For the first time in Korea, a study has used a machine learning model to segment leukemia by cytogenetic characteristics and applied it to personalized treatment strategies for elderly acute myeloid leukemia patients.

Acute myeloid leukemia is a blood cancer that is increasing as the population ages, with the average age of its onset being 65 to 67 years old. These elderly patients range from those who are well enough to be considered for high-intensity chemotherapy to those who are unsuitable for standard treatment due to decreased systemic performance status and need to opt for low-intensity treatment.

Therefore, a one-size-fits-all treatment cannot be applied, requiring special attention to treatment selection.

A trio of hematologists at the Catholic University of Korea St. Mary’s Hospital conducted a genetic classification of elderly acute leukemia patients who require caution about choosing chemotherapy due to declining physical performance based on an artificial intelligence learning model and found that the choice of treatment has a significant impact on survival prognosis.

The three researchers include Professors Cho Byung-sik (corresponding author) and Silvia Park (co-first author) from the Department of Hematology at Seoul St. Mary's Hospital, along with Professor Kim Tong-yoon (co-first author) from the Department of Hematology at Yeoeuido St. Mary's Hospital.

Between 2017 and 2021, the researchers conducted a comparative analysis of treatment outcomes for 279 elderly acute myeloid leukemia patients aged 60 years or older. Treatments included high-intensity chemotherapy, methylation inhibitor alone low-intensity chemotherapy, or methylation inhibitor plus venetoclax low-intensity chemotherapy, assessed based on genetic characteristics and survival rates.

Using the most commonly used European Leukemia Research Group (ELN) molecular risk classification, the researchers assessed the predictive power of each treatment group. They found that younger patients treated with high-intensity chemotherapy and hematopoietic stem cell transplantation matched the predictive power of the risk classification (low-, intermediate-, and high-risk), while older patients aged 60 years or more had significantly poorer survival prediction.

So, the team applied machine learning models to pattern the cytogenetic characteristics of complex and diverse leukemia cells across patients and grouped them into nine genomic populations of similar types.

When the survival prognosis of each treatment arm in these nine genomic groups was examined independently, high-intensity chemotherapy was not always superior to low-intensity chemotherapy, depending on the genomic characteristics of each group.

It was also confirmed that the recently demonstrated combination of methylation inhibitors and venetoclax among low-intensity treatments was not always superior to methylation inhibitors alone.

The researchers found that the genomic patterns of patients who responded well to high-intensity chemotherapy were not predictive of a good response to low-intensity chemotherapy and vice versa.

In conclusion, they proved that a personalized treatment strategy is ultimately needed for each patient, including the choice of treatment intensity and therapies, such as monotherapy or combination therapy, and that AI models can make personalized treatment strategies a reality.

Acute myeloid leukemia is characterized by various cytological and genetic alterations in leukemic cells, with the combination and extent of these alterations varying widely among patients.

Recent leukemia drug developments have expanded the options for low-intensity anticancer therapies, including hypomethylating agents alone and in combination with venetoclax. This targeted therapy inhibits B-cell lymphoma-2 protein (BCL-2).

A phase 3 international trial demonstrated superior efficacy (increased response rate, improved survival) of the combination compared to methylation inhibitors alone, leading the U.S. Food and Drug Administration to approve the combination of a methylation inhibitor (azacitidine) and venetoclax as first-line therapy in 2020 for elderly patients with acute myeloid leukemia who are refractory to standard anticancer therapy.

However, the additional cytogenic side effects of adding venetoclax cannot be overlooked, and there may be a group of patients for whom the benefit of combination therapy is uncertain, depending on the genetics of their leukemia and a strategy for selecting a less intensive treatment may be warranted.

High-intensity chemotherapy still has the highest probability of achieving complete remission in elderly patients with AML, so it is difficult to exclude high-intensity treatment strategies in treatment selection completely.

However, even if a patient is physically fit for high-intensity standard chemotherapy, depending on the genetic characteristics of the leukemia, high-intensity chemotherapy may not have a significant therapeutic effect compared to low-intensity treatment, so a patient-specific treatment intensity selection strategy is needed.

Professor Park, co-first author, explained, "This study explores the potential positive impact on patient survival by connecting the expanding array of leukemia treatments with the constantly emerging molecular and genetic information about leukemia."

Because the cytological and genetic variations in individual patients are so diverse and multiple simultaneous variations are common, it was impossible to reflect them with conventional statistical processing methods, and it was necessary to utilize machine learning models, she added.

To apply and implement the machine learning model in practice, Professor Kim, an expert in hematology and medical research statistics, built genetic statistics of blood cancer patients. In addition, ImpriMed Korea, a startup company that develops predictive models for anticancer drugs and has a collaborative relationship with Seoul St. Mary's Hospital, complemented the strengths and weaknesses of the analysis method.

"Another important aspect of the study is that it proved with objective data that customized treatment for each patient by cytogenetic characteristics affects the survival rate of patients," Professor Cho, corresponding author, said. "Although it is still the first step, we will do our best to provide the best treatment for elderly patients with acute myeloid leukemia, which has been increasing rapidly in recent years, by applying optimal treatment considering individual disease characteristics rather than one-size-fits-all treatment in the clinic."

The paper is titled, "Prognostic value of European Leukemia Net 2022 criteria and genomic clusters using machine learning in older adults with acute myeloid leukemia."