The PubMed, EMBASE, Science Citation Index, Cochrane Central Register of Controlled Trials, and Clinical Trial Registries databases were searched for relevant studies published up to April 2021. The search terms included “FG-4592,” “roxadustat,” “anemia,” “chronic kidney disease,” “chronic kidney failure,” “chronic renal insufficiency,” “chronic renal disease,” and “randomized controlled trial.” The search was limited to studies involving human subjects, and no language restrictions were applied. The citations of the included studies were scanned to identify additional relevant studies.

The inclusion criteria were as follows: (1) study design: randomized controlled trial (RCT); (2) population: anemia patients with CKD (>18 years old) with or without dialysis; (3) intervention: roxadustat compared with placebo or ESA (epoetin alfa or darbepoetin alfa); and (4) primary outcome: the safety of roxadustat. The secondary outcomes included the hemoglobin response rate, the change in hemoglobin, hepcidin, transferrin saturation, ferritin, total iron-binding capacity and total iron during treatment. The exclusion criteria were as follows: (1) studies that involved healthy individuals; (2) studies that included inappropriate comparisons or did not include a reference group; and (3) studies with research data that could not be extracted and analyzed.

Two reviewers (CL and ZF) independently extracted the data using a standardized, pre-established form. If one study contained more than one clinical trial (trials with national clinical trial numbers), we extracted the data from the study separately. Each trial was assessed using the Cochrane risk of bias tool. The standard criteria included the following domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other bias. Any disagreements were resolved by a third reviewer.

All data were analyzed using Review Manager (version 5.3) and R software (version 3.5.1). The effect size was assessed by relative risks (RRs) with 95% confidence intervals (CIs) for dichotomous outcomes and mean differences (MDs) with 95% CIs for continuous outcomes. Subgroup analyses were conducted to explore between-study heterogeneity. Publication bias was assessed using Begg's and Egger's tests. All P-values were two-sided, and a P < 0.05 indicated a statistically significant difference.

In the meta-analyses, trial sequential analysis (TSA) was used to reduce the risk of reaching a false-positive or false-negative conclusion (21). When the cumulative Z-curve crossed the trial sequential monitoring boundary or entered the futility area, a sufficient level of evidence for the anticipated intervention effect was reached, and no further trials were needed. If the Z-curve did not cross any of the boundaries and the required information size (RIS) had not been reached, evidence to reach a conclusion was insufficient, and more trials were needed to confirm the results (22). For this TSA, we estimated the RIS based on an RR reduction (RRR) of 20%. The type I error (α) = 0.05 (two-sided) and power (1 – β) = 0.80. The control event proportion was calculated from the comparator group (23). The TSA was conducted using TSA Version 0.9.5.10 Beta (www.ctu.dk/tsa).

The study retrieval process is outlined in Figure 1. In total, eleven studies (13–19, 24–27) (containing 12 clinical trials) including 6,631 patients met the inclusion criteria. The study characteristics are summarized in Table 1. These studies were published between 2015 and 2021. Seven studies (13–19) were phase 3 clinical trials, and the others were phase 2 clinical trials. Three studies (13, 14, 25) were conducted in the Chinese population, two (26, 27) in the American population and two (15, 24) in the Japanese population, while the others were multicenter studies (16–19). In NDD-CKD patients, the interventions were roxadustat vs. a placebo. In DD-CKD patients, the interventions were roxadustat vs. ESA (epoetin alfa or darbepoetin alfa).

The details of the risk of bias tool are shown in Figure 2. All studies had a low risk of selection bias. Randomized sequence generation and allocation concealment were reported adequately in most studies. Four studies were high-quality studies with a low risk of bias for all items.

All studies reported adverse events during the treatment process. In the NDD-CKD patient groups (13, 16, 17, 19, 24, 25, 27), the results showed that there was no significant difference in the total adverse events between the roxadustat and placebo groups (RR = 1.02, 95% CI = 1.00, 1.04, P = 0.08, I² = 0%; Figure 3A), and the TSA confirmed this result [the cumulative Z-curve crossed the futility boundary and entered the futility area; relative risk reduction (RRR) = 20%, α = 5%, β = 20%, I² =0%; Figure 3C]. In the DD-CKD patient groups (14, 15, 25, 26), the results showed that the frequency of total adverse events was also similar between the roxadustat and ESA treatment groups (RR = 1.22, 95% CI = 0.91, 1.64, P = 0.18, I² = 60%; Figure 3B), but the TSA did not confirm this result (the cumulative Z-curve just crossed the conventional boundary but did not cross the trial sequential monitoring boundary; RRR = 20%, α = 5%, β = 20%, I² = 60%; Figure 3D). No significant differences in hyperkalemia and infection incidences were found in the DD-CKD patient groups (Table 2). However, there was an increased risk of hyperkalemia in the roxadustat group vs. the placebo group in NDD-CKD patients (RR = 1.24, 95% CI = 1.02, 1.51, P = 0.03, I² = 0%, Table 2).

The hemoglobin response rate was defined as the percentage of patients who had a response to a trial regimen, which was defined as an increase from baseline of at least 1.0 g per deciliter in the hemoglobin level (13). All the included studies reported the hemoglobin response rate. The pooled results showed that roxadustat significantly increased the hemoglobin response rate compared with the placebo in the NDD-CKD patient group (RR = 8.07, 95% CI = 5.78, 11.27, P < 0.01, I² = 61%; Figure 4A), and it had an effect similar to that of ESA in the DD-CKD patient group (RR =1.05, 95% CI = 0.98, 1.13, P = 0.14, I² = 62%; Figure 4B).

We used the hemoglobin change from baseline to evaluate the effect of roxadustat on hemoglobin. Nine studies (13–19, 24, 25) reported comparisons of hemoglobin. If the studies reported the data for more than one dose, we chose the optimal dose (the most effective dose with the lowest risk of adverse events), such as the high dose (1.50–2.25 mg/kg) in the Chen et al. study (2017) (25), 100 mg TIW in the Akizawa et al. study (24) and mid-range dose (1.5–2.3 mg/kg) in the study of Chen et al. (25) in the roxadustat group to perform the pooled analysis. Among both the NDD-CKD and DD-CKD patients, the hemoglobin change from baseline was significantly increased in the roxadustat group compared with the placebo or ESA group (MD = 1.65, 95% CI = 1.37, 1.93, P < 0.01, I² = 92% and MD = 0.16, 95% CI = 0.05, 0.27, P < 0.01, I² = 0% respectively, Table 2).

In all the studies, IV iron preparations were strictly restricted, and oral iron preparations were generally permitted. As described above, if the studies reported the data for more than one dose, we chose the optimal dose (the most effective dose with the lowest risk of adverse events), such as the high dose (1.50–2.25 mg/kg) in the study of Chen et al. (25) for NDD-CKD patients, 100 mg TIW in the study of Akizawa et al. (24) and mid-range dose (1.5–2.3 mg/kg) in the study of Chen et al. (25) for DD-CKD patients in the roxadustat group to perform the pooled analysis. In the NDD-CKD patients, roxadustat significantly decreased hepcidin, ferritin and transferrin saturation and increased the total iron-binding capacity and transferrin compared with placebo (MD = −19.16, 95% CI = −37.67, −0.65, P = 0.04, I² = 96%, MD = −50.16, 95% CI = −71.19, −29.13, P < 0.01, I² = 55%, MD = −2.73, 95% CI = −5.23, −0.23, P = 0.03, I² =74%, MD = 70.22, 95% CI = 33.86, 106.58, P < 0.01, I² = 94% and MD = 0.65, 95% CI =0.26, 1.04, P < 0.01, I² = 95% respectively, Table 2). Total iron was not significantly changed (MD = −2.21, 95% CI = −9.24, 4.81, P = 0.54, I² = 20%, Table 2). In DD-CKD patients, roxadustat significantly increased the total iron-binding capacity, total iron and transferrin compared with ESA (MD = 42.56, 95% CI = 22.95, 62.18, P < 0.01, I² = 86%, MD = 16.28, 95% CI = 9.39, 23.17, P < 0.01, I² = 27% and MD = 0.42, 95% CI = 0.29, 0.54, P < 0.01, I² = 75% respectively, Table 2). Despite the fact that hepcidin and ferritin were decreased in the roxadustat group, the difference was not significant compared with the levels in the ESA group (MD = −13.62, 95% CI = −35.77, 8.53, P = 0.23, I² = 77% and MD = −17.46, 95% CI = −61.63, 26.71, P = 0.44, I² = 73% respectively, Table 2). The rate of oral iron therapy was also decreased, but the difference was not significant (Table 2).

Six studies reported the levels of total cholesterol, low-density lipoprotein (LDL), non-high-density lipoprotein (HDL) and the LDL/HDL ratio (13, 14, 16, 18, 19, 25). However, only the LDL levels data could be pooled and analyzed. The results showed that LDL levels decreased in the roxadustat group compared with the ESA group in DD-CKD patients but no significant change was observed compared with the placebo group in NDD-CKD patients (MD = −18.59, 95% CI = −21.77, −15.42, P < 0.01, I² = 0% and MD = −16.11, 95% CI = −32.15, −0.07, P = 0.05, I² = 98%, respectively, Table 2).

Three studies (15, 18, 26) included in this meta-analysis reported that the EPO treatment effect decreased when the patient was in a state of hyperinflammation (CRP > 4.9 mg/L), whereas the effect of roxadustat did not. However, the data could not be extracted and pooled for analysis.

No evidence of publication bias was observed in the outcomes of adverse events (P = 0.174 for Begg's test, P = 0.260 for Egger's test) in the DD-CKD patients but it was detected in the NDD-CKD patients (P = 0.024 for Begg's test, P = 0.529 for Egger's test). Regarding the hemoglobin response rate, publication bias was detected for this outcome in the NDD-CKD patients (P = 0.573 for Begg's test, P = 0.000 for Egger's test) while the funnel plot showed that there was no significant publication bias in the DD-CKD patients (P = 0.327 for Begg's test, P = 0.579 for Egger's test) (see Supplementary Material).

There are several limitations of our meta-analysis. First, four phase 2 trials were included, which decreased the level of evidence. Second, due to the limited amount of data, we could not analyze possible confounding factors by subgroup or sensitivity analyses. Third, the treatment durations were different among the included studies, resulting in between-study heterogeneity. Fourth, the included studies were conducted with a relatively small sample size; therefore, large sample size trials are required to further evaluate whether the results are reliable. Fifth, this meta-analysis only included the published literature, but not the literature supplemented by other resources, which decreased the level of evidence. Finally, long-term follow-up results were still lacking. Long observation periods are needed to further investigate the long-term effect of roxadustat and to confirm whether HIF-PHIs exhibit off-target activity in injured kidneys. Therefore, further high-quality RCTs are needed to confirm or refute this finding.