Effect of Probiotics on Oral Candidiasis: A Systematic Review and Meta-Analysis

Oral candidiasis (OC) is an increasing health problem due to the introduction of new drugs, population aging, and increasing prevalence of chronic illness. This study systematically reviews the effects of the oral intake of probiotics, prebiotics, and synbiotics on Candida spp. counts (colony-forming units (CFU)/mL) in oral and palatal samples. A literature search was conducted. Twelve studies, eight randomized clinical trials (RCTs), and four pre-post studies, resulted as eligible for the meta-analysis, which was performed through a Bayesian random-effects model. All studies analyzed probiotics, and none of them analyzed prebiotics or synbiotics. The treatments effects were measured in terms of odds ratio (OR) of OC (CFU/mL >102, 103, or 104). The meta-analytic OR was 0.71 (95% credibility interval (CrI): 0.37, 1.32), indicating a beneficial effect of treatment; the I2 index was 56.3%. Focusing only on RCTs, the OR was larger and more precise at 0.53 (95% CrI: 0.27, 0.93). The effect of treatment appeared to be larger on denture wearers. Our findings indicate that the intake of probiotics can have a beneficial effect on OC and that the effects could vary according to the patients’ characteristics. Due to the presence of medium–high-risk studies, the results should be interpreted with caution.


Introduction
Candida spp. represent a commensal yeast belonging to the normal microbiota localized on the surface of different body sites (skin, oral cavity, and the gastro-intestinal, uro-genital, and respiratory tracts) of human beings [1]. Candida spp. colonization of the mucus membranes occurs very early in life, usually at birth [2]. Under specific conditions, the fungus can switch from a harmless form into a pathogenic form that can lead to infections [3]. About 75% of healthy adults carry Candida spp. in the mouth; when there is a detection of a salivary Candida spp. count >400 colony-forming units of immunoglobulin A (IgA) production, increased production of anti-inflammatory cytokines, and induction of regulatory T cells [42]).
These probiotics' proprieties suggested their use for the treatment and prevention of many medical conditions (diarrhea, constipation, inflammatory bowel disease, irritable bowel syndrome, allergic disease), sometimes with excellent results [31]. In addition, probiotics also showed an antifungal action and were successfully used in mucosal candidiasis, as reported in an in vivo study by Wagner in 1997 [43]. Sookkhee et al., in 2001, studied the effects on Candida albicans growth of different lactic-acid bacteria isolated from the oral cavity of volunteers and found that two strains, Lactobacillus paracasei and Lactobacillus rhamnosus, had the strongest effect on the yeast [44].
Lactobacillus reuteri is a promising bacterium (especially DSM 17938 and ATCC PTA 5289) for its anti-Candida properties, confirmed by several studies. In one of these, Lactobacillus reuteri was demonstrated to be able to reduce Candida load in vivo through co-aggregation, modification of oral pH with production of lactic acid and other organic acids that inhibit the virulence of Candida cells, and production of H 2 O 2 [45].
In a recent in vitro study by Coman et al. (2014), the strains Lactobacillus rhamnosus IMC 501 and lactobacillus paracasei IMC 502, alone or in combination, showed an inhibitory effect on Candida spp. growth [46].
Recently, it was found that Lactobacillus rhamnosus GR-1 and Lactobacillus reuteri RC-14 modulate Candida glabrata virulence, through the complete inhibition of fungal biofilms [48].
In addition, Lactobacillus acidophilus ATCC 4356 was found to inhibit the biofilm formation of fungus through in vitro experiments [49]. Biofilm formation is probably reduced through the production of substances called "bacteriocins" by probiotics. Wannun et al. reported the isolation of a bacteriocin, called "fermencin SD11", from Lactobacillus fermentum SD11, a human oral Lactobacillus, which has a strong inhibitory effect on oral Candida cells [50].
In 1997, Wagner et al. showed that the administration of probiotics could be a prophylactic and therapeutic strategy for mucosal candidiasis [43]. They demonstrated that the presence of four strains of bacteria (Lactobacillus acidophilus, Lactibacillus reuteri, Lactobacillus casei GG, and Bifidobacterium animalis) in the gastro-intestinal tract of immunodeficient mice reduced the number of Candida albicans cells, as well as the incidence and severity of mucosal and systemic candidiasis, prolonging their survival [43].
In a murine model, Matsubara et al. inoculated Candida albicans in the oral cavity and subsequently administrated an antifungal drug (nystatin) or probiotics (Lactobacillus acidophilus and Lactobacillus rhamnosus). At the end of the experiment, colonization by yeast cells was lower in the group that received probiotics (particularly L. rhamnosus) than in the group treated with nystatin [51].
In conclusion, even if the mechanism of probiotics' antifungal effect remains to be fully elucidated, some authors explored it in vitro and in vivo studies, showing that these bacteria may contrast Candida spp. infection through different and synergistic mechanisms of action.
In this paper, we performed a systematic review and meta-analysis of clinical studies, randomized controlled trials (RCT), and pre-post intervention studies, with the aim of investigating the efficacy of probiotics (compared with a control treatment or placebo) on oral Candida spp. counts in subjects of any age, sex, nationality, or health status.

Literature Search
The PRISMA statement guidelines were followed for conducting and reporting a systematic review and meta-analysis [52].
A computerized search of the articles published from inception to 1 February 2019, was conducted in Embase, Medline/PubMed, Cochrane Library central, clinicaltrials.gov databases, and other individual journals sources (Brazilian Dental Journal, Indian Journal of Health Sciences, and Biomedical Research Kleu), using the following search string: (candidosis OR candidiasis OR oral Candida spp. OR thrush OR yeast infection) AND (probiotic OR prebiotic OR yogurt OR synbiotic OR Lactobacillus OR Bifidobacteria OR Saccaromyces OR Bacillus OR xylitol). In the PubMed database, we activated the filter "Humans"; in Embase, we activated the filter "Research articles"; in Cochrane Library, we activated the filter "Trials"; and, in clinicaltrials.gov, we activated the filter "recruitment: terminated or completed". No restrictions of language, country, duration of follow-up, and participants' characteristics (race, age, and sex) were imposed.

Study Selection
Two authors independently reviewed titles and abstracts of the collected articles, applying pre-defined inclusion /exclusion criteria. The inclusion criteria were as follows: (1) Randomized clinical trials or pre-post intervention studies; (2) Availability of full text; (3) Patients regardless of age, race, nationality, sex, and health status; (4) Comparison between oral intake of probiotics, prebiotics, or synbiotics (of any type and dosage) with a control treatment or a placebo in RCT; and between pre-and post-treatment conditions in pre-post intervention studies; (5) Outcome measurement expressed in CFU/mL of oral Candida spp. counts in saliva or palatal samples.
The exclusion criteria were as follows: (1) Studies with fewer than 10 participants; (2) Reviews, articles, and case reports; (3) Studies with incomplete outcome data; (4) Studies reporting results in a format which was not suitable for a meta-analysis, for example, without Candida spp. counts/carriage.

Data Extraction
The same two authors performed the analysis of the full texts and the data extraction, with the intervention of a third author in the case of poor agreement or discrepancies. Each reviewer independently recorded data in a predefined data extraction form. The following data were obtained from each selected trial: first author name, year of publication, study design, availability of a registered study protocol, setting (institution, city, and country), characteristics of the studied population (mainly age and health status), sample size, number of total participants at the end of follow-up, number of subjects, number of subjects in the treatment and control groups (for RCTs), experimental treatment (strain type or mixture type, dose in CFU/mL, and frequency of administration), control treatment, inclusion and exclusion criteria, follow-up duration, characteristics of the sample, and outcome measure expressed in Candida spp. counts (in CFU/mL).

Outcome Assessment
For each selected study, we calculated the odds ratio of OC (Candida spp. counts lower than a threshold of 10 2 , 10 3 , or 10 4 ) of treated subjects versus controls, as a measure of treatment effect.
For  [59], which reported a continuous outcome (mean of Candida spp. counts in CFU/mL), we firstly calculated the standardized mean difference (SMD) between the treatment group and the control group; then, we derived the OR and the corresponding standard error according to the Hasselblad and Hedges method [60]. The same approach was also used for two pre-post studies (Rane et al. 2018 [61]; Sutula et al. 2013 [62]). In these cases, we assumed independence between pre and post means, thus overestimating the standard errors. Rane et al. (2008) [61] reported the results of separated analyses conducted on three different samples of different age (50-59, 60-69, ≥70 years). A fixed-effects meta-analysis was performed on the three results in order to obtain an overall combined estimate and the corresponding standard error. Miyazima et al. (2017) [57] conducted an RCT with three treatment arms (placebo, treatment 1 with Lactobacillus acidophilus, and treatment 2 with Lactobacillus rhamnosus). We collapsed the results for the two experimental treatments in order to obtain an overall OR of treatment vs. placebo.
For the pre-post studies by Lopez-Jornet et al. 2018 [63] and Mendonca et al. 2012 [64], which reported the status of the patient in binary form, the McNemar OR was calculated [65].

Risk of Bias
The assessment of quality of the randomized clinical trials was performed using Review Manager 5.3 software, according to the Cochrane Handbook guidelines [66].
The two reviewers expressed, for each of the eight selected RCTs, their independent judgment (low risk, high risk, or unclear) on the following domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias.
The assessment of quality of the pre-post intervention studies was performed using the quality assessment tool for pre-post studies with no control group developed by the United States (US) National Heart, Lung, and Blood Institute (NHLBI) [67].
In the case of disagreement between the two authors, a third investigator was involved to resolve the controversy.

Statistical Analysis
A Bayesian random-effects meta-analysis model was specified to combine the results of the selected studies and, successively, to combine the results from the subset of the RCTs [68]. Let b i be the estimate of the log (OR) from the ith study, and s i be the estimate of the corresponding standard error (i = 1, 2, . . . , n). The random-effects meta-analysis assumes that the study-specific effects b i are mutually independent and follow the following model: where β is the overall meta-analytic effect, i is a random effect normally distributed with variance τ 2 , and ε i is an error term with known variance; u i and ε i are assumed to be independent. The meta-analysis model accounts for possible heterogeneity among studies through the random terms u i , and the variance τ 2 expresses the heterogeneity among studies. In the Bayesian formulation of the model, we need to specify prior distributions. Two non-informative priors were specified on the hyperparameters β and τ 2 . We used MCMC methods to obtain a sample from the joint posterior distribution of the parameters [69]. The posterior distribution of β was summarized in terms of mean and 95% credibility interval (CrI), i.e., the 2.5th and 97.5th percentiles of the posterior distribution. The posterior distribution of the I 2 index, which expresses the percentage of total variance captured by τ 2 , was summarized in terms of median and 95% CrI.
Sensitivity analyses were performed to evaluate the discrepancy between the results of the meta-analysis on the RCTs and the results obtained excluding the two RCTs that enrolled only subjects wearing dentures and the RCT that enrolled children.
When the meta-analysis included fewer than three studies, a fixed-effects models was adopted where only the within-study component of the variance was accounted for.

Study Selection
The initial search identified 2490 articles. In total, 87 studies were excluded for duplication. Of the remaining 2403 papers, 2385 were excluded after the screening for title and abstracts because they did not fulfil the inclusion criteria.
Among the 18 remained eligible articles, four were eliminated because the full text was not available, the outcome was inappropriate, or the results were not reported. Finally, 12 studies were included in the meta-analysis. Figure 1 illustrates the selection process, according to the PRISMA statement 2009. The presence of publication bias was evaluated by inspecting a funnel plot [70,71] and calculating the Begg's test [72]. All analyses were performed with R software (v3.5.1) [73] and the library OpenBUGS (v3.0.7) [74].

Study Selection
The initial search identified 2490 articles. In total, 87 studies were excluded for duplication. Of the remaining 2403 papers, 2385 were excluded after the screening for title and abstracts because they did not fulfil the inclusion criteria.
Among the 18 remained eligible articles, four were eliminated because the full text was not available, the outcome was inappropriate, or the results were not reported. Finally, 12 studies were included in the meta-analysis. Figure 1 illustrates the selection process, according to the PRISMA statement 2009.

Characteristics of the Included Studies
An overview of the included studies is reported in Table 2, which includes the reference, study design, setting, enrolled population, number of participants, intervention, comparison, follow-up, sample type, and outcome measurement.
The sample size ranged from 21 to 192. The total number of participants was 843, with an age between five and 100 years, with elderly people (70-100 years) being the most represented (n = 395). Three studies included only denture wearers [54,58,62], and one study involved schoolchildren [60].
The studies investigated a total of 16 strains (alone or in combination) of probiotics. The most represented were Lactobacillus spp., followed by Bifidobacterium spp., Saccaromyces spp., and in one case Propionibacterium spp., at doses from 5 × 10 5 to 5 × 10 9 CFU/mL one, two, or three times a day or less frequently (such as three times a week or every two weeks). None of the studies included prebiotics or synbiotics.
Most probiotics were given though lozenges or capsules, or with dairy products such as cheese, milk, or yogurt in some studies.
The length of the follow-up ranged from a minimum of four weeks to a maximum of 16 weeks. The reported outcomes measures were different among studies. Three studies reported the means of CFU/mL, two studies reported the presence (yes/no) of Candida spp. carriage, and seven studies reported the Candida spp. carriage (yes/no) based on different thresholds of Candida spp. counts (10 2 , 10 3 , or 10 4 ).
The sample size ranged from 21 to 192. The total number of participants was 843, with an age between five and 100 years, with elderly people (70-100 years) being the most represented (n = 395). Three studies included only denture wearers [54,58,62], and one study involved schoolchildren [60].
The studies investigated a total of 16 strains (alone or in combination) of probiotics. The most represented were Lactobacillus spp., followed by Bifidobacterium spp., Saccaromyces spp., and in one case Propionibacterium spp., at doses from 5 × 10 5 to 5 × 10 9 CFU/mL one, two, or three times a day or less frequently (such as three times a week or every two weeks). None of the studies included prebiotics or synbiotics.
Most probiotics were given though lozenges or capsules, or with dairy products such as cheese, milk, or yogurt in some studies.
The length of the follow-up ranged from a minimum of four weeks to a maximum of 16 weeks. The reported outcomes measures were different among studies. Three studies reported the means of CFU/mL, two studies reported the presence (yes/no) of Candida spp. carriage, and seven studies reported the Candida spp. carriage (yes/no) based on different thresholds of Candida spp. counts (10 2 , 10 3 , or 10 4 ).

Evaluation of the Risk of Bias for RCTs
The results of the risk of bias evaluation are reported in Figure 2. Risk of bias summary and graph. Green: low risk, yellow: unclear risk, red: high risk. In total, 50% of RCTs appropriately described the random sequence generation process, but only one In total, 50% of RCTs appropriately described the random sequence generation process, but only one reported the allocation concealment in detail. In most studies (87.5%), participants and personnel were blinded; in five studies, the outcome assessors were also blinded. Six studies out of eight had a low risk of incomplete data, with only one (Hatakka et al. 2007 [53]) being classified as having a high risk because 30% of participants dropped out during the study. Regarding selective reporting, one study (Burton et al., 2003 [60]) was judged to have a high risk because, in the article, the outcome of interest was incompletely reported. However, after e-mail contact with the authors, the reviewers received the data required. Finally, the reviewers assigned high risk to Keller et al., 2018 [55], because, as declared by authors, "fewer participants completed the study as projected, and the study was terminated before completion because of recruitment problems".

Evaluation of the Risk of Bias for Pre-Post Intervention Studies
On the basis of the quality appraisal criteria proposed by the NHLBI [68] for pre-post intervention studies with no control group, the study by Lopez et al. (2018) [64] was classified as having poor quality because the paper consisted of a correspondence letter which provided poor information about the trial. The other three studies were susceptible to some bias, but not sufficient to invalidate the results ( Figure 3). Thus, they were assigned to the fair quality category. reported the allocation concealment in detail. In most studies (87.5%), participants and personnel were blinded; in five studies, the outcome assessors were also blinded. Six studies out of eight had a low risk of incomplete data, with only one (Hatakka et al. 2007 [53]) being classified as having a high risk because 30% of participants dropped out during the study. Regarding selective reporting, one study (Burton et al., 2003 [60]) was judged to have a high risk because, in the article, the outcome of interest was incompletely reported. However, after e-mail contact with the authors, the reviewers received the data required. Finally, the reviewers assigned high risk to Keller et al., 2018 [55], because, as declared by authors, "fewer participants completed the study as projected, and the study was terminated before completion because of recruitment problems".

Evaluation of the Risk of Bias for Pre-Post Intervention Studies
On the basis of the quality appraisal criteria proposed by the NHLBI [68] for pre-post intervention studies with no control group, the study by Lopez et al. (2018) [64] was classified as having poor quality because the paper consisted of a correspondence letter which provided poor information about the trial. The other three studies were susceptible to some bias, but not sufficient to invalidate the results (Figure 3). Thus, they were assigned to the fair quality category.

Meta-Analysis Results
The results of the Bayesian random-effects meta-analysis conducted on the 12 evaluated studies are reported in the Figure 4.

Meta-Analysis Results
The results of the Bayesian random-effects meta-analysis conducted on the 12 evaluated studies are reported in the Figure 4. Overall, we estimated that the OC odds ratio in the treated patients was around 30% lower than in the controls (OR = 0.71; 95% CrI: 0.37, 1.32). However, it should be noted that the 95% confidence interval of the overall OR was not completely included in the beneficial effect region (OR < 1). The I 2 index, with a posterior median of 56.3% (95% CrI: 6.0%, 84.4%), suggests the presence of a relevant heterogeneity among studies.
When we restricted the meta-analysis to RCTs, a clear beneficial effect of treatment arose ( Figure  5). The estimated effect size was lower than in the previous analysis (OR = 0.53; 95% CrI: 0.27, 0.93), but the precision of the estimate was larger. As expected, this subset of studies appeared to be less heterogeneous (posterior median of I 2 = 32.2%; 95 CrI: 0.3%, 84.0%), even if a relevant discrepancy between results was still present.
In a sensitivity analysis, we performed a stratified meta-analysis distinguishing between RCTs performed on denture wearers and RCTs on non-denture wearers. The effect of treatment appeared to be larger in denture wearers, with an OR equal to 0.65 (95% CrI: 0.36, 1.17) for non-denture wearers ( Figure 6) versus an OR equal to 0.19 (95% CrI: 0.03, 1.29) for denture wearers. Overall, we estimated that the OC odds ratio in the treated patients was around 30% lower than in the controls (OR = 0.71; 95% CrI: 0.37, 1.32). However, it should be noted that the 95% confidence interval of the overall OR was not completely included in the beneficial effect region (OR < 1). The I 2 index, with a posterior median of 56.3% (95% CrI: 6.0%, 84.4%), suggests the presence of a relevant heterogeneity among studies.
When we restricted the meta-analysis to RCTs, a clear beneficial effect of treatment arose ( Figure 5).

Meta-Analysis Results
The results of the Bayesian random-effects meta-analysis conducted on the 12 evaluated studies are reported in the Figure 4. Overall, we estimated that the OC odds ratio in the treated patients was around 30% lower than in the controls (OR = 0.71; 95% CrI: 0.37, 1.32). However, it should be noted that the 95% confidence interval of the overall OR was not completely included in the beneficial effect region (OR < 1). The I 2 index, with a posterior median of 56.3% (95% CrI: 6.0%, 84.4%), suggests the presence of a relevant heterogeneity among studies.
When we restricted the meta-analysis to RCTs, a clear beneficial effect of treatment arose ( Figure  5). The estimated effect size was lower than in the previous analysis (OR = 0.53; 95% CrI: 0.27, 0.93), but the precision of the estimate was larger. As expected, this subset of studies appeared to be less heterogeneous (posterior median of I 2 = 32.2%; 95 CrI: 0.3%, 84.0%), even if a relevant discrepancy between results was still present.
In a sensitivity analysis, we performed a stratified meta-analysis distinguishing between RCTs performed on denture wearers and RCTs on non-denture wearers. The effect of treatment appeared to be larger in denture wearers, with an OR equal to 0.65 (95% CrI: 0.36, 1.17) for non-denture wearers ( Figure 6) versus an OR equal to 0.19 (95% CrI: 0.03, 1.29) for denture wearers. The estimated effect size was lower than in the previous analysis (OR = 0.53; 95% CrI: 0.27, 0.93), but the precision of the estimate was larger. As expected, this subset of studies appeared to be less heterogeneous (posterior median of I 2 = 32.2%; 95 CrI: 0.3%, 84.0%), even if a relevant discrepancy between results was still present.
In a sensitivity analysis, we performed a stratified meta-analysis distinguishing between RCTs performed on denture wearers and RCTs on non-denture wearers. The effect of treatment appeared to be larger in denture wearers, with an OR equal to 0.65 (95% CrI: 0.36, 1.17) for non-denture wearers ( Figure 6) versus an OR equal to 0.19 (95% CrI: 0.03, 1.29) for denture wearers. However, in interpreting this result, we should account for the fact that the meta-analysis on denture-wearing patients relied on only two studies.
A summary of the meta-analyses results is presented in Table 3. From the inspection of the funnel plot, no evidence of publication bias arose (Figure 8). These results were confirmed by the Begg's test (p = 0.80 for the meta-analysis on the RCTs, p = 1 for the meta-analysis on the 12 studies).  Table 3. Summary of the results from the Bayesian random-effect meta-analyses. CrI-credibility interval.

Type of Meta-Analysis
Meta-Analytic Estimate (OR) (95% CrI)  However, in interpreting this result, we should account for the fact that the meta-analysis on denture-wearing patients relied on only two studies.
When, in a second sensitivity analysis, we excluded from the RCTs the study by Burton et al. (2013) [60], which was conducted on schoolchildren, we observed a slight increase in effect size (OR = 0.44; 95% CrI: 0.25, 0.73) and a reduction in I 2 (posterior median of I 2 = 7.0%; 95% CrI: 0.2%, 76.2%) (Figure 7). However, in interpreting this result, we should account for the fact that the meta-analysis on denture-wearing patients relied on only two studies.
A summary of the meta-analyses results is presented in Table 3.
From the inspection of the funnel plot, no evidence of publication bias arose ( Figure 8). These results were confirmed by the Begg's test (p = 0.80 for the meta-analysis on the RCTs, p = 1 for the meta-analysis on the 12 studies).   A summary of the meta-analyses results is presented in Table 3. From the inspection of the funnel plot, no evidence of publication bias arose (Figure 8).

Discussion
Currently, fungal infections are widespread, especially in developed countries. A higher incidence of Candida spp. infections is associated with some predisposing factors such as the use of dentures, malnutrition, endocrine disorders, smoke, and some chronic diseases such as diabetes, HIV infection, and cancer [75]. The anti-OC treatment is mainly based on antifungal drugs, but different clinical types of OCs and the increasing number of multi-resistance phenotypes of Candida spp. represent current threats for public health. Consequently, the development of alternative therapeutic or complementary measures appears necessary to prevent the emergence of fungal resistance [76].
Many studies demonstrated that probiotics represent an efficient alternative treatment against Candida spp. infections. Moreover, they are easy to use and, thus, these products are usually well accepted by the patients [77]. The present study provides an overview of the literature on this issue, as well as a quantitative analysis that combines the results of independent studies of different design.
Both the meta-analysis on the 12 selected studies and the meta-analysis conducted on the subset of the RCTs indicated that the treatment had a beneficial effect on reducing oral Candida spp. counts.
As expected, the heterogeneity among studies was relevant because we combined studies of different design, which focused on different populations, used different treatments and doses, and were affected by different kinds and levels of bias. Our sensitivity analyses highlighted that part of the observed heterogeneity could be due to an actual difference of the treatment effect when used in different populations. For example, we found that the effect on denture wearers was larger than the effect estimated on non-denture wearers. The result on denture wearers relied only on two RCTs and should be interpreted with caution, but it is suggestive of a true difference. The larger reduction in the number of Candida spp. colonies in these patients could be caused by the direct application of probiotic products on the denture surface [57]. This hypothesis supports the idea that a lower effect of probiotics could be due to the low frequency of usage, number of probiotic cells, and delivery system, which exert an effect on the period of probiotics maintenance at the oral cavity. In this sense, the development of a mucoadhesive buccal drug delivery system [78], in order to enable the prolonged retention at the site of action, could improve the therapeutic outcome. An indication in favor of the relevance of the number of doses per day seems to arise also from the comparisons of the ORs in our meta-analysis. If we focus on RCTs on non-denture wearers, a larger effect was reported in Li et al. (2014) [60], where the treated patients received three doses per day instead of one or two. Three doses were administered also in Keller et al. 2018 [55], but this study was affected by recruitment problems, and the result relied on a very small number of subjects. These results were confirmed by the Begg's test (p = 0.80 for the meta-analysis on the RCTs, p = 1 for the meta-analysis on the 12 studies).

Discussion
Currently, fungal infections are widespread, especially in developed countries. A higher incidence of Candida spp. infections is associated with some predisposing factors such as the use of dentures, malnutrition, endocrine disorders, smoke, and some chronic diseases such as diabetes, HIV infection, and cancer [75]. The anti-OC treatment is mainly based on antifungal drugs, but different clinical types of OCs and the increasing number of multi-resistance phenotypes of Candida spp. represent current threats for public health. Consequently, the development of alternative therapeutic or complementary measures appears necessary to prevent the emergence of fungal resistance [76].
Many studies demonstrated that probiotics represent an efficient alternative treatment against Candida spp. infections. Moreover, they are easy to use and, thus, these products are usually well accepted by the patients [77]. The present study provides an overview of the literature on this issue, as well as a quantitative analysis that combines the results of independent studies of different design.
Both the meta-analysis on the 12 selected studies and the meta-analysis conducted on the subset of the RCTs indicated that the treatment had a beneficial effect on reducing oral Candida spp. counts.
As expected, the heterogeneity among studies was relevant because we combined studies of different design, which focused on different populations, used different treatments and doses, and were affected by different kinds and levels of bias. Our sensitivity analyses highlighted that part of the observed heterogeneity could be due to an actual difference of the treatment effect when used in different populations. For example, we found that the effect on denture wearers was larger than the effect estimated on non-denture wearers. The result on denture wearers relied only on two RCTs and should be interpreted with caution, but it is suggestive of a true difference. The larger reduction in the number of Candida spp. colonies in these patients could be caused by the direct application of probiotic products on the denture surface [57]. This hypothesis supports the idea that a lower effect of probiotics could be due to the low frequency of usage, number of probiotic cells, and delivery system, which exert an effect on the period of probiotics maintenance at the oral cavity. In this sense, the development of a mucoadhesive buccal drug delivery system [78], in order to enable the prolonged retention at the site of action, could improve the therapeutic outcome. An indication in favor of the relevance of the number of doses per day seems to arise also from the comparisons of the ORs in our meta-analysis.
If we focus on RCTs on non-denture wearers, a larger effect was reported in Li et al. (2014) [60], where the treated patients received three doses per day instead of one or two. Three doses were administered also in Keller et al. 2018 [55], but this study was affected by recruitment problems, and the result relied on a very small number of subjects.
A second relevant source of heterogeneity was related to the fact that different microbial probiotic strains could have different effects on the reduction of Candida spp. counts [58]. For example, Matsubara and colleagues found that, in a mice model, the treatment with Lactobacillus rhamnosus Lr-32 was more effective than the treatment with Lactobacillus acidophilus on the Candida spp. colonization levels [51]. Unfortunately, as the number of studies collected from the literature was too small to build a network of comparisons involving multiple treatments, we considered all treatments as having the same effect, which was clearly a very strong assumption.
Therefore, summarizing the obtained results, we can conclude that probiotics have a protective role in the Candida spp. infection and especially colonization. As previously reported, the anti-Candida properties can be explained in different ways, such as (a) through co-aggregation, modification of oral pH, and production of H 2 O 2 [45], (b) through releasing high amounts of lactic acid [47], and (c) through the complete inhibition of fungal biofilms [48,49]. However, these positive effects are highly linked to the administration method, the dosage, and the used probiotics strains. In addition, we did not find studies on prebiotics and synbiotics eligible for our meta-analysis. The effect of these products on the oral candidiasis must be better investigated in order to discover novel antifungal effects. In fact, some studies demonstrated that the combination of probiotics and prebiotics (synbiotics) can be very effective in infections [79,80].
Our results suggest planning a new clinical study to evaluate the real effectiveness of probiotics treatment in Candida spp. infection. The focal points of the study should be (1) the age stratification of the patients (old or adult), (2) the administration method (topic or oral), the type (lozenges or capsules), the dosage, and the treatment duration, (3) the choice of appropriate probiotic strains (Lactobacillus spp., Bifidobacterium spp., Saccaromyces spp., or Propionibacterium spp.), and (4) the length of the patient follow-up.
Our study also had other limitations. Firstly, the number of studies included in the meta-analysis was small, in particular when we focused on the RCTs. Secondly, some of the studies had a high risk of bias. Thirdly, with the aim of providing an overview of the literature, we did not apply strong exclusion criteria, at the price of a larger heterogeneity among studies. For the same reason, we sometimes had to adopt approximations to obtain a common comparable effect measure (OR) from the results reported in the original papers; this could have introduced a certain degree of bias in the meta-analysis.

Conclusions
In conclusion, our meta-analysis is one of the first that critically evaluated the impact of probiotics in oral candidiasis and, on the basis of the meta-analysis results, despite the high heterogeneity among studies, we are confident in declaring that the treatment can have a beneficial effect on reducing oral Candida spp. counts.