Data extraction and quality assessment

The primary search gave a total of 227 articles across the four databases. No relevant studies could be retrieved from grey literature search. After removing duplicates (n = 71), 156 titles and abstracts were screened for eligibility. Of these after screening out articles, 25 full-text articles were assessed, and 11 studies were excluded for not meeting the inclusion criteria (e.g., incomplete information, irrelevant findings) (Supplementary Table: 1 and 2). Ultimately, 14 studies were included in the final meta-analysis (Fig. 1, PRISMA flowchart).

Fig. 1

PRISMA Chart denoting the articles selection algorithm

Data was extracted (Supplementary Table: 3). Total of 14 paper with 15 study rows (as one study used 2 diagnostic methods and was used as two studies) were used for data extraction. Proportion (percent positivity) of samples was calculated (number of positive cases out of total number of samples tested/ Number of patients tested for Parvovirus B19). The risk of bias in the studies included was checked with The Joanna Briggs Institute Prevalence Critical Appraisal Tool [10] (Supplementary Table 4).

Statistical analysis for primary outcome of pooled prevalence of Parvovirus B19 in encephalitis /neurological disorders patients.

The 14 included studies spanned four continents, with studies conducted in Europe, Asia, and North America and South America. The sample sizes ranged from 20 to 887, and the number of positive Parvovirus B19 cases ranged from 0 to 10.7% of the total sample in each study. The tests used for diagnosis were under 3 broad categories, i.e. PCR, ELISA and NGS. [4, 12,13,14,15,16,17,18,19,20,21,22,23,24,25].

Geographic distribution

Five studies were conducted in India (Dey et al., 2024; Kumar et al., 2018; Pattabiraman et al., 2022; Rathore et al., 2022; Sonowal et al., 2024), three in Italy (Monticelli et al., 2018; Parisi et al., 2016), one in London (Barah et al., 2001) and the remaining studies were spread across the United States, Brazil, Japan, China and Poland. The variation in geographic representation underscores the potential for regional epidemiological differences in Parvovirus B19 prevalence and the possible impact of diagnostic capabilities in different settings.

Diagnostic methods

Except one study done on serum samples [26] (all other studies were based on CSF sample testing. Most studies used PCR to detect Parvovirus B19 DNA in patient samples, typically in cerebrospinal fluid (CSF) (11 out of 15 total studies). Two studies used serological methods (ELISA) to detect IgM antibodies [12, 26] indicating recent Parvovirus B19 infection. Notably, serological methods showed a lower positivity rate compared to PCR, which may be due to the heterogeneity in sample types (one in CSF and one in Serum) and diagnostic thresholds of ELISA used. The use of IgM/ ELISA which traces the recent infection and may differ in sensitivity may have contributed to this difference.

Pooled prevalence and subgroup analysis

The pooled prevalence of Parvovirus B19 across all studies was 3% (95% CI: 2–4%) based on the random-effects model and 3% (95% CI: 3–4%) based on the common-effects model. The forest plot (Fig. 2) visually represents the effect sizes from individual studies and the overall pooled estimate. The forest plot depicted the subgroup analysis based on diagnostic tests used. The pooled proportion based on PCR as diagnostic test gave an overall value of 3% which was higher as compared to ELISA subgroup (1.00%) and NGS ((2.00%).

Fig. 2

Forest Plot with 14 studies included and the subgroup analysis based on diagnostic tests. The effect sizes (proportions) from individual studies have been shown as squares with size of square box showing weight of the study and horizontal line denotes the CI. The overall pooled estimate from common effect and Random Effect model has been shown as Diamonds

The prevalence estimates varied widely across studies, with the highest reported prevalence of 10.7% in the study by Dey et al. (2024) [15] from India and the lowest at 0.0% in the study by Parisi et al. (2016) [13] from Italy. Heterogeneity among the studies was significant (I2 = 57.4%, p < 0.01), indicating substantial variability in the effect sizes, which may be due to differences in diagnostic methods, patient populations, or study design. Most of the heterogenicity was attributed to subgroup of Diagnostic method- PCR as most of the pooled prevalence was derived from those studies.

This value measures the percentage of total variation across studies attributable to heterogeneity rather than chance, which indicates a moderate to high level of variability between the studies. About 57% of the observed differences in virus positivity rates could be due to true heterogeneity among studies rather than random error. This also supports the use of a mixed-effects model, as it accounts for variability across studies in the pooled prevalence estimate.

The I2 values for each subgroup of different diagnostic methods is as follows. PCR: I2 = 59.7%; ELISA: I2 = 0% and NGS: I2 = 0%. The high I2 value in PCR group shows a lot of heterogeneity in studies using PCR. This difference in PCR subgroup may come from various PCR types (Multiplex, conventional, nested and real time PCR), different protocols, populations, or study conditions. The 0% I2 values for both ELISA and NGS indicate no significant heterogeneity. with very little variation between them.

The tau-squared (τ2) values calculated as ‘0’ for subgroup NGS and ELISA indicates little variation between studies within these subgroups, showing that studies using the same diagnostic method (ELISA, or NGS) had similar virus positivity rates. However, in this particular meta-analysis this could not be explained due to fewer studies in these sub-groups. However, τ2 = 0.3697 in PCR group suggests that about 37% of heterogenicity in PCR sub-group is accounted for due to reasons other than random sampling error.

Sensitivity analysis

Sensitivity analyses revealed that the overall prevalence estimate was robust, with minimal changes in pooled prevalence when individual studies were excluded. Subgroup analyses indicated that studies using PCR for detection reported higher prevalence rates (3%) compared to studies using ELISA (1%) or NGS (2%). However, this difference did not reach statistical significance, likely due to the small number of studies using ELISA or NGS.

Sensitivity analysis after Leave-one-out method was used to make a forest plot (Fig. 3), The pooled effect estimates remain significant across all analyses (ranging from approximately 0.0197 to 0.0237). Confidence intervals did not include 0 (none of the CI touched the line of null effect), confirming a robust association. The estimates were all clustered within a narrow range, suggesting that no single study could influence the overall effect disproportionately. This forest plot supports the robustness of the meta-analysis results and highlights that while some studies contribute more to heterogeneity, their exclusion does not drastically affect the pooled prevalence.

Fig. 3

Forest Plot with Sensitivity analysis after Leave-one-out method

Meta-regression analysis

Moderate heterogeneity was detected among the included studies, as evidenced by the I2 statistic of 59.7% in subgroup based meta-analysis as shown in Fig. 2. This suggests that the prevalence of Parvovirus B19 in encephalitis patients varies widely across different settings and study designs. While meta-regression analysis explored potential sources of this heterogeneity, none of the examined covariates—sample size, publication year, or geographic region—were found to significantly influence effect sizes.

On individual analysis of best fit model for individual covariates following were the results based on Regression coefficients (Supplementary Table 5).

1.

Geographical region Country as a moderator was checked by mixed effect meta-regression model. However, there we no significant moderation by the country variable. The pooled effect size remained consistent across all countries.

2.

Publication year—Publication year was checked by mixed effect meta-regression model. There was no meaningful trend over the time. Year as a moderator did not significantly affect the effect size.

3.

Sample size: Sample size as a moderator was checked by mixed effect meta-regression model. No significant variation in pooled effect size was found on using sample size as moderator.

4.

Diagnostic method: Diagnostic method as a moderator was used in mixed effect meta-regression model. No significant variation in pooled effect size was found on using different diagnostic method as moderator.

Further analysis by the combined effects of multiple moderators (sample size, year of publication, country, and diagnostic method) on the variability in effect sizes across 15 studies was assessed by Mixed-Effects Model which Combines fixed effects (overall pooled effect size and moderator effects) and random effects (heterogeneity across studies). With this model, sample size, diagnostic method and publication year were not found to be significant predictors of variability in effect size, however country was found to be a significant moderator overall, with significant effects for UK and Japan. Hence most of the heterogeneity was explained by country (76.30%) and combined moderators explains for about 80.71% of heterogenicity in effect sizes. ((Supplementary Table 6).

To explain further due to moderate heterogeneity as detected among the included studies, as evidenced by the I2 statistic of 57.4% in subgroup based meta-analysis contributed by PCR diagnostic group, we compared the heterogeneity metrics such as I2, τ2 and R2 with and without PCR as reference (Supplementary Table 7). We found that heterogeneity exists within PCR subgroup but is not due to diagnostic method itself. When analyzed along with other diagnostic methods, heterogeneity is fully explained by sampling variability (I2 = 0%). The inclusion of publication year and diagnostic method as moderators did not explain the observed heterogeneity within PCR subgroup.

A meta-regression plot was created with proportion (prevalence from individual studies on Y-axis and Year of publication on X-axis). Trend lines for indicating specific diagnostic method for PCR, ELISA and NGS with pooled prevalence line were also added. (Fig. 4). The findings from this plot denotes that publication year does not have significant effect on pooled prevalence. There are slight differences in prevalence between PCR, ELISA and NGS but are statistically non-significant, but shows a slight increasing trend with years. This also confirms the findings of the meta-regression. These finding suggests that the observed variability in prevalence rates may be driven by unmeasured factors, such as differences in diagnostic protocols, patient selection criteria, or study type etc. Slight increased trend in prevalence rate might be due to increased sensitivity of molecular methods used for screening.

Fig. 4

A meta-regression plot with proportion (prevalence from individual studies on Y-axis and Year of publication on X-axis). Trend lines for PCR, ELISA and NGS with pooled prevalence line shown in blue, red, purple and black color

Publication bias evaluation

Funnel plot was made with the studies used for metanalysis (Fig. 5). Visual inspection of the funnel plot revealed symmetry suggesting that studies with smaller or insignificant results are not systematically missing from the analysis. The plot also indicates a good overall consistency of effect sizes across studies. The spread of the points, particularly at the base of the funnel, reflects the moderate-to-substantial heterogeneity observed in the meta-analysis.

Fig. 5

Funnel Plot for publication bias assessment. Individual dots denote each study with Proportion on X-axis and Standard Error on Y-axis

To further explain Egger’s plot was created (Fig. 6) which visualizes the regression used to assess potential funnel plot asymmetry, which can indicate publication bias or small-study effects in a meta-analysis. It denoted slight asymmetry, with only a mild slope in the regression line. To quantify the asymmetry, egger’s regression test for funnel plot asymmetry was done. Model used was weighted regression with multiplicative dispersion using standard error as predictor. The test gave t = 4.1204 with p = 0.0012, suggesting that the intercept (b = 0.0007) is significantly different from Zero indicating asymmetry in the funnel plot (Supplementary Table 8). This publication bias may be due to selective reporting of the significant findings, differences in the study quality or methodology or true heterogeneity amongst studies. However, the small intercept denotes that publication bias might be there but its impact on overall results is likely to be limited. To assess further, a trim and fill analysis was done to estimate the number of potentially missing studies and their impact. The Fig. 7 shows Funnel plot with Trim and Fill method. The Trim and Fill method identifies 3 studies missing (likely due to publication bias) on the left side of the funnel plot indicating smaller effect studies to be missing from the meta-analysis. The standard error of the estimate of number of studies (2.6402) indicates the level of uncertainty in this number. The adjusted pooled effect size is 0.0211 with a narrow confidence interval [0.0187, 0.0235]. The pooled estimate is statistically significant (p < 0.05) (Supplementary Table 9). After accounting for publication bias, the meta-analysis still finds a minimum effect on the pooled effect size, with no heterogeneity remaining after adjustment.

Fig. 6

Egger’s plot to assess Funnel Plot asymmetry. Individual dots denote each study with Pro Standard Error on X-axis Effect Size/Standard Error on Y-axis. Mild Slope in regression line denotes asymmetry in the funnel plot

Fig. 7

Inverted Funnel plot with Trim and Fill method with confidence interval Individual dots denote each study with Proportion on X-axis and Standard Error on Y-axis