Researchers have presented at ASH 2018 in San Diego the model of using machine learning - a technique for automating the creation of computer models - to develop a new system to predict how long patients with myelodysplastic syndromes are likely to live.

In tests, the system outperformed the current gold standard prognostic tool, suggesting the new model could offer patients and doctors a better and more personalised tool to understand a patient’s risk and inform treatment.

Myelodysplastic syndromes are a type of bone marrow cancer in which the bone marrow fails to manufacture enough healthy blood cells, leading to anaemia, bleeding, or infection.

Patients diagnosed with myelodysplastic syndromes show a wide range of symptoms and may live for only a few months or for decades.

About one-third of patients develop acute myeloid leukaemia (AML), a more aggressive type of blood cancer.

Predicting a patient’s risk of dying or developing AML is crucial, both to help patients understand their disease and to help doctors determine a course of treatment.

High-risk patients are generally treated with a stem cell transplant, which can cure the disease but carries significant risks, while other, less risky, treatments are recommended for patients with a better prognosis.

The best course of treatment for patients at intermediate risk can be unclear because individual clinical trials define risk thresholds in different ways.

“All treatment guidelines are driven by risk, which means that if we get the risk wrong, we get the treatment wrong,” said lead study author Aziz Nazha, MD, of the Cleveland Clinic. “Improving and personalising our prognostic models can help to delineate patients who are at higher versus lower risk - which is particularly challenging for those who fall into the intermediate range - and match them with the appropriate treatment.”

Currently, doctors use the Revised International Prognostic Scoring System (IPSS-R) to assess risk for patients with myelodysplastic syndromes.

However, the IPSS-R underestimates or overestimates risk in up to one-third of patients, according to Dr. Nazha.

To improve prognostic tools, Dr. Nazha’s team developed a sophisticated machine learning algorithm that uses genomic and clinical data to determine a patient’s prognosis.

They trained the system using patient data from Cleveland Clinic and Munich Leukaemia Laboratory (1,471 patients total) and validated it in a separate collection of patient data from Moffitt Cancer Centre (831 patients).

In head-to-head comparisons using patient medical records, the new model correctly predicted a patient’s likelihood of surviving for a given length of time relative to another patient 74 percent of the time, compared to 67 percent of the time for IPSS-R.

The model correctly predicted a patient’s likelihood of developing AML relative to another patient 81 percent of the time, compared to 73 percent of the time for IPSS-R.

Like any decision-support tool, the model is intended to inform human clinicians, not to replace or compete with them, Dr. Nazha said.

To further improve the model, the researchers are gathering feedback from clinicians and working to incorporate more outcomes, such as quality of life, into the model.

They are also developing ways for the model to update the assessment of risk in response to changing conditions, such as when new test results are available or treatments are completed.

“This project started out of a frustration voiced by many of my patients who want to know what their own risk is and how their prognosis might differ from that of other patients,” said Dr. Nazha.

“We wanted to build a personalised prediction tool that can give insights about a specific outcome for a specific patient.”

Source: ASH