assessment of computed radiography, digital radiography, and screen-film for outpatient chest examinations. J Digit Imaging 2002; 15:161–169. DOI 10.1007/s10278-002-0026-3.
Copyright © 2021 American College of Occupational and Environmental Medicine