Reply

Our international, randomized controlled trial showed that the addition of low-cost high-resolution microendoscopy (HRME) to standard Lugol’s chromoendoscopy (LCE) increased diagnostic yield and saved unnecessary biopsies in esophageal squamous cell neoplasia (ESCN) detection across global settings¹; however, specificity was significantly lower among novice microendoscopists who tended to overcall neoplasia resulting in false positive results.

In response to He et al² regarding follow-up of missed lesions, all six subjects (0.7%) with neoplasia missed by the endoscopist on HRME underwent biopsy of the lesion regardless of the HRME read per standard of care. The HRME read as neoplastic or non-neoplastic was used for calculation of diagnostic yield and did not impact the standard of care which required biopsy of LCE voiding areas. As such, no patients were harmed. As mentioned, two of these subjects had moderate (borderline) dysplasia, and three were missed by novices.

We agree that implementation of HRME in under-resourced global settings without experts may effectively be supported by computer-assisted detection (CADe) algorithms. These tools help endoscopists interpret “optical biopsy” images and have the potential to standardize HRME interpretation, reduce interobserver variability, and overcome lack of experience (microendoscopy novices). We are currently conducting studies to determine the impact of a CADe algorithm on ESCN detection. For the current study, where qualitative image interpretation was done, we trained all the endoscopists on standardized criteria evaluating nuclear size, density, and pleomorphism, and all study endoscopists were tested on image interpretation before study commencement.

Regarding proflavine, this agent has been used for fluorescence imaging in the European, Asian, and Australian gastrointestinal literature³ and is monitored as an Investigational New Drug by the U.S. Food and Drug Administration. In >1500 subjects imaged to date, no adverse events nor an increased progression rate have been noted in follow-up when compared to controls. Proflavine hemisulfate has been clinically used as an antibacterial agent in neonatal care. Triple dye, a combination of brilliant green, proflavine hemisulfate, and gentian violet, is routinely used as a topical antibacterial agent on the umbilical stump of newborn babies, with studies categorizing toxicity as rare.^4,5 The concentration of proflavine used for our study (0.01% w/v) is lower than that of proflavine component in commercial triple dye, (0.11% w/v; Kerr Triple Dye, VistaPharm) used in neonatal care. Likewise, investigational in vivo human studies of confocal microscopy for gastrointestinal cancer detection used topical acriflavine at 0.05% concentration.⁶

In response to Zuo and Ji,⁷ per institutional review board requirements and standard of care, all sites imaged with HRME were required to be biopsied, regardless of the endoscopist HRME read. Therefore, any decision about whether to biopsy or not was theoretical, and decision to biopsy was not affected by which group the subject was randomized to. Furthermore, we performed per-biopsy and per-patient analysis to account for the multifocal nature of ESCN.

Our team showed the addition of HRME was more cost-effective than screening for ESCN using LCE alone in high-risk regions.⁸ Although the cost-effectiveness of HRME may diminish in global settings with lower biopsy costs, the benefit and reusability of HRME remains significant in many regions with limited access to pathologists or in low-resource settings where pathology costs exceed endoscopy costs.

Lastly, with regards to data provided by Zuo and Ji,⁷ because there are no published data on this prototype microendoscope, the comparison may be “apples to oranges.” That is, it cannot be definitively determined whether their device shares the technical specifications, image resolution, or performance capabilities of the HRME used in our study. Image resolution and optical design can substantially impact image quality and performance. To compare diagnostic performance across devices, including CADe algorithm performance, further data on hardware specifications and software training and validation are needed. Transparent reporting of these parameters is essential for any meaningful cross-study comparison.