Development of a risk assessment model for cardiac injury in patients newly diagnosed with acute myeloid leukemia based on a multicenter, real-world analysis in China

In this study, we developed and validated a personalized predictive nomogram and risk score for the risk of cardiac injury before chemotherapy in patients with newly diagnosed AML, using cost-effective and readily available variables to help hematologists identify high-risk patients. The prediction model includes four variables: WBC, RBC, NT-proBNP, and NPM1 mutations. Internal and external validation showed that our nomogram and risk scores have good predictive performance.

In recent years, the field of cardio-oncology has made significant progress in elucidating the intricate relationship between tumors and the heart. While much attention has been devoted to studying cardiac injury resulting from anti-tumor therapy, injury caused by the tumor itself is also coming into the limelight [25]. A large Danish HF Cohort study reported that when diagnosed with cancer, their all-cause mortality rate was higher than that of without HF [26]. Xiao W et al. found that newly diagnosed AL patients did experience some cardiac-related lesions prior to chemotherapy, and this injury persisted after adjusting for confounding factors. In addition, cardiac enzyme abnormalities were more severe in the hyperleukocytic leukemia subgroup (WBC count ≥ 100 × 10⁹/L), and the proportion of blasts was positively correlated with cardiac injury [20]. In agreement with our result, a retrospective study from Harvard Medical School also found that AL was associated with cardiac alterations before chemotherapy [27], which together with other similar studies [28, 29], suggests that leukemia itself may cause direct damage to the heart prior to anti-tumor therapy.

From an etiological perspective, several factors may explain the development of cardiac injury in AL patients before chemotherapy. First, these two entities have overlapping risk factors, such as hypertension, diabetes, genetic risk factors, etc. [30]. Second, both share common systemic pathogenic pathways and mechanisms, such as abnormal production of pro-inflammatory cytokines, metabolic reprogramming in the tumor environment, clonal hematopoiesis of uncertain potential, crosstalk between stromal cells and the extracellular environment, etc. [8]. Finally, the direct infiltration of leukemic cells into the heart has been identified as another potential mechanism [11, 12, 31‐33]. Additionally, the investigation of cytokine networks in AML has gained increasing attention in recent years. Some pro-inflammatory mediators such as IL-1β, TNF-α and IL-6 tend to increase AML aggressiveness and can promote the survival and proliferation of AML cells [34]. And these can also cause heart damage [35, 36].

As previously stated, the bidirectional relationship between AML and cardiac injury status makes the diagnosis of cardiac injury potentially rather delayed, in part because symptoms of the former are attributable to the latter and vice versa [37]. Certain studies have suggested that the assessment of global longitudinal strain peak during systole using speckle tracking echocardiography may sensitively monitor early cardiotoxic alterations [38, 39]. However, this approach is limited to hospitals equipped with the necessary resources. Furthermore, cardiac biomarkers, including troponin, BNP, and cardiac enzymes [40] or other potential biomarkers such as glycogen phosphorylase BB [41], total antioxidant status, circulating microRNAs [42] and endothelial dysfunction [43] should be comprehensively assessed. The complexity and specialization involved make it easy for non-cardiac doctors to miss tests when focusing on the patient’s pre-treatment cardiac status and therefore slightly less clinically implementable.

We developed and validated a nomogram of cardiac injury risk before chemotherapy in patients with newly diagnosed AML. The model demonstrated good predictive performance. Nomograms combine several modeling algorithms to calculate continuous probabilities for specific outcomes. Among the many predictive tools currently available, nomograms have higher accuracy and better discriminative features [44]. We used LASSO regression and the multivariate logistic regression in the screening process, considering the collinearity and interaction of the screened variables. Besides, we performed a complete evaluation of the model for discrimination, clinical use and calibration, as well as an external validation of the model. Moreover, for further convenience of clinical use, we also established a risk score, and risk stratification was performed.

In clinical practice, AML patients are applied chemotherapy regimens very quickly after being definitively diagnosed, which often directly exacerbates the occurrence of adverse cardiac effects [19]. Generally, we attribute these cardiac injuries to the treatment, ignoring the heart’s injury state. Therefore, it is essential to promptly recognize the risk of cardiac injury in patients with incipient AML. In this study, we provide a prediction model that can help clinicians identify patients with incipient AML at high risk of cardiac injury and help them choose more appropriate treatment options promptly. Our nomogram model and risk score are routine clinical variables that are readily available to hematology clinicians, and as such, they can be easily applied in practice.

Our model includes four indicators: NT-proBNP, NPM1 mutations, WBC, and RBC. The level of NT-proBNP was closely related to the state of the heart [45]. AML with NPM1 mutations, although considered to have a good prognosis, results in low patient survival and high relapse rates due to its frequent coexistence with FLT3-ITD [46]. In addition, it often combines to exhibit higher leukocyte levels, percentage of BM blasts, cell invasiveness, and an incidence of extramedullary involvement [47]. AML patients with leukocytosis were prone to vascular stagnation and accumulation. In addition, the increase in metabolic byproducts in the state of blast cells will also increase the risk of cardiac injury [48]. The erythropoietic activity of AML patients was more active, especially those with BM involvement [49]. Furthermore, under certain conditions of excess oxidative stress, erythrocytes may switch the redox balance, promoting oxidative stress and being detrimental to resident and circulating adjacent cell types [50]. The latest research found that RBCs may be carriers of cytokines, which can induce significant damage to endothelial function, thereby causing vascular dysfunction [51]. This provides a new idea for cardiac injury in AML state.

Although we have developed a predictive model for the risk of cardiac injury before chemotherapy in patients with newly diagnosed AML patients, and tried to predict the risk with a minimum number of markers, this real-world study still has certain limitations. First, although hs-TNI is a more sensitive indicator of the degree of myocardial damage, it was combined with the diagnostic value of CK-MB in order to investigate this scientific question as it is not routinely examined in AML patients. Second, we used multiple imputations to replace missing values. This has the potential to lead to bias. Third, compared with other published disease prediction models, nomogram makes it difficult to deal with some nonlinear effects in the algorithm [52‐54]. In the future, we can consider collaborating with other professionals to optimizing the model algorithm and verifying the model effect with multiple calibrations to screen the optimal model. Fourth, our sample size is small and focuses on Asian people, which affects the extrapolation of the model to some extent. Although there is much room for improvement in this study, we believe that as cardiac oncology develops, this interdisciplinary model of collaboration will eventually bridge the gap between the two fields. It still needs multi-center, large-sample clinical studies or other related work to test it before it can be widely accepted or applied.