Derivation of a Risk Prediction Model for Venous Thromboembolism in Adult Patients with Acute Myeloid Leukemia

Background: Venous thromboembolism (VTE) is a well-established complication of malignancy associated with high morbidity. Many validated scoring systems exist to predict risk of VTE in patients with solid tumors. However, there are no models that can prospectively identify patients at high risk for the development of VTE in acute myeloid leukemia (AML). The purpose of this study was to determine risk factors for the development of VTE in a large cohort of AML patients and to develop a risk stratification model to predict VTE risk in this patient population. Methods: Adult patients presenting to the University of Wisconsin-Madison with newly-diagnosed AML from 1/1/2010 through 12/31/2021 were included in this retrospective cohort study. The primary end point was VTE within one year of diagnosis. Univariate logistic regression models were used to assess the association of each variable with development of VTE. Multivariate analysis was then completed using 16 candidate variables including demographics (sex, race, age at diagnosis), comorbidities (hypertension, diabetes, coronary artery disease, asthma/COPD), body mass index (BMI), Eastern Cooperative Oncology Group performance status (ECOG PS), history of solid organ or hematologic malignancy, smoking status, indwelling venous catheter, history of VTE, and baseline laboratory values (white blood cell count, hemoglobin, platelet count). Non-normally distributed continuous candidates were log10-transformed, then stepwise selection was used on a training set to choose predictive variables to construct a risk score to predict VTE within one year of AML diagnosis. Results: 326 patients with AML were included in the analysis. The median age at diagnosis was 60 years (range 19-92 years), 55% of the study population was male, and 94% were Caucasian race. 59 patients (18%) experienced VTE within one year of diagnosis. Age at diagnosis (OR 0.98, 95% CI, 0.963-0.998, p=0.033), BMI (OR 1.058, 95% CI, 1.02-1.10, p=0.005), hypertension (OR 0.522, 95% CI, 0.29-0.96, p=0.035) and platelets (OR 3.16, 95% CI, 1.44-6.94, p = 0.004) were individually associated with VTE. In a subgroup analysis of 278 patients with next generation sequencing data available, mutations in DNA methylation genes (DNMT3A, DNMT1, IDH1, IDH2, TET2) were associated with an increased risk of VTE (OR 1.911, 95% CI 1.017-3.591, p=0.0442, Table 1). A multivariate logistic regression model identified six independent predictors of VTE including platelet count >150,000 (OR 1.64, 95% CI 0.65-4.14, p=0.2953), BMI >25 (OR 3.10, 95% CI 1.22-7.86, p=0.0173), hypertension (OR 2.17, 95% CI 1.03-4.55, p=0.0410), history of other cancer (OR 1.72, 95% CI, 0.74-4.01, p=0.2109), diabetes (OR 2.05, 95% CI 0.73-5.73, p=0.1718), and ECOG PS (OR 1.33, 95% CI, 0.66-2.65, p=0.4263). Subsequently, a risk score was constructed from the multivariable model (3*platelets>150 + 8*BMI>25 + 5*NoHTN + 3*NoOtherCancer + 4*Diabetes + 1*ECOG0). Patients were categorized into three groups: low (score<9), moderate (9score<12), and high (score) risk of developing VTE (Table 2). Rates of VTE within one year of AML diagnosis were 3.2%, 16.4%, and 23.6% in the low, moderate, and high risk groups respectively. The model had an external AUC=0.704 when applied to our test set. Conclusions: Our study demonstrates a high rate of early VTE in patients with AML at 18%, highlighting the importance of identifying those at higher risk which would allow for early screening, patient education, and consideration of thromboprophylaxis. On multivariable analysis, we found that normal or elevated platelet count >150, BMI 25, and diabetes were associated with an increased risk of early VTE. Surprisingly, higher ECOG PS, history of malignancy, and hypertension were associated with a decreased risk of early VTE in our cohort. A risk prediction model incorporating these findings was able to identify patients with low, moderate, and high risk of VTE. This model may help with clinical decision making, such as the consideration of thromboprophylaxis for patients with a moderate or high risk score. However, prospective validation of this model in a larger cohort of patients is needed. In summary, we propose a risk prediction model for the development of VTE in AML that can now be validated in order to provide optimal supportive care for this patient population.