Reply to Wang and Zhang.

1. An additional updated literature search had been conducted by the authors (D. Menzies and H. Hussain) to identify additional studies that had been published since the last systematic review and might otherwise have been included in the individual patient data (IPD). Of 523 titles identified, 30 were selected for full text review. Thirteen were additional reports on the cohorts that were already in the IPD data set, and 11 were excluded, leaving 6 studies with a total of 613 patients. However, the characteristics of these 613 patients were very similar to those of the patients included in our analysis, and considering the large data set already included in our study (almost 9000 patients), we are confident that the addition of these studies would not have significantly impacted the current results. 2. The use of >1 quinolone or injectable was a marker of failure—we believed that including these patients in our analysis would have biased our results. 3. We presented the results of the multivariate analysis using mean values from members of the same cohort. In each table where these analyses were presented, we added footnotes with the information of how often the imputation was needed—for all covariables <15% of the data were missing. Also as we stated in the Methods, we performed a sensitivity analysis using probabilistic imputation [3], and the results using this method were virtually identical (estimates and 95% confidence intervals). 4. Although we agree that a major challenge of our study was potential confounding, we did adjust for indicators of