Second hematopoietic stem cell transplantation as salvage therapy for relapsed acute myeloid leukemia/myelodysplastic syndromes after a first transplantation

Abstract

Second allogeneic hematopoietic stem-cell transplantation (HSCT2) is a therapeutic option for patients with acute myeloid leukemia (AML)/myelodysplastic syndrome (MDS) relapsing after a first transplant (HSCT1). However, patients allocated to HSCT2 may be a selected group with better prognosis and the added efficacy of HSCT2 is not well established. We retrospectively analyzed 407 consecutive patients with relapsed AML/MDS after HSCT1. Sixty-two patients had HSCT2 (15%) and 345 did not. The 2-year cumulative incidence rates of non-relapse mortality and relapse after HSCT2 were 26% (95% confidence interval [95% CI]: 17-39%) and 50% (95% CI: 39-65%), respectively. The 5-year overall survival rates were 25% (95% CI: 14-36%) and 7% (95% CI: 4-10%) in the HSCT2 and no-HSCT2 groups, respectively. Multivariate analysis identified female gender (hazard ratio [HR]=0.31, P=0.001), short remission duration after HSCT1 (HR=2.31, P=0.05), acute graft-versus-host disease after HSCT1 (HR=2.27, P=0.035), HSCT2 from a haplo-identical donor (HR=13.4, P=0.001) or matched unrelated donor (HR=4.53, P=0.007) and relapse after HSCT1 in earlier years (HR=2.46, P=0.02) as factors predicting overall survival after HSCT2. Multivariate analysis of all patients including HSCT2 as a timedependent variable identified relapse within 6 months after HSCT1 (HR=2.32, P<0.001), acute graft-versus-host disease before relapse (HR=1.47, P=0.005), myeloablative conditioning in HSCT1 (HR=0.67, P=0.011), female gender (HR=0.71, P=0.007), relapse in earlier years (HR=1.33, P=0.031) and not having HSCT2 (HR=1.66, P=0.010) as predictive of overall survival after relapse. In conclusion, HSCT2 is associated with longer survival compared to non-transplant treatments and may be the preferred approach in a subset of patients with relapsed AML/MDS after HSCT1.

Introduction

Allogeneic hematopoietic stem cell transplantation (HSCT) is an effective treatment with a curative potential for patients with acute myeloid leukemia (AML) or myelodysplastic syndromes (MDS). The outcome of HSCT has improved markedly in the last decades due to a significant reduction in the rate of non-relapse mortality after stem cell transplants.¹ However, relapsed disease remains the major cause of treatment failure.² Marked changes have been introduced in modern HSCT over the past decades, including transplants in older patients, more common use of unrelated donors as well as haplo-identical donors, and the use of peripheral blood stem cells as the source of stem cells. Novel conditioning regimens and regimens for the prevention of graft-versus-host disease (GvHD) have also been introduced. However, these changes did not change the rate of disease relapse substantially, and the prognosis for relapsed patients following HSCT remains dismal with a long-term survival rate of about 10-15%.^3-5 There is no established standard-of-care therapy for patients relapsing after HSCT. The spectrum of management includes palliative care, withdrawal of immune-suppression therapy, low-dose or intensive chemotherapy, targeted treatments, donor lymphocyte infusion, a second allogeneic transplantation (HSCT2), or a combination of these therapies.^2-5 Prolonged survival can be achieved only by patients who have a second complete remission and are supported by a form of cellular therapy such as donor lymphocyte infusion or HSCT2.^6,7 The role of HSCT2 in the treatment of relapsed AML or MDS patients still needs to be determined, including the indications and predictive factors for, and outcomes after, a second transplant in comparison with those for non-transplant treatments. Several studies have shown that the major predictive factors of outcome of HSCT2 are the duration of remission after the first HSCT and the status of disease at HSCT2.^7-17 Age, gender, choice of stem cell donor and acute and/or chronic GvHD prior to and following HSCT2 have also been described as important predictive factors.^2-17 The longterm survival after HSCT2 is estimated as 25-30%.^2-17 However, patients addressed to HSCT2 may be a selected group with a better prognosis than those not given a second transplant. In addition, survival time between relapse and HSCT2 may bias in favor of HSCT2, as early deaths are not included in the analysis of HSCT2. In the current study we describe the outcomes of patients relapsing after a first allogeneic HSCT whether they did or did not have HSCT2. We consider HSCT2 as a time-dependent variable to reduce these biases.

Methods

Study design and data collection

This is a retrospective, single-center analysis. The study included adult patients with AML or MDS who relapsed after a first allogeneic HSCT from an HLA-matched sibling or unrelated donor between the years 2000-2018. Patients given a first HSCT from haplo-identical donors were not included, because of their small number during that period and to the different biology of relapse after haploidentical transplants. Patients were divided into two subgroups according to whether they did or did not have HSCT2. Only patients who underwent HSCT2 from a different donor were considered in the HSCT2 subgroup.

All patients provided written informed consent authorizing the use of their personal information for research purposes and the study was approved by the institutional review board.

Conditioning regimens

The conditioning regimen was selected at the attending physicians’ discretion. Dose intensity was defined as myeloablative, reduced toxicity (intermediate intensity) or reduced intensity according to standard criteria.^18,19 GvHD prophylaxis consisted of cyclosporine with short-term methotrexate or mycophenolate mofetil in most cases.

Antithymocyte globulin was allowed at the attending physicians’ discretion. No ex-vivo manipulation of cells was used.

Evaluation of outcomes

Active disease was defined as no complete remission (CR) or complete remission with incomplete count recovery (CRi) in AML and >5% marrow blasts in MDS. Disease relapse and transplant engraftment were defined according to standard hematologic criteria. In the analysis of outcomes after HSCT2, non-relapse mortality was defined as death of any cause in the absence of disease recurrence. Leukemia-free survival was defined as survival without relapse. Overall survival was calculated from the day of HSCT2 until death of any cause or the date of last follow-up. In the analysis of outcomes in the entire group of patients all outcomes were calculated from the day of relapse. Acute GvHD was graded and staged by the consensus criteria.²⁰ Chronic GvHD was graded and staged according to National Institute of Health consensus criteria.²¹

Statistical analysis

This study had two parts. In the first part we analyzed the group of patients who underwent HSCT2. Overall survival and leukemia-free survival were analyzed by the Kaplan-Meier method.²² Non-relapse mortality, relapse, and acute and chronic GvHD were evaluated by cumulative incidence analysis considering competing risks.²³ Univariate analysis of predictive factors was done by log-rank tests for overall survival and leukemia-free survival and by the Gray test for the other outcomes. Multivariate analysis was done using the Cox proportional-hazard method. In the second part of the study, we evaluated the role of HSCT2 in the entire group of patients who relapsed after their first HSCT. The primary endpoint of this part was overall survival after relapse. The two treatment groups (HSCT2 and no-HSCT2) were compared by the χ² method for qualitative variables, and the Mann-Whitney test for continuous parameters. Outcomes were analyzed by the same methodologies. Multivariate analyses were performed using Cox proportional hazards with stepwise backward selection including performing HSCT2 as a time-dependent variable. We also used a landmark analysis at 60 days after relapse and included only patients alive at the landmark. Statistical analyses were performed with SPSS 24.0 (SPSS Inc., Chicago, IL, USA) and R 3.4.1 software packages (Vienna, Austria; URL https://www.R-project.org/).

Results

Patients’ characteristics

The study included 407 patients with AML (n=338) and MDS (n=69) who relapsed after a first allogeneic HSCT that was carried out during the years 2000-2018. The median year of relapse was 2012 (range, 2001-2021). The patients’ characteristics are outlined in Table 1. A total of 62 patients were given HSCT2. Among the other 345 patients, 98 patients had cellular therapy that was not considered HSCT2, including donor lymphocyte infusion (n=40), granulocyte colony-stimulating factor-mobilized donor lymphocyte infusion given after salvage chemotherapy (n=50) or a second transplant from the same donor (n=7). The patients’ median age was 56 years (range, 18-78); 49 years (range, 18-76) in the HSCT2 group and 58 years (range, 18-78) in the no-HSCT2 group (P<0.001). Forty-two percent of patients were males in the HSCT2 group and 61% in the no-HSCT2 group (P=0.006). The conditioning regimens for the first HSCT were myeloablative (n=153), reduced intensity (n=118) and reduced toxicity (n=136) with a higher rate of myeloablative conditioning and a lower rate of reduced intensity conditioning in those receiving HSCT2. The GvHD prophylaxis regimen at first HSCT was cyclosporine A/methotrexate in most patients. In the HSCT2 and no-HSCT2 groups, 44% and 37% of the patients, respectively, were in first complete remission/MDS with no blasts and 11% and 12% were in second complete remission. Data on measurable residual disease at the time of HSCT were not available as it was not routine practice to determine residual disease in AML at the beginning of the study. The median time from the first HSCT to first relapse was 4.5 months (range, 0.4-143.1 months). It was longer in the HSCT2 group (10.5 months [range, 1.3-143]) than in the no-HSCT2 group (3.8 months [range, 0.4-112]) (P<0.001). Of all relapses, 31% and 65% occurred within the first 6 months after the first HSCT in the HSCT2 and no-HSCT2 groups, respectively. A similar percent of patients had acute GvHD and/or chronic GvHD prior to relapse.

Table 1.Patients’ characteristics.

Characteristics of the second hematopoietic stem cell transplantation

The median time from relapse to HSCT2 was 4.5 months (range, 0.3-91). Eighteen patients (29%) underwent HSCT2 within 3 months of relapse. The median time from first to second HSCT was 23 months (range, 3.3-146). Seventeen patients (27%) underwent HSCT2 within less than 1 year of the first HSCT. At the time of HSCT2, 29 patients (47%) were in complete remission, and 33 patients (53%) had active disease. The donor for HSCT2 was a different donor in all second transplants; either a second HLA-matched sibling (n=13, 21%), a matched unrelated donor (n=39, 63%), or a haplo-identical donor (n=10, 16%). The conditioning regimen in the matched transplants included a sequential salvage and reduced intensity conditioning (FLAMSA-like) in six patients (treosulfan-based in 2 of these patients), fludarabine with 2 days of busulfan in seven patients, fludarabine with 4 days of busulfan in five patients, and fluradabine-treosulfan in 34 patients. Among the ten recipients of haplo-identical transplants, five were given thiotepa/busulfan/fludarabine with post-transplant cyclophosphamide and five were given T-cell-depleted transplants. In all, conditioning intensity was determined as low intensity in 29 patients and intermediate intensity in 33 patients according to the redefined European Society for Blood and Marrow Transplantation (EBMT) criteria.

Outcome afer a second hematopoietic stem cell transplantation

Fifty-seven of the 62 patients who were given HSCT2 following relapse engrafted at a median of 12 days (range, 9-32). Five patients died prior to engraftment. The median follow-up was 55 months (range, 8-188 months). The 100-day cumulative incidence of acute GvHD was 31% (95% confidence interval [95% CI]: 21-46%) and the 1-year cumulative incidence of chronic GvHD was 15% (95% CI: 8-27%). In all, at last follow-up, 17 patients were alive, and 45 patients had died. Sixteen patients died of non-relapse causes including GvHD (n=3), infection (n=4), and organ toxicity (n=9). The 2-year cumulative incidence of non-relapse mortality was 26% (95% CI: 17-39%) (Figure 1A). Forty-nine patients relapsed after HSCT2, with a 2-year cumulative incidence of relapse of 50% (95% CI: 39-65%) (Figure 1B). Four of these patients were alive after a second relapse, three of them in the long-term. The 2-year and 5-year overall survival rates were 38% (95% CI: 26-50%) and 25% (95% CI: 14-36%), respectively (Figure 1C). The 2-year and 5-year leukemia-free survival rates were 24% (95% CI: 13-34%) and 18% (95% CI: 8-29%), respectively. Seven patients were given maintenance treatment after HSCT2, including azacytidine (n=5) and sorafenib (n=2). Two of these patients are long-term survivors.

Figure 1.Figure 1. Outcomes after second allogeneic hematopoietic stem cell transplantation. (A) Non-relapse mortality. (B) Relapse. (C) Overall survival. HSCT2: second hematopoietic stem cell transplantation.

Table 2.Univariate analysis of factors predicting 2-year overall survival after second hematopoietic stem cell transplantation.

Figure 2.Overall survival afer relapse. (A) Overall survival in 345 patients not given a second hematopoietic stem cell transplant (HSCT2) compared with overall survival after HSCT2 in 62 recipients (calculated from the day of the second transplant). (B) Landmark analysis at 60 days after relapse according to whether patients did or did not have a HSCT2.

Factors predicting 2-year overall survival afer second hematopoietic stem cell transplantation

Table 2 outlines the univariate analysis of factors predicting 2-year overall survival after HSCT2. A 2-year overall survival advantage was statistically significant for females compared to males (P=0.001), AML compared to MDS (P=0.04), favorable European LeukemiaNetwork (ELN²⁴) risk (P=0.02), a prior remission duration from HSCT to first relapse that was longer than 6 months (P=0.05), and a matched sibling as a donor at HSCT2 (P=0.03) compared to an unrelated or haplo-identical donor. Interestingly, the survival of patients with active disease at HSCT2 was not statistically significantly different from that of patients in complete remission. However, six of the 33 patients with active disease at HSCT2 had not had therapy prior to HSCT. When these patients were excluded the 2-year overall survival was 14% (95% CI: 6-34) compared with 40% (95% CI: 22-59) in patients in complete remission (P=0.10). The use of treosulfan in the conditioning regimen was also associated with better outcome in the univariate analysis (P=0.002). Survival in more recent years has improved, but this difference has not reached statistical significance. Post-transplant maintenance therapy was given to a small group of patients so a meaningful evaluation of its role was not possible.

Table 3 outlines the multivariate analysis of factors predicting survival after a second HSCT. The analysis identified female gender as a factor predicting improved survival rate (HR=0.31, P=0.001). Short remission after first HSCT (HR=2.31, P=0.05), acute GvHD after first HSCT (HR=2.27, P=0.035) and HSCT2 from a haplo-identical donor (HR=13.04, P=0.001) or matched unrelated donor (HR=4.53, P=0.007) predicted lower survival rates. Earlier year of relapse after the first HSCT (in or before 2012) was associated with inferior survival (HR=2.46, P=0.012).

Table 3.Multivariate analysis of factors predicting survival after second hematopoietic stem cell transplantation.

Outcome of patients who did not have a second hematopoietic stem cell transplant

A total of 345 patients did not have second HSCT following relapse. The median follow-up for patients alive was 57 months (range, 8-223). At the last follow-up, 31 patients were alive. The median survival was 3.3 months. Figure 2A shows the Kaplan-Meier survival curves from relapse. The 2-year and 5-year overall survival rates were 13% (95% CI: 9-16%) and 7% (95% CI: 4-10%), respectively. The 5-year overall survival of patients given cellular therapy other than HSCT2 from a different donor was 14% (95% CI: 7-21) compared to 25% (95% CI: 14-36) in the HSCT2 group (P=0.05). In the comparative group of patients who had HSCT2, the survival was calculated from the time of HSCT2 in order not to overestimate the advantage of survival from relapse to HSCT2.

Factors predicting 2-year overall survival after relapse following hematopoietic stem cell transplantation

We analyzed data from 407 consecutive patients who relapsed after a first HSCT. The median follow-up of patients alive after relapse was 60 months (range, 8-222). At last follow-up, 47 patients were alive. The 2-year overall survival rate for the entire group was 18% (95% CI: 14-22). Univariate analysis of factors predicting 2-year overall survival after first relapse is presented in Table 4. Improved survival was seen in females (P=0.003), those with favorable ELN risk (P=0.008), patients in complete remission at HSCT (P=0.05), patients given myeloablative conditioning (P=0.006), those given methotrexate as GvHD prophylaxis (P=0.001), not a female donor to male recipient combination at first HSCT (P=0.04) and relapse in more recent years (after 2012, P=0.002). A significant advantage in 2-year overall survival rate was seen in patients with relapse occurring more than 6 months after HSCT (P<0.001). To reduce the bias of time to HSCT2 we analyzed only patients having HSCT2 within 6 months of relapse and found that having HSCT2 was a highly significant predictive factor in this univariate analysis (P=0.0002).

For multivariate analysis we used a Cox proportional hazard model with HSCT2 entered as a time-dependent variant. Patients who did not have HSCT2 had an inferior survival (HR=1.66, P=0.010). Other factors predicting an inferior outcome were relapse within the first 6 months after HSCT (HR=2.32, P<0.001) and acute GvHD before relapse (HR=1.47, P=0.005). Relapse in earlier years (before 2012) was associated with inferior overall survival (HR=1.33, P=0.031). Myeloablative conditioning after first HSCT and female gender predicted improved outcome (Table 5). Since only 15% had HSCT2, the performance of HSCT2 was the only post-HSCT2 factor included (as a time-dependent factor), while subgroups of second transplant characteristics could not be analyzed in this group.

To further explore the role of HSCT2 we used a landmark analysis set at 60 days after relapse. At this landmark, 65 patients were in remission, 208 patients were alive with active disease, nine patients had already undergone HSCT2 due to relapse or graft failure, 112 patients had died, and four patients were lost to follow-up. A total of 33 patients who were alive at the 2-month landmark underwent HSCT2 within the following 4 months. The 2-year and 5-year overall survival rates were 35% (95% CI: 18-51%) and 22% (95% CI: 8-32%), respectively. Among all other patients alive at the 2-month landmark, 2-year and 5-year overall survival rates were 23% (95% CI: 18-29%) and 13% (95% CI: 8-18%), respectively (P=0.03) (Figure 2B). Of the latter, 24 patients underwent HSCT2 later in their disease course (>6 months).

Discussion

A second HSCT is a potentially curative treatment for patients with relapsed AML or MDS after a first transplant. The current single-center study shows that approximately 25% of HSCT2 recipients achieve long-term survival. A similar outcome has been found in several other retrospective studies^6-17 and is better than expected with no additional cellular therapy.^3,6,25 In the current study we analyzed the results of HSCT2 in the context of all relapsing AML or MDS patients with the intent to explore the independent effect of HSCT2. Despite our policy to offer a second transplant to patients relapsing after a first HSCT, only 15% of our patients did eventually undergo HSCT2. This group included patients who were younger and with a longer time to relapse, and as such with a better expected survival. The median time from relapse to HSCT2 was 4.5 months, while 28% of all patients had already died in the first 2 months after relapse and could not have been considered for HSCT2. After adjusting for these biases by multivariate analysis and by considering HSCT2 as a time-dependent variable, HSCT2 remained an independent positive prognostic factor for survival after relapse. In addition, we also performed a landmark analysis of patients alive 2 months after first relapse and treated with HSCT2 compared with other treatments and showed a similar survival advantage for HSCT2. These analyses are based on retrospective data and may not completely adjust for unknown considerations that led the attending physicians to select patients for HSCT2. A randomized study comparing HSCT2 to other treatments may be the only way to prove the advantage of the former but is unlikely to be performed because of the high probability of physicians’ reluctance to include patients in such a study with the possibility of deferring a curative approach. In the absence of such studies the current analysis supports an independent advantage of HSCT2.

The definition of a second HSCT is not well established. HSCT2 with a different donor from the one used for the first transplant can obviously be defined as a second transplant. However, when using the same donor, peripheral blood stem cells left from the original HSCT or recollected can be used to support non-myeloablative salvage chemotherapy, with no or minimal immune suppression. This can be defined as a second transplant or, more appropriately, as donor lymphocyte infusion (mobilized donor lymphocyte infusion) or a form of cellular therapy.²⁶ To circumvent these differences in definition, we included in the subgroup of patients who underwent HSCT2 only those patients who received the second transplant from a different donor (another HLA-matched sibling, a matched unrelated donor, or a haplo-identical donor). The selection of a different donor is based on the assumption that this may provide a graft-versus-leukemia effect that was not induced by the first transplant or that may overcome resistance mechanisms against the first given immune system. However, there are no data to support HSCT2 from a different donor being more effec-tive.^2,8,10,11,13 Thus, our data may underestimate the advantage of HSCT2. In addition, other forms of cellular therapies, such as donor lymphocyte infusion or mobilized donor lymphocyte infusion can also be associated with long-term survival and there are currently no data to support HSCT2 over these other forms of cellular therapy.^{7,1 5}

Table 4.Univariate analysis of factors predicting 2-year overall survival after first relapse.

The other poor prognostic factors for survival after relapse identified by the multivariate analysis were short duration of prior remission and prior acute GvHD. Female patients and patients given myeloablative conditioning at the first HSCT had better outcomes. In line with most previous studies,^6-10,25 our study showed that relapse within 6 months after the first HSCT was consistently associated with poor outcome, also among patients who were able to proceed with HSCT2. This reflects an aggressive biology of the underlying leukemia and, in our series, overrode the role of other prognostic factors such as cytogenetics and ELN classification. The results of all therapies in these patients are dismal and such patients are often offered palliative care alone. However, a small fraction of these patients (approximately 10%) did enjoy long survival and they should not be automatically deferred from an intensive treatment approach. The better outcome of patients given myeloablative conditioning is not consistent in all studies.¹⁵ Patients initially given myeloablative conditioning may be fitter for additional treatments even after relapse. In addition, the anti-leukemia effect following reduced intensity condition is more dependent on the graft-versus-leukemia effects.²⁷ Patients relapsing after reduced intensity conditioning may, therefore, respond less to immune manipulations. This may also explain the worse prognosis of patients with acute GvHD before relapse. We previously reported better outcomes in female recipients.⁵ This was also seen in the current study but not in other large series. There is no definite biological explanation for this observation and it merits further study. The prognostic factors for poor survival in the group of patients who had a HSCT2 included male gender, short duration from first HSCT to relapse and prior acute GvHD. We also found that HSCT2 from a second unrelated donor and in particular from a haplo-identical donor was associated with inferior survival. An inferior outcome of HSCT from a second unrelated donor has been described in earlier reports.⁹ A recent EBMT report showed that haploidentical second transplants may be associated with lower survival due to increased non-relapse mortality.² However, a subsequent EBMT report found no difference between outcomes following transplants from haploidentical or unrelated HSCT2 donors.²⁸ The group of second haplo-identical transplants in the current study was too small to enable definite conclusions. In addition, about half of the transplants were T-cell-depleted while the recent EBMT studies included patients conditioned with post-transplant cyclophosphamide, which may be much safer. Due to the small numbers and in order to create a less heterogeneous group, we did not include patients with a first HSCT from a haplo-identical donor in our study. In these patients a different haplo-identical donor is usually required for a second haplo-identical HSCT to overcome the possibility of leukemia immune escape by loss of the unshared haplotype.²⁹ We did include HSCT2 from a haplo-identical donor as we wanted to explore all HSCT2 options and the potential role of the graft-versus-leukemia effect from a mismatched donor transplant after failure of a matched donor transplant. In all, it seems there is no graft-versus-leukemia advantage from haplo-identical transplantation that could justify preferential switching to a haplo-identical donor in a second transplant.

Table 5.Multivariate analysis of factors predicting survival after relapse.

The status of disease at HSCT2 has been shown in multiple studies to be an important predictive factor for outcome, with patients transplanted in remission having significantly better outcome. We did not find such an association in the current study possibly because of the small number of patients. However, some of the patients with active disease at HSCT2 had not been previously treated at the time of the HSCT2. When these patients were excluded the 2-year overall survival of this group was much lower at 14%, but still some were salvaged with HSCT2. We used treosulfan-based conditioning in the majority of HSCT2. Treosulfan has shown some advantages compared to other regimens in patients with active disease.³⁰ In the current series, treosulfan was indeed associated with a survival advantage in the HSCT2 setting in the univariate analysis, but not in the multivariate analysis. Other studies have not shown an advantage for any conditioning regimen in HSCT2.² It seems that a small subset of patients who do not achieve a stringent remission with salvage chemotherapy prior to HSCT2 can still benefit from a second transplant, but this subgroup should be defined better in larger studies.

With modern transplantation techniques, non-relapse mortality after HSCT2 is relatively acceptable. The current study, in line with other studies, has shown that the outcomes following post-transplant relapse and HSCT have improved in recent years.^4,7 However, the major obstacle to cure remains a very high incidence of relapse, with a 2-year cumulative incidence of 50% in the current series. The chances of prolonged survival after a post-HSCT2 relapse are very low.³¹ Novel maintenance therapies and immune therapies need to be explored in an attempt to improve the survival.

In conclusion, relapse of AML or MDS following HSCT is associated with a relatively poor outcome. A second HSCT can be curative in a subset of patients, in particular those with longer remission after the first HSCT.

Footnotes

• Received August 1, 2022
• Accepted November 28, 2022

Correspondence

A. Shimoni avichai.shimoni@sheba.health. gov.il

Disclosures

No conflicts of interest to disclose.

Contributions

YY and AS designed the study, analyzed the data and wrote the manuscript. NS, ID, JC, AA, RY, AN, and AS collected patients’ data. YY and AS performed the statistical analysis. All authors revised the manuscript and approved the last version.

Data-sharing statement

The data are available from the corresponding author upon reasonable request.