Peripheral blood cells chimerism after unrelated cord blood transplantation in children: kinetics, predictive factors and impact on post-transplant outcome

Authors


Summary

This study aimed to describe kinetics of complete donor chimerism occurrence (cDC, >99·9% donor) after unrelated cord blood transplantation (UCBT), to identify its predictive factors and its impact on post-transplant outcome. Ninety-four children who received single UCBT after a myeloablative conditioning regimen had blood chimerism evaluation at predefined post-transplant dates, using a real-time polymerase chain reaction method with 0·1% sensitivity. Cumulative incidence of cDC at 1 year post-transplantation was 61·8%. Three predictive factors were identified in multivariate analysis: history of malignant disease (P = 0·03), older age (above 2·16 years, the first quartile of age, P = 0·0055) and higher level of cord/recipient human leucocyte antigen mismatch (4/6 vs. 5-6/6, < 0·001) increased the probability of post-transplant cDC. Although graft cell dose had a strong impact on haematological recovery, it did not apparently influence cDC occurrence. Early cDC (i.e. more than 99·9% donor chimerism on days 15–30 post-transplant) appeared useful to predict engraftment (P = 0·003) as well as acute and chronic graft-versus-host disease (GvHD). Severe acute or chronic GvHD never occurred in patients with DC ≤99·9%, suggesting than even minimal residual host haematopoiesis is associated with a very low risk of GvHD after UCBT.

Evaluation of peripheral blood cell chimerism is an important component of patient follow-up after allogeneic haematopoietic stem cell transplantation (HSCT). Real-time quantitative polymerase chain reaction (qPCR) currently allows post-transplant detection of donor and/or host genomic-specific markers in the host nucleated blood cells with a 0·1% sensitivity (Alizadeh et al, 2002). In the context of bone marrow or peripheral blood transplantation, many studies have highlighted that the level of donor chimerism influences the occurrence of graft-versus-host disease (GvHD) and, in the case of patients with malignant diseases, the risk of disease relapse (Bader et al, 2004a; Lamba et al, 2004; Saito et al, 2008; Svenberg et al, 2009; Rupa-Matysek et al, 2011). During the two last decades, the use of unrelated umbilical cord blood (UCB) as an alternative source of HSCT has increased substantially, especially for children (Rocha et al, 2001; Gluckman et al, 2004; Eapen et al, 2007; Barker et al, 2011). However, data on chimerism after UCB transplantation (UCBT) are very scarce (Prasad et al, 2008; Moscardó et al, 2009; Berglund et al, 2012) and the published studies did not uniformly use qPCR.

We present a cohort of 94 children who received UCBT after a myeloablative conditioning (MAC) regimen and had longitudinal chimerism follow-up at predefined post-transplant timepoints, using qPCR. This enabled us to describe kinetics of complete donor chimerism (cDC) occurrence after UCBT, its predictive factors and its impact on haematological recovery, GvHD, transplant-related mortality, relapse and overall survival.

Patients and methods

Criteria for patient selection

This study was prospectively conducted between 2001 and 2010 in a single French paediatric transplant centre. All children under 18 years of age who received a first single UCBT after MAC for either a non-malignant disorder or a malignant disease in complete remission were eligible, except those transplanted for severe combined immunodeficiency. Ninety-four children met the inclusion criteria and all were included in the chimerism study. Informed consent was obtained from parents/guardians.

Blood chimerism

Definition

The primary endpoint of our study was the occurrence of a cDC over time: cDC was reached when more than 99·9% of peripheral blood nucleated cells were of donor origin. Additionally, a secondary endpoint was defined that also included near-complete donor chimerism (n-cDC). This secondary endpoint was achieved when more than 99% nucleated blood cells were of donor origin. Blood chimerism was tested on peripheral blood nucleated cells without any attempt to separately analyse specific sorted cell lines.

Quantification

Thirty-one bi-allelic short deletion/insertion sequence polymorphisms were used as specific genetic markers for initial primer screening and quantitative chimerism, with a detection limit of 0·1% marker-specific cells (Alizadeh et al, 2002). Once informative assays were identified in the screening test, one or more of these assays were selected to quantify the percentage of DNA representative of a single genome of interest in a chimeric DNA sample, using an established relative quantification method for qPCR (Livak & Schmittgen, 2001). All samples were analysed in duplicate using an ABI 7500 Real Time PCR system (Applied Biosystems, Foster City, CA, USA).

Evaluation time

For each patient, we evaluated post-transplant blood chimerism at predefined dates: 15, 30, 60, 90, 120 and 180 d, then 1, 1·5, 2, 2·5 and 3 years. The cumulative incidence of donor chimerism occurrence over time was then calculated. Children without any blood reconstitution at a given date were considered to have a 0% donor chimerism. In addition, we defined the concept of early chimerism as the highest donor chimerism during the first month following the transplant.

Cord blood transplant selection

Human leucocyte antigen (HLA) compatibility between recipient and cord blood transplant units was established on 6 HLA antigens/alleles (HLA-A, -B, -DR). Cord blood units were selected following HLA matching at antigen level (low or intermediate resolution) for HLA-A and -B and allele level matching for HLA-DRB1.

A cord blood unit was considered adequate for transplantation if it was 4-6/6 HLA identical with the patient and contained more than 3 × 107 nucleated cells/kg recipient body weight. After thawing, the infused transplanted cell doses were evaluated by the total nucleated cell count as well as by CD34+ and CD3+ cells counts.

Statistical methods

Occurrence of donor chimerism was analysed through the use of cumulative incidence curves for estimating incidence over time (Gooley et al, 1999). We considered as competing events either relapse for malignant disease or death for malignant and non-malignant diseases. Fine and Gray model was used to assess prognostic factors (Fine, 2001). Covariates tested in the multivariate analysis as potential prognostic factors were the following: age at transplant (first included as a continuous variable and then grouped into quartiles), malignant versus non-malignant disease, total body irradiation (TBI, yes versus no), HLA compatibility between donor and recipient (4/6 vs. 5-6/6) and graft cell dose per kg of recipient body weight. Infused graft cell dose was described by three variables: total nucleated cell count, CD34+ and CD3+ cell dose. Therefore the influence of transplant cell dose was investigated by performing three different multivariate analyses, each of these analyses using one of the variables described above. We present the multivariate results with their levels of statistical significance (P), their hazard ratio (HR), and the 95% confidence interval of their hazard ratio (95% CI). P values were considered significant when <0·05. Other post-transplant outcomes evaluated in this study were graft failure, autologous recovery (more than 80% blood cells from recipient origin), GvHD, relapse risk, transplant-related mortality (TRM), overall survival, neutrophil recovery and platelet recovery.

Results

Patient characteristics and main post-transplant outcomes

Patient characteristics and main post transplant outcomes are detailed in Table 1. Median age of the 94 children was 6·3 years (interquartile range [IQR]: 2·16–9·85). Sixty-nine had a malignant disease in complete remission and 25 a non-malignant disease. Among 69 with malignant disease, 43 had acute lymphoblastic leukaemia, 16 acute myeloid leukaemia, eight lymphoma and two a juvenile myelomonocytic leukaemia. The 25 non-malignant diseases were 11 metabolic disorders, seven acquired aplastic anaemias, six primary immune deficiencies and one osteopetrosis. The HLA donor/recipient compatibility was 4/6 for 42·6%, 5/6 for 44·7%, and 6/6 for 12·8% of the patients. Forty-nine children received TBI during their conditioning as opposed to 45 who did not. Median infused cell dose was 5·78 × 107/kg for total nucleated cells and 1·7 × 105/kg for CD34+ cells. GvHD prophylaxis of all patients consisted of ciclosporin, steroid and antithymocyte globulin (ATG), 2·5 mg/kg per day × 3 d pre-transplant (rabbit ATG, Thymoglobuline®, Genzyme, France).

Table 1. Patient and transplant characteristics
  1. UCBT, umbilical cord blood transplantation; SEM, standard error of the mean; IQR, interquartile range; CMV, cytomegalovirus; TBI, total body irradiation; HLA, human leucocyte antigen; aGvHD, acute graft-versus-host disease; cGvHD, chronic graft-versus-host disease; TRM, transplant-related mortality.

  2. a

    86 patients alive at Day 100 were assessable.

  3. b

    69 patients with malignant disease were assessable.

Patients characteristics (n = 94)
Gender (Male/Female)53/41
Age at UCBT (years)
Mean ± SEM6·64 ± 0·48
Median (IQR)6·30 (2·16–9·85)
Diagnosis
Malignant/Non-malignant disease69/25
Recipient CMV serology: ±39/55
Transplantation procedure (n = 94)
Conditioning regimen
TBI versus no TBI49/45
Graft/Recipient HLA compatibility, %
0 mismatch (6/6)12 (12·8)
1 mismatch (5/6)42 (44·7)
2 mismatch (4/6)40 (42·6)
Collected total nucleated cell dose (×107/kg)
Mean ± SEM9·49 ± 0·75
Median (IQR)7·29 (5·11–11·21)
Infused total nucleated cell dose (×107/kg)
Mean ± SEM7·39 ± 0·58
Median (IQR)5·78 (3·99–8·95)
Infused CD34+ cell dose (×105/kg)
Mean ± SEM2·4 ± 0·2
Median (IQR)1·7 (1·0–2·9)
Infused CD45+ cell dose (×107/kg)
Mean ± SEM4·20 ± 0·34
Median (IQR)3·36 (2·29–4·80)
Infused CD3+ cell dose (×106/kg)
Mean ± SEM11·33 ± 1·07
Median (IQR)8·22 (6·08–13·10)
Main post-transplant outcomes, %
aGvHD grade ≥224/94 (25·5)
aGvHD grade ≥38/94 (8·5)
cGvHDa10/86 (11·6)
Graft failure/Autologous recovery11/94 (11·7)
Relapse (malignant disease)b19/69 (27·5)
TRM, %
Day 100 cumulative incidence5 ± 2
1-year cumulative incidence13 ± 4
5-year cumulative incidence15 ± 4
Overall survival, %
5-year overall survival68 ± 5

Relapse risk, TRM at Day 100 and overall 5 years survival were 27·5%, 5% and 68%, respectively. Graft failure or autologous recovery was observed in 11 patients (11·7%). In 24 patients (25·5%), the evolution was complicated by an acute GvHD ≥2 and for eight of them by a severe acute GvHD (≥3). Among the 86 children that could be evaluated at Day 100, 10 (11·6%) developed a chronic GvHD.

Cumulative incidence of donor chimerism occurrence, overall description

Figure 1 shows the occurrence probabilities of cDC (>99·9% donor chimerism) and n-cDC (>99% donor chimerism) over time, with relapse and TRM considered as the two potential competing risks. The cumulative incidence 1 year after transplant was 61·8% (±5·2%) for cDC, and 81·9% (±4·5%) for n-cDC.

Figure 1.

Cumulative incidence of complete donor chimerism occurrence over time. Occurrence of >99·9% donor blood chimerism. Occurrence of >99% donor blood chimerism.

Predictive factors for donor chimerism

Table 2 presents the results of the multivariate analysis. The significant predictive factors that increase the probability of post-transplant cDC (>99·9% donor chimerism) were a history of malignant disease (P = 0·03), older age at transplant (P < 0·003) and a higher level of HLA mismatch between graft and recipient (P ≥ 0·001). As shown in Fig 2A–C, cumulative incidences of cDC at 1 year post-transplant were 72·5% for malignant disease versus 31·2% for non malignant disease (Fig 2). It was 23·6% for children <2·16 years old (first quartile of age) compared to 67·9%, 71·4% and 83·6% for the subsequent age quartiles and 83·6% for 4/6 HLA compatibility compared to 48·2% for 5/6 or 6/6 HLA compatibility. By contrast, TBI and infused CD34+ cells dose did not significantly influence the probability of post-transplant cDC in the multivariate analysis.

Table 2. Multivariate analysis of predictive factors for complete donor chimerism
Predictive factorEndpoint
More than 99·9% donor chimerismMore than 99% donor chimerism
HR (95%CI)P-valueHR (95%CI)P-value
  1. HR, hazard ratio. 95%CI, 95% confidence interval; UCBT, umbilical cord blood transplantation; TBI, total body irradiation; HLA, human leucocyte antigen.

  2. a

    Reference group for odds ratio estimation.

  3. b

    Odds ratios are given for each additional 1 × 105 CD34+ cell dose.

Diagnosis
Malignant versus non malignant diseasea2·41 (1·11–5·26)0·032·94 (1·67–5·17)<0·001
Age at UCBT
First quartilea (<2·16 years) versus older4·29 (1·64–11·21)<0·0032·34 (1·30–4·21)<0·005
TBI versus no TBIa1·20 (0·69–2·07)0·520·73 (0·49–1·09)0·12
Graft/recipient HLA compatibility
4/6 versus 5-6/6 HLA identitiesa2·18 (1·38–3·44)<0·0011·47 (1·04–2·07)0·03
Infused CD34+ cell doseb1·09 (0·97–1·24)0·151·05 (0·97–1·14)0·24
Figure 2.

Predictive factor for occurrence of >99·9% donor chimerism. (A) Influence of initial disease (malignant versus non malignant disease). (B) Influence of recipient/cord blood human leucocyte antigen (HLA) compatibility (4/6 vs. 5-6/6 HLA identities). (C) Influence of age at umbilical cord blood transplantation (quartile distribution).

The same factors also proved significant in explaining the occurrence of a > 99% donor chimerism. Replacing the CD34+ cell dose in the multivariate analysis by other cell dose evaluations, such as total infused nucleated cell count and CD3+ cell dose per kg of recipient body weight did not substantially change the results (see Table SIa, b). In these models, the transplanted cell dose was not a predictive factor and the same significant factors were found except that the influence of diagnosis on cDC and the influence of age at UCBT on a > 99% donor chimerism did not reach the level of statistical significance.

Predictive factors for haematologic recovery

We conducted the same multivariate analysis in order to identify the predictive factors for haematopoietic recovery after transplant (neutrophil count >0·5 × 109/l and platelet count >50 × 109/l). Competing risks for haematological recovery were again TRM and relapse. The results are detailed in Table 3. Predictive factors for haematological recovery were very different from those described above for donor chimerism. As expected, the most significant factor for haematological recovery was the infused CD34+ cell, total nucleated cell and CD3+ doses per kg of recipient body weight (see also Table SIIa, b).

Table 3. Multivariate analysis of predictive factors for haematopoietic recovery
Predictive factorEndpoint
Neutrophil count >0·5 × 109/lPlatelet count >50 × 109/l
HR (95%CI)P-valueHR (95%CI)P-value
  1. HR, hazard ratio. 95%CI, 95% confidence interval; UCBT, umbilical cord blood transplantation; TBI, total body irradiation; HLA, human leucocyte antigen.

  2. a

    Reference group for odds ratio estimation.

  3. b

    Odds ratios are given for each additional 1 × 105 CD34+ cell dose.

Diagnosis
Malignant versus non malignant diseasea1·96 (1·01–3·80)0·051·25 (0·58–2·68)0·57
Age at UCBT
First quartilea (<2·16 years) versus older0·74 (0·37–1·47)0·390·91 (0·42–1·98)0·81
TBI versus no TBIa0·70 (0·40–1·21)0·201·16 (0·63–2·12)0·64
Graft/recipient HLA compatibility
4/6 versus 5-6/6 HLA identitiesa0·73 (0·50–1·06)0·100·66 (0·41–1·09)0·10
Infused CD34+ cell doseb1·13 (1·02–1·25)0·021·12 (1–1·26)0·04

Influence of complete donor chimerism (>99·9%) on the main clinical outcome after UCBT

We evaluated the association between cDC (>99·9% donor chimerism) and several post-transplant outcomes (Table 4). Patients who achieved cDC, any time after transplant, had a significantly higher risk of acute and chronic GvHD. On the other hand, none of the patients who never achieved post-transplant cDC developed severe acute GvHD or chronic GvHD. Achieving cDC during the first month post-UCBT was also a strong predictive criteria for acute and chronic GvHD. Among 57 children who had 99·9% or less donor chimerism during the first post-transplant month, only one experienced further grade 3–4 acute GvHD. Additionally, among children who achieved post-transplant cDC during the first post-UCBT month, none experienced graft failure or autologous reconstitution later on. At last, we were unable to detect any significant relationship between cDC and relapse, TRM or overall survival.

Table 4. Influence of complete donor chimerism (>99·9% donor) on the main clinical outcome after UCBT
 More than 99·9% donor during first month post-UCBT
 Yes (total number = 37) (%)No (total number = 57) (%) P
Graft failure/autologous reconstitution0 (0)11 (19·3)0·003
aGvHD grade ≥217 (45·9)7 (12·3)<0·001
aGvHD grade ≥37 (18·9)1 (1·8)0·006
cGvHDa7 (20·6)3 (5·8)0·05
Relapseb7 (20·6)12 (34·3)0·20
TRM7 (18·9)7 (12·3)0·38
Overall mortality12 (32·4)18 (31·6)0·93
 More than 99·9% donor, any time after UCBT
 Yes (total number = 58) (%)No (total number = 36) (%) P
  1. UCBT, umbilical cord blood transplantation; aGvHD, acute graft-versus-host disease; cGvHD, chronic graft-versus-host disease; TRM, transplant-related mortality.

  2. a

    Percentages are calculated in 86 patients alive at day 100.

  3. b

    Percentages are calculated in 69 patients with malignant disease.

Graft failure/autologous reconstitution0 (0)11 (30·6)<0·001
aGvHD grade ≥222 (37·9)2 (5·6)<0·001
aGvHD grade ≥38 (13·8)0 (0)0·02
cGvHDa10 (18·2)0 (0)0·01
Relapseb13 (26)6 (31·6)0·64
TRM9 (15·5)5 (13·9)0·83
Overall mortality19 (32·8)11 (30·6)0·82

Influence of first month chimerism on haematologic recovery

Finally, we assessed whether there was a significant relationship between early chimerism (i.e. the higher donor chimerism observed during the first month post-UCBT) and haematological recovery. Figure 3A shows the cumulative incidence of neutrophil recovery (neutrophil count >0·5 × 109/l) depending on whether or not an early cDC chimerism occurred. Figure 3B displays the same feature according to occurrence of early >99% donor chimerism. We found that early cDC was favourably correlated to neutrophil recovery. However, no significant relationship was found between early chimerism and platelet recovery.

Figure 3.

Influence of first post-transplant month chimerism on haematopoetic recovery. (A) cumulative incidence of neutrophil recovery (>0·5 × 109/l) in patients who did or did not achieve more than 99·9% donor chimerism during the first month after transplant. (B) cumulative incidence of neutrophil recovery (>0·5 × 109/l) in patients who did or did not achieve more than 99% donor chimerism during the first month after transplant.

Discussion

The aim of this study was to describe the kinetics of cDC occurrence after UCBT, and identify its predictive factors and impact on major post-transplant outcomes. For this purpose, we constructed cumulative curves for estimating cDC occurrence over time and multivariate models of potential predictive factors. Three predictive factors for cDC were identified in this study: a history of malignant disease, older age at transplant and a higher level of HLA mismatch between graft and recipient.

Children with a non-malignant disease achieve cDC after UCBT less frequently than children with malignant disease. This feature has been already described after bone marrow or peripheral blood stem cell transplantation (Svenberg et al, 2009; Park et al, 2011) and can be partly explained by the lack of prior chemotherapy in non-malignant diseases. Indeed, in the context of malignant diseases, several studies have suggested that the amount of chemotherapy given before HSCT influences post-transplant donor chimerism (Mohty et al, 2007; Saito et al, 2008; Berglund et al, 2012). Berglund et al (2012) also suggested that a history of malignant disease increased the probability of cDC in their study of 50 UCBT, although this did not prove statistically significant in their multivariate analysis. In addition to the absence of prior chemotherapy, other factors such as young age and conditioning regimen without TBI may also explain the low probability of cDC in patients with non malignant disease (Stikvoort et al, 2013). In the largest study of transplantation for Hurler syndrome, median age at transplant was 16·7 months and near all children received a Busulfan-based myeloablative conditioning regimen (Boelens et al, 2013). In the 10/10 HLA matched sibling and unrelated donor setting, mixed chimerism (i.e. <95% donor) was seen in 30–50% of these young children. Interestingly, in the same study, mixed chimerism was less frequent after UCBT and although the impact of HLA-mismatch level on chimerism was not reported, it is noteworthy that more than 80% of UCBTs were <6/6 HLA identical to recipient.

In our study, the occurrence of cDC was more frequent after 4/6 than after 5/6 or 6/6 HLA-matched UCBT and was not apparently influenced by the transplanted cell dose. One could underline that this is quite inconsistent with a large burden of scientific evidence demonstrating that better cord/recipient HLA compatibility and higher transplanted cell dose improve haematological recovery (Locatelli et al, 1999; Wagner et al, 2002; Gluckman et al, 2004; Eapen et al, 2014). Indeed, our results (Table 3) also demonstrated that the transplanted cell dose significantly improves neutrophil and platelet recovery. These findings suggest that haematological recovery and cDC occurrence may be two different concepts with different sets of influencing factors. These two concepts are not independent as we found a significant relationship between early chimerism and neutrophil recovery (and not platelet recovery). However, haematological recovery may depend on the capacity of transplanted cells to engraft in a given host whereas cDC reflects the allogeneic reactivity of the transplanted immune system. In this context, a higher level of HLA mismatch may enhance alloreactivity against the host residual haematopoietic cells and may result in a more complete donor chimerism. In their study of chimerism after bone marrow or peripheral blood transplantation for non-malignant diseases, Stikvoort et al (2013) found that stable mixed chimerism was more frequent when the donor was a sibling compared to unrelated volunteer. They argued that, because sibling donor cells are a closer HLA and minor histocompatibility antigen match than unrelated donor cells, they are less prone to induce immune-mediated killing of recipient cells. Another concordant argument is provided by a study of chimerism after double UCBT (Verneris et al, 2012). After double UCBT, one unit predominates by Day 100 in around 90% of patients, due to graft-versus-graft interactions but, on rare occasions, recipients can live with “dual chimerism”, i.e. persistence of both units with a minimum haematopoietic contribution of 5% by each unit. Recipients of two HLA 6/6 matched units (with each other and the recipient) were more likely to have dual chimerism, less aGvHD and more relapses, suggesting less alloreactivity of these well matched units (Verneris et al, 2012).

A strong association between donor chimerism after bone marrow or peripheral blood allogeneic transplantation and the risk of relapse has been well documented. Children with leukaemia who show increasing mixed chimerism are at increasing risk of relapse (Bader et al, 1997, 1998, 2004b; Barrios et al, 2003). Both the present study and another post-UCBT study (Berglund et al, 2012) were unable to detect any association between donor chimerism and the relapse risk. This does not mean that UCBT differs from other stem cell sources in this setting. It could just reflect an insufficient power of our study, which included malignant and non-malignant diseases. Although analysis of the potential relationship between chimerism and relapse was restricted to children with haematological malignant diseases, this analysis was done in only 69 children who had different diseases and various haematological status at time of transplant. Moreover, cDC was defined by reaching >99·9% donor but not necessarily by sustaining this high level of donor engraftment. Much of the literature on chimerism as a predictor of relapse focuses on either failure to attain cDC or declining chimerism as risk factors. Consequently, we have to consider that description of the relationship between chimerism and relapse is not a reasonable objective of the study presented here.

On the contrary, we found a clear correlation between cDC and the risk of GvHD. This relationship has been extensively described after bone marrow or peripheral blood transplant (Mohty et al, 2007; Svenberg et al, 2009; Rupa-Matysek et al, 2011; Nikolousis et al, 2013; Stikvoort et al, 2013). We demonstrated here that this correlation is particularly strong after UCBT. In our study, none of the children who never achieved cDC (including those in the 99–99·9% range) developed severe acute GvHD or chronic GvHD. Given that the sensitivity threshold for detection of residual host cell was 0·1%, we suggest that even the presence of very few host cells can reflect a low risk of severe GvHD. In this study we used two thresholds to define complete (>99·9%) and near-complete (>99%) donor chimerism. The negative predictive value was the same for both definitions but the positive predictive value was clearly better for >99·9% (Table 4, Table SIII). Moreover, we show that very few patients who had 99·9% or less donor chimerism during the first month after UCBT further developed severe GvHD. Thus, we assume that the concept of early post-transplant chimerism, i.e. the highest donor chimerism during the first month after UCBT, could be a useful tool for predicting the occurrence of severe GvHD thereafter.

In conclusion, three predictive factors for cDC were identified in this study: a history of malignant disease, older age at transplant and a higher level of HLA mismatch between graft and recipient. Although graft cell dose had a strong impact on haematological recovery, it did not apparently influence cDC occurrence. Moreover, early cDC appeared useful to predict engraftment as well as acute and chronic GvHD.

Authorship contributions

Elodie Elkaim and Gerard Michel designed the study, reviewed all medical records, participated in the statistical analysis and wrote the manuscript. Anderson Loundou performed statistical analysis. Christophe Picard contributed substantially to the acquisition of data and participated in the writing of the manuscript. Claire Galambrun, Vincent Barlogis, Catherine Curtillet, Claire Oudin, Isabelle Thuret, Hervé Chambost recruited patients and revised the manuscript.

Ancillary