Table 4.
Results from re‐analysing data from the network meta‐analysis to compare interventions for smoking cessation. Bayesian approaches apply an empirically based inverse‐gamma(1.39, 0.13) prior for τ 2 in a non‐pharmacological meta‐analysis with a subjective outcome.
| Method | μ AB | (95% CI) | μ AC | (95% CI) | μ AD | (95% CI) | τ 2 | (95% CI) |
|---|---|---|---|---|---|---|---|---|
| Conventional (DL estimation) | 0.46 | ( − 0.24, 1.16) | 0.70 | (0.27, 1.12) | 0.98 | (0.13, 1.84) | 0.60 | (0.21, 1.32)1 |
| Data augmentation by DL | 0.35 | ( − 0.03, 0.72) | 0.59 | (0.38, 0.80) | 0.85 | (0.34, 1.37) | 0.09 | (0.19, 1.07)1 |
| Data augmentation by ML | 0.42 | ( − 0.13, 0.96) | 0.64 | (0.32, 0.97) | 0.92 | (0.23, 1.61) | 0.31 | (0.19, 1.07)1 |
| Data augmentation by REML | 0.42 | ( − 0.14, 0.99) | 0.65 | (0.31, 1.00) | 0.93 | (0.22, 1.64) | 0.35 | (0.19, 1.07)1 |
| MCMC | 0.43 | ( − 0.16, 1.05)2 | 0.66 | (0.31, 1.04)2 | 0.94 | (0.20, 1.70)2 | 0.37 | (0.18, 0.83)2 |
1
The confidence interval for τ 2 is obtained iteratively via the Q‐profile method 23.
2
Posterior medians and 95% credible intervals are reported for the log odds ratios μ AB,μ AC and μ AD and for the common heterogeneity variance τ 2.
DL, DerSimonian and Laird; ML, maximum likelihood; REML, restricted maximum likelihood; MCMC, Markov chain Monte Carlo;