Improving the Specificity of Autofluorescence Bronchoscopy

The problem of intraepithelial neoplasia is generally felt to involve the large central airways. Autofluorescence bronchoscopy (AFB) was developed in the 1990s as a method to detect airway dysplasia and carcinoma in situ. The premise behind this was that a large proportion of lung cancers were of squamous cell origin and arose in the central airways. Ideally, earlier detection of these could lead to earlier treatment and prevention of invasive disease. Despite the technology being available for over a decade, the clinical uptake of AFB has been limited. Although in some centers it is used routinely, in Canada only 4% of pulmonary trainees reported any exposure to AFB in their training. This number only climbed to 25% if an interventional pulmonologist was present at their center.1 Similarly, in a primarily US-based survey, only 10% of training programs offered training in AFB.2 Various reasons may account for this poor uptake, including lack of knowledge of the natural history of intraepithelial neoplasia, the absence of the demonstrated impact of AFB on hard clinical outcomes, and the poor specificity of this test. There is no doubt that autofluorescence increases the ability to detect the precursors to squamous cell carcinoma. Several randomized and nonrandomized trials have shown relative increases in sensitivity of 2-fold to 3-fold over white light bronchoscopy in detecting moderate/severe dysplasia or worse.3–5 Unfortunately, AFB suffers from a problem of poor specificity because of the multiple areas of false-positive autofluorescence. These are usually caused by inflammation, or occur at an earlier biopsy site, but an endoscopist is still required to perform a biopsy of the abnormal areas, the vast majority of which yield results of no clinical significance. The prevalence of preinvasive lesions (moderate dysplasia or worse) in biopsy samples taken from visually abnormal areas after inspection by white light and autofluorescence (positive predictive value) ranges from 4.3% to 20.3%.3–5 Even in the hands of an experienced endoscopist, this can prolong the procedure and lead to the use of more sedation. There are also the downstream effects of generating work for the pathologists and increasing costs to the healthcare system. The issue of specificity in autofluorescence studies can be clouded by the methods used by the researchers, in particular the number of control biopsies performed on normal areas. Specificity is calculated as the proportion of those who are disease-free, and who are labeled negative by the diagnostic test—in other words, the number of true-negative results divided by the sum of the true-negative and false-positive results. As the number of true-negative results in studies of AFB is largely determined by the number of control biopsies of visually normal areas, performing more control biopsies increases the number of true-negative results and alters the proportion of true-negatives to false-positives, thereby increasing the specificity. This is the main reason behind the wide range of specificities reported for AFB, from 4% to 94% in various studies.6 Regardless of the true specificity of AFB, the addition of autofluorescence to white light bronchoscopy has been shown to generally decrease the specificity compared with white light alone. In a summary of the literature on AFB published as part of the ACCP guidelines on bronchial intraepithelial neoplasia, 13 of 15 studies with specificity data showed the specificity of AFB to be lower than with white light alone.6 This decrease in the relative specificity of AFB is the result of the large number of false-positive areas of autofluorescence or low positive predictive value. This is the real problem of the specificity of AFB. In this issue of the Journal, Reinders et al describe a new technology aimed at improving the specificity of AFB by reducing the number of false-positives.7 They describe using noncontact spectral measurements to help predict which lesions may represent a high-grade lesion (severe dysplasia, carcinoma in situ, or invasive cancer). Spectral measurements were taken of areas of abnormal autofluorescence, which were then used as surrogates or estimates of certain tissue vascular properties, the rational being that there are differences in these vascular properties between high-grade and low-grade lesions. With this approach, receiver operating characteristic curves were then calculated to evaluate the discriminatory ability of these spectral properties. Several tissue vascular properties were evaluated, including blood volume, fraction of deoxygenated blood, and blood volume of deoxygenated blood. The area under the curve for the latter property showed a value of 0.83 comparing high-grade with low-grade lesions. The vast majority (66 of 74) of the high-grade lesions biopsied were, however, invasive malignancies, and one has to wonder whether autofluorescence or additional spectral measurements would have been required at all to detect these. If a patient has gross endobronchial disease, the decision to biopsy is clear. Nevertheless, when the assessment is limited to patients with carcinoma in situ or severe dysplasia only, the results are still impressive, with an area under the curve of 0.87. However, with only 8 biopsies in this category, more data are needed to determine whether the results would hold up when looking at mainly dysplastic lesions or a screening population. In addition, as the spectral measurement results were not used in the decision to proceed with biopsy, calculation of specificity and positive predictive value is not possible, but these data can be used in future studies to help the bronchoscopist decide whether a lesion is likely to represent a false-positive or one that requires biopsy to exclude a high-risk lesion. A similar attempt at improving specificity using spectral data has been reported by Edell et al8 using fluorescence-reflectance bronchoscopy. An area under the curve of 0.735 for moderate dysplasia or worse was calculated in a true screening population. Numerous other competing and possibly complementary technologies are also being developed to address the issue of specificity, including narrow band imaging (NBI), high-resolution videobronchoscopy, optical coherence tomography (OCT), Raman spectroscopy, and confocal microscopy. NBI is already available commercially, but suffers from the same problem of still requiring the gold standard tissue biopsy for confirmation of a neoplastic lesion. NBI, similar to AFB, has been shown to have increased sensitivity in detecting dysplasia compared with white light alone,9 but no direct comparisons to AFB have been published. OCT, Raman spectroscopy, and confocal microscopy are still in the early phases of development, and are not yet commercially available. OCT has recently been shown to detect differences in epithelial thickness between carcinoma in situ and invasive cancer when compared with dysplasia.10 Confocal microscopy has been shown in gastrointestinal endoscopy to provide high-resolution images of mucosal histology, and some studies have shown up to 99% accuracy in predicting neoplastic changes.11 This latter technique may offer the most potential for discriminating between high-grade and low-grade lesions, but at this point in time true endomicroscopes have only been incorporated into larger gastroscopes and colonoscopes, and not bronchoscopes. Certainly, the addition of noncontact spectral measurements, as shown by Reinders et al in this study, may represent an improvement to AFB. Whether this will lead to changes in clinical practice or screening for lung cancer remains to be determined. Additional studies in high-risk patients undergoing screening rather than diagnostic bronchoscopy will need to be carried out before the true impact of this technology can be assessed. Along the way, there will be significant evolution from competing technologies. Which technology will win out remains to be determined, but the true benefit of these technologies will be realized when the need for biopsy can be significantly reduced or avoided altogether.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.