Impact of age on the cumulative risk of transformation in patients with chronic myelomonocytic leukaemia

Abstract In older patients with chronic myelomonocytic leukaemia (CMML) and limited life expectancy due to age and or comorbidities, it is particularly important to consider the risk of transformation for individualised treatment decisions. There is limited information on potential differences between younger and older CMML patients regarding the cumulative risk of transformation as well as haematological, molecular and biologic characteristics. We analysed data from the Austrian Biodatabase for CMML (ABCMML) to compare these parameters in 518 CMML patients. Categorisation of patients into 3 age‐related groups: <60 years, 60‐79 years and ≥80 years, showed a significantly lower risk of transformation at higher age by competing risk analysis, with a 4‐year risk of 39%, 23% and 13%, respectively (P < .0001). The lower probability of transformation was associated with a lower percentage of blast cells in the peripheral blood (PB) of older patients. Furthermore, we provide a simple score based on age, PB blasts and platelet counts that allowed us to define subgroups of CMML patients with a different cumulative transformation risk, including a low‐risk group with a transformation risk of only 5%. Our findings may facilitate reasonable treatment decisions in elderly patients with CMML.


| INTRODUC TI ON
Chronic myelomonocytic leukaemia (CMML) is a hematopoietic malignancy with a male predominance and a median age at diagnosis of 70 ranging from 16 to 93 years. 1 This clonal disease is characterised by overlapping features of myelodysplastic syndromes (MDS) and myeloproliferative neoplasms (MPN), and an inherent risk of transformation to secondary acute myeloid leukaemia (sAML). [2][3][4][5][6][7] The median overall survival of CMML patients is about 30 months.
Allogeneic hematopoietic stem cell transplantation, which is the only curative therapy, is rarely feasible because of age and/or comorbidities. In patients ineligible for stem cell transplantation, intensive chemotherapy results in low response rates and short response duration. 8 Hydroxyurea is often used to control myeloproliferation in CMML. 9 The cytidine analogues azacytidine (AZA) and decitabine (5-aza-2'-deoxycytidine) have demonstrated some efficacy in delaying disease course in advanced CMML and were approved for the treatment of CMML. [10][11][12] In the "Austrian Biodatabase for Chronic Myelomonocytic Leukemia" (ABCMML), we retrospectively and prospectively collect haematological, clinical, molecular and biologic information of patients with CMML from different centres in a real-life setting to elaborate epidemiologic, prognostic and predictive information for specific subgroups and to get insights into the pathophysiology of the disease. We have recently demonstrated that regarding patient characteristics and established prognostic parameters our patient cohort is consistent with other published CMML patient series and therefore may be an adequate data source for further research. 13 In CMML patients with limited life expectancy due to old age and or comorbidities, it is particularly important to consider the aggressiveness of the disease and the risk of transformation for individualised treatment decisions. There is limited information on potential differences between younger and older CMML patients regarding risk of transformation as well as haematological, molecular and biologic characteristics. In this study, therefore, we analysed data from the ABCMML to compare these parameters in 518 CMML patients categorised into sub-groups according to different age.

| PATIENTS AND ME THODS
The ABCMML was established to retrospectively collect real-life data, including epidemiologic, haematological, biochemical, immunophenotypic, cytogenetic, molecular and biologic parameters in 606 patients with CMML from 14 different centres seen between 1988 and 2018. Details of the biodatabase regarding patient characteristics and sample numbers, data sources and methods have been published recently. 13 Clinical and laboratory parameters were obtained at time of referral to centres that, in most instances, coincided with the time of bone marrow biopsies. Data were checked for accuracy and consistency. Data curation included the extraction of discrete data elements from patient records, a check for accuracy and consistency of data and a verification that baseline

Abstract
In older patients with chronic myelomonocytic leukaemia (CMML) and limited life expectancy due to age and or comorbidities, it is particularly important to consider the risk of transformation for individualised treatment decisions. There is limited information on potential differences between younger and older CMML patients regarding the cumulative risk of transformation as well as haematological, molecular and biologic characteristics. We analysed data from the Austrian Biodatabase for CMML (ABCMML) to compare these parameters in 518 CMML patients. Categorisation of patients into 3 age-related groups: <60 years, 60-79 years and ≥80 years, showed a significantly lower risk of transformation at higher age by competing risk analysis, with a 4-year risk of 39%, 23% and 13%, respectively (P < .0001). The lower probability of transformation was associated with a lower percentage of blast cells in the peripheral blood (PB) of older patients. Furthermore, we provide a simple score based on age, PB blasts and platelet counts that allowed us to define subgroups of CMML patients with a different cumulative transformation risk, including a low-risk group with a transformation risk of only 5%. Our findings may facilitate reasonable treatment decisions in elderly patients with CMML.

K E Y W O R D S
age, blast cell count, chronic myelomonocytic leukaemia, cytogenetics, mutations, score data were reflective of CMML that was strictly defined according to WHO criteria. 6 Patients with a history of antecedent CMML and 20% or more blasts in peripheral blood (PB) and/or bone marrow (BM) were uniformly considered as CMML-derived AML. PB and/ or BM samples were taken after written informed consent was provided by patients. Internal review board approval was obtained at each institution.

| Semisolid in vitro cultures
Colony-forming unit-granulocyte-macrophage (CFU-GM) growth was assessed in semisolid cultures without growth factors as previously described in one central laboratory. 15 Cultures were plated in duplicates or triplicates, respectively, at 25-100 ×

| Molecular analysis
Molecular analyses were performed using next-generation sequencing (NGS) with amplicon-based target enrichment. Genomic DNA was isolated from MNC fractions of PB or BM samples according to standard procedures. The gene panels, databases and software tools for variant annotation and interpretation used have been reported elsewhere. 13 Only pathogenic variants were taken and a VAF of 5% or higher was considered as positive for analysis. 16

| Statistical analysis
For description of categorical features, counts and percentages are given, and medians plus ranges for continuous variables. As a measure of correlation, Kendall's tau was used. Tests for differences by age are based on Kendall's tau.
The relation of age with the actual risk of transformation was described by penalised splines (psplines) without as well as with stratification by the Mayo risk score, WBC, cytogenetic risk and ASXL1.
The risk of transformation was analysed in a competing risk framework, 17 regarding death without transformation as competing event.
Differences were tested according to Gray. 18 The development of the proposed score was internally validated by bootstrapping each step of the construction.
Two-sided P-values < .05 were considered significant.
Confidence limits refer to a 95% level. In line with the essentially exploratory nature of the study, no adjustment for multiple testing was applied.
All analyses were conducted with the statistics software R 3.6.2, including the packages "survival" and "cmprsk". 19-21

| RE SULTS
Adequate data were obtained from 518 out of the 606 registered patients with CMML. The patient characteristics of the total cohort have been described previously. 13 Since the impact of age on the cumulative risk of transformation was a main focus of this study, we first looked for potential differences in CMML-related features between older and younger CMML patients. In Tables 1 and 2, comparisons of laboratory and molecular characteristics in patients with CMML categorised into age younger than 60 years vs age 60-79 vs 80 years or older are shown. Older patients ≥60 years had a lower number of peripheral blood blasts as compared to younger patients (see also Table S1), whereas all other parameters were neither substantially nor significantly different between the age categories.
Since for the individualised management of patients with CMML with comorbidities, it is critical to know the patient's actual risk to experience transformation into AML, the possibility of dying without transformation (because of age and comorbidities) was considered as competing risk.
For "descriptive purposes", age was divided into three categories (<60, 60-79, ≥80 years). These were chosen to best present the estimated relation between age and the actual risk of transformation into sAML ( Figure S1). As displayed in Figure 1, age exhibits an essential (log)-linear relation to the concerned risk, with a risk reduction of roughly 70% between adjacent age categories, after multivariable adjustment for potential confounders, namely the Mayo prognostic model for CMML (abbreviation used: Mayo score, which is based on the parameters haemoglobin <10 g/dL, PLT <100 G/L, absolute monocyte count >10 G/L and circulating immature monocytic cells 22 ), WBC, cytogenetic risk, ASXL1. There was a significantly declining risk to actually experience transformation with increasing age. To cross-check the division into three categories, also subdivision into four groups was analysed, but it TA B L E 1 Comparison of epidemiologic, laboratory, biologic and molecular characteristics in patients with CMML stratified by age younger than 60 years vs 60-79 years vs 80 years or older  Univariable and multivariable competing risk regression models with established single prognostic factors and one established score (Mayo risk score) were calculated. As can be seen in Figure S3 and In order to substantiate potential age effects, also the correlation of each prognostic factor with age was estimated by Kendall's tau (see Table S2). Most of these showed at most slightly lower values at older age, significant only regarding Hb, peripheral blasts and ASXL1.
Models for the actual risk to experience transformation considering age with one established prognostic factor at a time showed that higher age maintains a substantial and significant lowering impact on risk also in combination with each of the considered prognostic indicators (see Table S3), regardless of being categorised or not, while only platelets, peripheral blasts, ASXL1 and the Mayo score reached significance in combination with age. In the analysed combinations, the risk to transform to sAML was reduced by roughly 30% per ten additional years of age.
To make the previously described results clinically applicable, they were combined in a score for actual risk of transformation (ART score) based on a multivariable Fine and Gray model with age, peripheral blasts and platelets as risk indicators. ASXL1 mutation was considered as potential predictor, but it did not enhance the identification of patient subgroups with minimal actual transformation risk and was therefore not included in the score. For descriptive purposes and clinical convenience, the ART score was divided into three categories (low, intermediate and high) see Table 3. Figure 3A and the Table S5 show the estimated risk of AML transformation over time depending on the risk categories. Figure 3B and TA B L E 2 Risk ratios, 95% confidence limits and P-values for univariable competing risk regressions regarding AML transformation Figure S3A,B show the estimated risk of AML transformation over time predicted by the Mayo Score.
The ART score including age as one component is (necessarily but only) moderately correlated with age (tau = −0.45) (see Table S7 and Figure S4A,B). In the age group ≥80, about 40% are assigned to low risk, while in the age group <60 no one is categorised into "low," but 94% into "high." Example calculations using the ART score are shown in Table S4.

| Proving of the stability by the Bootstrap method
As internal validation of the proposed ART score, several steps in  Tables S8 and S9)

| D ISCUSS I ON
The management of cancer in older aged people is becoming a common challenge due to the ageing of the population. 23 There are many interconnected variables that may contribute to the complex situation in older patients suffering from haematological malignancies. CMML is a typical example for a malignancy, in which the risk of dying from progression of blood cancer (sAML) has to be viewed in relation to the risk of mortality unrelated to sAML. The primary aim of CMML treatment with HMA is to delay transformation into sAML. Since HMA treatment may be associated with significant inconvenience, particularly for elderly patients, it is clinically highly desirable to know if and when the patient will experience this complication in the future. The actual risk to experience transformation to sAML depends on both, the transformative aggressiveness of the disease and on the competing risk of dying before transformation, mainly by comorbidities and sometimes also by the consequences of severe CMMLrelated cytopenias. By competing risk regression as introduced by Fine and Gray, we analysed the impact of age and further features on the actual risk of transformation.
Age was found to have a major impact on the actual risk (also called cumulative incidence) of transformation of CMML to sAML, that is with rising age, the actual risk to transform declines. This seems to be partly, but not solely, due to the increased mortality by the comorbidities. Splitting patients into three age categories showed a significantly declining transformation risk with higher age.
In our sample, CMML patients 80 years or older have a cumulative transformation incidence at four years of only 13%. The favourable effect of older age persists, even if other clinical or molecular variables are taken into account and combining age with other features was expected to offer an even better discrimination of patients with low actual transformation risk.
Several prognostic parameters and risk scores have been reported for patients with CMML. 22,[24][25][26][27][28][29][30][31] Their main focus was to predict survival, time to transformation and to a lesser extent leukaemia-free survival. This is reasonable, as the main focus of these papers was to identify high-risk patients, who would require intensive therapy, assuming that they will be alive long enough to be at substantial risk of transformation. Although not explicitly stated, one could say that they aimed at estimating the severity of the dis- Furthermore, we cannot exclude a selection bias favouring referral of younger patients with more clinically aggressive disease by physicians in the community.
In order to arrive at a stable estimate regarding the actual transformation risk, we used peripheral blasts instead of bone marrow blasts for several reasons. The determination of peripheral blasts, not requiring a bone marrow puncture, allows repeated application of the ART score. The determination of the sum of blasts and promonocytes seems to be more reliable in PB than in BM, when different pathologists are involved in the evaluation of CMML patients, 35 which was the case in our study and would necessarily be in general clinical application of the score. Also in the multivariable analysis reported by the Mayo group 22 for their prognostic model, PB blasts but not BM blasts had a significant impact on leukaemiafree survival.
One limitation, however, was, that because of incomplete data on the exact blast cell content in bone marrow we could not conclusively check in the present study, if the bone marrow blast percentage would add prognostic power to the suggested score.
Furthermore, the proposed ART score was developed in an exploratory analysis on retrospective data. Although all crucial steps of the development were internally validated by bootstrapping, a validation study by an independent study group would be highly appreciated and required to confirm our observation. In a real-life collection such as the ABCMML, not all data which would be obtained in a systematic prospective study are available. However, real-life data have been recently recognised as an important way to get insights into the natural history of rare diseases. 36 On the other hand, the ABCMML provides information derived from molecular and from functional studies and therefore allows a more comprehensive view and deeper insight into the complex pathophysiology of this haematological malignancy. 13

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest to disclose in this study.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author, [KG], upon reasonable request.