TABLE 2.
Multivariate analyses of cardiotoxicity and pathological complete response rate (pCR) by fitting two linear fixed‐effects meta‐regression models with the modern stepwise variable selection method a .
| Covariate | Regression coefficient estimate | Estimated standard error | p value c | 95% confidence interval |
|---|---|---|---|---|
| Cardiotoxicity (log RR)q | ||||
| Intercept | −0.1205 | 0.2762 | 0.6625 | −0.6619−0.4208 |
| Trials with HR(+) participants >57.20% | 1.0875 | 0.4225 | 0.0100 | 0.2595–1.9154 |
| Trials with a control arm containing lapatinib | −1.3250 | 0.8431 | 0.1161 | −2.9775−0.3275 |
| Fixed‐effects meta‐regression model (QE = 1.0755, df = 5, p = 0.9563; I 2 = 0.00% < 50%, R 2 = 0.8121) b | ||||
| pCR (log OR) | ||||
| Intercept | 0.5582 | 0.1376 | <0.0001 | 0.2886–0.8278 |
| Trials with HR(+) participants >58.02% | −0.4911 | 0.1994 | 0.0138 | −0.8819 to −0.1003 |
| Fixed‐effects meta‐regression model (QE = 6.2820, df = 6, p = 0.3924; I 2 = 4.49% < 50%, R 2 = 0.3812) b | ||||
A multivariable linear meta‐regression analysis of 7 randomized clinical trials with 8 observations was conducted using the escalc() and rma() functions in the ‘metafor’ package, version 3.8‐1 (2022‐08‐27), of the statistical software R, version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria), with the modern stepwise variable selection procedure, to fit the meta‐regression models of the log(hazard ratio) of cardiotoxicity and the log(odds ratio) of the pathological complete response (pCR) rate, respectively, where “log” was the natural logarithm.
The coefficient of determination, R 2, was the squared Pearson correlation between the observed and predicted responses from all included observations, which helped us assess the GOF of the fitted linear meta‐regression model.
The boldfaced p values reached statistical significance at α = 0.05.