Background
Methods
Mixed effects models for longitudinal data
IPD meta-analyses based on the mixed effects models
Two-step marginal multivariate random effects analyses
-
Step-1. Analyse the IPD from N longitudinal trials separately, using the mixed effects models (1), and obtain the REML estimates of the model parameters (\( {\widehat{\boldsymbol{\beta}}}_i,{\widehat{\boldsymbol{D}}}_i,{\widehat{\boldsymbol{\Sigma}}}_i \)), i = 1, 2, … , N. In addition, obtain the variance-covariance matrix estimates of \( {\widehat{\boldsymbol{\beta}}}_i \), \( {\widehat{\boldsymbol{S}}}_i\left({\widehat{\boldsymbol{\beta}}}_i\right) \) for i = 1, 2, … , N.
-
Step-2. Conduct synthesizing analyses for the regression coefficient parameters \( \Big({\widehat{\boldsymbol{\beta}}}_i \), \( {\widehat{\boldsymbol{S}}}_i\left({\widehat{\boldsymbol{\beta}}}_i\right)\Big) \) that are the primary objects of interests in these meta-analyses, dropping statistical information about the within-studies covariance parameters (\( {\widehat{\boldsymbol{D}}}_i,{\widehat{\boldsymbol{\Sigma}}}_i \)),
Remark A
Remark B
Remark C
Results
Simulation studies
Simulation 1
N = 5 | N = 10 | N = 15 | N = 20 | |||||
---|---|---|---|---|---|---|---|---|
One-step | Two-step | One-step | Two-step | One-step | Two-step | One-step | Two-step | |
μ2 = 0.14 | ||||||||
Mean | 0.135 | 0.135 | 0.146 | 0.146 | 0.135 | 0.136 | 0.130 | 0.130 |
SE | 0.866 | 0.867 | 0.612 | 0.613 | 0.495 | 0.496 | 0.434 | 0.435 |
\( \widehat{\mathrm{SE}} \) | 0.867 | 0.866 | 0.608 | 0.606 | 0.495 | 0.494 | 0.428 | 0.427 |
Coverage Rate | 0.949 | 0.950 | 0.948 | 0.947 | 0.949 | 0.947 | 0.950 | 0.948 |
Power | 0.052 | 0.053 | 0.056 | 0.058 | 0.058 | 0.059 | 0.071 | 0.070 |
μ4 = 0.83 | ||||||||
Mean | 0.827 | 0.826 | 0.836 | 0.836 | 0.828 | 0.829 | 0.825 | 0.824 |
SE | 0.415 | 0.416 | 0.287 | 0.287 | 0.236 | 0.236 | 0.206 | 0.205 |
\( \widehat{\mathrm{SE}} \) | 0.410 | 0.410 | 0.287 | 0.287 | 0.234 | 0.234 | 0.202 | 0.202 |
Coverage Rate | 0.950 | 0.950 | 0.951 | 0.951 | 0.948 | 0.947 | 0.939 | 0.941 |
Power | 0.533 | 0.533 | 0.825 | 0.825 | 0.936 | 0.937 | 0.978 | 0.978 |
μ5 = −0.46 | ||||||||
Mean | −0.461 | −0.461 | −0.460 | −0.460 | −0.460 | −0.460 | −0.461 | −0.461 |
SE | 0.032 | 0.032 | 0.023 | 0.023 | 0.018 | 0.018 | 0.016 | 0.016 |
\( \widehat{\mathrm{SE}} \) | 0.032 | 0.032 | 0.022 | 0.022 | 0.018 | 0.018 | 0.016 | 0.016 |
Coverage Rate | 0.949 | 0.948 | 0.949 | 0.947 | 0.953 | 0.955 | 0.951 | 0.949 |
Power | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 |
μ7 = 0.14 | ||||||||
Mean | 0.139 | 0.139 | 0.141 | 0.141 | 0.143 | 0.143 | 0.144 | 0.144 |
SE | 0.178 | 0.178 | 0.122 | 0.122 | 0.100 | 0.100 | 0.088 | 0.088 |
\( \widehat{\mathrm{SE}} \) | 0.176 | 0.176 | 0.123 | 0.123 | 0.100 | 0.100 | 0.087 | 0.087 |
Coverage Rate | 0.948 | 0.947 | 0.949 | 0.949 | 0.949 | 0.949 | 0.940 | 0.942 |
Power | 0.127 | 0.128 | 0.211 | 0.215 | 0.295 | 0.297 | 0.382 | 0.381 |
μ9 = 0.05 | ||||||||
Mean | 0.050 | 0.050 | 0.050 | 0.050 | 0.050 | 0.050 | 0.050 | 0.050 |
SE | 0.020 | 0.020 | 0.014 | 0.014 | 0.011 | 0.011 | 0.010 | 0.010 |
\( \widehat{\mathrm{SE}} \) | 0.020 | 0.020 | 0.014 | 0.014 | 0.011 | 0.011 | 0.010 | 0.010 |
Coverage Rate | 0.947 | 0.947 | 0.951 | 0.951 | 0.949 | 0.952 | 0.941 | 0.940 |
Power | 0.734 | 0.735 | 0.947 | 0.947 | 0.995 | 0.995 | 0.999 | 0.999 |
μ10 = −0.06 | ||||||||
Mean | −0.060 | −0.060 | −0.060 | −0.060 | −0.060 | −0.060 | −0.059 | −0.059 |
SE | 0.042 | 0.042 | 0.030 | 0.030 | 0.024 | 0.024 | 0.021 | 0.021 |
\( \widehat{\mathrm{SE}} \) | 0.042 | 0.042 | 0.029 | 0.029 | 0.024 | 0.024 | 0.021 | 0.021 |
Coverage Rate | 0.949 | 0.949 | 0.948 | 0.947 | 0.949 | 0.950 | 0.946 | 0.945 |
Power | 0.301 | 0.302 | 0.540 | 0.543 | 0.700 | 0.701 | 0.811 | 0.811 |
Simulation 2
N = 5 | N = 10 | N = 15 | ||||
---|---|---|---|---|---|---|
τ2 = 0.10 | τ2 = 0.20 | τ2 = 0.10 | τ2 = 0.20 | τ2 = 0.10 | τ2 = 0.20 | |
μ2 = 0.14 | ||||||
Mean | 0.141 | 0.144 | 0.141 | 0.144 | 0.145 | 0.143 |
SE | 0.694 | 0.709 | 0.694 | 0.709 | 0.564 | 0.565 |
\( \widehat{\mathrm{SE}} \) | 0.680 | 0.692 | 0.680 | 0.692 | 0.553 | 0.562 |
Coverage Rate | 0.946 | 0.947 | 0.946 | 0.947 | 0.947 | 0.950 |
Power | 0.061 | 0.057 | 0.061 | 0.057 | 0.059 | 0.057 |
μ4 = 0.83 | ||||||
Mean | 0.828 | 0.831 | 0.828 | 0.831 | 0.832 | 0.833 |
SE | 0.306 | 0.323 | 0.306 | 0.323 | 0.252 | 0.268 |
\( \widehat{\mathrm{SE}} \) | 0.307 | 0.325 | 0.307 | 0.325 | 0.251 | 0.265 |
Coverage Rate | 0.951 | 0.950 | 0.951 | 0.950 | 0.948 | 0.947 |
Power | 0.773 | 0.725 | 0.773 | 0.725 | 0.912 | 0.877 |
μ5 = −0.46 | ||||||
Mean | −0.459 | −0.458 | −0.459 | −0.458 | −0.461 | −0.460 |
SE | 0.104 | 0.143 | 0.104 | 0.143 | 0.084 | 0.117 |
\( \widehat{\mathrm{SE}} \) | 0.103 | 0.143 | 0.103 | 0.143 | 0.084 | 0.117 |
Coverage Rate | 0.944 | 0.947 | 0.944 | 0.947 | 0.949 | 0.949 |
Power | 0.992 | 0.890 | 0.992 | 0.890 | 1.000 | 0.976 |
μ7 = 0.14 | ||||||
Mean | 0.142 | 0.141 | 0.142 | 0.141 | 0.138 | 0.140 |
SE | 0.163 | 0.193 | 0.163 | 0.193 | 0.134 | 0.157 |
\( \widehat{\mathrm{SE}} \) | 0.163 | 0.192 | 0.163 | 0.192 | 0.133 | 0.157 |
Coverage Rate | 0.947 | 0.948 | 0.947 | 0.948 | 0.947 | 0.951 |
Power | 0.140 | 0.115 | 0.140 | 0.115 | 0.181 | 0.146 |
μ9 = 0.05 | ||||||
Mean | 0.051 | 0.050 | 0.051 | 0.050 | 0.048 | 0.050 |
SE | 0.101 | 0.140 | 0.101 | 0.140 | 0.082 | 0.116 |
\( \widehat{\mathrm{SE}} \) | 0.100 | 0.141 | 0.100 | 0.141 | 0.082 | 0.116 |
Coverage Rate | 0.944 | 0.945 | 0.944 | 0.945 | 0.947 | 0.947 |
Power | 0.091 | 0.071 | 0.091 | 0.071 | 0.091 | 0.076 |
μ10 = −0.06 | ||||||
Mean | −0.061 | −0.058 | −0.061 | − 0.058 | − 0.060 | −0.061 |
SE | 0.106 | 0.145 | 0.106 | 0.145 | 0.087 | 0.119 |
\( \widehat{\mathrm{SE}} \) | 0.105 | 0.145 | 0.105 | 0.145 | 0.086 | 0.119 |
Coverage Rate | 0.945 | 0.944 | 0.945 | 0.944 | 0.945 | 0.947 |
Power | 0.096 | 0.074 | 0.096 | 0.074 | 0.110 | 0.084 |
N = 5 | N = 10 | N = 15 | ||||
---|---|---|---|---|---|---|
τ2 = 0.10 | τ2 = 0.20 | τ2 = 0.10 | τ2 = 0.20 | τ2 = 0.10 | τ2 = 0.20 | |
μ2 = 0.14 | ||||||
Mean | 0.140 | 0.142 | 0.144 | 0.136 | 0.136 | 0.139 |
SE | 1.026 | 1.026 | 0.693 | 0.702 | 0.483 | 0.559 |
\( \widehat{\mathrm{SE}} \) | 0.989 | 1.004 | 0.681 | 0.690 | 0.475 | 0.561 |
Coverage Rate | 0.947 | 0.950 | 0.948 | 0.946 | 0.948 | 0.953 |
Power | 0.057 | 0.052 | 0.058 | 0.058 | 0.061 | 0.053 |
μ4 = 0.83 | ||||||
Mean | 0.825 | 0.822 | 0.823 | 0.828 | 0.873 | 0.836 |
SE | 0.546 | 0.584 | 0.377 | 0.404 | 0.283 | 0.329 |
\( \widehat{\mathrm{SE}} \) | 0.537 | 0.574 | 0.376 | 0.402 | 0.257 | 0.328 |
Coverage Rate | 0.948 | 0.950 | 0.950 | 0.947 | 0.907 | 0.947 |
Power | 0.347 | 0.316 | 0.591 | 0.548 | 0.892 | 0.717 |
μ5 = −0.46 | ||||||
Mean | −0.459 | −0.459 | −0.460 | −0.459 | −0.460 | −0.461 |
SE | 0.147 | 0.204 | 0.103 | 0.143 | 0.073 | 0.118 |
\( \widehat{\mathrm{SE}} \) | 0.143 | 0.200 | 0.102 | 0.143 | 0.073 | 0.117 |
Coverage Rate | 0.931 | 0.932 | 0.944 | 0.944 | 0.944 | 0.945 |
Power | 0.874 | 0.625 | 0.994 | 0.890 | 1.000 | 0.974 |
μ7 = 0.14 | ||||||
Mean | 0.145 | 0.145 | 0.136 | 0.135 | 0.138 | 0.141 |
SE | 0.295 | 0.359 | 0.207 | 0.245 | 0.147 | 0.202 |
\( \widehat{\mathrm{SE}} \) | 0.291 | 0.346 | 0.206 | 0.245 | 0.145 | 0.201 |
Coverage Rate | 0.943 | 0.936 | 0.947 | 0.950 | 0.946 | 0.949 |
Power | 0.088 | 0.083 | 0.105 | 0.082 | 0.167 | 0.110 |
μ9 = 0.05 | ||||||
Mean | 0.051 | 0.052 | 0.051 | 0.055 | 0.049 | 0.050 |
SE | 0.185 | 0.263 | 0.133 | 0.184 | 0.093 | 0.151 |
\( \widehat{\mathrm{SE}} \) | 0.178 | 0.252 | 0.128 | 0.181 | 0.091 | 0.149 |
Coverage Rate | 0.925 | 0.927 | 0.935 | 0.940 | 0.941 | 0.943 |
Power | 0.086 | 0.080 | 0.084 | 0.070 | 0.089 | 0.070 |
μ10 = −0.06 | ||||||
Mean | −0.062 | − 0.063 | − 0.060 | − 0.060 | − 0.060 | − 0.059 |
SE | 0.150 | 0.205 | 0.106 | 0.146 | 0.075 | 0.118 |
\( \widehat{\mathrm{SE}} \) | 0.147 | 0.203 | 0.105 | 0.145 | 0.074 | 0.118 |
Coverage Rate | 0.934 | 0.937 | 0.942 | 0.943 | 0.947 | 0.948 |
Power | 0.084 | 0.074 | 0.096 | 0.079 | 0.130 | 0.083 |
Applications to IPD meta-analysis for new generation antidepressants
One-step method (Fixed effects model) | Two-step method (Fixed effects model) | Two-step method (Random effects model) | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Estimate | 95%CI | P-value | Estimate | 95%CI | P-value | Estimate | 95%CI | P-value | ||||
μ
1
| 2.100 | −0.407 | 4.606 | 0.101 | 1.353 | −0.884 | 3.590 | 0.236 | −0.040 | −2.313 | 2.234 | 0.973 |
μ
2
| −0.029 | −2.813 | 2.755 | 0.984 | 0.284 | −2.183 | 2.750 | 0.822 | 0.193 | −2.311 | 2.697 | 0.880 |
μ
3
| 2.829 | 1.574 | 4.084 | < 0.001 | 2.790 | 1.537 | 4.043 | < 0.001 | 2.933 | 1.660 | 4.207 | < 0.001 |
μ
4
| 1.865 | 0.604 | 3.125 | 0.004 | 1.812 | 0.553 | 3.072 | 0.005 | 1.576 | 0.297 | 2.856 | 0.016 |
μ
5
| 0.979 | −0.289 | 2.247 | 0.130 | 0.985 | −0.282 | 2.252 | 0.128 | 0.846 | −0.442 | 2.134 | 0.198 |
μ
6
| −0.547 | −0.663 | −0.430 | < 0.001 | −0.511 | −0.615 | −0.407 | < 0.001 | −0.456 | −0.571 | −0.341 | < 0.001 |
μ
7
| 1.022 | 0.487 | 1.558 | < 0.001 | 1.019 | 0.486 | 1.551 | < 0.001 | 1.081 | 0.544 | 1.618 | < 0.001 |
μ
8
| 0.551 | 0.014 | 1.088 | 0.044 | 0.536 | 0.002 | 1.071 | 0.049 | 0.577 | 0.038 | 1.115 | 0.036 |
μ
9
| 0.110 | −0.431 | 0.651 | 0.689 | 0.104 | − 0.435 | 0.642 | 0.705 | 0.141 | −0.402 | 0.684 | 0.610 |
μ
10
| 0.193 | 0.137 | 0.249 | < 0.001 | 0.196 | 0.140 | 0.251 | < 0.001 | 0.189 | 0.116 | 0.261 | < 0.001 |
μ
11
| 0.138 | 0.082 | 0.194 | < 0.001 | 0.141 | 0.085 | 0.197 | < 0.001 | 0.150 | 0.077 | 0.222 | < 0.001 |
μ
12
| 0.045 | −0.011 | 0.102 | 0.117 | 0.045 | −0.011 | 0.102 | 0.115 | 0.049 | −0.024 | 0.123 | 0.185 |
μ
13
| −0.046 | −0.175 | 0.082 | 0.479 | −0.063 | −0.177 | 0.051 | 0.278 | −0.066 | −0.190 | 0.058 | 0.294 |