Critical update, systematic review, and meta‐analysis of oral erythroplakia as an oral potentially malignant disorder

Abstract Background Oral erythroplakia has been classically considered as the potentially malignant disorder with the highest rate of malignant development into squamous cell carcinoma. This critical systematic review and meta‐analysis aim to estimate the malignant development rate of oral erythroplakia and identify the associated risk factors. Methods We performed a bibliographic search in PubMed, Scopus, Web of Science, Embase, and LILACS, with keywords “erythroplakia,” “erythroplasia,” “malignant transformation,” “malignant development,” “malignization,” “carcinogenesis,” “oral cancer,” “oral squamous cell carcinoma,” “mouth neoplasm,” and “prognosis.” Meta‐analysis was conducted using a random‐effects model. Results Ten observational studies with 441 patients met the inclusion criteria, whose mean malignant development rate was 12.7% and with a mean follow‐up period of patients of 6.66 years. In the initial biopsy, 42.8% of oral erythroplakia were already squamous cell carcinoma. The buccal mucosa was the most frequent location of oral erythroplakia, but the floor of the mouth was the most common site of malignant development. All patients who underwent malignant development showed epithelial dysplasia on the initial diagnostic biopsy. Conclusion Overall malignant development rate of OE in the meta‐analysis was 19.9%. We could not associate any specific clinicopathological feature with the malignant development. The presence of epithelial dysplasia in the initial biopsy remains the worst prognostic factor. Further observational studies on OE are needed, with well‐established diagnostic criteria and good clinical follow‐up, in order to identify the true risk of malignant development of oral erythroplakia and the related risk factors.

database LILACS, and the Conference Proceedings Citation Index databases. Searches combined thesaurus (e.g., MeSH and EMTREE) and free terms to maximize sensitivity. The algorithms used in each database contained the following keywords: "erythroplakia," "erythroplasia," "malignant transformation," "malignant development," "malignization," "carcinogenesis," "oral cancer," "oral squamous cell carcinoma," "mouth neoplasm," and "prognosis." These words were subjected to syntax adaptation for each database. All of the databases were searched from inception to February 2022 (upper limit).
We performed a second search, introducing each keyword in an unstructured fashion to check whether every article on the topic was retrieved. Potentially relevant articles that any of the authors were already familiar with and reference lists from the retrieved articles were also comprehensively checked. All retrieved references were managed using Mendeley v.1.19.9 (Elsevier, Amsterdam, The Netherlands).

| Eligibility criteria
Criteria for eligibility of the studies included in the qualitative and quantitative analysis were as follows: (1) original research articles published in English language, (2) longitudinal observational studies, and (3) studies that analyses the MD of OE. The exclusion criteria were as follows: (1) letters to the editor, case reports, conference abstracts, clinical trials, etc.; (2) previous systematic reviews and meta-analysis; (3) studies which reported on patients with diagnosis of erythroleukoplakia; (4) studies without a complete clinicopathological diagnosis of OE; and (5) studies that did not report specific data on OE.

| Data selection and extraction
Articles were selected in two phases by two authors (APJ and MPS), first by screening titles and abstracts that apparently meet inclusion criteria, and then by reading the full texts to assess their final inclusion. During the calibration exercise, reviewers thoroughly discussed the criteria and applied them to a sample of 50% of the retrieved studies to determine the inter-examiner agreement. After adequate agreement was achieved (κ score = 0.78), all of the studies were independently read by the reviewers.
Any discrepancies between the researchers were resolved by a third or fourth researcher blinded to the study hypothesis (ILIM and JMA).
The information was retrieved by two investigators (APJ and MPS) using a custom-made extraction sheet. The recorded data included: the first author and year of publication, country, study design, sample size, gender and age of patients, tobacco/alcohol/betel consumption, number and location of OE, histopathological data, follow-up period, number of MD cases, and time until MD.

| Statistical analysis
In the main meta-analysis, we computed the prevalence MD events and that of total OE case by dividing the number of events by the sample size of the study. Then, we weighted the study-specific log prevalence by the inverse of their variance to compute a pooled prevalence. For retrieved studies, the Clopper-Pearson interval was applied to estimate 95% confidence intervals (CIs). Combined pooled proportions were estimated with a random-effect model (Der-Simonian and Laird method). A further pre-planned meta-analysis to acknowledge the effect of clinicopathological factors on the MD of OE was considered, but this was discarded due to poor reporting of primary data. Therefore, we opted to perform a meta-analysis of prevalence with the degrees of ED/presence of SCC in initial biopsy at diagnosis time-point.
For statistical heterogeneity analysis, Cochran Q (χ 2 ) and Higgins I 2 test parameters were calculated. Cochran's Q test p < 0.1 was considered significant to assume apparent heterogeneity. The Higgins I 2 statistic cut-off points of 25%, 50%, and 75% were considered to indicate low, moderate, and high heterogeneity, respectively. 13 We assessed publication bias, first visually, using funnel plots, and then, more formally, using the test proposed by Egger et al. 14

| Quality assessment and risk of bias
Risk of bias (RoB) was assessed using the Quality in Prognosis Studies-QUIPS tool, supported by Cochrane Prognosis Methods Group for prognosis studies. QUIPS considers the following domains: (1) study participation, (2) study attrition, (3) prognostic factor measurement, (4) outcome measurement, (5) study confounding, and (6) statistical analysis and reporting. 15 RoB was qualified as low, moderate, or high for each domain. Each item scored as high adds 3 points, while when scored as moderate 2 and as low 1 to overall quality assessment for each study. Studies were categorized as high quality when the overall score was >13. RoB was assessed by two authors (MPS and APJ). Discrepancies between authors were resolved by all participating authors by consensus.

| Bibliographical research
We identified 974 registers through the aforementioned search, whose abstracts were reviewed for contents relevant to the topic of this study, resulting in 633 exclusions. A total of 341 papers were then retrieved and, after careful consideration, 318 of them were excluded because their information was not useful for this study.
The remaining 23 (2.37%) registers were checked according to the inclusion/exclusion criteria set for this investigation.

| Malignant development
Only five studies 18,19,[22][23][24] stated the follow-up period of patients. In these, mean MD rate was 12.7%, with a mean follow-up period of 6.66 years (Table 1). In the three studies that reported cases of OE with MD, 19,23,24 mean MD rate was 21.2%, with a mean follow-up period of 8.25 years and a mean time until MD of 3.17 years. Conversely, in the two studies that did not report cases of MD, mean follow-up period was 5.07 years 18,22 (Table 1).
In the five studies without a follow-up period of patients, 16,17,20,21,25 mean MD of OE was 51.10%. Besides, 68.38% of cases of these studies were already SCC at the time of diagnosis.
Regarding the presence of ED, different evaluation systems were used; thus, we divided ED into low-grade ED (previous mild to moderate) and high-grade ED (previous moderate to severe), following the WHO proposal. 26 In the initial biopsy, 47.9% of cases showed ED: 76.6% low-grade ED and 23.4% high-grade ED. [16][17][18][19][20][21][22]24 A total of 42.8% of the OE corresponded to SCC at the initial biopsy (Table 3).
Only Feng et al. 24 reported the clinical data of patients who underwent MD: 8 females and 9 males, all of them older than 50 years. Also, 42% and 50% of patients were tobacco and alcohol consumers, respectively. 17,24 The most common area of MD was the FOM (25.6%), followed by the tongue (24.3%), the retromolar area (14.6%), the gingiva (13.4%), the soft palate (13.4%), and the buccal mucosa (8.5%). Only Feng et al. 24 stated that 47.1% of cases with MD showed high-grade ED and 52.9% low-grade ED on the initial diagnostic biopsy.

| Meta-analysis
In the main meta-analysis, three studies 19,23,24 testing the same hypothesis (prevalence of MD in OE) were included in the quantitative synthesis. The combined MD rate was 19.9% (95% CI = À1.6 to 41.4; I 2 = 91.74%, p-value by Q test = 0.00001) ( Figure 1A). A remarkable asymmetry with a lack of studies on the middle of the plot was seen ( Figure 1B); nonetheless, Egger's regression test neglected the existence of publication bias (p-value of the intercept = 0.276).
In the secondary analysis, we performed a pooled analysis on the prevalence of initial histopathological diagnosis: SCC, high-grade ED, low-grade ED, and absence of ED (Table 4). Globally, studies harbored a substantial degree of heterogeneity and no individual study seemed to represent an influential point that increased heterogeneity dramatically. Presence of ED was almost six times higher than absence of ED (41.2% vs. 7.19%). Observation of the asymmetry of the funnel plots ( Figure 2) and the statistical tests conducted for the same purpose confirmed the absence of "small-study" effects on the pooled prevalence of the initial diagnosis of SCC (p Egger = 0.549) and low-grade ED (p Egger = 0.685). The exceptions were high-grade ED (p Egger = 0.001) and absence of ED (p Egger = 0.002), for which bias could not be ruled out.

| Risk of bias
The QUIPS tool showed that RoB was high in three studies [16][17][18] and low in seven (18)(19)(20)(21)(22)(23)(24) (Figure S2). The most potential biases were inappropriate statistical analyses and poor assessment of study cofounding. Nevertheless, the quality of some studies was also suboptimal in the remaining four domains.

| DISCUSSION
Erythroplakia is a premalignant disorder first described as a raised, velvety red plaque in the male and female genital mucosa. 1,27 The first oral cases were published in the late twentieth century, 28,29 rapidly becoming a matter of controversy.
In 1955, Blau & Hyman 30 made the first review on OE, with lesions on the tongue, the buccal mucosa, and the lip. Later, Shear 31 proposed three clinical forms of OE (homogeneous, erythroleukoplakia, and speckled), with different prognostic aspects; which were later approved by several authors. 10,16 In 2007, "erythroleukoplakia" was considered an OPMD independent of OE, 32 appearing as such in the latest WHO Classification of Head and Neck Tumours. 26 Recently, erythroleukoplakia has been once again reconsidered as a clinical form of nonhomogenous of OL, 9 removing it as an individual OPMD.
Taking into account the current definition of OPMD: "any oral mucosal abnormality that is associated with a statistically increased risk of developing oral cancer", 9 we believe that some aspects of previous reviews about the MD of OE are somewhat questionable. 6 We think that clinical follow-up of patients is necessary to assess the true MD of OE as an OPMD. Therefore, when we considered the studies without a follow-up period, we achieved a "wrong" MD rate of 50.51%, similar to that of previous reviews (44.9%-50%). 6 All these findings question whether OE is the OPMD with the highest risk of MD, as previously stated. 9,32 We believe that OE should only be considered as an OPMD when (1) there is a clinical follow-up period of patients, and (2) the initial biopsy rules out the presence of an SCC. Since the follow-up period has been longer in the studies with cases of MD (8.25 vs. 5 years) (Table 1), we think patients with OE should be actively followed for life.
Except for the study by Chuang et al., 23 gender and age distribution were similar in patients with OE and those who suffer MD (Table 1). 41,42 The buccal mucosa has been the most frequent location of OE; which could be related to some particular carcinogenic habits, such as consumption of betel. 22,25 However, the absence of cases with MD in some studies 22,25 suggest that some of these cases may not be true OE.
We also observed interesting differences between the location OE and where MD occurred. Some infrequent areas of OE like the and histopathological differential diagnosis with all mucous pathologies clinically similar to OE, preferably SCC, but also oral lichenoid disease (OLD), erythematous candidiasis, lupus erythematosus, pemphigoid, and so forth. 9 Performing a good clinicopathological correlation is key to reach the final diagnosis of OE; thus, it is mandatory to get a good biopsy of all suspicious oral "red plaques" in order to (1) rule out other diagnoses and (2) assess the existence and/or degree of ED.
In 1972, Shear 31 proposed two histopathological types of OE: a "neoplastic" one, associated with SCC, carcinoma in situ or ED; and an "inflammatory" one, related to Candida infection, prosthetic stomatitis, immunological processes, and so forth. In our opinion, this classification is misleading and should not be used. Currently, we cannot consider as an OPMD an oral lesion that is already a SCC in the histopathological analysis; or can all "red plaque-like" lesions that may appear on the oral mucosa be considered as an OE. 9 Indeed, some multifocal cases of OE reviewed on this study, 22  multiple OPMD, such as OL, OLD, and proliferative multifocal/verrucous leukoplakia. [44][45][46] Unfortunately, we had several limitations during the performance of this systematic review and meta-analysis. First, is the low number of observational studies on the MD of OE, and with good clinical follow-up period of patients; and second, is the lack of standardization and changes in the diagnostic criteria for OE over the years. This jeopardized the association between the MD of OE and different clinicopathological features.