Table 5: GRADE (Grading of Recommendations, Assessment, Development and Evaluation)23 rating of confidence in the evidence of the 10 RCTs with 4841 participants.
| Summary of findings for primary outcome (CDAD) | Rating* | Overall quality | Comments | ||
|---|---|---|---|---|---|
| Median RR (95% CrI)(prediction interval) | Absolute risk | ||||
| Median (range) risk for CDAD in placebo group across studies | Expected risk for CDAD in probiotics group (95% CrI) (prediction interval) | ||||
| 0.25 (0.08 to 0.47) (0.02 to 1.34) |
6 per 100 patients (0 to 24) |
1.5 per 100 patients (0.5 to 3) (0.1 to 8.0) |
Very low | Apparently strong beneficial effect, with median RR = 0.25, but surrounded by wide prediction interval owing to heterogeneity of effect between studies | |
| Quality assessment domain | |||||
| Risk of bias | +++ | 6 studies had B or C rating (see Appendix 1) | |||
| Inconsistency | ++ | Wide range of risk for CDAD in placebo group. Wide prediction interval and wide interval around I2 statistic. Small number of cases of CDAD overall reduces credibility of results. | |||
| Indirectness | +++ | Although Lactobacillus probiotics were used in all studies, there was variation in strain and dosage used. Varying definitions of outcome across studies. | |||
| Imprecision | ++ | Small sample in several studies as CDAD was typically the secondary outcome. Few cases of CDAD in most studies. | |||
| Publication bias | Unclear | 6 studies were industry funded. The funnel plot (Figure 4), although based on few studies, suggested no asymmetry in scatter of points. | |||
| Large effect | Unclear | Although pooled estimate suggests a large effect, there remained much uncertainty in individual study estimates as reflected by wide prediction interval and credibility analysis results | |||
| Dose response | Not enough evidence | Most studies did not consider dose response | |||
| Residual confounding | Not an issue for RCTs | ||||
Note: CDAD = Clostridium-difficile-associated diarrhea, CrI = credible interval, RCT = randomized controlled trial, RR = risk ratio, SE = standard error.
*The overall quality is determined by downgrading, from a rating of "high" (4 plus signs), for any concerns in each of the first 5 domains as follows: 0 (no concerns), -1 (serious concerns) or -2 (very serious concerns), or upgrading in the case of the remaining 3 domains.