- Email: [email protected]
Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer
Oral Oncology xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Oral Oncology journal homepage: www.elsevier.com/locate/oraloncology
Review
Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer Xiaojing Ye a, Jing Zhang a,b, Yaqin Tan a, Guanying Chen a, Gang Zhou a,b,⇑ a The State Key Laboratory Breeding Base of Basic Science of Stomatology (Hubei-MOST) and Key Laboratory of Oral Biomedicine Ministry of Education, School and Hospital of Stomatology, Wuhan University, Wuhan, PR China b Department of Oral Medicine, School and Hospital of Stomatology, Wuhan University, Wuhan, PR China
a r t i c l e
i n f o
Article history: Received 20 May 2015 Received in revised form 3 September 2015 Accepted 4 September 2015 Available online xxxx Keywords: Oral precancer Oral cancer Brush biopsy DNA-image cytometry Early diagnosis
s u m m a r y The early diagnosis of oral precancer and cancer is crucial and could have the highest impact on improving survival rates. A meta-analysis was conducted to compare the accuracy between the OralCDx brush biopsy and DNA-image cytometry in diagnosing both conditions. Bibliographic databases were systematically searched for original relevant studies on the early diagnosis of oral precancer and oral cancer. Study characteristics were evaluated to determine the accuracy of the two screening strategies. Thirteen studies (eight of OralCDx brush biopsy and five of DNA-image cytometry) were identified as having reported on 1981 oral mucosa lesions. The meta-analysis found that the area under the summary receiver operating characteristic curves of the OralCDx brush biopsy and DNA-image cytometry were 0.8879 and 0.9885, respectively. The pooled sensitivity, specificity, and diagnostic odds ratio of the OralCDx brush biopsy were 86% (95% CI 81–90), 81% (95% CI 78–85), and 20.36 (95% CI 2.72–152.67), respectively, while these modalities of DNA-image cytometry were 89% (95% CI 83–94), 99% (95% CI 97–100), and 446.08 (95% CI 73.36–2712.43), respectively. Results of a pairwise comparison between each modality demonstrated that specificity, area under the curve (AUC), and Q⁄ index of DNA-image cytometry was significantly higher than that of the OralCDx brush biopsy (Z = 2.821, p < 0.05; Z = 1.711, p < 0.05; Z = 1.727, p < 0.05), but no significant difference in sensitivity was found (Z = 1.520, p > 0.05). In conclusion, the meta-analysis of the published studies indicated that DNA-image cytometry is more accurate than the OralCDx brush biopsy in diagnosing oral precancer and oral cancer. Ó 2015 Elsevier Ltd. All rights reserved.
Introduction Oral cancer is a fatal disease of which >90% is oral squamous cell carcinoma (OSCC), the sixth most common cancer in the world [1,2]. For oral cancer, the mortality and over 5-year survival rates remain unchanged, which can be largely attributed to the delayed detection, despite recent therapeutic advances [3]. Oral cancer is often not diagnosed until the advanced stages rather than when there are premalignant lesions, when cancer cells have already become aggressive [4,5]. A combination of risk reduction and early detection should further reduce the incidence of oral cancer and improve survival rates [6]. However, oral precancer cannot be adequately identified by visual inspection alone and is easily missed. Even when clinically diagnosed, the exact stage remains unknown. Therefore, the early diagnosis of oral precancer and oral cancer is ⇑ Corresponding author at: Department of Oral Medicine, School and Hospital of Stomatology, Wuhan University, Luoyu Road 237, Wuhan, PR China. Tel.: +86 27 87686213. E-mail address: [email protected] (G. Zhou).
crucial and might have the highest impact on improving survival rates. As the gold standard, scalpel biopsy with histopathological diagnosis appears to be the most accepted method by which to reliably evaluate suspicious oral lesions [7]. However, this method is not acceptable as a detection tool for early identification of oral precancer and cancer because of the invasiveness, especially when the lesions appear in seemingly ‘‘normal” or asymptomatic oral mucosa [8]. Thus, early detection of oral precancer is key in reducing the number of deaths from oral cancer, and painless, noninvasive, and accurate early diagnosis is needed [9]. Brush cytology, a simple, non-invasive and relatively painless technique that is well accepted by patients, is used for diagnosing many tumors, such as cervical cancers, laryngeal tumors, and bladder cancer [10,11]. Brush cytology is no longer limited to conventional smear-staining analysis using quantitative cytomorphology. The improvements in liquid-based exfoliative cytology and molecular analysis techniques have led to renewed interest in brush cytology [12]. Soon after its creation, brush cytology was overlooked because of sampling techniques, evaluating methods,
http://dx.doi.org/10.1016/j.oraloncology.2015.09.002 1368-8375/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
2
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
and false negatives [13]. Within the last few years, diagnostic cytopathology has emerged as a result of the application of adjunctive diagnostic methods such as liquid-based cytology, OralCDx brush biopsy system, DNA-image cytometry, and AgNOR (nucleolar organizer regions-associated proteins) staining analysis [12,14,15]. These adjunctive diagnostic methods improve the quality and reliability of brush cytology as a diagnostic technique for diagnosing cancer [10]. OralCDx is a brush biopsy technique using computer-assisted sample analysis. OralCDx brush collects cells from the three cell layers of the epithelium of the oral mucosa—the superficial, intermediate, and based layers [16]. These samples are fixed onto a glass slide and sent to a laboratory where they are stained, scanned and analyzed microscopically using a computer-based imaging system that can rank cells on the basis of degree of abnormal morphology [17,18]. A cytopathologist interprets the computerized results which are reported as ‘‘negative or benign”, ‘‘positive”, or ‘‘atypical” [18]. DNA-image cytometry was recently introduced for the very early diagnosis of malignant transformation of squamous epithelial cells [19]. It is used to detect chromosomal aneuploidy, called ‘‘DNA-aneuploidy”, which is abnormal nuclear DNA content [20]. After Feulgen staining of the slides used in cytological diagnosis, chromosomal aneuploidy can be observed from DNA-cytology. DNA aneuploidy is an indicator of numerical chromosomal changes and its emergence is often a critical step in carcinogenesis [21]. Oral dysplastic lesions have a higher risk of malignant progression and DNA aneuploidy might help to identify those lesions that are at an increased risk of malignancy [20,22]. A DNA stemline was defined as the G0/G1 cell-phase fraction of a proliferating cell population [20]. Although extensive research has been conducted on OralCDx brush biopsy and DNA-image cytometry for the detection of oral precancer and oral cancer, no direct comparison has been made between these two noninvasive diagnostic tools. Thus, a metaanalysis was performed to compare the influence and diagnostic accuracy of these two adjuvant techniques in brush cytology. In addition, the future clinical and research implications relating to brush cytology diagnostic testing were explored.
Materials and methods Literature searching A literature search of Medline, Embase, PubMed, Elsevier, and Web of Science databases from 1980 to 2014 were performed to identify relevant diagnostic test studies in English language. The major keywords searched for were ‘oral precancerous condition, oral cancer, oral leukoplakia, oral lichen planus, oral submucous fibrosis, oral erythroplakia, oral erythroleukoplakia, oral verrucous hyperplasia’, and ‘diagnosis early, cytodiagnosis, computerassisted, bush biopsy, brush cytology, DNA-image cytometry, DNA-aneuploidy’. And this search was combined using the Boolean operator ‘AND’ or ‘OR’. The abstracts of literature and review were repeated on three different occasions by the same author in an attempt to reduce human error. In addition, the reference lists of all known primary and review papers were scanned for relevant citations. Unpublished data were not attempted to find. Literatures with no detail data were excluded in this study.
Inclusion criteria and exclusion criteria Reports considered eligible for inclusion were those that investigated the value of brush cytology. This meta-analysis review
included both prospective and retrospective studies. The inclusion criteria and exclusion criteria were as following: Inclusion criteria 1. Oral precancer that was primary and untreated. 2. OralCDx brush biopsy or DNA-image cytometry were used to diagnose disease. 3. Reference standard of histopathology obtained by scalpel biopsy. 4. A disease-positive case defined as a histologically confirmed oral precancer or oral cancer, while a disease-negative case was defined as any other histologically confirmed diagnosis. Exclusion criteria 1. Not computer-assisted method was used. 2. Less than ten patients were studied both in histopathology and cytology. 3. Cytological specimens was not obtained directly from the lesion in situ by brush tools. 4. Published data of sensitivity and specificity which could not be calculated or extracted. Data collection Methodological quality was assessed by one un-blinded author and was repeated on two separate occasions. Any issues of uncertainty were discussed with a second author and agreement was obtained at several consensus meetings. For the histological diagnosis the final agreed diagnosis presented in the primary paper was accepted. We used a standard form to extract data on relevant characteristics. The quality assessment of diagnostic accuracy studies (QUADAS) was used to evaluate the quality, applicability, and reporting of the studies [23]. However, studies were not excluded on the basis of quality. Data from all articles included in the quantitative analysis were abstracted by 2 authors independently. Data were only collected if a 2 2 contingency table could be constructed from the data presented. These data were double-checked against any stated summary statistics in each article, such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV). In certain cases, the raw data were reconstituted from these summary statistics which determined the whole number ratios. Regarding the test of OralCDx brush biopsy, we considered the presence of any level of concern for dysplasia or cancer to be positive, including ‘‘atypical”, ‘‘mild, moderate, severe dysplasia”, results such as ‘‘benign” or ‘‘normal” were considered negative. ‘‘DNA-euploid” or ‘‘DNA-non-aneuploid” was taken as positive results of DNA-image cytometry, and negative results were considered to be ‘‘DNA-aneuploid”. Gold standard results positive for cancer or any amount of dysplasia were also considered positive. Statistical analysis Heterogeneity Potential sources of heterogeneity were assessed by using the likelihood ratio X2 test and if p < 0.05 it was considered having apparent heterogeneity. I2 index is a measure of the percentage of total variation across studies due to heterogeneity beyond chance and takes values between 0% and 100%. Its values over 50% indicate heterogeneity. Threshold effects between studies were evaluated by Spearman correlation coefficients p. Publication bias was assessed by Deeks plots, if p < 0.05, then publication bias was existed [24].
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Sensitivity and specificity Pooled sensitivity and specificity were calculated with 95% CIs (confidence intervals) separately for each study. The 95% CIs of the pooled sensitivity and specificity were presented which were estimated by the random effects model (REM) if heterogeneity was exist, otherwise fixed effect model (FEM) was used. Participants were classified according to the presence or absence of dysplasia given by the histopathology and OralCDx brush cytology, or the existence of DNA aneuploidy. Likelihood ratios and diagnostic OR The likelihood ratios (LRs) indicate by how much a given test would raise or lower the probability of having pathology. A LR of 1 indicates that the test had no predictive value for the outcome of interest. A positive LR of >1 increases the probability of having disease, and the greater the LR the larger the increase. Conversely, a negative LR of <1 decreases the probability of having disease, and the smaller the LR the larger the decrease [25]. Therefore, a high LR with a positive test result and low LR with a negative test result reflect a test’s diagnostic capacity. If diagnostic accuracy was supposed to be high, a LR of >10 or <0.1 would be asked for a positive and negative test result, respectively [25]. Diagnostic OR was defined as the ratio of the sensitivity odds over the odds of 1-specificity. The impact of covariates on the diagnostic odd ratio was assessed by summary receiver operating characteristics regression. Summary receiver-operating characteristic curve analysis A summary receiver-operating characteristic (SROC) curve analysis was performed using the statistical package MetaDiSc version 1.4 [26]. This package requires the input of true-positive, falsenegative, false-positive and true-negative cases, respectively, for each study included in the meta-analysis. This package then displays the SROC curve graphically by plotting the actual and predicted sensitivity across a range of values of 1-specificity. The pooled sensitivity and specificity and the log odds ratios are generated as global measures of accuracy. The pooled sensitivity and specificity are sufficient to summarize the data if the SROC curve does not show a trade-off between these two parameters. If a trade-off is evident, the SROC curve needs to be used to summarize the data. Q⁄ index is the sensitivity of the intersection of the SROC curve and a straight line (sensitivity = specificity). When Q⁄ index is larger, the accuracy of the diagnostic test is greater. A reliable statistic method in pooling studies is on the meaning of the area under the SROC curve (AUC), which summarizes the diagnostic accuracy as a single number. The AUC is closer to 1, indicating that the higher the accuracy of diagnostic tests and the diagnostic value of the check. Forest plots were constructed using meta-analytical models to pool ratios of proportions. Inter-study heterogeneity was assessed with Cochran’s Q test, and the percentage of total variation across studies due to heterogeneity was evaluated by the I2 measure. The potential influence of study characteristics on inter-study heterogeneity of sensitivity and specificity was explored using subgroup meta-analyses and meta-regression. Z test was applied to analyze whether the sensitivity, specificity, AUC, and Q⁄ index was significantly different between these two tests, if p < 0.05, it was considered as statistically significant. All statistical analyses were performed with MetaDiSc 1.4, Review Manager 5., and SPSS 19.0.
3
relevant. Seven studies were excluded because the methods used for diagnosis were either conventional brush cytology or other analysis techniques [27–33]. Studies that used paraffinembedded tissue or that were data deficient were excluded [22,34]. One study was barred because it was unclear whether the method was coupled with computer assistance to diagnose oral disease [35]. Another was excluded for testing the oral leukoplakia after treatment [36]. Thus, 13 of the 24 articles met the inclusion criteria [16,17,19,37–46]. These 13 studies formed the focus of this meta-analysis and comprised eight studies on OralCDx brush biopsy and five for DNA-image cytometry. Collectively, the 13 articles represented 1,981 clinically oral precancer or oral cancer lesions that were investigated by either OralCDx brush biopsy or DNA-image cytometry and detected by histopathology. The publication year, sample size, technique, staining, country and published journal name were abstracted from the primary articles (Table 1). Also, the numbers of true-positive (TP), false-negative (FN), false-positive (FP) and true-negative (TN) results were extracted in Table 1. The type of study for each eligible paper, three basic blinding questions, and recruitment are summarized (Supplementary Table 1). Only one study was retrospective; the others were prospective [38]. In 11 studies, cytological and histological specimens were measured independently; and two studies were not clearly documented [38,39]. And additional characteristics of these 13 studies are summarized (Supplementary Table 2). Only five studies indicated the use of tobacco and/or alcohol. Nine of the 13 studies described the location of the lesion and nearly studies assumed the patients’ gender and age. The Feulgen staining technique was used in five studies. Of these, three used the Papanicolaou staining technique to be evaluated by a cytotechnologist before Feulgen staining [19,37,43]. Three studies used only the Papanicolaou staining technique and one study used hematoxylin and eosin staining [16]. The staining techniques used in the remaining studies were not identified. Blinding in varying degrees was sufficient in a majority of the studies and well reported. There were nine studies that blinded the clinician or pathologist, who evaluated the cytology specimens from the histopathology results and vice versa. The method of recruitment was reported in all but one study [43]. Standard histopathological techniques were used in all studies. Only those lesions that underwent both examinations were included in this meta-analysis. Meta-analysis Heterogeneity and quality assessment X2 test and I2 index of the likelihood ratios of these two diagnostic techniques were evaluated (Supplementary Table 3). The inconsistency (I2 > 70%, P for Cochrane Q test <0.1) detected in the meta-analysis of OralCDx brush biopsy was high. The statistically relative heterogeneity of DOR of DNA-image cytometry was evaluated (I2 = 55.7%, P for Cochrane Q test <0.1). And threshold analysis was conducted to explore whether the high inconsistency was caused by different threshold. Considering the Spearman correlation coefficient 0.595 (p = 0.120), and 0.000 (p = 1.000), no threshold existed in either of the two diagnostic methods. The main features of the methodological qualities of the included studies and the quality assessment of diagnostic accuracy studies (QUADAS) records were summarized (Fig. 2). The majority of studies applied the ideal reference standard were of high quality.
Result Literature retrieval In total, 4391 studies were identified for online screening (Fig. 1). Twenty-four articles were thought to be potentially
Pooled sensitivity, specificity The individual and pooled sensitivities as well as specificities with 95% confidence intervals (CIs) of each study were extracted and calculated (Table 2). Using the random effects model, OralCDx brush biopsy had a pooled sensitivity of 86% (95% CI 81–90) and
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
4
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Identification
Records identified through database searches (n=4391)
Total records identified, duplicates removed (n=358)
Screening
Records screened (n=87)
Full-text articles assessed for eligibility (n=70)
Eligibility
Failed inclusion criteria (n=46)
Included in qualitative synthesis (n=24)
Included
Excluded from quantitative synthesis (n=11) 1.Not computer-assisted (n=7) 2.Not detailed data (n=2) 3.After treatment (n=1) 4.Method unclear (n=1) Included in quantitative synthesis (n=13) Fig. 1. Flow diagram for selection of studies included in this meta-analyses.
Table 1 Study characteristics of included studies. Published time
Literature
Patients
Technique
Staining
Country
TP
FP
TN
FN
1999 2001
Sciubba Remmerbach
945 251
OralCDx brush biopsy DNA-image cytometry
U.S. Germany
131 53
14 0
182 195
0 2
2004 2004
Poate Maraki
112 98
OralCDx brush biopsy DNA-image cytometry
UK Germany
15 17
19 2
9 79
6 0
2004 2009 2009
Scheifele Hohlweg-Majert Remmerbach
103 69 47
OralCDx brush biopsy OralCDx brush biopsy DNA-image cytometry
Germany Germany Germany
24 25 18
4 6 0
66 23 27
2 15 2
2011 2012 2012 2013 2014 2014
Mehrotra Reddy Seijas-Naya Kammerer Ma Casparis
85 60 24 70 52 200
OralCDx brush biopsy OralCDx brush biopsy OralCDx brush biopsy DNA-image cytometry DNA-image cytometry OralCDx brush biopsy
Papanicolaou Feulgen Papanicolaou Not mentioned Feulgen Papanicolaou Not mentioned Not mentioned Feulgen Papanicolaou Not mentioned H&E Papanicolaou Feulgen Feulgen Papanicolaou
India India Spain Germany China Germany
26 7 9 21 19 17
1 5 0 0 3 19
47 39 13 57 27 15
5 9 2 9 3 2
TP: true positive; FP: false positive; TN: true negative; FN: false negative.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
5
Fig. 2. Methodological quality graph and summary: review authors’ judgements about each methodological quality item for each included study.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
6
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Table 2 Pooled sensitivity and specificity of both OralCDx brush biopsy and DNA-image cytometry. Study
Sen [95% Conf. Interval.]
A 1999 2004 2004 2009 2011 2012 2012 2014
1.000 0.714 0.923 0.625 0.839 0.438 0.727 0.895
0.972 0.478 0.749 0.458 0.663 0.198 0.390 0.669
1.000 0.887 0.991 0.773 0.945 0.701 0.940 0.987
0.929 0.321 0.943 0.207 0.979 0.886 0.923 0.441
0.883 0.159 0.860 0.080 0.889 0.754 0.640 0.272
0.960 0.524 0.984 0.397 0.999 0.962 0.998 0.621
Pooled Sen
0.858
0.812
0.895
0.814
0.775
0.848
B 2001 Remmerbach (DNA) 2004 Maraki 2009 Remmerbach 2013 Kammerer (DNA) 2014 Ma
0.964 1.000 0.900 0.700 0.864
0.875 0.805 0.683 0.506 0.651
0.996 1.000 0.988 0.853 0.971
1.000 0.975 1.000 1.000 0.900
0.981 0.914 0.872 0.937 0.735
1.000 0.997 1.000 1.000 0.979
Pooled Sen
0.889
0.826
0.935
0.987
0.970
0.996
Sciubba Poate Scheifele Hohlweg-Mejert Mehrotra Reddy Seijas-Naya Casparis
Spe [95% Conf. Interval.]
A represented OralCDx brush biopsy and B represented DNA-image cytometry.
A
Diagnostic OR (95% CI)
B
2001 Remmerbach(DNA) 2004 Maraki 2009 Remmerbach 2013 Kammerer(DNA) 2014 Ma
0.01
1
Diagnostic Odds Ratio
8,367.40 (395.72 - 176,927.15) 1,113.00 (51.14 - 24,221.00) 407.00 (18.46 - 8,971.90) 260.26 (14.51 - 4,667.39) 57.00 (10.37 - 313.43)
Random Effects Model Pooled Diagnostic Odds Ratio = 446.08 (73.36 to 2712.43) Cochran-Q = 9.04; df = 4 (p = 0.0601) 100.0 Inconsistency (I-square) = 55.7 % Tau-squared = 2.3073
Fig. 3. Diagnostic odd ratio of OralCDx brush biopsy (A) and DNA-image cytometry (B) were showed.
pooled specificity of 81% (95% CI 78–85), while the pooled sensitivity and specificity of DNA-image cytometry were 89% (95% CI 83–94) and 99% (95% CI 97–100). Although the sensitivity of DNA-image cytometry was higher than that of OralCDx brush biopsy, the difference was not statistically significant (Z = 1.520, p > 0.05). The specificity of DNA-image cytometry was significantly higher than that of OralCDx brush biopsy (Z = 2.821, p < 0.05).
DOR and the likelihood ratios The DOR plots are presented in Fig. 3. Heterogeneity was considered as a result of the non-threshold effect because DOR of each study was not linear with the pooled DOR. DOR estimates for OralCDx brush biopsy and DNA-image cytometry were 20.36 (95% CI 2.72–152.67) and 446.08 (95% CI 73.36–2712.43), respectively. Estimates of LRs for individual studies were shown in Supplementary
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
7
Fig. 4. Summary receiver operating characteristic curves displaying the fitted joint variation of sensitivity and specificity of OralCDx brush biopsy (A) and DNA-image cytometry (B). The dots represent the observed sensitivity and specificity in individual studies.
Table 4. The pooled estimates of positive LRs for DNA-image cytometry increased to 38.081 (95% CI 9.64–150.39) compared to OralCDx brush biopsy (4.472; 95% CI 1.489–13.427). The opposite trend was evaluated on the pooled estimates of negative LRs. SROC curve, AUC, and Q* index The overall relationship between sensitivity and 1-specificity was estimated from an unweighted SROC mode (Fig. 4). Results
were shown to generally cluster in the top left-hand corner of the ROC space and the pooled mean sensitivity and specificity laid very close to the SROC curve. Consequently, the pooled sensitivity and specificity were considered sufficient summary measures for this meta-analysis. In addition, the area under the SROC curve was evaluated. The AUCs of OralCDx brush biopsy and DNA-image cytometry were 0.8879 and 0.9885, respectively, a statistically significant difference (Z = 1.711, p < 0.05). The Q⁄ index
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
8
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
estimates for DNA-image cytometry (Q⁄ = 0.9548) was higher than that of OralCDx brush biopsy (Q⁄ = 0.8186) (Z = 1.727, p < 0.05). Regression analysis and publication bias Meta-regression was used to explore possible sources of heterogeneity. Staining technique, study size, and design method were analyzed by meta-regression. Finally, study size was discovered to be closely related to heterogeneity among studies. To evaluate publication bias, Deeks plots were designed using the log diagnostic odd ratios of individual studies against their sample size. There was publication bias when evaluating the study of OralCDx brush biopsy (p = 0.01), but no publication bias in the study of DNA-image cytometry (p = 0.06). Discussion This was the first meta-analysis to compare OralCDx brush biopsy with DNA-image cytometry, two computer-assisted diagnostic tools for clinically suspected oral precancer and oral cancer. The results showed that both of methods have a significant potential for diagnosis with high sensitivity and specificity, but that DNA-image cytometry has higher sensitivity and specificity than OralCDx brush biopsy. Because of the higher specificity, AUC, and Q⁄ index of SROC, it is proposed that DNA-aneuploidy cytology is more accurate in diagnosing oral precancer and oral cancer. Until now, scalpel biopsy with histological assessment was considered to be the only accepted method by which to definitely evaluate suspicious oral lesions [47]. Considering the invasiveness and low-patient acceptance of incision biopsy, respectively simple and repeatable methods were needed [48]. Brush cytology, as a diagnostic method, has numerous advantages [10]. First, it offers the opportunity for early diagnosis by the clinician and decreases the time from diagnosis to treatment [49]. Second, it is a simple, non-invasive, and relatively painless technique that is well accepted by patients [50]. Finally, to those very sick patients with suspected oral cancer or those who are allergic to local anesthesia, brush cytology can be a beneficial diagnostic method [51]. However, it has certain limitations [10]. For example, it has been reported that this technique is not able to differentiate tumor subtypes [10]. Further, brush cytology sampling is diagnosed by a cytologist and human error and bias can exist to some extent [52]. Thus, the clinician should analyze the results by combining clinical manifestations. Both OralCDx brush biopsy and DNA-image cytometry can assess cellular benignity or malignancy using computer-assisted analysis which can improve their quality and reliability as a cancer diagnostic technique [10,53]. The improved accuracy is attributed to the ease of obtaining full transepithelial cellular samples [53]. In addition, computer-assisted analysis can reduce human cognitive bias. The evaluation of smears using an image analysis system has been adapted specifically to detect oral epithelial abnormalities [16]. Patton et al. found that OralCDx brush biopsy, as an adjunctive technique, was useful in assessing dysplastic changes in clinically suspicious lesions [18]. In addition, compared with conventional brush biopsy, OralCDx computer can search the brush biopsy specimen for a combination of abnormal cellular morphology and abnormal keratinization, which uniquely characterize dysplasia and carcinoma of the oral epithelium [17]; however, Patton et al. concluded that there were insufficient data that met the inclusion criteria to be able assess the availability in innocuous mucosal lesions [18]. It is known that molecular and genetic changes might precede both clinical and microscopic morphological changes and might be present in histologically benign tissue [54]. Furthermore,
DNA-aneuploidy is an internationally accepted marker for carcinoma transformation of cells [55,56]. Previous result suggests that DNA aneuploidy arises in even histopathologically mild oral epithelial dysplasia, and such lesions will transform into oral squamous cell carcinoma [57]. Remmerbach et al. studied the reliability of oral brush cytology and its DNA cytometric analysis in the early detection of oral cancer. The results of the study were substantial because by combining both techniques, the sensitivity increased to 98.2% and the specificity was 100%, with a positive predictive value of 100% and a negative value of 99.5% [37]. Maraki et al. also found that the sensitivity of cytological diagnosis combined with DNA-image cytometry was 100%, and specificity was 97.4% [19]. Remmerbach et al. also concluded in the clinical setting that DNA-aneuploidy might be used to detect histologically obvious malignancy, from 1 to 15 months prior to histology [37]. Sudbo et al. analyzed archival material and reported that nuclear DNA in the cells of oral leukoplakia could be used to predict the risk of oral epithelial dysplasia up to 5 years before histological diagnosis [57]. There remain some deficiencies in the methodological assessment of diagnostic tests. Some of the clinical characteristic data of the primary studies used in this meta-analysis were undocumented, but some were useful and important, such as smoking or tobacco chewing, which are factors that could cause oral cancer. [19,39,43]. In addition, there was selection bias for including only articles written in English but ignoring those unpublished or published studies and articles written in other languages. Furthermore, only five studies classified their results, including inadequate categories and reported those cases with incomplete transepithelial cells, which could be beneficial information in the development of brush technology [16,41]. From the diagnostic meta-analysis, the Q⁄ value of SROC curve analysis that considers the nonlinear relationship between sensitivity and specificity, is a reflection of the accuracy of the combined index on the diagnostic tests [58]. Thus, the results are more scientific and accurate and the diagnostic tests can be intuitively compared through a graphical area using the AUC value [58]. In this meta-analysis, the curve of DNA-image cytometry was closer to the upper left corner than that of OralCDx brush biopsy. In addition, the Q⁄ index estimates for DNA-image cytometry was statistically significantly higher than that of OralCDx brush biopsy; however, there is relatively less primary research about DNAimage cytometry, which might result in an overestimate its diagnostic value. Acha et al. claimed that brush biopsy will never prevail over the classic biopsy, and that a biopsy should be done on all clinically suspicious lesions even with a benign cytological diagnosis [53]. However, the cytological study of oral precancer and oral cancer is rapid, non-aggressive and relatively painless compared with that of oral scalpel biopsy and is well accepted by patients and suitable for routine application in population screening programs, for the early analysis of suspect lesions, and for pre-and post-treatment monitoring of confirmed malignant lesions.
Conclusion The results of this meta-analysis suggest that DNA-image cytometry has a highly significant potential over OralCDx brush biopsy as an accurate and simple diagnostic tool for clinically suspected oral precancer and oral cancer.
Conflict of interest statement The authors declare no conflict of interest.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Acknowledgement This work was supported by Grants from the National Natural Science Foundation of China (Nos. 81170972 and 81371147) to Prof. Zhou Gang. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.oraloncology. 2015.09.002. References [1] Adams D. Oral cancer: early diagnosis. Dent Today 2014;33:8. [2] Gaitan-Cepeda LA, Peniche-Becerra AG, Quezada-Rivera D. Trends in frequency and prevalence of oral cancer and oral squamous cell carcinoma in Mexicans. A 20 years retrospective study. Med Oral, Patol Oral Cir Bucal 2011;16:e1–5. [3] Guneri P, Epstein JB. Late stage diagnosis of oral cancer: components and possible solutions. Oral Oncol 2014;50:1131–6. [4] Mehrotra R, Gupta DK. Exciting new advances in oral cancer diagnosis: avenues to early detection. Head Neck Oncol 2011;3:33. [5] Jafari A, Najafi S, Moradi F, Kharazifard M, Khami M. Delay in the diagnosis and treatment of oral cancer. J Dent 2013;14:146–50. [6] Hunter KD, Yeoman CM. An update on the clinical pathology of oral precancer and cancer. Dent Update 2013;40. p. 120–2, 5–6. [7] Baykul T, Yilmaz HH, Aydin U, Aydin MA, Aksoy M, Yildirim D. Early diagnosis of oral cancer. J Int Med Res 2010;38:737–49. [8] O’Sullivan E. Improving early diagnosis of oral cancer. J Oral Maxillofac Surg: Off J Am Assoc Oral Maxillofac Surg 2004;62:115. [9] Warnakulasuriya S, Reibel J, Bouquot J, Dabelsteen E. Oral epithelial dysplasia classification systems: predictive value, utility, weaknesses and scope for improvement. J Oral Pathol Med: Off Publ Int Assoc Oral Pathol Am Acad Oral Pathol 2008;37:127–33. [10] Diniz-Freitas M, Garcia-Garcia A, Crespo-Abelleira A, Martins-Carneiro JL, Gandara-Rey JM. Applications of exfoliative cytology in the diagnosis of oral cancer. Medicina oral: organo oficial de la Sociedad Espanola de Medicina Oral y de la Academia Iberoamericana de Patologia y Medicina Bucal 2004;9:355–61. [11] Cikojevic D, Gluncic I, Pesutic-Pisac V. Role of exfoliative cytology in diagnosis of laryngeal tumors. Acta Cytol 2007;51:767–72. [12] Perez-Sayansm M, Somoza-Martin JM, Barros-Angueira F, Reboiras-Lopez MD, Gandara-Vila P, Gandara Rey JM, et al. Exfoliative cytology for diagnosing oral cancer. Biotech Histochem: Off Publ Biol Stain Comm 2010;85:177–87. [13] Nichols ML, Quinn Jr FB, Schnadig VJ, Zaharopoulos P, Hokanson JA, Des Jardins L, et al. Interobserver variability in the interpretation of brush cytologic studies from head and neck lesions. Arch Otolaryngol – Head Neck Surg 1991;117:1350–5. [14] Navone R, Pentenero M, Gandolfo S. Liquid-based cytology in oral cavity squamous cell cancer. Curr Opin Otolaryngol Head Neck Surg 2011;19:77–81. [15] Delavarian Z, Mohtasham N, Mosannen-Mozafari P, Pakfetrat A, Shakeri MT, Ghafoorian-Maddah R. Evaluation of the diagnostic value of a modified liquidbased cytology using OralCDx brush in early detection of oral potentially malignant lesions and oral cancer. Med Oral, Patol Oral Cir Bucal 2010;15: e671–6. [16] Reddy SG, Kanala S, Chigurupati A, Kumar SR, Poosarla CS, Reddy BV. The sensitivity and specificity of computerized brush biopsy and scalpel biopsy in diagnosing oral premalignant lesions: a comparative study. J Oral Maxillofac Pathol: JOMFP 2012;16:349–53. [17] Sciubba JJ, U.S. Collaborative OralCDx Study Group. Improving detection of precancerous and cancerous oral lesions: computer-assisted analysis of the oral brush biopsy. J Am Dent Assoc 1999;130:1445–57. [18] Patton LL, Epstein JB, Kerr AR. Adjunctive techniques for oral cancer examination and lesion diagnosis: a systematic review of the literature. J Am Dent Assoc 2008;139:896–905. quiz 93-4. [19] Maraki D, Becker J, Boecking A. Cytologic and DNA-cytometric very early diagnosis of oral cancer. J Oral Pathol Med: Off Publ Int Assoc Oral Pathol Am Acad Oral Pathol 2004;33:398–404. [20] Giaretti W, Pentenero M, Gandolfo S, Castagnola P. Chromosomal instability, aneuploidy and routine high-resolution DNA content analysis in oral cancer risk evaluation. Future Oncol 2012;8:1257–71. [21] Rajagopalan H, Lengauer C. Aneuploidy and cancer. Nature 2004;432:338–41. [22] Torres-Rendon A, Stewart R, Craig GT, Wells M, Speight PM. DNA ploidy analysis by image cytometry helps to identify oral epithelial dysplasias with a high risk of malignant progression. Oral Oncol 2009;45:468–73. [23] Whiting PF, Weswood ME, Rutjes AW, Reitsma JB, Bossuyt PN, Kleijnen J. Evaluation of QUADAS, a tool for the quality assessment of diagnostic accuracy studies. BMC Med Res Methodol 2006;6:9. [24] Deeks S. Clinical implications of virological ‘‘failure”: interview with Steven Deeks, M.D., San Francisco General Hospital. Interview by John S. James. AIDS Treat News 1998:1–6.
9
[25] Jaeschke R, Guyatt GH, Sackett DL, The Evidence-Based Medicine Working Group. Users’ guides to the medical literature: III. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients? Jama 1994;271:703–7. [26] Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostic test into a summary ROC curve: data-analytic approaches and some additional considerations. Stat Med 1993;12:1293–316. [27] Remmerbach TW, Weidenbach H, Muller C, Hemprich A, Pomjanski N, Buckstegge B, et al. Diagnostic value of nucleolar organizer regions (AgNORs) in brush biopsies of suspicious lesions of the oral cavity. Anal Cell Pathol: J Eur Soc Anal Cell Pathol 2003;25:139–46. [28] Mehrotra R, Singh MK, Pandya S, Singh M. The use of an oral brush biopsy without computer-assisted analysis in the evaluation of oral lesions: a study of 94 patients. Oral Surg, Oral Med, Oral Pathol, Oral Radiol, Endodont 2008;106:246–53. [29] Guneri P, Epstein JB, Kaya A, Veral A, Kazandi A, Boyacioglu H. The utility of toluidine blue staining and brush cytology as adjuncts in clinical examination of suspicious oral mucosal lesions. Int J Oral Maxillofac Surg 2011;40:155–61. [30] Koch FP, Kunkel M, Biesterfeld S, Wagner W. Diagnostic efficiency of differentiating small cancerous and precancerous lesions using mucosal brush smears of the oral cavity – a prospective and blinded study. Clin Oral Investigat 2011;15:763–9. [31] Babshet M, Nandimath K, Pervatikar S, Naikmasur V. Efficacy of oral brush cytology in the evaluation of the oral premalignant and malignant lesions. J Cytol/Indian Acad Cytol 2011;28:165–72. [32] Graveland AP, Bremmer JF, de Maaker M, Brink A, Cobussen P, Zwart M, et al. Molecular screening of oral precancer. Oral Oncol 2013;49:1129–35. [33] Mulki S, Shetty P, Pai P. Oral rinse as a simpler approach to exfoliative cytology: a comparative study. J Clin Diagn Res: JCDR 2013;7:3036–8. [34] Femiano F, Scully C. DNA cytometry of oral leukoplakia and oral lichen planus. Med Oral, Patol Oral Cir Bucal 2005;10(Suppl. 1):E9–E14. [35] Mishra M, Mohanty J, Sengupta S, Tripathy S. Epidemiological and clinicopathological study of oral leukoplakia. Indian J Dermatol, Venereol Leprol 2005;71:161–5. [36] Schwarz F, Maraki D, Yalcinkaya S, Bieling K, Bocking A, Becker J. Cytologic and DNA-cytometric follow-up of oral leukoplakia after CO2- and Er:YAG-laser assisted ablation: a pilot study. Lasers Surg Med 2005;37:29–36. [37] Remmerbach TW, Weidenbach H, Pomjanski N, Knops K, Mathes S, Hemprich A, et al. Cytologic and DNA-cytometric early diagnosis of oral cancer. Anal Cell Pathol: J Eur Soc Anal Cell Pathol 2001;22:211–21. [38] Poate TW, Buchanan JA, Hodgson TA, Speight PM, Barrett AW, Moles DR, et al. An audit of the efficacy of the oral brush biopsy technique in a specialist oral medicine unit. Oral Oncol 2004;40:829–34. [39] Scheifele C, Schmidt-Westhausen AM, Dietrich T, Reichart PA. The sensitivity and specificity of the OralCDx technique: evaluation of 103 cases. Oral Oncol 2004;40:824–8. [40] Hohlweg-Majert B, Deppe H, Metzger MC, Schumm S, Hoefler H, Kesting MR, et al. Sensitivity and specificity of oral brush biopsy. Cancer Investigat 2009;27:293–7. [41] Seijas-Naya F, Garcia-Carnicero T, Gandara-Vila P, Couso-Folgueiras E, PerezSayans M, Gandara-Vila R, et al. Applications of OralCDx(R) methodology in the diagnosis of oral leukoplakia. Med Oral, Patol Oral Cir Bucal 2012;17:e5–9. [42] Kammerer PW, Koch FP, Santoro M, Babaryka G, Biesterfeld S, Brieger J, et al. Prospective, blinded comparison of cytology and DNA-image cytometry of brush biopsies for early detection of oral malignancy. Oral Oncol 2013;49:420–6. [43] Remmerbach TW, Meyer-Ebrecht D, Aach T, Wurflinger T, Bell AA, Schneider TE, et al. Toward a multimodal cell analysis of brush biopsies for the early detection of oral cancer. Cancer 2009;117:228–35. [44] Ma JM, Zhou TJ, Wang R, Shan J, Wu YN, Song XL, et al. Brush biopsy with DNAimage cytometry: a useful and noninvasive method for monitoring malignant transformation of potentially malignant oral disorders. Eur Arch Oto-rhinolaryngol: Off J Eur Fed Oto-Rhino-Laryngol Soc 2014. [45] Casparis S, Borm JM, Tomic MA, Burkhard A, Locher MC. Transepithelial brush biopsy – oral CDx(R) – a noninvasive method for the early detection of precancerous and cancerous lesions. J Clin Diagn Res: JCDR 2014;8:222–6. [46] Mehrotra R, Mishra S, Singh M, Singh M. The efficacy of oral brush biopsy with computer-assisted analysis in identifying precancerous and cancerous lesions. Head Neck Oncol 2011;3:39. [47] Richards D. Adjunctive tests cannot replace scalpel biopsy for oral cancer diagnosis. Evid Dent 2015;16:46–7. [48] Brocklehurst P, Kujan O, O’Malley LA, Ogden G, Shepherd S, Glenny AM. Screening programmes for the early detection and prevention of oral cancer. Cochrane Database Syst Rev 2013;11:CD004150. [49] Gupta S, Shah JS, Parikh S, Limbdiwala P, Goel S. Clinical correlative study on early detection of oral cancer and precancerous lesions by modified oral brush biopsy and cytology followed by histopathology. J Cancer Res Ther 2014;10:232–8. [50] Driemel O, Kosmehl H, Rosenhahn J, Berndt A, Reichert TE, Zardi L, et al. Expression analysis of extracellular matrix components in brush biopsies of oral lesions. Anticancer Res 2007;27:1565–70. [51] Collins BM. The oral brush biopsy: an adjunct to early oral cancer detection. Pennsylvania Dent J 2002;69:35–7. [52] Bocking A, Sproll C, Stocklein N, Naujoks C, Depprich R, Kubler NR, et al. Role of brush biopsy and DNA cytometry for prevention, diagnosis, therapy, and followup care of oral cancer. J Oncol 2011;2011:875959.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
10
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
[53] Acha A, Ruesga MT, Rodriguez MJ, Martinez de Pancorbo MA, Aguirre JM. Applications of the oral scraped (exfoliative) cytology in oral cancer and precancer. Med Oral, Patol Oral Cir Bucal 2005;10:95–102. [54] Thomson PJ. Field change and oral cancer: new evidence for widespread carcinogenesis? Int J Oral Maxillofac Surg 2002;31:262–6. [55] Nadjari B, Kersten A, Ross B, Motherby H, Krallmann R, Sundmacher R, et al. Cytologic and DNA cytometric diagnosis and therapy monitoring of squamous cell carcinoma in situ and malignant melanoma of the cornea and conjunctiva. Anal Quant Cytol Histol/Int Acad Cytol Am Soc Cytol 1999;21:387–96.
[56] Nadjari B, Motherby H, Pooschke T, Pooschke S, Gabbert HE, Simon D, et al. DNA aneuploidy as a specific marker of neoplastic cells in FNAB of the thyroid. Anal Quant Cytol Histol/Int Acad Cytol Am Soc Cytol 1999;21:481–8. [57] Sudbo J, Kildal W, Risberg B, Koppang HS, Danielsen HE, Reith A. DNA content as a prognostic marker in patients with oral leukoplakia. N Engl J Med 2001;344:1270–8. [58] Walter SD. Properties of the summary receiver operating characteristic (SROC) curve for diagnostic test data. Stat Med 2002;21:1237–56.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
Contents lists available at ScienceDirect
Oral Oncology journal homepage: www.elsevier.com/locate/oraloncology
Review
Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer Xiaojing Ye a, Jing Zhang a,b, Yaqin Tan a, Guanying Chen a, Gang Zhou a,b,⇑ a The State Key Laboratory Breeding Base of Basic Science of Stomatology (Hubei-MOST) and Key Laboratory of Oral Biomedicine Ministry of Education, School and Hospital of Stomatology, Wuhan University, Wuhan, PR China b Department of Oral Medicine, School and Hospital of Stomatology, Wuhan University, Wuhan, PR China
a r t i c l e
i n f o
Article history: Received 20 May 2015 Received in revised form 3 September 2015 Accepted 4 September 2015 Available online xxxx Keywords: Oral precancer Oral cancer Brush biopsy DNA-image cytometry Early diagnosis
s u m m a r y The early diagnosis of oral precancer and cancer is crucial and could have the highest impact on improving survival rates. A meta-analysis was conducted to compare the accuracy between the OralCDx brush biopsy and DNA-image cytometry in diagnosing both conditions. Bibliographic databases were systematically searched for original relevant studies on the early diagnosis of oral precancer and oral cancer. Study characteristics were evaluated to determine the accuracy of the two screening strategies. Thirteen studies (eight of OralCDx brush biopsy and five of DNA-image cytometry) were identified as having reported on 1981 oral mucosa lesions. The meta-analysis found that the area under the summary receiver operating characteristic curves of the OralCDx brush biopsy and DNA-image cytometry were 0.8879 and 0.9885, respectively. The pooled sensitivity, specificity, and diagnostic odds ratio of the OralCDx brush biopsy were 86% (95% CI 81–90), 81% (95% CI 78–85), and 20.36 (95% CI 2.72–152.67), respectively, while these modalities of DNA-image cytometry were 89% (95% CI 83–94), 99% (95% CI 97–100), and 446.08 (95% CI 73.36–2712.43), respectively. Results of a pairwise comparison between each modality demonstrated that specificity, area under the curve (AUC), and Q⁄ index of DNA-image cytometry was significantly higher than that of the OralCDx brush biopsy (Z = 2.821, p < 0.05; Z = 1.711, p < 0.05; Z = 1.727, p < 0.05), but no significant difference in sensitivity was found (Z = 1.520, p > 0.05). In conclusion, the meta-analysis of the published studies indicated that DNA-image cytometry is more accurate than the OralCDx brush biopsy in diagnosing oral precancer and oral cancer. Ó 2015 Elsevier Ltd. All rights reserved.
Introduction Oral cancer is a fatal disease of which >90% is oral squamous cell carcinoma (OSCC), the sixth most common cancer in the world [1,2]. For oral cancer, the mortality and over 5-year survival rates remain unchanged, which can be largely attributed to the delayed detection, despite recent therapeutic advances [3]. Oral cancer is often not diagnosed until the advanced stages rather than when there are premalignant lesions, when cancer cells have already become aggressive [4,5]. A combination of risk reduction and early detection should further reduce the incidence of oral cancer and improve survival rates [6]. However, oral precancer cannot be adequately identified by visual inspection alone and is easily missed. Even when clinically diagnosed, the exact stage remains unknown. Therefore, the early diagnosis of oral precancer and oral cancer is ⇑ Corresponding author at: Department of Oral Medicine, School and Hospital of Stomatology, Wuhan University, Luoyu Road 237, Wuhan, PR China. Tel.: +86 27 87686213. E-mail address: [email protected] (G. Zhou).
crucial and might have the highest impact on improving survival rates. As the gold standard, scalpel biopsy with histopathological diagnosis appears to be the most accepted method by which to reliably evaluate suspicious oral lesions [7]. However, this method is not acceptable as a detection tool for early identification of oral precancer and cancer because of the invasiveness, especially when the lesions appear in seemingly ‘‘normal” or asymptomatic oral mucosa [8]. Thus, early detection of oral precancer is key in reducing the number of deaths from oral cancer, and painless, noninvasive, and accurate early diagnosis is needed [9]. Brush cytology, a simple, non-invasive and relatively painless technique that is well accepted by patients, is used for diagnosing many tumors, such as cervical cancers, laryngeal tumors, and bladder cancer [10,11]. Brush cytology is no longer limited to conventional smear-staining analysis using quantitative cytomorphology. The improvements in liquid-based exfoliative cytology and molecular analysis techniques have led to renewed interest in brush cytology [12]. Soon after its creation, brush cytology was overlooked because of sampling techniques, evaluating methods,
http://dx.doi.org/10.1016/j.oraloncology.2015.09.002 1368-8375/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
2
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
and false negatives [13]. Within the last few years, diagnostic cytopathology has emerged as a result of the application of adjunctive diagnostic methods such as liquid-based cytology, OralCDx brush biopsy system, DNA-image cytometry, and AgNOR (nucleolar organizer regions-associated proteins) staining analysis [12,14,15]. These adjunctive diagnostic methods improve the quality and reliability of brush cytology as a diagnostic technique for diagnosing cancer [10]. OralCDx is a brush biopsy technique using computer-assisted sample analysis. OralCDx brush collects cells from the three cell layers of the epithelium of the oral mucosa—the superficial, intermediate, and based layers [16]. These samples are fixed onto a glass slide and sent to a laboratory where they are stained, scanned and analyzed microscopically using a computer-based imaging system that can rank cells on the basis of degree of abnormal morphology [17,18]. A cytopathologist interprets the computerized results which are reported as ‘‘negative or benign”, ‘‘positive”, or ‘‘atypical” [18]. DNA-image cytometry was recently introduced for the very early diagnosis of malignant transformation of squamous epithelial cells [19]. It is used to detect chromosomal aneuploidy, called ‘‘DNA-aneuploidy”, which is abnormal nuclear DNA content [20]. After Feulgen staining of the slides used in cytological diagnosis, chromosomal aneuploidy can be observed from DNA-cytology. DNA aneuploidy is an indicator of numerical chromosomal changes and its emergence is often a critical step in carcinogenesis [21]. Oral dysplastic lesions have a higher risk of malignant progression and DNA aneuploidy might help to identify those lesions that are at an increased risk of malignancy [20,22]. A DNA stemline was defined as the G0/G1 cell-phase fraction of a proliferating cell population [20]. Although extensive research has been conducted on OralCDx brush biopsy and DNA-image cytometry for the detection of oral precancer and oral cancer, no direct comparison has been made between these two noninvasive diagnostic tools. Thus, a metaanalysis was performed to compare the influence and diagnostic accuracy of these two adjuvant techniques in brush cytology. In addition, the future clinical and research implications relating to brush cytology diagnostic testing were explored.
Materials and methods Literature searching A literature search of Medline, Embase, PubMed, Elsevier, and Web of Science databases from 1980 to 2014 were performed to identify relevant diagnostic test studies in English language. The major keywords searched for were ‘oral precancerous condition, oral cancer, oral leukoplakia, oral lichen planus, oral submucous fibrosis, oral erythroplakia, oral erythroleukoplakia, oral verrucous hyperplasia’, and ‘diagnosis early, cytodiagnosis, computerassisted, bush biopsy, brush cytology, DNA-image cytometry, DNA-aneuploidy’. And this search was combined using the Boolean operator ‘AND’ or ‘OR’. The abstracts of literature and review were repeated on three different occasions by the same author in an attempt to reduce human error. In addition, the reference lists of all known primary and review papers were scanned for relevant citations. Unpublished data were not attempted to find. Literatures with no detail data were excluded in this study.
Inclusion criteria and exclusion criteria Reports considered eligible for inclusion were those that investigated the value of brush cytology. This meta-analysis review
included both prospective and retrospective studies. The inclusion criteria and exclusion criteria were as following: Inclusion criteria 1. Oral precancer that was primary and untreated. 2. OralCDx brush biopsy or DNA-image cytometry were used to diagnose disease. 3. Reference standard of histopathology obtained by scalpel biopsy. 4. A disease-positive case defined as a histologically confirmed oral precancer or oral cancer, while a disease-negative case was defined as any other histologically confirmed diagnosis. Exclusion criteria 1. Not computer-assisted method was used. 2. Less than ten patients were studied both in histopathology and cytology. 3. Cytological specimens was not obtained directly from the lesion in situ by brush tools. 4. Published data of sensitivity and specificity which could not be calculated or extracted. Data collection Methodological quality was assessed by one un-blinded author and was repeated on two separate occasions. Any issues of uncertainty were discussed with a second author and agreement was obtained at several consensus meetings. For the histological diagnosis the final agreed diagnosis presented in the primary paper was accepted. We used a standard form to extract data on relevant characteristics. The quality assessment of diagnostic accuracy studies (QUADAS) was used to evaluate the quality, applicability, and reporting of the studies [23]. However, studies were not excluded on the basis of quality. Data from all articles included in the quantitative analysis were abstracted by 2 authors independently. Data were only collected if a 2 2 contingency table could be constructed from the data presented. These data were double-checked against any stated summary statistics in each article, such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV). In certain cases, the raw data were reconstituted from these summary statistics which determined the whole number ratios. Regarding the test of OralCDx brush biopsy, we considered the presence of any level of concern for dysplasia or cancer to be positive, including ‘‘atypical”, ‘‘mild, moderate, severe dysplasia”, results such as ‘‘benign” or ‘‘normal” were considered negative. ‘‘DNA-euploid” or ‘‘DNA-non-aneuploid” was taken as positive results of DNA-image cytometry, and negative results were considered to be ‘‘DNA-aneuploid”. Gold standard results positive for cancer or any amount of dysplasia were also considered positive. Statistical analysis Heterogeneity Potential sources of heterogeneity were assessed by using the likelihood ratio X2 test and if p < 0.05 it was considered having apparent heterogeneity. I2 index is a measure of the percentage of total variation across studies due to heterogeneity beyond chance and takes values between 0% and 100%. Its values over 50% indicate heterogeneity. Threshold effects between studies were evaluated by Spearman correlation coefficients p. Publication bias was assessed by Deeks plots, if p < 0.05, then publication bias was existed [24].
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Sensitivity and specificity Pooled sensitivity and specificity were calculated with 95% CIs (confidence intervals) separately for each study. The 95% CIs of the pooled sensitivity and specificity were presented which were estimated by the random effects model (REM) if heterogeneity was exist, otherwise fixed effect model (FEM) was used. Participants were classified according to the presence or absence of dysplasia given by the histopathology and OralCDx brush cytology, or the existence of DNA aneuploidy. Likelihood ratios and diagnostic OR The likelihood ratios (LRs) indicate by how much a given test would raise or lower the probability of having pathology. A LR of 1 indicates that the test had no predictive value for the outcome of interest. A positive LR of >1 increases the probability of having disease, and the greater the LR the larger the increase. Conversely, a negative LR of <1 decreases the probability of having disease, and the smaller the LR the larger the decrease [25]. Therefore, a high LR with a positive test result and low LR with a negative test result reflect a test’s diagnostic capacity. If diagnostic accuracy was supposed to be high, a LR of >10 or <0.1 would be asked for a positive and negative test result, respectively [25]. Diagnostic OR was defined as the ratio of the sensitivity odds over the odds of 1-specificity. The impact of covariates on the diagnostic odd ratio was assessed by summary receiver operating characteristics regression. Summary receiver-operating characteristic curve analysis A summary receiver-operating characteristic (SROC) curve analysis was performed using the statistical package MetaDiSc version 1.4 [26]. This package requires the input of true-positive, falsenegative, false-positive and true-negative cases, respectively, for each study included in the meta-analysis. This package then displays the SROC curve graphically by plotting the actual and predicted sensitivity across a range of values of 1-specificity. The pooled sensitivity and specificity and the log odds ratios are generated as global measures of accuracy. The pooled sensitivity and specificity are sufficient to summarize the data if the SROC curve does not show a trade-off between these two parameters. If a trade-off is evident, the SROC curve needs to be used to summarize the data. Q⁄ index is the sensitivity of the intersection of the SROC curve and a straight line (sensitivity = specificity). When Q⁄ index is larger, the accuracy of the diagnostic test is greater. A reliable statistic method in pooling studies is on the meaning of the area under the SROC curve (AUC), which summarizes the diagnostic accuracy as a single number. The AUC is closer to 1, indicating that the higher the accuracy of diagnostic tests and the diagnostic value of the check. Forest plots were constructed using meta-analytical models to pool ratios of proportions. Inter-study heterogeneity was assessed with Cochran’s Q test, and the percentage of total variation across studies due to heterogeneity was evaluated by the I2 measure. The potential influence of study characteristics on inter-study heterogeneity of sensitivity and specificity was explored using subgroup meta-analyses and meta-regression. Z test was applied to analyze whether the sensitivity, specificity, AUC, and Q⁄ index was significantly different between these two tests, if p < 0.05, it was considered as statistically significant. All statistical analyses were performed with MetaDiSc 1.4, Review Manager 5., and SPSS 19.0.
3
relevant. Seven studies were excluded because the methods used for diagnosis were either conventional brush cytology or other analysis techniques [27–33]. Studies that used paraffinembedded tissue or that were data deficient were excluded [22,34]. One study was barred because it was unclear whether the method was coupled with computer assistance to diagnose oral disease [35]. Another was excluded for testing the oral leukoplakia after treatment [36]. Thus, 13 of the 24 articles met the inclusion criteria [16,17,19,37–46]. These 13 studies formed the focus of this meta-analysis and comprised eight studies on OralCDx brush biopsy and five for DNA-image cytometry. Collectively, the 13 articles represented 1,981 clinically oral precancer or oral cancer lesions that were investigated by either OralCDx brush biopsy or DNA-image cytometry and detected by histopathology. The publication year, sample size, technique, staining, country and published journal name were abstracted from the primary articles (Table 1). Also, the numbers of true-positive (TP), false-negative (FN), false-positive (FP) and true-negative (TN) results were extracted in Table 1. The type of study for each eligible paper, three basic blinding questions, and recruitment are summarized (Supplementary Table 1). Only one study was retrospective; the others were prospective [38]. In 11 studies, cytological and histological specimens were measured independently; and two studies were not clearly documented [38,39]. And additional characteristics of these 13 studies are summarized (Supplementary Table 2). Only five studies indicated the use of tobacco and/or alcohol. Nine of the 13 studies described the location of the lesion and nearly studies assumed the patients’ gender and age. The Feulgen staining technique was used in five studies. Of these, three used the Papanicolaou staining technique to be evaluated by a cytotechnologist before Feulgen staining [19,37,43]. Three studies used only the Papanicolaou staining technique and one study used hematoxylin and eosin staining [16]. The staining techniques used in the remaining studies were not identified. Blinding in varying degrees was sufficient in a majority of the studies and well reported. There were nine studies that blinded the clinician or pathologist, who evaluated the cytology specimens from the histopathology results and vice versa. The method of recruitment was reported in all but one study [43]. Standard histopathological techniques were used in all studies. Only those lesions that underwent both examinations were included in this meta-analysis. Meta-analysis Heterogeneity and quality assessment X2 test and I2 index of the likelihood ratios of these two diagnostic techniques were evaluated (Supplementary Table 3). The inconsistency (I2 > 70%, P for Cochrane Q test <0.1) detected in the meta-analysis of OralCDx brush biopsy was high. The statistically relative heterogeneity of DOR of DNA-image cytometry was evaluated (I2 = 55.7%, P for Cochrane Q test <0.1). And threshold analysis was conducted to explore whether the high inconsistency was caused by different threshold. Considering the Spearman correlation coefficient 0.595 (p = 0.120), and 0.000 (p = 1.000), no threshold existed in either of the two diagnostic methods. The main features of the methodological qualities of the included studies and the quality assessment of diagnostic accuracy studies (QUADAS) records were summarized (Fig. 2). The majority of studies applied the ideal reference standard were of high quality.
Result Literature retrieval In total, 4391 studies were identified for online screening (Fig. 1). Twenty-four articles were thought to be potentially
Pooled sensitivity, specificity The individual and pooled sensitivities as well as specificities with 95% confidence intervals (CIs) of each study were extracted and calculated (Table 2). Using the random effects model, OralCDx brush biopsy had a pooled sensitivity of 86% (95% CI 81–90) and
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
4
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Identification
Records identified through database searches (n=4391)
Total records identified, duplicates removed (n=358)
Screening
Records screened (n=87)
Full-text articles assessed for eligibility (n=70)
Eligibility
Failed inclusion criteria (n=46)
Included in qualitative synthesis (n=24)
Included
Excluded from quantitative synthesis (n=11) 1.Not computer-assisted (n=7) 2.Not detailed data (n=2) 3.After treatment (n=1) 4.Method unclear (n=1) Included in quantitative synthesis (n=13) Fig. 1. Flow diagram for selection of studies included in this meta-analyses.
Table 1 Study characteristics of included studies. Published time
Literature
Patients
Technique
Staining
Country
TP
FP
TN
FN
1999 2001
Sciubba Remmerbach
945 251
OralCDx brush biopsy DNA-image cytometry
U.S. Germany
131 53
14 0
182 195
0 2
2004 2004
Poate Maraki
112 98
OralCDx brush biopsy DNA-image cytometry
UK Germany
15 17
19 2
9 79
6 0
2004 2009 2009
Scheifele Hohlweg-Majert Remmerbach
103 69 47
OralCDx brush biopsy OralCDx brush biopsy DNA-image cytometry
Germany Germany Germany
24 25 18
4 6 0
66 23 27
2 15 2
2011 2012 2012 2013 2014 2014
Mehrotra Reddy Seijas-Naya Kammerer Ma Casparis
85 60 24 70 52 200
OralCDx brush biopsy OralCDx brush biopsy OralCDx brush biopsy DNA-image cytometry DNA-image cytometry OralCDx brush biopsy
Papanicolaou Feulgen Papanicolaou Not mentioned Feulgen Papanicolaou Not mentioned Not mentioned Feulgen Papanicolaou Not mentioned H&E Papanicolaou Feulgen Feulgen Papanicolaou
India India Spain Germany China Germany
26 7 9 21 19 17
1 5 0 0 3 19
47 39 13 57 27 15
5 9 2 9 3 2
TP: true positive; FP: false positive; TN: true negative; FN: false negative.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
5
Fig. 2. Methodological quality graph and summary: review authors’ judgements about each methodological quality item for each included study.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
6
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Table 2 Pooled sensitivity and specificity of both OralCDx brush biopsy and DNA-image cytometry. Study
Sen [95% Conf. Interval.]
A 1999 2004 2004 2009 2011 2012 2012 2014
1.000 0.714 0.923 0.625 0.839 0.438 0.727 0.895
0.972 0.478 0.749 0.458 0.663 0.198 0.390 0.669
1.000 0.887 0.991 0.773 0.945 0.701 0.940 0.987
0.929 0.321 0.943 0.207 0.979 0.886 0.923 0.441
0.883 0.159 0.860 0.080 0.889 0.754 0.640 0.272
0.960 0.524 0.984 0.397 0.999 0.962 0.998 0.621
Pooled Sen
0.858
0.812
0.895
0.814
0.775
0.848
B 2001 Remmerbach (DNA) 2004 Maraki 2009 Remmerbach 2013 Kammerer (DNA) 2014 Ma
0.964 1.000 0.900 0.700 0.864
0.875 0.805 0.683 0.506 0.651
0.996 1.000 0.988 0.853 0.971
1.000 0.975 1.000 1.000 0.900
0.981 0.914 0.872 0.937 0.735
1.000 0.997 1.000 1.000 0.979
Pooled Sen
0.889
0.826
0.935
0.987
0.970
0.996
Sciubba Poate Scheifele Hohlweg-Mejert Mehrotra Reddy Seijas-Naya Casparis
Spe [95% Conf. Interval.]
A represented OralCDx brush biopsy and B represented DNA-image cytometry.
A
Diagnostic OR (95% CI)
B
2001 Remmerbach(DNA) 2004 Maraki 2009 Remmerbach 2013 Kammerer(DNA) 2014 Ma
0.01
1
Diagnostic Odds Ratio
8,367.40 (395.72 - 176,927.15) 1,113.00 (51.14 - 24,221.00) 407.00 (18.46 - 8,971.90) 260.26 (14.51 - 4,667.39) 57.00 (10.37 - 313.43)
Random Effects Model Pooled Diagnostic Odds Ratio = 446.08 (73.36 to 2712.43) Cochran-Q = 9.04; df = 4 (p = 0.0601) 100.0 Inconsistency (I-square) = 55.7 % Tau-squared = 2.3073
Fig. 3. Diagnostic odd ratio of OralCDx brush biopsy (A) and DNA-image cytometry (B) were showed.
pooled specificity of 81% (95% CI 78–85), while the pooled sensitivity and specificity of DNA-image cytometry were 89% (95% CI 83–94) and 99% (95% CI 97–100). Although the sensitivity of DNA-image cytometry was higher than that of OralCDx brush biopsy, the difference was not statistically significant (Z = 1.520, p > 0.05). The specificity of DNA-image cytometry was significantly higher than that of OralCDx brush biopsy (Z = 2.821, p < 0.05).
DOR and the likelihood ratios The DOR plots are presented in Fig. 3. Heterogeneity was considered as a result of the non-threshold effect because DOR of each study was not linear with the pooled DOR. DOR estimates for OralCDx brush biopsy and DNA-image cytometry were 20.36 (95% CI 2.72–152.67) and 446.08 (95% CI 73.36–2712.43), respectively. Estimates of LRs for individual studies were shown in Supplementary
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
7
Fig. 4. Summary receiver operating characteristic curves displaying the fitted joint variation of sensitivity and specificity of OralCDx brush biopsy (A) and DNA-image cytometry (B). The dots represent the observed sensitivity and specificity in individual studies.
Table 4. The pooled estimates of positive LRs for DNA-image cytometry increased to 38.081 (95% CI 9.64–150.39) compared to OralCDx brush biopsy (4.472; 95% CI 1.489–13.427). The opposite trend was evaluated on the pooled estimates of negative LRs. SROC curve, AUC, and Q* index The overall relationship between sensitivity and 1-specificity was estimated from an unweighted SROC mode (Fig. 4). Results
were shown to generally cluster in the top left-hand corner of the ROC space and the pooled mean sensitivity and specificity laid very close to the SROC curve. Consequently, the pooled sensitivity and specificity were considered sufficient summary measures for this meta-analysis. In addition, the area under the SROC curve was evaluated. The AUCs of OralCDx brush biopsy and DNA-image cytometry were 0.8879 and 0.9885, respectively, a statistically significant difference (Z = 1.711, p < 0.05). The Q⁄ index
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
8
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
estimates for DNA-image cytometry (Q⁄ = 0.9548) was higher than that of OralCDx brush biopsy (Q⁄ = 0.8186) (Z = 1.727, p < 0.05). Regression analysis and publication bias Meta-regression was used to explore possible sources of heterogeneity. Staining technique, study size, and design method were analyzed by meta-regression. Finally, study size was discovered to be closely related to heterogeneity among studies. To evaluate publication bias, Deeks plots were designed using the log diagnostic odd ratios of individual studies against their sample size. There was publication bias when evaluating the study of OralCDx brush biopsy (p = 0.01), but no publication bias in the study of DNA-image cytometry (p = 0.06). Discussion This was the first meta-analysis to compare OralCDx brush biopsy with DNA-image cytometry, two computer-assisted diagnostic tools for clinically suspected oral precancer and oral cancer. The results showed that both of methods have a significant potential for diagnosis with high sensitivity and specificity, but that DNA-image cytometry has higher sensitivity and specificity than OralCDx brush biopsy. Because of the higher specificity, AUC, and Q⁄ index of SROC, it is proposed that DNA-aneuploidy cytology is more accurate in diagnosing oral precancer and oral cancer. Until now, scalpel biopsy with histological assessment was considered to be the only accepted method by which to definitely evaluate suspicious oral lesions [47]. Considering the invasiveness and low-patient acceptance of incision biopsy, respectively simple and repeatable methods were needed [48]. Brush cytology, as a diagnostic method, has numerous advantages [10]. First, it offers the opportunity for early diagnosis by the clinician and decreases the time from diagnosis to treatment [49]. Second, it is a simple, non-invasive, and relatively painless technique that is well accepted by patients [50]. Finally, to those very sick patients with suspected oral cancer or those who are allergic to local anesthesia, brush cytology can be a beneficial diagnostic method [51]. However, it has certain limitations [10]. For example, it has been reported that this technique is not able to differentiate tumor subtypes [10]. Further, brush cytology sampling is diagnosed by a cytologist and human error and bias can exist to some extent [52]. Thus, the clinician should analyze the results by combining clinical manifestations. Both OralCDx brush biopsy and DNA-image cytometry can assess cellular benignity or malignancy using computer-assisted analysis which can improve their quality and reliability as a cancer diagnostic technique [10,53]. The improved accuracy is attributed to the ease of obtaining full transepithelial cellular samples [53]. In addition, computer-assisted analysis can reduce human cognitive bias. The evaluation of smears using an image analysis system has been adapted specifically to detect oral epithelial abnormalities [16]. Patton et al. found that OralCDx brush biopsy, as an adjunctive technique, was useful in assessing dysplastic changes in clinically suspicious lesions [18]. In addition, compared with conventional brush biopsy, OralCDx computer can search the brush biopsy specimen for a combination of abnormal cellular morphology and abnormal keratinization, which uniquely characterize dysplasia and carcinoma of the oral epithelium [17]; however, Patton et al. concluded that there were insufficient data that met the inclusion criteria to be able assess the availability in innocuous mucosal lesions [18]. It is known that molecular and genetic changes might precede both clinical and microscopic morphological changes and might be present in histologically benign tissue [54]. Furthermore,
DNA-aneuploidy is an internationally accepted marker for carcinoma transformation of cells [55,56]. Previous result suggests that DNA aneuploidy arises in even histopathologically mild oral epithelial dysplasia, and such lesions will transform into oral squamous cell carcinoma [57]. Remmerbach et al. studied the reliability of oral brush cytology and its DNA cytometric analysis in the early detection of oral cancer. The results of the study were substantial because by combining both techniques, the sensitivity increased to 98.2% and the specificity was 100%, with a positive predictive value of 100% and a negative value of 99.5% [37]. Maraki et al. also found that the sensitivity of cytological diagnosis combined with DNA-image cytometry was 100%, and specificity was 97.4% [19]. Remmerbach et al. also concluded in the clinical setting that DNA-aneuploidy might be used to detect histologically obvious malignancy, from 1 to 15 months prior to histology [37]. Sudbo et al. analyzed archival material and reported that nuclear DNA in the cells of oral leukoplakia could be used to predict the risk of oral epithelial dysplasia up to 5 years before histological diagnosis [57]. There remain some deficiencies in the methodological assessment of diagnostic tests. Some of the clinical characteristic data of the primary studies used in this meta-analysis were undocumented, but some were useful and important, such as smoking or tobacco chewing, which are factors that could cause oral cancer. [19,39,43]. In addition, there was selection bias for including only articles written in English but ignoring those unpublished or published studies and articles written in other languages. Furthermore, only five studies classified their results, including inadequate categories and reported those cases with incomplete transepithelial cells, which could be beneficial information in the development of brush technology [16,41]. From the diagnostic meta-analysis, the Q⁄ value of SROC curve analysis that considers the nonlinear relationship between sensitivity and specificity, is a reflection of the accuracy of the combined index on the diagnostic tests [58]. Thus, the results are more scientific and accurate and the diagnostic tests can be intuitively compared through a graphical area using the AUC value [58]. In this meta-analysis, the curve of DNA-image cytometry was closer to the upper left corner than that of OralCDx brush biopsy. In addition, the Q⁄ index estimates for DNA-image cytometry was statistically significantly higher than that of OralCDx brush biopsy; however, there is relatively less primary research about DNAimage cytometry, which might result in an overestimate its diagnostic value. Acha et al. claimed that brush biopsy will never prevail over the classic biopsy, and that a biopsy should be done on all clinically suspicious lesions even with a benign cytological diagnosis [53]. However, the cytological study of oral precancer and oral cancer is rapid, non-aggressive and relatively painless compared with that of oral scalpel biopsy and is well accepted by patients and suitable for routine application in population screening programs, for the early analysis of suspect lesions, and for pre-and post-treatment monitoring of confirmed malignant lesions.
Conclusion The results of this meta-analysis suggest that DNA-image cytometry has a highly significant potential over OralCDx brush biopsy as an accurate and simple diagnostic tool for clinically suspected oral precancer and oral cancer.
Conflict of interest statement The authors declare no conflict of interest.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
Acknowledgement This work was supported by Grants from the National Natural Science Foundation of China (Nos. 81170972 and 81371147) to Prof. Zhou Gang. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.oraloncology. 2015.09.002. References [1] Adams D. Oral cancer: early diagnosis. Dent Today 2014;33:8. [2] Gaitan-Cepeda LA, Peniche-Becerra AG, Quezada-Rivera D. Trends in frequency and prevalence of oral cancer and oral squamous cell carcinoma in Mexicans. A 20 years retrospective study. Med Oral, Patol Oral Cir Bucal 2011;16:e1–5. [3] Guneri P, Epstein JB. Late stage diagnosis of oral cancer: components and possible solutions. Oral Oncol 2014;50:1131–6. [4] Mehrotra R, Gupta DK. Exciting new advances in oral cancer diagnosis: avenues to early detection. Head Neck Oncol 2011;3:33. [5] Jafari A, Najafi S, Moradi F, Kharazifard M, Khami M. Delay in the diagnosis and treatment of oral cancer. J Dent 2013;14:146–50. [6] Hunter KD, Yeoman CM. An update on the clinical pathology of oral precancer and cancer. Dent Update 2013;40. p. 120–2, 5–6. [7] Baykul T, Yilmaz HH, Aydin U, Aydin MA, Aksoy M, Yildirim D. Early diagnosis of oral cancer. J Int Med Res 2010;38:737–49. [8] O’Sullivan E. Improving early diagnosis of oral cancer. J Oral Maxillofac Surg: Off J Am Assoc Oral Maxillofac Surg 2004;62:115. [9] Warnakulasuriya S, Reibel J, Bouquot J, Dabelsteen E. Oral epithelial dysplasia classification systems: predictive value, utility, weaknesses and scope for improvement. J Oral Pathol Med: Off Publ Int Assoc Oral Pathol Am Acad Oral Pathol 2008;37:127–33. [10] Diniz-Freitas M, Garcia-Garcia A, Crespo-Abelleira A, Martins-Carneiro JL, Gandara-Rey JM. Applications of exfoliative cytology in the diagnosis of oral cancer. Medicina oral: organo oficial de la Sociedad Espanola de Medicina Oral y de la Academia Iberoamericana de Patologia y Medicina Bucal 2004;9:355–61. [11] Cikojevic D, Gluncic I, Pesutic-Pisac V. Role of exfoliative cytology in diagnosis of laryngeal tumors. Acta Cytol 2007;51:767–72. [12] Perez-Sayansm M, Somoza-Martin JM, Barros-Angueira F, Reboiras-Lopez MD, Gandara-Vila P, Gandara Rey JM, et al. Exfoliative cytology for diagnosing oral cancer. Biotech Histochem: Off Publ Biol Stain Comm 2010;85:177–87. [13] Nichols ML, Quinn Jr FB, Schnadig VJ, Zaharopoulos P, Hokanson JA, Des Jardins L, et al. Interobserver variability in the interpretation of brush cytologic studies from head and neck lesions. Arch Otolaryngol – Head Neck Surg 1991;117:1350–5. [14] Navone R, Pentenero M, Gandolfo S. Liquid-based cytology in oral cavity squamous cell cancer. Curr Opin Otolaryngol Head Neck Surg 2011;19:77–81. [15] Delavarian Z, Mohtasham N, Mosannen-Mozafari P, Pakfetrat A, Shakeri MT, Ghafoorian-Maddah R. Evaluation of the diagnostic value of a modified liquidbased cytology using OralCDx brush in early detection of oral potentially malignant lesions and oral cancer. Med Oral, Patol Oral Cir Bucal 2010;15: e671–6. [16] Reddy SG, Kanala S, Chigurupati A, Kumar SR, Poosarla CS, Reddy BV. The sensitivity and specificity of computerized brush biopsy and scalpel biopsy in diagnosing oral premalignant lesions: a comparative study. J Oral Maxillofac Pathol: JOMFP 2012;16:349–53. [17] Sciubba JJ, U.S. Collaborative OralCDx Study Group. Improving detection of precancerous and cancerous oral lesions: computer-assisted analysis of the oral brush biopsy. J Am Dent Assoc 1999;130:1445–57. [18] Patton LL, Epstein JB, Kerr AR. Adjunctive techniques for oral cancer examination and lesion diagnosis: a systematic review of the literature. J Am Dent Assoc 2008;139:896–905. quiz 93-4. [19] Maraki D, Becker J, Boecking A. Cytologic and DNA-cytometric very early diagnosis of oral cancer. J Oral Pathol Med: Off Publ Int Assoc Oral Pathol Am Acad Oral Pathol 2004;33:398–404. [20] Giaretti W, Pentenero M, Gandolfo S, Castagnola P. Chromosomal instability, aneuploidy and routine high-resolution DNA content analysis in oral cancer risk evaluation. Future Oncol 2012;8:1257–71. [21] Rajagopalan H, Lengauer C. Aneuploidy and cancer. Nature 2004;432:338–41. [22] Torres-Rendon A, Stewart R, Craig GT, Wells M, Speight PM. DNA ploidy analysis by image cytometry helps to identify oral epithelial dysplasias with a high risk of malignant progression. Oral Oncol 2009;45:468–73. [23] Whiting PF, Weswood ME, Rutjes AW, Reitsma JB, Bossuyt PN, Kleijnen J. Evaluation of QUADAS, a tool for the quality assessment of diagnostic accuracy studies. BMC Med Res Methodol 2006;6:9. [24] Deeks S. Clinical implications of virological ‘‘failure”: interview with Steven Deeks, M.D., San Francisco General Hospital. Interview by John S. James. AIDS Treat News 1998:1–6.
9
[25] Jaeschke R, Guyatt GH, Sackett DL, The Evidence-Based Medicine Working Group. Users’ guides to the medical literature: III. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients? Jama 1994;271:703–7. [26] Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostic test into a summary ROC curve: data-analytic approaches and some additional considerations. Stat Med 1993;12:1293–316. [27] Remmerbach TW, Weidenbach H, Muller C, Hemprich A, Pomjanski N, Buckstegge B, et al. Diagnostic value of nucleolar organizer regions (AgNORs) in brush biopsies of suspicious lesions of the oral cavity. Anal Cell Pathol: J Eur Soc Anal Cell Pathol 2003;25:139–46. [28] Mehrotra R, Singh MK, Pandya S, Singh M. The use of an oral brush biopsy without computer-assisted analysis in the evaluation of oral lesions: a study of 94 patients. Oral Surg, Oral Med, Oral Pathol, Oral Radiol, Endodont 2008;106:246–53. [29] Guneri P, Epstein JB, Kaya A, Veral A, Kazandi A, Boyacioglu H. The utility of toluidine blue staining and brush cytology as adjuncts in clinical examination of suspicious oral mucosal lesions. Int J Oral Maxillofac Surg 2011;40:155–61. [30] Koch FP, Kunkel M, Biesterfeld S, Wagner W. Diagnostic efficiency of differentiating small cancerous and precancerous lesions using mucosal brush smears of the oral cavity – a prospective and blinded study. Clin Oral Investigat 2011;15:763–9. [31] Babshet M, Nandimath K, Pervatikar S, Naikmasur V. Efficacy of oral brush cytology in the evaluation of the oral premalignant and malignant lesions. J Cytol/Indian Acad Cytol 2011;28:165–72. [32] Graveland AP, Bremmer JF, de Maaker M, Brink A, Cobussen P, Zwart M, et al. Molecular screening of oral precancer. Oral Oncol 2013;49:1129–35. [33] Mulki S, Shetty P, Pai P. Oral rinse as a simpler approach to exfoliative cytology: a comparative study. J Clin Diagn Res: JCDR 2013;7:3036–8. [34] Femiano F, Scully C. DNA cytometry of oral leukoplakia and oral lichen planus. Med Oral, Patol Oral Cir Bucal 2005;10(Suppl. 1):E9–E14. [35] Mishra M, Mohanty J, Sengupta S, Tripathy S. Epidemiological and clinicopathological study of oral leukoplakia. Indian J Dermatol, Venereol Leprol 2005;71:161–5. [36] Schwarz F, Maraki D, Yalcinkaya S, Bieling K, Bocking A, Becker J. Cytologic and DNA-cytometric follow-up of oral leukoplakia after CO2- and Er:YAG-laser assisted ablation: a pilot study. Lasers Surg Med 2005;37:29–36. [37] Remmerbach TW, Weidenbach H, Pomjanski N, Knops K, Mathes S, Hemprich A, et al. Cytologic and DNA-cytometric early diagnosis of oral cancer. Anal Cell Pathol: J Eur Soc Anal Cell Pathol 2001;22:211–21. [38] Poate TW, Buchanan JA, Hodgson TA, Speight PM, Barrett AW, Moles DR, et al. An audit of the efficacy of the oral brush biopsy technique in a specialist oral medicine unit. Oral Oncol 2004;40:829–34. [39] Scheifele C, Schmidt-Westhausen AM, Dietrich T, Reichart PA. The sensitivity and specificity of the OralCDx technique: evaluation of 103 cases. Oral Oncol 2004;40:824–8. [40] Hohlweg-Majert B, Deppe H, Metzger MC, Schumm S, Hoefler H, Kesting MR, et al. Sensitivity and specificity of oral brush biopsy. Cancer Investigat 2009;27:293–7. [41] Seijas-Naya F, Garcia-Carnicero T, Gandara-Vila P, Couso-Folgueiras E, PerezSayans M, Gandara-Vila R, et al. Applications of OralCDx(R) methodology in the diagnosis of oral leukoplakia. Med Oral, Patol Oral Cir Bucal 2012;17:e5–9. [42] Kammerer PW, Koch FP, Santoro M, Babaryka G, Biesterfeld S, Brieger J, et al. Prospective, blinded comparison of cytology and DNA-image cytometry of brush biopsies for early detection of oral malignancy. Oral Oncol 2013;49:420–6. [43] Remmerbach TW, Meyer-Ebrecht D, Aach T, Wurflinger T, Bell AA, Schneider TE, et al. Toward a multimodal cell analysis of brush biopsies for the early detection of oral cancer. Cancer 2009;117:228–35. [44] Ma JM, Zhou TJ, Wang R, Shan J, Wu YN, Song XL, et al. Brush biopsy with DNAimage cytometry: a useful and noninvasive method for monitoring malignant transformation of potentially malignant oral disorders. Eur Arch Oto-rhinolaryngol: Off J Eur Fed Oto-Rhino-Laryngol Soc 2014. [45] Casparis S, Borm JM, Tomic MA, Burkhard A, Locher MC. Transepithelial brush biopsy – oral CDx(R) – a noninvasive method for the early detection of precancerous and cancerous lesions. J Clin Diagn Res: JCDR 2014;8:222–6. [46] Mehrotra R, Mishra S, Singh M, Singh M. The efficacy of oral brush biopsy with computer-assisted analysis in identifying precancerous and cancerous lesions. Head Neck Oncol 2011;3:39. [47] Richards D. Adjunctive tests cannot replace scalpel biopsy for oral cancer diagnosis. Evid Dent 2015;16:46–7. [48] Brocklehurst P, Kujan O, O’Malley LA, Ogden G, Shepherd S, Glenny AM. Screening programmes for the early detection and prevention of oral cancer. Cochrane Database Syst Rev 2013;11:CD004150. [49] Gupta S, Shah JS, Parikh S, Limbdiwala P, Goel S. Clinical correlative study on early detection of oral cancer and precancerous lesions by modified oral brush biopsy and cytology followed by histopathology. J Cancer Res Ther 2014;10:232–8. [50] Driemel O, Kosmehl H, Rosenhahn J, Berndt A, Reichert TE, Zardi L, et al. Expression analysis of extracellular matrix components in brush biopsies of oral lesions. Anticancer Res 2007;27:1565–70. [51] Collins BM. The oral brush biopsy: an adjunct to early oral cancer detection. Pennsylvania Dent J 2002;69:35–7. [52] Bocking A, Sproll C, Stocklein N, Naujoks C, Depprich R, Kubler NR, et al. Role of brush biopsy and DNA cytometry for prevention, diagnosis, therapy, and followup care of oral cancer. J Oncol 2011;2011:875959.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002
10
X. Ye et al. / Oral Oncology xxx (2015) xxx–xxx
[53] Acha A, Ruesga MT, Rodriguez MJ, Martinez de Pancorbo MA, Aguirre JM. Applications of the oral scraped (exfoliative) cytology in oral cancer and precancer. Med Oral, Patol Oral Cir Bucal 2005;10:95–102. [54] Thomson PJ. Field change and oral cancer: new evidence for widespread carcinogenesis? Int J Oral Maxillofac Surg 2002;31:262–6. [55] Nadjari B, Kersten A, Ross B, Motherby H, Krallmann R, Sundmacher R, et al. Cytologic and DNA cytometric diagnosis and therapy monitoring of squamous cell carcinoma in situ and malignant melanoma of the cornea and conjunctiva. Anal Quant Cytol Histol/Int Acad Cytol Am Soc Cytol 1999;21:387–96.
[56] Nadjari B, Motherby H, Pooschke T, Pooschke S, Gabbert HE, Simon D, et al. DNA aneuploidy as a specific marker of neoplastic cells in FNAB of the thyroid. Anal Quant Cytol Histol/Int Acad Cytol Am Soc Cytol 1999;21:481–8. [57] Sudbo J, Kildal W, Risberg B, Koppang HS, Danielsen HE, Reith A. DNA content as a prognostic marker in patients with oral leukoplakia. N Engl J Med 2001;344:1270–8. [58] Walter SD. Properties of the summary receiver operating characteristic (SROC) curve for diagnostic test data. Stat Med 2002;21:1237–56.
Please cite this article in press as: Ye X et al. Meta-analysis of two computer-assisted screening methods for diagnosing oral precancer and cancer. Oral Oncol (2015), http://dx.doi.org/10.1016/j.oraloncology.2015.09.002