Twenty-five studies (n=4,767) were included. These seemed to comprise 24 RCTs and a study of unknown design.
Twenty-four trials were graded A and one was rated D.
BCG was associated with a statistically significantly lower likelihood of tumour recurrence compared with no BCG (OR 0.61, 95% CI: 0.46, 0.80, p<0.001; based on 24 trials). There was evidence of statistically heterogeneity (p<0.00001; I-squared 76.1%). The results were similar after excluding the only D-rated trial. Drug-induced cystitis, dysuria, frequency/urgency and systemic side-effects were significantly more common in patients given BCG than in patients given chemotherapy and immunotherapy (no data were presented).
The subgroup analysis showed that BCG maintenance therapy was associated with a statistically significantly lower likelihood of tumour recurrence than no BCG (OR 0.47, 95% CI: 0.28, 0.78, p=0.004; based on 1,070 patients in 8 trials). There was evidence of statistical heterogeneity (p=0.002). No statistically significant difference was shown in tumour recurrence in studies that compared no BCG maintenance with no BCG treatment (OR 0.47, 95% CI: 0.28, 0.78, p=0.002; based on 1,070 patients in 8 trials).
The subgroup analysis also showed that BCG was associated with a statistically significantly lower likelihood of tumour recurrence than TUR alone (OR 0.35, 95% CI: 0.20, 0.59, p<0.0001; based on 1,100 patients in 9 trials). There was evidence of statistical heterogeneity (p=0.002). No statistically significant difference in the likelihood of tumour recurrence was shown between BCG and chemotherapy treatment (chemotherapy alone or BCG plus immunotherapy/chemotherapy) (OR 0.88, 95% CI: 0.58, 1.35, p=0.57; based on 1,793 patients in 10 trials). There was evidence of statistical heterogeneity (p=0.0005).
There was no significant effect of BCG when the studies were stratified by BCG strain.
For patients with papillary cancer, BCG was associated with a significantly lower likelihood of tumour recurrence than no BCG (OR 0.50, 95% CI: 0.33, 0.75, p=0.0008; based on 1,397 patients in 10 trials). There was evidence of statistical heterogeneity (p=0.004).