Correlation Between Isocitrate Dehydrogenase Gene Aberrations and Prognosis of Patients with Acute Myeloid Leukemia: A Systematic Review and Meta-Analysis

The development of effective therapies for most subtypes of acute myeloid leukemia (AML) has remained sluggish for decades. Among the many reasons, an essential one is the substantial heterogeneity of this malignancy (1). Prognosis of AML can be divided into three risk stratifications according to cytogenetic karyotypes reported in National Comprehensive Cancer Network (NCCN) guidelines: favorable, intermediate, and unfavorable risk. Although clinical outcome in most AML patients with favorable or poor karyotypes could be predicted (2), those with intermediate-risk karyotypes (IR-AML) need more molecular markers to determine their prognosis and guide personalized therapies. Fortunately, next-generation sequencing (NGS) techniques provided new opportunities for discovering more mutational profiles in AML genomes, such as the genes encoding DNA methyltransferase 3A (DNMT3A), isocitrate dehydrogenase 1 and 2 (IDH1/2), as well as Tet oncogene family member 2 (TET2;ref.3), which are the key for the modification of DNA methylation and involved in the pathogenesis of leukemia (4, 5).

Several studies and reviews have reported poor prognosis of patients with DNMT3A (6) and TET2 (7) mutation. However, prognostic assessment of IDH mutations was still controversial and needs to be further evaluated. Notably, a meta-analysis performed by Feng and colleagues (8) containing 15 studies indicated that mutant IDH1 was significantly related to shorter overall survival (OS), whereas in a study from Zhou and colleagues (9) including 13 studies, the IDH2 mutation was observed to improve OS. Besides, IDH mutations are particularly common in cytogenetically normal AML patients (CN-AML; ref. 9). For these reasons, prognostic availability of IDH mutations in IR-AML patients should be further explored. In addition, several studies, including the one by Wagner and colleagues (10), reported the association of the IDH1 single-nucleotide polymorphism (SNP) rs11554137 with noticeably poor prognosis in CN-AML patients. In addition, a large sample study by Papaemmanuil and colleagues (11) systematically perfected existing molecular classification system and especially pointed out that IDH2 (R172) but not IDH2 (R140) mutation was related to favorable prognosis in AML patients. Therefore, it is necessary to further assess IDH1 SNP rs11554137 and different subtypes of IDH2 mutation in AML.

In the other side, IDH mutations conferred a new enzymatic function, resulting in accumulation of 2-hydroxyglutarate (2-HG), which plays a vital role in changing DNA methylation of cells, impairing cell differentiation and probably contributing to aberrantly epigenetic mechanism of pathogenesis in AML (1). On the basis of the mechanism of IDH mutations, there have been 3 types of target drugs (IDH305 to IDH1R132 mutation, clinicaltrials.gov NCT02381886; AG-120 to IDH1 mutation, clinicaltrials.gov NCT02074839, NCT02632708, NCT02677922 and AG-221 to IDH2 mutation, clinicaltrials.gov NCT01915498, NCT02632708) generated and undergoing clinical trials (12–14). However, due to controversial prognosis of IDH mutations, it is crucial to determine a clear insight of prognostic value of each subtype in these mutations, which will contribute to more accurate and personalized medicine.

In this study, we identified whether each subtype of IDH mutations [IDH1, IDH2 (R140) and R (172)] could influence prognosis of AML patients.

Eligible studies have these criteria: (i) cohort studies published in English or Chinese, (ii) restricted to human studies mainly reporting the prognostic impact of IDH mutations on adult AML patients, (iii) included information in terms of survival and treatment outcome, and (iv) simultaneously described prognosis-related details comparing characteristics between patients harboring IDH mutations or not. Studies were excluded if they: (i) only emphasized on pediatric leukemia, (ii) reported data which were unavailable or insufficient, (iii) were reviews, case reports, editorials and letters, and (iv) had overlapping cohorts.

We performed a literature search without limitation in regions. The primary sources were the electronic databases of PubMed, Embase, and Cochrane as well as Chinese databases including WanFang Database and China National Knowledge Internet (CNKI), with the publication date from January 1, 2010 to June 26, 2016. The terms included “AML,” “acute myeloid leukemia,” “IDH,” “IDH1,” “IDH2,” “IDH1/2,” “ isocitrate dehydrogenase” and “IDH1 SNP.” In addition, manual searches of reference list were also performed.

The titles and abstracts of all potential studies were initially browsed independently by two reviewers (Q.Y. Xu and Y. Li) to screen and narrow down the studies according to inclusion and exclusion criteria. Any discrepancy was resolved by discussion or consultation with another investigator not involved in the initial procedure. After candidate studies were selected, full-length articles were reviewed to identify eligible studies and the ultimately identified studies were determined by quality assessment.

The methodologic quality of primary manuscripts was evaluated separately by two reviewers (Q.Y. Xu and Y. Li), according to the Newcastle-Ottawa-Scale (NOS; ref. 15), which is used for quality assessment of cohort studies and case–control studies. Studies scoring six or more were considered to be with high quality. Any disparities between investigators were addressed by discussion.

Related information from the identified studies were extracted and summarized independently by two of the authors (Q.Y. Xu and Y. Li). Any disagreement was solved by discussion or consultation with a third reviewer.

The extracted data contained the first author, study characteristics (including NOS scores, publication year, journal, region, sample size, age, tumor types, incidence of different IDH gene aberrations, mutation detection therapies, therapeutic regimens and data types; Table 1) and participant characteristics [including sex ratio, laboratory examination results, French-American-British (FAB) classification, cytogenetic risk categories, sample size with normal karyotype, incidence of mutant nucleophosmin 1 gene (NPM1) and fms-like tyrosine kinase 3-internal tandem duplication gene (FLT3-ITD; Supplementary Table S1)].

Furthermore, survival and treatment outcome information was also incorporated into this meta-analysis, including relative risk (RR) for complete remission (CR) rates as defined according to recommended criteria (16) and HR for overall survival (OS, defined from diagnosis or the date of entry onto the studies to death or alive at last follow-up; refs. 10, 17), event-free survival (EFS, defined from diagnosis or the date of entry onto the studies to treatment failure, relapse, death or last follow-up in CR; refs. 17, 18), as well as cumulative incidence of risk (CIR; ref. 19). Data were preferentially extracted from multivariate analyses. However, in some studies without multivariate results provided, data were obtained from univariate analyses, or calculated from numeric reports or Kaplan–Meier survival curves using the methods previously proposed by Tierney and colleagues (20).

The Stata 12.0 statistical software was used for the meta-analysis. Pooled RRs or HRs less than 1.00 indicated a better therapeutic effect or prognosis in AML patients with IDH gene aberrations, compared with those harboring wild-type IDH gene, and it would be considered statistically significant that 95% CIs did not cover 1. A P value less than 0.05 also meant statistical significance.

The heterogeneity among primary studies was evaluated by using the Q test. A P value greater than 0.10 was considered to be without heterogeneity or with slight heterogeneity, whereas P < 0.10 indicated the existence of significant heterogeneity. Besides, I-square (I²) < 30%, 30%–50%, 50%–75%, and >75% were defined as low, moderate, substantial, and considerable heterogeneity, respectively (21). The random effect model, which was admitted to be more conservative, was chosen if significant heterogeneity was observed. Otherwise, the fixed-effect model was used. Moreover, to find the source of significant heterogeneity, sensitivity analyses were performed to determine whether an individual study affected the aggregate result. Subgroup analyses were also used for exploring the potential source of significant heterogeneity based on the following aspects: region of subjects, mean or median age, mutation detection therapy, data type, and therapeutic regimens.

Visual inspection of the funnel plot was initially done to assess publication bias. We next used the Egger (22) and Begg tests (23) to further evaluate it. A P value less than 0.05 indicates the existence of publication bias.

All analyses were based on published studies; therefore, no ethical approval and patient consent were required.

The procedure of selecting studies was shown in Figure 1. First, 502 studies were retrieved from PubMed, Embase, Cochrane, and Chinese databases (WanFang and CNKI), 197 of which were removed because of overlapping datasets. We screened the remaining 305 studies by browsing their titles and abstracts, excluded 225 studies for no association with our interest and chose 80 reports including articles and conference literatures for further evaluation. Of the 80 full-text studies, 47 were ruled out for the following reasons: 6 studies were only involved in pediatric leukemia; 3 reports were reviews; the data of 35 studies were incomplete or unavailable for analysis; the patient cohorts of 3 literatures overlapped with 3 other articles. The reviewers had perfect agreement in identifying the remaining 33 studies using the aforementioned eligibility criteria and all of identified manuscripts were with high quality (Table 1).

The characteristics of 33 studies were summarized in Table 1 (6, 10, 11, 16–19, 24–49): 13 studies from Europe and Australia, 10 from Asia, 8 from America and 2 collaborated studies from Asia, Africa, and America as well as America, Japan, and Germany, respectively, with a total sample size of 12,747 cases. The frequency of IDH1 mutation was 2.02%–9.30% in AML patients and 10.91%–15.96 % in CN-AML patients, respectively, whereas the frequency of IDH2 mutation varied from 5.05% to 14.76% in AML subjects and from 5.85% to 19.27% in CN-AML cases, respectively.

As shown in Supplementary Table S1, IDH gene aberrations were closely associated with elderly patients in 12 studies (16, 18, 25, 26, 30, 32, 33, 38, 41–43, 47, 48) and with higher platelet count in 9 studies (16, 18, 26, 27, 29, 36, 38, 42, 43). In addition, in the aspect of FAB classification, patients with M1 harbored higher percentage of IDH mutations (6, 17, 18, 24, 25, 29, 30, 39, 42), whereas in cases with M4, lower percentage was observed (17, 24, 29, 47). Besides, the frequency of IDH mutations was higher in cases with normal karyotype in 9 studies (6, 16–18, 24, 25, 29, 30, 39, 40, 42, 48). Finally, higher percentage of mutant NPM1 was always correlated with higher frequency of IDH gene aberrations in 14 studies (6, 16–19, 24, 25, 30, 36, 38–40, 42, 43, 47, 48).

When mutant IDH1 and IDH2 were considered as a single factor - IDH mutation, pooled HRs for OS in AML patients were 1.05 (95% CI, 0.89–1.23; P = 0.5836; heterogeneity: I²= 65.1%, P = 0.000; Fig. 2A). Because of significant heterogeneity observed among the selected studies, we conducted a sensitivity test and found that omitting any single study did not influence the result of OS. Hence, subgroup analyses were proposed in Table 2. Among AML patients under 60 years of age, the combined HRs of OS were 0.99 (95% CI, 0.85–1.16, P = 0.9292; heterogeneity: I²= 0.0%, P = 0.662; Fig. 2B). Except for prognostic insignificance for OS, the summary HRs for EFS were 0.97 (95% CI, 0.80–1.18; P = 0.7713; heterogeneity: I²= 0.0%, P = 0.674; Fig. 2C).

For prognostic influence of IDH mutation on CN-AML patients, pooled HRs of OS and EFS were 1.09 (95% CI, 0.85–1.40, P = 0.4944; heterogeneity: I²= 52.3%, P = 0.078) and 1.14 (95% CI, 0.69–1.88; P = 0.6393; heterogeneity: I²= 67.5%, P = 0.027) (Supplementary Fig. S1A–S1C), respectively. As heterogeneity was observed, we also performed a similar sensitivity analysis. When the study by Nomdedéu and colleagues (34) was ruled out, the summary HRs of OS and EFS decreased to 1.03 (95% CI, 0.88–1.20, P = 0.7306; heterogeneity: I²= 0.0%, P = 0.568) and 0.91 (95% CI, 0.67–1.23, P = 0.5460; heterogeneity: I²= 0.0%, P = 0.795; Supplementary Fig. S1B and S1D), respectively. Therefore, this study might be the source of significant heterogeneity, which needs further discussion.

These pooled HRs of OS and EFS suggested that when IDH1 and IDH2 mutation were analyzed together rather than separately, prognostic impact of these mutations were insignificant in AML or CN-AML patients. Interestingly, among IR-AML patients, the combined HRs of CIR was 1.44 (95% CI, 1.18–1.76, P = 0.0003; heterogeneity: I²= 0.0%, P = 0.429; Fig. 2D).

Becasue of prognostic ineffectiveness of IDH mutation, it was hypothesized that IDH1 and IDH2 mutations might be associated with different prognosis for AML patients. Indeed, patients with mutant IDH1 had a significant disadvantage of OS (HR: 1.17; 95% CI, 1.05–1.31, P = 0.0047; heterogeneity: I²= 0.0%, P = 0.570; Fig. 3A) and EFS (HR, 1.29; 95% CI 1.07–1.56, P = 0.0110; heterogeneity: I²= 2.8%, P = 0.391; Fig. 3B). Patients with IDH1 SNP rs11554137 also had shorter OS (HR, 1.34; 95% CI, 1.03–1.75, P = 0.0294; heterogeneity: I²= 0.0%, P = 0.396; Fig. 3C).

Among CN-AML patients, the summary HRs for OS and EFS were 1.21 (95% CI, 1.01–1.46, P = 0.0388; heterogeneity: I²= 4.4%, P = 0.393; Fig. 3D) and 1.56 (95% CI, 1.23–1.98, P = 0.0002; heterogeneity: I²= 22.2%, P = 0.273; Fig. 3E), respectively. Finally, for patients with IDH1 mutation, the CR rates were also lower (RR, 1.21; 95% CI, 1.02–1.44, P = 0.0289; heterogeneity: I²= 0.0%, P = 0.710; Fig. 3F).

Interestingly, the prognosis of patients with IDH2 mutation was significantly favorable in OS (HR, 0.78; 95% CI, 0.66–0.93, P = 0.0053; heterogeneity: I²= 28.7%, P = 0.199; Fig. 4A), especially in IR-AML patients (HR, 0.65; 95% CI 0.49–0.86, P = 0.0026; heterogeneity: I²= 0.0%, P = 0.950; Fig. 4B).

However, among CN-AML patients, the combined HRs for OS were 0.99 (95% CI, 0.70–1.42, P = 0.9867; heterogeneity: I²= 52.1%, P = 0.051; Supplementary Fig. S2A) and became 0.85 (95% CI, 0.65–1.10, P = 0.2114; heterogeneity: I²= 0.0%, P = 0.608; Supplementary Fig. S2bB) with the study by Boissel and colleagues (19) excluded. In addition, the summary HRs for EFS in CN-AML cases were 0.89 (95% CI, 0.66–1.21, P = 0.4671; heterogeneity: I²=0.00%, P = 0.933; Supplementary Fig. S2C).

For patients with the IDH2 (R140) mutation, the combined HRs for OS were 0.83 (95% CI 0.60–1.16, P = 0.2813; heterogeneity: I²= 69.0%, P = 0.006; Supplementary Fig. S3). As exclusion of any single study would not alter the aggregate result, subgroup analyses were also performed in Table 2.

For patients harboring the IDH2 (R172) mutation, the pooled HRs of OS were 0.72 (95% CI, 0.41–1.27, P = 0.2579; heterogeneity: I²= 64.3%, P = 0.024) and the sensitivity analysis showed that the study by Green and colleagues (30) brought great heterogeneity. After this study was removed, the pooled HRs for OS were 0.59 (95% CI, 0.35–0.99, P = 0.0457; heterogeneity: I²= 30.9%, P = 0.227; Supplementary Fig. S4A and S4B).

In the aspect of treatment outcome of patients with mutant IDH2, the pooled RRs of CR rates were 1.23 (95% CI, 0.96–1.59, P = 0.1004; heterogeneity: I²= 59.2%, P = 0.023) and were slightly lowered by excluding the study by Boissel and colleagues (ref.19; RR: 1.15; 95% CI, 0.94–1.40, P = 0.1767; heterogeneity: I²= 38.0%, P = 0.153; Supplementary Fig. S5A and S5B). Interestingly, the summary RRs of CR rates for the mutant IDH2 (R172) were 2.14 (95% CI, 1.61–2.85; P = 0.0000; heterogeneity: I²= 16.4%, P = 0.302; Supplementary Fig. S5C).

In CN-AML patients with IDH mutations, sensitivity analysis revealed that the study by Nomdedéu and colleagues (34) was the source of heterogeneity in pooled HRs for OS and EFS. The follow-up period in Nomdedéu and colleagues (34) (12–72 months) was significantly shorter than that of other studies (approximately 10 years), which brought relatively prolonged OS expectancy of patients with wild-type IDH gene, leading to outlier HR and generating great heterogeneity.

Among CN-AML cases with mutant IDH2, the great heterogeneity was derived from the study by Boissel and colleagues (19) in OS and CR rates analysis due to its lower percentage (5.85%) and fewer samples (12 cases) of IDH2 mutation than that in the remaining studies, which might be unrepresentative and bring randomness to HR for OS and RR for CR rates.

The prognostic value of IDH2 (R172) mutation was unclear due to significant heterogeneity, which might be from the study of Green and colleagues (30). In fact, the prognostic significance of IDH2 (R172) mutation in this manuscript was really different from four other studies. However, this study could not be excluded for the following two reasons: First, although the frequency of IDH2 (R172) mutation in Green and colleagues (30) was the lowest (2.01%) compared with other studies [Patel and colleagues (35), 2.30%; Koszarska and colleagues (38), 2.65%; Willander and colleagues (41), 2.64%; Papaemmanuil and colleagues (11), 2.53%], the sample size of cases with IDH2 (R172) mutation in this study was large, accounting for 32.22% of the total number of cases with mutant IDH2 (R172) in this analysis. Besides, the study by Green and colleagues (30) also had enough cases, accounting for 36.89% of the total. Therefore, the possibility of contingency derived from small samples might be low. Finally, the follow-up period of Green and colleagues (30) was 10 years, which was long enough to minimize bias from shorter duration of follow-up.

With the advent of significant heterogeneity in OS for IDH and IDH2 (R140) mutations, we therefore performed subgroup analyses (Table 2). Although the most important source of heterogeneity might be from different prognostic value of subtypes of IDH mutations, we did not performed related subgroup analyses for clear prognosis of each mutation has been shown above.

As shown in Table 2, the original regions of samples, mutation direction methods, therapeutic schedules and data types had no effect on OS for mutant IDH and IDH2 (R140). However, in the aspect of mean/median age, younger people (mean/median age ≤ 50 years) with IDH mutations were with more consistency (HR, 0.91; 95% CI, 0.80–1.02, heterogeneity: I²= 0.0%, P = 0.553) than other age groups (> 50 years, HR, 1.00; 95% CI 0.73–1.37, heterogeneity: I²= 74.4%, P = 0.001, and unknown; HR, 1.36; 95% CI 1.10–1.68, heterogeneity: I² = 0.0%, P = 0.635; P = 0.005). In addition, the IDH2 (R140) mutation was associated with prolonged OS among younger patients (mean/median age ≤ 50 years, HR, 0.64; 95% CI, 0.49–0.82, heterogeneity: I²= 0.0%, P = 0.906) when compared with older patients (mean/median age > 50 years, HR, 1.11; 95% CI, 0.37–3.33, heterogeneity: I²= 84.7%, P = 0.010).

Egger and Begg tests were performed to evaluate publication bias and funnel plot symmetry was examined. No evident publication bias was observed based on the visual distribution of funnel plot (Supplementary Fig. S6A–S6C) and P values in Egger and Begg tests (Supplementary Table S2).

As shown in Supplementary Table S3, the analyses without heterogeneity (I² = 0.00%) had the same HRs or RRs and 95% CIs in random effect model and fixed effect model. Additionally, HRs, RRs and 95% CI from the analyses with heterogeneity (I² > 0.00%) were slightly changed from random effect model to fixed effect model but it had no impact on prognostic analyses.

In this study, IDH mutations showed obviously different prognostic significance because of various subtypes. When IDH1 and IDH2 mutation were analyzed together rather than separately, there was no prognostic availability and great heterogeneity was observed in our analysis, both of which could be explained by the diverged impact of IDH1 and IDH2 mutations on survival.

Cases with mutant IDH1 had reduced OS (HR, 1.17; P = 0.0047) and EFS (HR, 1.29; P = 0.0110) than those harboring wild type, especially in CN-AML patients (HR for OS: 1.21, P = 0.0388; EFS: 1.56, P = 0.0002). IDH1 SNP rs11554137 also conferred shorter OS (HR, 1.34; P = 0.0294). Therefore, these results suggested that mutant IDH1 and IDH1 SNP rs11554137 might contribute to adverse prognosis, similar to FLT3-ITD mutation (50), and would be allowed to perform more accurate molecular risk stratification of AML.

Likewise, mutant IDH2 might also be one of prognostic markers for its significant association with better prognosis (HR for OS: 0.78; P = 0.0053), especially in the subset of IR-AML cases (HR for OS: 0.65; P = 0.0026). Interestingly, among CN-AML patients, mutant IDH2 had no impact on OS (P = 0.2114) and EFS (P = 0.4671). The discrepancy of survival between CN-AML and IR-AML cases might originate from cytogenetically abnormal karyotype belonging to intermediate risk, such as isolated trisomy 8, t (9; 11) and some nondefined karyotypes. Abnormal karyotype frequencies were listed as follows: 24.44% versus 28.95% (mutant IDH2 vs. wild-type IDH2) in Chou and colleagues (29) and 31.58% versus 17.81% (mutant IDH2 vs. wild-type IDH2) in Ley and colleagues (6, 39). Notably, trisomy 8 alone was significantly associated with IDH mutations (24, 29). However, little is known about survival of patients with the combination of isolated trisomy 8 and IDH mutations, and more data are required to distinguish prognostic differences between IR-AML cases with normal and those with abnormal karyotypes when harboring mutant IDH. Furthermore, the IDH2 (R140) mutation could improve OS among younger patients (mean/median age ≤ 50 years, HR: 0.64, P = 0.0005) after conducting subgroup analyses. In addition, since several studies have reported favorable prognosis of IDH2 (R172) mutation, particularly in studies by Papaemmanuil and colleagues (11) and by Paschka and colleagues (European Hematology Association, 2016), we conducted a meta-analysis to evaluate prognostic value of this mutation. The impact of IDH2 (R172) mutation on survival had significant heterogeneity, showing it still remains controversial about its prognostic value and more studies are required to clarify its role.

Despite the diverged survival significance between IDH1 and IDH2 mutations, CR rates in patients with IDH mutations were consistently lower, particularly for those with mutant IDH1 (RR, 1.21; P = 0.0289) and IDH2 (R172; RR, 2.14; P = 0.0000). Interestingly, Green and colleagues (30) found that CR rates of the IDH2 (R140) mutation were relatively high (RR, 0.47; P = 0.0032). Except for declining CR rates, there was increasing CIR in IR-AML patients with IDH mutations (HR, 1.44; P = 0.0003).

Our results have identified the prognostic value of each subtype of IDH mutations in AML patients, which might contribute to the application and therapeutic evaluation of target drugs in clinic.

As mentioned in Supplementary Table S1, IDH mutations appear to be notably connected with mutant NPM1. Besides, FLT3-ITD and NPM1 mutations have been compiled into the NCCN guidelines. In this regard, it is worth discussing the prognosis of IDH mutations in consideration of the other two.

In cases with wild NPM1, IDH1 mutations were related to reduced OS (refs. 6, 39; P = 0.0030) and EFS (ref. 17; P = 0.044), whereas IDH2 mutation was associated with prolonged OS (ref. 30; P = 0.0206) and EFS (refs. 6, 39; P = 0.0283). In the study by Green and colleagues (30), patients with the IDH2 (R140) mutation showed a better OS (P = 0.0002) when limited to those with wild FLT3-ITD. Furthermore, for IR-AML cases with wild-type NPM1 and FLT3-ITD, IDH1 mutations were also related to remarkably shorter OS (P < 0.05) and EFS (P < 0.05; refs. 6, 18, 39).

Our results were consistent with two other studies published previously: a meta-analysis (8) including 15 studies involved in IDH1 mutation in AML and another one (9) covering 13 studies associated with IDH1 and IDH2 mutation in nonpromyelocytic AML. Both found that IDH1 mutation was correlated with poor prognosis, whereas opposite prognosis was observed in IDH2 mutation. Our meta-analysis contained 33 publications and included 12,747 cases in total, a larger scale study that could significantly increase the statistical power and accurately assess the prognostic impact of IDH mutations on AML patients.

Another significant feature of our research is that the populations included were broadly from Europe, Asia, Australia, and America. Hence, our meta-analysis promised an extensive utilization of IDH mutations in the prognosis of AML patients.

More importantly, we investigated more endpoints of survival and treatment outcome. IDH mutations had no impact on OS and EFS if integrated as a single factor. For mutant IDH1, we not only performed analyses about OS and EFS in AML and CN-AML cases, but also found that IDH1 SNP rs11554137 was correlated with poor prognosis for AML patients, a finding previously not reported in the two meta-analyses. We also noted that mutant IDH1 was more unfavorable for OS in CN-AML cases, which was not shown in the two previous analyses. We found that IDH2 mutation was not only associated with prolonged OS but more linked to favorable prognosis among IR-AML patients, which was also not shown in the two previous articles. We found that mutant IDH2 (R140) but not IDH2 (R172) was correlated with longer OS in subgroup analyses, a novel finding as well. In addition, we found that IDH mutations were related to CIR and treatment outcome. In particular, the CR rates were lower for patients with the IDH2 (R172) mutation, a finding not shown before.

We performed in-depth subgroup analyses according to different clinical and methodological features to synthetically estimate the prognostic influence of IDH and IDH2 (R140) mutations and investigated potential sources of heterogeneity. It was obvious that characteristics of patients (mean/median age ≤ 50 years) were more uniform than those older than 50 years and the IDH2 (R140) mutation in younger cases was notably related to better OS, reflecting that older age might be a more powerful factor in contributing to poor prognosis than the IDH2 (R140) mutation does.

Our meta-analysis has several limitations. First, although comprehensive studies were selected from three major databases and Chinese databases, other relevant studies, especially those published in non-English or non-Chinese language or not published in public might be overlooked. Second, our analyses were based on retrospective cohort studies. Therefore, it was difficult to precisely control selection, attrition, information, and confounding bias. Third, some data were obtained from univariate analyses or calculated from Kaplan–Meier survival curves and numeric reports, which might be in slight disparity with the fact. In addition, these results were from univariate analyses and might not be as stable as multifactor-derived data. Finally, although we extracted HRs or RRs from multivariate analyses as many as possible, various confounding factors existed and should be taken into account in different studies, thereby leading to heterogeneity of pooled HRs or RRs to some extent.

No potential conflicts of interest were disclosed.

The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conception and design: Q. Xu, Y. Li, L. Yu

Development of methodology: Q. Xu, Y. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Q. Xu, Y. Li, Y. Xu, Y.Y. Li, W. Li, Z. Yao, X. Chen, S. Huang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Q. Xu, Y. Li

Writing, review, and/or revision of the manuscript: Q. Xu, Y. Li, Y.H. Li, L. Yu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Lv, Y. Jing, L.L. Wang

Study supervision: L. Yu

This work was supported by the National Natural Science Foundation of China (81670162, 81370635, 81170518, 81270611, 81570137, 81470010 and 81400135), Capital Medical Development Scientific Research Fund (SF2001–5001-07), Beijing Natural Science Foundation (7151009), National Public Health Grant Research Foundation(No.201202017), the Capital of the Public Health Project (Z111107067311070).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.