Identification of Predictors of Anthracycline-Induced Cardiotoxicity in Cancer Patients
Summary
Anthracycline chemotherapy has been effective antineoplastic treatment for a number of cancers, including breast cancer, soft tissue sarcomas, lymphomas, and leukemia. However, the increased survivorship of cancer patients attributed to anthracycline chemotherapy medicines such as doxorubicin and epirubicin has been mitigated by a rise in treatment-related cardiotoxicity, with an incidence of about 6% to 9%. Though the mechanism of cardiotoxicity is not well understood, the formation of free oxygen radicals, DNA damage, and apoptosis of myocytes are stipulated to yield clinical manifestation of impaired left ventricular systolic function and eventual progression to heart failure. The clinical expression of cardiotoxicity usually develops in the first year following treatment but may remain sub-clinical for many years up to 10-20 years. It is postulated that the cumulative dose of anthracycline medicines is the most cardiotoxic risk factor; however, recent studies suggest that there is no safe dose to mitigate cardiotoxicity.
Studies explored the use of serial measurements such as tissue Doppler imaging (TDI) echocardiographic parameters to characterize cardiac function during and long after anthracycline chemotherapy. Other studies suggested co-administration of therapeutic medicines such as Ranolaz, Nebivolol, and Dantrolene with anthracycline chemotherapy to attenuate cardiotoxic effects. To date, management of patients requiring anthracycline chemotherapy remains challenging due to uncertainty of predictors of cardiotoxicity. Studies assessed independent risk factors such as ethnicity, age, and co-morbidities in addition to cumulative anthracycline dose. However, no study has resulted in prevention and detection of cancer-treatment cardiotoxicity.
The aims of the study are: 1) Design and implement a machine learning model to appraise predictors of cardiotoxicity prior to initiating chemotherapy, and 2) Describe and quantify characteristics associated with cardiac dysfunction.
The proposed informatics approach utilizes deep learning, a subfield of machine learning, methods to identify relationships and patterns from data without the need to define a priori and is capable of learning meaningful abstractions from the input data. Deep learning has been applied successfully to gain improved insights in the diagnosis of diseases such as cancers, infectious diseases, and adverse event prediction in heart failure patients. Deep learning methods have the potential of visualizing complex patterns hidden in high dimensional data and are well suited for identifying predictors of cardiotoxicity in patients requiring anthracycline chemotherapy.
The accuracy of the proposed deep learning predictive model depends on type and variety of data that input into the training model. Study data will be collected from the following UAMS data sources: 1) The Arkansas clinical data repository (AR-CDR), which houses patients’ historical clinical data, including demographics, diagnoses, laboratory results, medications, procedures, vital signs, hospital and ICU admissions, and social histories, 2) The Cancer Institute Tumor registry, which contains patients’ cancer diagnoses and types, tumor staging information, chromosomal mutations, therapeutic modalities, and related laboratory studies, and 3) The MUSE Cardiology Information system, which maintains records of patients’ ECG information. In addition, the study cohort will include all cancer patients that received any of the anthracycline’s chemotherapy medicines such as doxorubicin, daunorubicin, mitoxantrone, and epirubicin.
Keywords:
- Chemotherapy
- Anthracycline
- Cardiotoxicity
- Deep Learning
- machine learning
- Cancer
- clinical informatics
Researchers:
- AHMAD BAGHAL (Author)
- Fred Prior
- Malek Al-Hawwas