Methods
Search strategy
Two independent investigators implemented a systematic search in PubMed, Embase, and Chinese National Knowledge Infrastructure (CNKI) databases systematically, with the last search updated on 30 September 2020. The following terms “(acute myeloid leukemia) or (acute myeloblastic leukemia) or (acute myelocytic leukemia) or (acute myelogenous leukemia) or (acute nonlymphoblastic leukemia) or AML”, “FLT3 or CD135 or (fms-like tyrosine kinase 3) or (fetal liver kinase-2) or (fetal liver kinase-3) or (human stem cell tyrosine kinase-1)”, “(tyrosine kinase domain mutation) or (TKD mutation) or D835 or I836”, and “prognosis or prognoses or (prognostic factors) or (prognostic implication) or (prognostic element)” were retrieved in PubMed entries in the National Institutes of Health and European Embase databases without any limitation applied. The key words “急性髓系白血病or急性髓性白血病or急性非淋巴细胞性白血病”, “FLT3”, and “TKD突变” were retrieved in CNKI. The reference lists in retrieved studies and relevant reviews were also manually searched for more eligible studies.
Selection criteria
All literature studies involved in AML and FLT3-TKD were electronically retrieved for the next filter. Afterward, prospective cohort studies were identified according to their titles and abstracts. The full texts of the literature studies that fulfilled the inclusion criteria were perused to validate their eligibility. Inclusion criteria were as follows: (a) the evaluation of association between prognosis of AML and FLT3-TKD; (b) untreated patients with AML were included in study; (c) complete original materials with specific explanation of sample size; (d) they provided data of all enrolled subjects on either or both of overall survival (OS) and disease-free survival (DFS) after a period of follow-up in the study; (e) with survival information based on the FLT3 status: FLT3-TKD and wild type; and (f) prospective cohort study focusing on human being. Exclusion criteria were as follows: (a) not conforming to inclusion criteria; (b) abstract, review article, letter, comment, and editorial; (c) duplicate publication of previous publications; (d) family-based studies of pedigrees; (e) without detailed FLT3 status data (FLT3-TKD and wild type); (f) with incomplete specific explanation or without specific explanation of sample size; and (g) studies were excluded if they focused exclusively on acute promyelocytic leukemia (APL) (M3). For multiple publications from a same population, the largest study was included only to exclude duplicate studies or overlapping data. According to the inclusion and exclusion criteria, the two independent investigators accomplished study selection independently by screening title, abstract, and full text. Any dissent was solved by discussion. If agreement could not be reached, then a third researcher was consulted. Four studies were discussed and excluded after discussion (11–14).
Data extraction
The data of the eligible studies were extracted in duplicate by two independent researchers. The data extracted comprised first author’s name, publication year, diagnostic criteria for AML, the resource of the subjects, genotyping methods, the number of subjects, the FLT3 status (FLT3-TKD or wild type) of subjects, the number of OS and/or DFS from all subjects (if any), hazard ratio (HR) with 95% confidence interval (CI) of OS and/or DFS, and the baseline data of all subjects in all included prospective cohort studies [age, sex, patient source, other mutations (if any), the usage of chemotherapeutics (if any), and so on] (
Table S1
). HR and its 95% CI were extracted directly or calculated the observed minus expected (O-E) according to the ratio of event or were extracted by using Engauge Digitizer according to the Cox curve (15). Various patient source descents were classified as Caucasian and Asian. Two investigators would check the extracted data and reach to consensus on all collected data. If a dispute existed, then original data of the included studies would be rechecked and be discussed again to reach consensus. If the dispute still existed, then the third investigators would be appointed as the decider to adjudicate the disagreements.
Quality assessment
The Newcastle-Ottawa (NOS) scale was used to score the strength of all included studies by the three independent investigators (16). The scale has nine items classified into three major categories: selection (four items), comparability (two items), and outcome (three items) (Supplement 1. Method). In this scoring system, selection, comparability, and outcome categories could be awarded a maximum of four, two, and three points, respectively. High quality was considered as six or more points that each cohort study scored. Any discrepancies were resolved among authors. The results of quality assessment are displayed in
Table S2
.
Publication bias and sensitivity analysis
Potential publication bias was checked by Begg’s funnel plots (17) and Egger’s test (18). An asymmetric plot with p-value less than 0.05 was considered a significant publication bias. Moreover, sensitivity analysis was performed on the pooled HRs to evaluate the effect of each study, in which the results of the meta-analysis were recalculated after removal of each study in a turn.
Data analysis
Among entire conduction of the meta-analysis, we strictly abided by the PRISMA checklists as a guideline (19). All statistical analyses were performed with Stata 16.0 software (StataCorp, College Station, TX, USA). A two-tailed p < 0.05 was considered significant except for specified conditions, where a certain p-value was declared. HR and corresponding 95% CI were applied to assess the prognostic impact of FLT3-TKD in patients with AML. Furthermore, HR > 1 was considered as poorer prognosis in patients with FLT3-TKD than patients with FLT3-WT, whereas HR < 1 was considered as beneficial prognosis in patients with FLT3-TKD than patients with FLT3-WT. The heterogeneity of the studies was assessed by I2 statistic (I2 = 0%–25%, no heterogeneity; I2 = 25%–50%, moderate heterogeneity; I2 = 50%–75%, large heterogeneity; and I2 = 75%–100%, extreme heterogeneity) (20). When the heterogeneity was statistically significant (I2 > 50%), the random-effects model was used for assessing information; otherwise, the fixed-effects was conducted (21). Heterogeneity was analyzed by meta-regression model including age and patient source, and subgroup analysis was stratified by age and patient source. In addition, Begg’s and Egger’s tests were performed to detect potential publication bias. Furthermore, sensitivity analysis was conducted to determine the stability of findings in meta-analysis.
Discussion
In this meta-analysis, 20 prospective cohort studies (n = 10,970) on FLT3-TKD in AML were included: 9,744 subjects with FLT3-WT and 1,226 subjects with FLT3-TKD (22–41). The incidence of FLT3-TKD in this meta-analysis was 11.2%, which was nearly consistent with previous studies (approximately 7%–10% of all AML cases) (2). Our results indicated that, within 20 cohort studies (n = 10,970) included, the pooled HR of DFS was 1.12 (95% CI: 0.90–1.41; I2 = 70.7%, p = 0.000) and OS was 0.98 (95% CI: 0.76–1.27; I2 = 79.9%, p = 0.000), which revealed no significant effect of FLT3-TKD on DFS and OS of patients with AML by random effect models. However, meta-regressions demonstrated that patient source associated with the prognosis effect of FLT3-TKD in patients with AML. To be specific, FLT3-TKD represented a beneficial prognosis of DFS (HR = 0.56, 95% CI: 0.37–0.85) and OS (HR = 0.63, 95% CI: 0.42–0.95) for Asians with AML, whereas FLT3-TKD represented an adverse prognosis of DFS for Caucasians with AML (HR = 1.34, 95% CI: 1.07–1.67). However, the results of DFS from Caucasians ought to be interpreted with caution due to the heterogeneity (I2 = 72.9%, p = 0.000).
In general, FLT3-TKD reveals no significant effects on DFS and OS of patients with AML, which is consistent with the controversial statue nowadays. The controversies of prognostic impacts on FLT3-TKD in patients with AML were supported by two-sided laboratory evidence of FLT3-TKD. On one hand, many studies indicated that FLT3-TKD associated a beneficial prognosis of AML. In some cases, patients with FLT3-TKD companying with other mutations, such as NPM1 mutation, showed a favorable prognosis (27) with the reasons that the localization and signaling of FLT3-TKD was changed by NPM1c in AML (42). On the other hand, other research studies manifested that FLT3-TKD was considered as a harmful mutation in prognosis of AML. Since FLT3-ITD was first recognized as a frequently mutated gene in AML in 1996 (43), growing studies indicated that a lot of FLT3-ITD–positive patients with AML relapse with the appearance of FLT3-TKD after initial response to FLT3 inhibitor treatments. Moreover, several FLT3 inhibitors including Sorafenib and Quizartinib potently had effects on inhibiting FLT3-ITD but were not effective toward FLT3-TKD (2, 44, 45). The phenomenon was plausibly interpreted as the coexistence of two kinds of FLT3 mutations and the presence of FLT3-TKD in a very low level at initial stage of disease, which subsequently became prevalent after FLT3-ITD–positive leukemic cells, are eliminated (46).
Our results indicated that FLT3-TKD represented a beneficial prognosis for Asians with AML, whereas it represented an adverse prognosis of DFS for Caucasians with AML, but with heterogeneity. First, in the Caucasian subgroup, the pooled HR of DFS was 1.34 (95% CI: 1.07–1.67; I2 = 72.9%, p = 0.000), and the pooled HR of OS was 1.11 (95% CI: 0.85–1.44; I2 = 80.5%, p = 0.000) (
Figure 4
). There were 635 Asians from four different countries and 10,335 Caucasians from eight different countries in this meta-analysis, which revealed that Caucasians outnumbered Asians. Hence, we speculated that confounding factors from large sample capacity of the Caucasian subgroup accounted for the heterogeneity in Caucasians (I2 = 72.9%, p = 0.000). In this situation, if related confounding factors were well controlled, then the conclusion might be more convinced. However, it was indicated that Caucasians reveals a distinct genetic alteration profiles of AML than Eastern Asian population (47), but what we focused on is the ratio of FLT3-TKD in AML, which would not be influence by population base too much at this meta-analysis. Second, according to different therapeutic guidelines of AML, the chemotherapy regimens for patients with AML vary internationally. On one hand, included studies from Japan accounted for a large population in the Asian subgroup in this meta-analysis. The conventional “3 + 7” induction is regarded as basic regimens for complete remission (CR) with anthracyclines for 3 days and standard dose of cytarabine for 7 days. We noticed that the dose of cytarabine and anthracyclines in Japan was less than the dose in Caucasian countries according to practical guideline for AML (48, 49). Cytarabine of 100 mg/m2 was recommended for 7 days in Japan, whereas cytarabine of 100–200 mg/m2 was recommended for 7 days in Caucasian countries; and daunorubicin of 50 mg/m2 was recommended for 3 days in Japan, whereas daunorubicin of 60–90 mg/m2 for 3 days was recommended in Caucasian countries (48, 49). Hence, we speculated prognosis of AML in different countries or regions might attribute to the dose of cytarabine and daunorubicin, which consist of the conventional “3 + 7” induction. On another hand, anthracyclines have been used extensively with standard dose of cytarabine to induce remission of patients with AML worldwide (48, 50). However, clinicians in China tend to attach great importance to homoharringtonine (HHT) and apply HHT-based induction regimens to induce the CR of patients with AML (not APL), which was considered as another discrepancy between Asian countries and Caucasian countries in the treatments of AML (48, 50). Concretely, clinical hematologists and oncologists in China are apt to replace anthracyclines with HHT in CR induction of AML or added HHT upon prime chemotherapy regimens for higher CR rate of AML. HHT is a kind of alkaloid deriving from genus Cephalotaxus and exerts selective antileukemia effects on patients with AML, especially on these carrying FLT3-ITD and elderly patients with AML (51–53). What is noteworthy is that the FLT3-TKD status associates with chemotherapy regimens, especially in relapse (2), whereas some salvage therapies for relapsed AML include HHT as the basic members of chemotherapy regimens, such as HAA regimen (54). Hence, we speculated that prognosis of AML in different ethnicities may also due to the clinical usage of HHT partially. Third, race diversity is the result of different genetic backgrounds, not only the gene encoding FLT3. Patients with AML usually were with genomic anomaly or sporadic gene mutations (42). Currently, the contribution of different genetic backgrounds to the occurrence and progression of AML remains unclear. In conclusion, FLT3-TKD exerts impacts on contrasting prognosis of AML in different ethnicities due to multiple reasons, which deserve further explorations.
FLT3-TKD is considered as a potentially metabolically related mutation in AML. FLT3-TKD associates with cytoplasmic Src family tyrosine kinases by increasing the phosphorylation of activating tyrosines, such as FGR and HCK (7). In addition, Src family members are involve in multiple nutrients metabolisms, including glucose (8), lipid (9), and glutamine (10), indicating that FLT3-TKD is a potentially metabolically related mutation. As for glucose metabolism, Src is able to regulate cyclin B1, N-cadherin, and E-cadherin under high glucose as a response to hyperglycemia in colorectal cancer (8). For glutamine metabolism, epidermal growth factor receptor (EGFR)-promoted tumor progression is considered as being Src signaling pathway related by influencing glutamine metabolism in multiple malignancies (10). For lipid metabolism, Src is able to being recruited by CDCP1 into lipid rafts, which affect HGF receptor Met via the activation of STAT3 (9). Overall, FLT3-TKD is considered as being metabolically related based on the functions of Src family tyrosine kinases, which can be used to explain FLT3-TKD status in AML.
Our meta-analysis has several strengths. First, we selected 20 eligible prospective cohort studies, which was considered as an ideal epidemiological method to predict prognosis of AML. In addition, selection bias was well controlled by two independent investigator and an unlimited literature search. Furthermore, most included studies were of high quality with regard to quality assessment of the NOS scale (16). Moreover, no evident publication bias was identified by either Begg’s funnel plot or Egger’s test. Finally, we conducted sensitivity analysis by deleting a single study in every model, and we did not find obvious abnormal studies contributing to the pooled HR.
Begg’s funnel plot was used to detect publication bias in this meta-analysis (
Figure 5
): As is shown in
Figure 5
, publication bias is not evident when DFS was regarded as an evaluated end point (
Figure 5A
); however, it indicated a publication bias when OS was regarded as an evaluated end point (
Figure 5B
). Indeed, OS reflects the multiple influences to individuals, not just AML. The OS of AML is able to attribute to multiple factors, which make the manuscript easy to publish. Moreover, OS can be concluded according to the status of the patients with AML (live/dead), whereas DFS is usually diagnosed on the basis of medical examination, such as bone marrow biopsy. The difference contributed to the difficulty of obtaining data from AML, which also may lead to the publication bias. All in all, DFS of AML is considered as a better evaluated index in the prognosis AML, and DFS of AML in this meta-analysis did not indicate a publication bias.
There are several limitations of this study that should be acknowledged. First, there was a heterogeneity in Caucasians (I2 = 72.9%, p = 0.000), which may attribute to the fact that Caucasians outnumbered Asians: 635 Asians from four different countries and 10,335 Caucasians from eight different countries in this meta-analysis. In our opinion, the heterogeneity from Caucasian group was from some unknown confounding factors in the united prospective cohorts of Caucasians. In this situation, if related confounding factors were well controlled, then the conclusion might be more convinced. In addition, we failed to get the specific information about chemotherapy from 10,970 participants including chemotherapy regimens, chemotherapy dose, and chemotherapy time, because different types of treatment may exert distinct impacts on the prognosis of the patients with AML. However, FLT3-TKD does not result from general chemotherapies, so the FLT3-TKD in AML in this study is regarded as being from individuals’ genetic backgrounds. Furthermore, because of the limitation from the extracted data, we were unable to perform more stratification analysis according to other confounding factors. Although confounding factors work on the ending events, it is considered as a common problem in clinical research studies because we cannot predict everything before the start.
In conclusion, our results showed that FLT3-TKD revealed no significant effect on DFS and OS of patients with AML. However, meta-regressions demonstrated that patient source associated with the prognosis effect of FLT3-TKD in patients with AML. To be specific, FLT3-TKD represented a beneficial prognosis of DFS and OS for Asians with AML, whereas FLT3-TKD represented an adverse prognosis of DFS for Caucasians with AML. However, the results of DFS from Caucasians ought to be interpreted with caution due to the heterogeneity. This meta-analysis provided new information about the distinct prognosis of patients with AML between Asians and Caucasians. From our perspectives, the Caucasians with FLT3-TKD at the initial diagnostic stage of AML could be recommended the Asians dose of cytarabine and daunorubicin (cytarabine of 100 mg/m2 + daunorubicin of 50 mg/m2) in conventional “3 + 7” induction, so that they could receive a better prognosis of AML for survivals. An adequately designed prospective study including a large population with AML with clear FLT3 gene statue is needed to confirm our results.