Clinical and Cytogenetical Characteristics-Related Nomograms for Assessing Outcome of Acute Myeloid Leukemia Patients Post IA Induction


 Background: Acute myeloid leukemia(AML) is a highly heterogeneous hematological malignancy. Despite, increase in treatment options for AML over the past decade, prognosis for AML remain dismal. Numerous prognostic models have been developed for this disease, however nomograms predicting long-term survival in AML patients after induction chemotherapy(IA regimen) have not been described.Method: We constructed nomograms to predict disease-free survival(DFS) and overall survival(OS) by analyzing the cohort of patients with de novo non-M3 AML patients who underwent induction chemotherapy between June 2008 to August 2019. We utilized univariable and multivariable Cox proportional hazards regression analyses to obtain the selected variables for the nomograms. The discriminative ability and calibration were tested using C statistics, calibration plots, and Kaplan-Meier curves.Results: A total of 360 patients who underwent induction chemotherapy with IA regimen were included in the study. Of these 55% were male with a median age was 48 years. Using the univariate and multivariate analyses, the following variables were identified in the prediction of DFS: age(HR, 1.770; 95%CI, 1.160-2.702; P = 0.008), Hb(HR, 0.634; 95%CI, 0.462-0.870; P = 0.005), albumin(HR, 0.473; 95%CI, 0.363-0.615; P < 0.001) and cyto/molecular risk group(intermediate vs. favorable: HR, 1.614; 95%CI, 1.128-2.309; P = 0.001; poor vs. favorable: HR, 2.459; 95%CI, 1.645-3.676; P < 0.001) at the time of diagnosis, and allo-HSCT treatment(HR, 0.341; 95%CI, 0.234-0.497; P < 0.001). Factors which predicted OS were Hb (HR, 0.616; 95%CI, 0.438-0.866; P = 0.005), albumin (HR, 0.448; 95%CI, 0.340-0.591; P < 0.001), cyto/molecular risk group (intermediate vs. favorable: HR, 1.558; 95%CI, 1.068-2.275; P = 0.021; poor vs. favorable: HR, 2.348; 95%CI, 1.523-3.620; P < 0.001), allo-HSCT treatment (HR, 0.256; 95%CI, 0.166-0.396; P < 0.001), and age (HR, 1.528; 95%CI 0.980-2.382; P = 0.061). The discriminative ability and calibration of the nomograms revealed good predictive ability as indicated by the C statistics (0.715 for DFS and 0.731 for OS). Conclusion: Independent predictors of survival and relapse risk after IA regimen for AML can be utilized to obtain survival nomograms. These nomograms were able to predict DFS and OS while having good calibration accuracy and discriminative ability on internal validation.

Despite the presence of several clinicopathological factors used to predict prognosis, the accuracy of these prognosticative models remains a challenge. Thus, there have been various prognostication models developed recently to address this issue. Sorror et al. developed a novel AML composite model, which included cytogenetic risk, age, regimen intensity and comorbidities, to predict 1-year mortality. [11] Shouval, et al. constructed an auto-AML nomogram, which contained age and cytogenetic risk, to predict leukemia-free survival from autologous stem cell transplantation (auto-SCT). [18] Considering that these models were constructed based on different populations and outcomes, the availability of predictive models is restrictive. Therefore, we aimed to develop and validate nomograms, which include clinicopathological factors, for estimate the probability of disease-free survival(DFS) and overall survival(OS) in patients with non-M3 AML undergoing "3+7" IA regimen as induction chemotherapy.

Patient Population and Data Collection
We conducted a retrospective study, based on data collected by review of electronic medical records for 360 patients treated at the First A liated Hospital of Wenzhou Medical University between June 2008 to August 2019. All patients were newly diagnosed with de novo non-M3 AML. Only patients who received conventional "3+7" IA(IDA 8-10 mg/m2 at day 1-3 and Ara-C 100 mg/m2 at day 1-7) regimen as initial induction chemotherapy were included in the study group. Secondary AML patients with preceding hematological disorders were excluded. Patients with missing values on follow-up data were not included in the study. This study was approved by the institutional review board. No additional patient informed consent was required due to retrospective nature of this study. All data were collected and analyzed in accordance with the Declaration of Helsinki.
Variables concerning clinicopathological characteristics were reviewed, including age, sex, WBC count, hemoglobin(Hb), platelet(PLT), serum albumin, blasts at bone marrow(BM) and peripheral blood(PB), morphological immunophenotypic, cytogenetic and molecular features of myeloid blasts at the time of diagnosis before chemotherapy. Additionally, variables concerning subsequent therapeutic data were extracted, such as initial chemotherapy regimen, allogeneic hematopoietic stem cell transplantation(allo-HSCT) treatment and remission status after treatment. De novo AML was diagnosed and classi ed according to FAB [19] and ELN 2017 criteria [3].

Outcome
The outcomes of interest in this study were OS and DFS. OS was calculated from the date of diagnosis to the death or last follow-up, and DFS was calculated from the date of complete remission(CR) to relapse, death or last follow-up. The criteria of CR included low residual blast percentages (<5), morphologically normal hematopoiesis, recovery of peripheral blood cell counts to normal levels, and without extramedullary disease. [20][21][22][23] Relapse was de ned as reoccurrence of high blast percentages (≥5%) in BM unrelated to recovery from prior chemotherapy [3,24].

Statistical Analysis
Numerical variables were reported as medians with interquartile ranges (IQRs), and categorical variables were reported as absolute and relative frequencies. Numerical predictors (e.g., WBC, Hb and PLT) were categorized by receiver operating characteristic(ROC) curve, and the optimal cut-off was made by combination of Youden index and clinical use.
The associations of each variable with DFS and OS were rst assessed by univariable cox proportional hazards regression analysis for investigating the independent risk factors. All variables associated with DFS and OS at a signi cant level were subsequently analyzed by stepwise multivariate analysis. Variables with a p value <0.05 in multivariable analysis and variables with clinical importance identi ed by previously published articles [3,4,11,12,14,15], were selected to incorporate in the nomograms to predict the probability of 1-year, 2-year and 3-year DFS and OS rates.
The nomograms were then constructed based on proportionally turning each regression coe cient in multivariate analysis to a 0-to 100-point scale. The performance of nomograms were measured using the concordance index(C-index) as per Harrell et al. [25] To further assess discriminative ability of the model, the Kaplan-Meier curves of DFS and OS were plotted, strati ed by the quartile of the total points calculated from the nomograms. The calibration of nomograms were evaluated by calibration curve, with bootstrap resampling to decrease the over t bias. Statistical analyses was performed with R version 3.6.1(https://www.r-project.org/). Signi cance level was de ned as p value <0.05.

Baseline Characteristics
A total of 360 AML patients were enrolled in this study who baseline characteristics as listed in Table 1

Independent Prognostic Factors
The results of univariate and multivariate analyses of DFS and OS are summarized in Table 2 and Table  3. The univariate and multivariate Cox proportional hazards regression modeling found: age(HR, 1.770; Although the association of age and OS was not signi cant (HR, 1.528; 95%CI 0.980-2.382; P = 0.061), age was still selected as candidate for nomogram due to its clinical importance [5].

Discussion
De novo non-M3 AML patients have a high overall mortality rate, ranging from 60% to 76% at 5-year. [7][8][9][10] Accurate prognostic models for individuals are needed, not only to select appropriate treatment regimens but also to inform patients regarding their long-term outcomes (which may aid them in their goals of care decisions). Mohamed et al. developed an AML composite model, to estimate the risk of death within 1 year after initial chemotherapy. However, models to predict the outcomes over a longer period currently do not exist. In this study, we constructed 2 nomograms that predict 1-year, 2-year and 3-year DFS and OS for AML patients after initial IA regimen. Moreover, these nomograms were validated with good discriminative ability indicated by a C-index of 0.715 for DFS and 0.731 for OS. According to Figure 3, the predicted probabilities of 3-year survival by the nomograms were similar to actual 3-year survival. Taken together, these results support that nomograms proposed in our study could predict the risk of relapse and survival for non-M3 AML patients.
In addition, nomograms proposed in our study incorporated previously reported variables found to be associated with prognosis of AML, including patient-speci c and AML-speci c features. A number of studies identi ed that increased age, poor-risk cytogenetics, anemia, hypoalbuminemia and treatment without allo-HSCT were correlated with inferior survival. [5,8,15,[26][27][28][29][30] Our ndings corroborated with those results. WBC count proved as a predictor of death in newly diagnosed AML in some studies. [13,31] However, we found that WBC count had a slight correlation with DFS(p = 0.065; Table), likely due to widespread use of leukapheresis and hydroxyurea pre-induction chemotherapy in recent years.
While the cost and life expectancy of the AML varies with the baseline characteristics of the patient and available therapies [32][33][34], accurate prognostic model for patients with AML will be helpful to clinicians.
For instance, as shown in Figure 1, allo-HSCT could increase 100 points for AML prediction, and correspondingly increase DFS and OS proportionally. Therefore, individualized prediction nomograms may play a role in choosing treatment regimen in the future.
Despite this, our study has some limitations. Firstly, the sample size was small due to the single-center nature of the study. There is a need to con rm the generalizability by conducting further multicenter studies in the future. Additionally, retrospective nature of the study and use of electronic medical records for data collection are other potentially limitations. Finally, despite the internal validation by bootstrap validation, external validation remains to be seen.

Conclusions
In this study, we identi ed several independent prognostic variables, including age, Hb, albumin and cyto/molecular risk group at the time of diagnosis, and allo-HSCT treatment. Based on these associated factors, we developed nomograms to predict DFS and OS. Additionally, the nomograms displayed good calibration accuracy and discriminative ability on internal validation. Further studies are required to externally validated the nomograms for de novo non-M3 AML patients in the future which may serve as accurate prediction tools in this patient cohort.

Declarations
Ethics approval and consent to participate This study was approved by institutional review board of the First A liated Hospital of Wenzhou Medical University. Consent was waived by the institutional review board due to retrospective nature of this study, yet con dentialities of patients were protected.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests  Nomograms predicting disease-free(A) and overall survival(B) in de novo non-M3 AML patients after IA regimen chemotherapy.

Figure 2
Kaplan-Meier curves of disease-free(A) and overall survival(B) in patients with non-M3 AML According to quartiles of total points calculated by nomograms. P-values were based on the log-rank test.

Figure 3
Calibration curve comparing predicted and actual 3-year disease-free(A) and overall survival proportion(B).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.