Search strategy and selection criteria
A total of 4,673 articles were initially identified. Of the 3,570 articles remaining after excluding duplicate publications, 3501 were excluded after screening the title and abstract because they were not relevant. An additional 62 articles were excluded during the full-text review owing to data proceeding from simulated patients, the subjects of the study being age over 18, lack of prospect to extract blood concentrations, insufficient data on clinical outcomes, clinical data not having been presented by Bu AUC strata or Bu not having been administered 4 times daily for 4 days, among other reasons. Consequently, a total of 13 studies involving 548 patients met the inclusion criteria and, accordingly, were included for meta-analysis [2, 13-17, 20, 23-27, 29]. The literature selection process is summarized in Figure 1.
A summary of descriptions of included studies is reported in Table 1, the studies were published between 1996 and 2017. Nine [13-17, 23, 25, 26, 29]were prospective studies and four [13-17, 23, 25, 26, 29]were retrospective studies. Six studies were conducted in Europe [14, 16, 20, 24, 25, 27] , six studies were in United States [2, 13, 15, 17, 26, 29]and one  was in Japan. Bu concentrations were measured by high-performance liquid chromatography by means of ultraviolet detection , while the remainder [2, 13-17, 23-27] were measured by gas chromatography with mass spectrometry detection.
Evaluation of efficacy
Table 2 displays a summary of outcomes for each study. Table 3 display summaries of meta-analysis for efficacy, Forest plots are shown in Figure 2. Raw data were shown in supplementary data (Table S1 and Figures S1–S12).
Our meta-analysis demonstrated that there were no significant first dose AUC cut-off values for efficacy. We found the cut-off level (AUC mean) of < 900 µM × min to be significantly associated with higher incidence of graft failure (RR=3.666, 95% CI: 1.419, 9.467).
Subgroup analyses showed that the incidence of graft failure significantly decreased above a cut-off level with mean AUC 900 µM × min in the subgroup of administration by an IV infusion route alone (RR=9.718; 95% CI: 1.499- 62.989)，There were no significant differences at other cut-off levels (Table 5 ).
Sensitivity analysis on each study’s effect on the summary estimates for efficacy was shown in supplementary data (Table S3), which illustrated that our results were not driven by any single study, as the RRs remained stable.
Evaluation of Safety
A summary of primary and subgroup analysis for safety are shown in Table 4 and Table 5. Forest plots are shown in Figure 3 and Figure 4. Raw data were shown in supplementary data (Table S2 and Figures S13-16).
The definitions of VOD varied across the 10 studies (Table 2), the incidence of VOD ranged from 4.8% to 40%. On average, VOD occurred between 1 and 29 days after HSCT. Our meta-analysis demonstrated a significantly lower incidence of VOD with mean AUC below cut-off levels of 1350 µM × min (RR=0.370, 95% CI: 0.205-0.666) and 1500 µM × min (RR=0.409, 95% CI: 0182-0.920). In terms of the relationship between first dose AUC and clinical outcomes, our meta-analysis demonstrated there were no significant differences at all cut-off values for VOD.
Subgroup analyses showed that the rate of VOD significantly decreased below a cut-off level with mean AUC 1350 µM × min in the subgroup of without VOD prophylaxis therapy (RR=0.349; 95% CI: 0.182- 0.670), administration by an IV infusion route alone (RR=0.378; 95% CI: 0.158-0.906) or not (either administration by an IV infusion route or by oral) (RR=0.363; 95% CI: 0.163-0.805). There were no significant differences at other cut-off levels.
For others toxic effects, the relationship of Bu AUC with graft versus-host disease (GVHD) was not found, although two studies [30, 31] reported a higher incidence of GVHD when Bu/cyclophosphamide was combined with melphalan. Regarding neurotoxicity, as benzodiazepine or phenytoin was routinely given for seizure prophylaxis, the incidence of neurotoxicity was relatively low. We could not pool the data to perform a meta-analysis. Therefore, an association between AUC and other toxic effects could not be evaluated.
On each study’s effect on the summary estimates showed that exclusion of studies by Wallet al, Bouligand et al and Tran et al  resulted in an insignificant difference at a cut-off level of 1500 µM × min Raw data were shown in supplementary data (Table S4).
The quality assessment of the included studies is presented in Supplementary Table S5. Overall, the subjects included were representative, and ascertainment of exposure was confirmed by secure record, seven studies were comparable on basis of main factors[13, 17, 20, 23, 26, 27, 29], and six studies were comparable on two or more factors[2, 14-16, 24, 25]. Outcome assessment was based on pharmacy and medical records, the follow-up period was sufficient for outcomes to occur, and adequacy of follow-up of cohorts. According to the NOS tool, the quality assessment showed that two studies [17, 26]were scored 6 stars, four studies 7 stars[20, 25, 27, 29], three studies[13, 16, 23] 8 stars, and four studies[2, 14, 15, 24] 9 stars. No study was excluded after rating because the study quality was always above 5 stars.