Histology Verification Demonstrates That Biospectroscopy Analysis of Cervical Cytology Identifies Underlying Disease More Accurately than Conventional Screening : Removing the Confounder of Discordance

Gajjar, Ketan and Ahmadzai, Abdullah and Valasoulis, George and Trevisan, Julio and Founta, Christina and Nasioutziki, Maria and Loufopoulos, Aristotelis and Kyrgiou, Maria and Stasinou, Sofia Melina and Karakitsos, Petros and Paraskevaidis, Evangelos and Da Gama-Rose, Bianca and Martin-Hirsch, Pierre Leonard and Martin, Francis Luke (2014) Histology Verification Demonstrates That Biospectroscopy Analysis of Cervical Cytology Identifies Underlying Disease More Accurately than Conventional Screening : Removing the Confounder of Discordance. PLoS ONE, 9 (1): e82416. ISSN 1932-6203

Abstract

Background Subjective visual assessment of cervical cytology is flawed, and this can manifest itself by inter- and intra-observer variability resulting ultimately in the degree of discordance in the grading categorisation of samples in screening vs. representative histology. Biospectroscopy methods have been suggested as sensor-based tools that can deliver objective assessments of cytology. However, studies to date have been apparently flawed by a corresponding lack of diagnostic efficiency when samples have previously been classed using cytology screening. This raises the question as to whether categorisation of cervical cytology based on imperfect conventional screening reduces the diagnostic accuracy of biospectroscopy approaches; are these latter methods more accurate and diagnose underlying disease? The purpose of this study was to compare the objective accuracy of infrared (IR) spectroscopy of cervical cytology samples using conventional cytology vs. histology-based categorisation. Methods Within a typical clinical setting, a total of n = 322 liquid-based cytology samples were collected immediately before biopsy. Of these, it was possible to acquire subsequent histology for n = 154. Cytology samples were categorised according to conventional screening methods and subsequently interrogated employing attenuated total reflection Fourier-transform IR (ATR-FTIR) spectroscopy. IR spectra were pre-processed and analysed using linear discriminant analysis. Dunn’s test was applied to identify the differences in spectra. Within the diagnostic categories, histology allowed us to determine the comparative efficiency of conventional screening vs. biospectroscopy to correctly identify either true atypia or underlying disease. Results Conventional cytology-based screening results in poor sensitivity and specificity. IR spectra derived from cervical cytology do not appear to discriminate in a diagnostic fashion when categories were based on conventional screening. Scores plots of IR spectra exhibit marked crossover of spectral points between different cytological categories. Although, significant differences between spectral bands in different categories are noted, crossover samples point to the potential for poor specificity and hampers the development of biospectroscopy as a diagnostic tool. However, when histology-based categories are used to conduct analyses, the scores plot of IR spectra exhibit markedly better segregation. Conclusions Histology demonstrates that ATR-FTIR spectroscopy of liquid-based cytology identifies the presence of underlying atypia or disease missed in conventional cytology screening. This study points to an urgent need for a future biospectroscopy study where categories are based on such histology. It will allow for the validation of this approach as a screening tool.