Corticosteroid and antiviral therapy for Bell's palsy: A network meta-analysis

One author (NP) located studies in MEDLINE (from 1966 to August 2010) and EMBASE (from 1950 to September 2010) using PubMed and Ovid search engines. Search terms used were as follows: (Bell's palsy or idiopathic facial palsy) and (antiviral agents or acyclovir or valacyclovir), limited to randomized controlled trials. Search strategies for both databases are described in the additional file 1.

Abstracts and/or full papers of identified studies were reviewed by one author (NP) and checked by another author (TA). Studies were included if they were RCTs, and studied subjects aged 18 years or older with sufficient data. Non-English papers were excluded from the review. Where eligible papers had insufficient information, corresponding authors were contacted by e-mail for additional information. The reference lists of the retrieved papers were also reviewed to identify relevant publications. Where there were multiple publications from the same study group, the most complete and recent results were used.

Data extraction was independently performed in duplicate by PN and AT using a standardized data extraction form, which included study design, sample size, patient characteristics (i.e., age, gender), type of intervention and comparator, outcomes, and follow-up time. Any disagreement was resolved by discussion.

Quality of studies was also independently assessed by PN and AT based on a modified Jadad score which takes into account randomization technique, allocation concealment, blinding, intention to treat, and patient attrition[24]. Each item was graded 2, 1, or 0 for appropriately, partially, and inappropriately described methods. Any disagreement between the two reviewers was discussed and resolved by consensus.

Complete recovery was defined as a score ≤2 on the House-Brackman Facial Recovery scale[12, 13, 25], ≥ 8 on the Facial Palsy Recovery Index[8], > 36 points on theYanagihara score[10], or 100 on the Sunnybrook scale[9].

For direct meta-analysis, the odds ratio (OR) and variance of each study were estimated and pooled. Heterogeneity of ORs was assessed using Cochran's Q test and I². If heterogeneity was present, ORs were pooled using the random effects model, i.e. Der-Simonian and Laird method. Meta-regression was applied to assess whether age, gender, and QA were sources of heterogeneity, if these data were available. Contour enhanced funnel plots were used to detect publication bias due to small study effects [26‐29]. Trim and fill meta-analysis was applied to impute the number of missing studies [30].

For network meta-analysis, treatment groups were considered in a mixed effects hierarchical model with logit link function using the xtmelogit command in STATA[31]. The treatments were included in the model as fixed effects whereas the study was included as a random effect. Likelihood estimates were used for estimation of parameters in the model. Goodness of fit of the model was assessed using the Hosmer-Lemeshow Chi-square test. The pooled ORs and 95% confidence intervals (CI) were estimated by exponential coefficients of treatments. Discrepancy of treatment effects between direct and indirect meta-results was then assessed using the standardized normal method (Z), i.e. by dividing the difference by its standard error[20, 32]. Number needed to treat/harm (NNT/NNH) was estimated using probability of complete recovery and ORs derived from the mixed effects hierarchical model, where the ORs were converted to risk ratios following the method of Zhang et al[33]. All analyses were performed using STATA version 11.0. P values < 0·05 were considered as statistically significant except for the heterogeneity test where <0·10 was used.

Among 6 studies, 3 studies[8, 12, 13] compared recovery rates within 3 months between Acyclovir plus Prednisolone versus Prednisolone alone (table 2). Pooled treatment effects were heterogeneous (Chi-sqaure = 6·30 (d.f. = 2) p = 0·043; I²= 68.3%). Using the random effects model gave a pooled OR of 1·39 (95% CI: = 0·52 - 3·75), i.e. patients who received Acyclovir plus Prednisolone were about 40% more likely to recover within 3 month than patients who received Prednisolone alone, although this did not reach statistical significance. Only 2 studies[9, 10] compared recovery rates between Valacyclovir plus Prednisolone and Prednisolone. They were homogeneous and the pooled OR with fixed effect model was 1·17 (95% CI: 0·75 - 1·81).

Combining 5 studies[8‐10, 12, 13] to assess the effect of AVT (Acyclovir/Valacyclovir) plus Prednisolone versus Prednisolone found moderate heterogeneity (Chi-sqaure = 7.78 (d.f. = 4) p = 0.100; I² = 48.6%). The pooled OR with a random effect model was 1.21 (95% CI: 0.77 - 1.89), i.e. AVT plus Prednisolone had 21% higher chance of complete recovery than Prednisolone alone but this was non-significant. Meta-regression suggested that quality assessment score and disease severity at baseline might be a source of heterogeneity, reducing I² from 48.6% to 32.5% and 15.5% respectively although both variables were non-significantly associated with treatment effects. Pooling studies with quality assessment scores > 8 (median) suggested no benefit of AVT plus Prednisolone compared with Prednisolone alone (pooled OR = 1.01, 95% CI: 0.74 - 1.37). Two studies[8, 10] had more severe patients at baseline (i.e., mean scores were 3 by facial palsy recovery index and 15 by Yanagihara score) compared with the other 3 studies[9, 12, 13] (i.e., mean scores were 3.6 to 4 by HB), see table 1. Pooling within the severe group only suggested the possibility of a treatment effect but this was non-significant (OR = 2.04, 95% CI: 0.93 - 4.46) while treatment effect was close to the null in the less severe group (OR = 0.94, 95%CI: 0.66 - 1.33). Contour enhanced funnel plots suggested asymmetry, i.e., four studies lay in the non-sigificant area (white, p > 5%) but the study with the largest treatment effect and SE lay in the positive-moderate significant area (1% ≤ p ≤ 5%, see Figure 2), suggesting that the asymmetry might be due to missing, non-significant studies. "metatrim" analysis indicated that only one negative study with borderline significance was missing; adjusting for this presumed missing studies resulted in an ORof 1.08 (95% CI: 0.66 - 1.76).

Three studies[8, 12, 13] reported the recovery rate between direct comparisons of Acyclovir plus Prednisolone versus Prednisolone after receiving treatments at 4 to 9 months (table 2). Treatment effects were statistically heterogeneous (Chi-square = 6·07 (d.f. = 2), p = 0·048; I2 = 67·0%), and the random effects pooled OR was 1·63 (95% CI: 0·47 - 5·75). Pooling effects of Valacyclovir plus Prednisolone versus Prednisolone was performed based on only 2 heterogeneous studies[9, 10]. The random effects pooled OR was 1·58 (95% CI: 0·55 - 4·48). There was only 1 study each that compared Acyclovir[12] or Valacyclovir[9] alone versus Prednisolone and this was insufficient for pooling.

All six studies[8‐10, 12, 13, 25] were able to contribute to a network meta-analysis, for a total of 1,805 patients, as described in table 2. Overall, 177 (9·8%), 207 (11·5%), 567 (31·4%), and 328 (18·2%) received only Acyclovir, Valacyclovir, Prednisolone, or placebo whereas 206 (11·4%) and 320 (17·7%) received Acyclovir plus Prednisolone and Valacyclovir plus Prednisolone, respectively. One-thousand two-hundred and forty-three patients had complete recovery (68·9%) within three months.

All treatment comparisons and results of the network meta-analysis for 3 month outcomes are displayed in Figure 3a. Each line in the figure represents a randomized comparison, in which the head and the tail of that line refer to intervention and reference groups respectively. Figures on the lines refer to estimated indirect ORs for that comparison; ORs less than one indicate that the intervention group had lower recovery than the reference group, whereas ORs higher than one mean that the intervention group had higher recovery than the reference group. The mixed-effects model with a random intercept and a constant slope was applied to assess treatment effects. The estimated variance between studies was 0·17 (95% CI: 0·04 - 0·76). Hosmer-Lemeshow test found that the model was a good fit (Chi-square = 12·99, p = 0·072).

Treatment effects for all comparisons are shown in Figure 3a and table 3. Adding Acyclovir or Valacyclovir to Prednisolone did not significantly increase recovery rate when compared with Prednisolone alone, although the point estimates suggest a potential benefit of the combination, i.e., ORs were 1·24 (95% CI: 0·79 - 1·94) and 1·02 (95% CI: 0·73 - 1·42) for Acyclovir and Valacyclovir, respectively. For monotherapy, either Acyclovir (OR = 0·44, 95% CI: 0·28 - 0·68) or Valacyclovir (OR = 0·60, 95% CI: 0·42 - 0·87) led to significantly lower recovery than Prednisolone alone and this was not significantly different compared with placebo (OR = 1·25, 95% CI: 0·78 - 1·98 for Placebo versus Acyclovir; OR = 0·91, 95% CI: 0·63 - 1·31 for Placebo versus Valacyclovir). These data indicate that Prednisolone led to higher recovery rates than not receiving Predisolone. We therefore combined all Prednisolone groups (regardless of additional AVT) compared with non-Prednisolone (i.e., Acyclovir, Valacyclovir, and Placebo). The estimated OR was 1·94 (95% CI: 1·55 - 2·42), i.e., Prednisolone-based treatment increase the chance of recovery by 3 months by 2 fold compared to non-Prednisolone-based treatment.

These effects were largely consistent when recovery was judged at later time points (Figure 3b and table 3). Acyclovir (OR = 1·74, 95% CI: 0·93 - 3·24) or Valacyclovir (OR = 1·15, 95% CI: 0·78 - 1·69) added to Prednisolone tended to improve recovery rates compared to receiving Prednisolone alone but the effects were non-significant. Mono-therapy with either Acyclovir (OR = 0·33, 95% CI: 0·18 - 0·68) or Valacyclovir (OR = 0·55, 95% CI: 0·37 - 0·81), or placebo (OR = 0·62, 95% CI: 0·43 - 0·88) led to significantly lower recovery than Prednisolone mono-therapy. Compared with Acyclovir, Acyclovir or Valacyclovir plus Prednisolone was more beneficial than Acyclovir alone, with ORs of 5·21 (95% CI: 2·48 - 10·94) and 3·45 (95% CI: 1·78 - 6·68) respectively; trends were similar for Valacyclovir. Overall, the effect of prednisolone-based regimens compared to those without prednisolone was similar to that at 3 months, with OR of 2·04 (95% CI 1·57 - 2·64).

Results of direct and network analyses are compared in the last column of table 3. Direction of treatment effects between both methods were similar for all comparisons, although the direct approach provided higher treatment effects in both positive and negative directions. The degree of discrepancy between the 2 methods was very small except for Acyclovir versus Prednisolone, where Z was large and significant at 3 and 6 months (1·77, p = 0·077; and -2·93, p = 0·002 respectively). Precision of treatment effects was generally higher in the network meta-analysis than the direct one.

Finally, estimates of NNT and NNH for recovery at 3 and 6 months are given in table 4. We found that 6 patients need to be treated with Acyclovir and prednisolone compared to placebo in order to gain 1 extra recovery and this had a tight confidence interval. By contrast, the benefit of combination AT and prednisolone compared to prednisolone alone was less certain, with 26 patients needing to be treated with Acyclovir plus predinisolone in order to gain 1 extra patient with recovery at 3 months, and this had wide confidence intervals.