2:40 PM
 | 
Jul 10, 2018
 |  BC Extra  |  Preclinical News

Models could help predict AML risk

In a Nature paper, a team led by researchers at University Health Network and Wellcome Sanger Institute developed two models to predict healthy subjects’ likelihood for developing acute myelogenous leukemia that could provide the basis for an AML risk diagnostic.

The first method was based on recurrent mutations in genes associated with AML that accumulate in clonally expanding hematopoietic stem and progenitor cells.

The researchers sequenced AML-associated genes in 124 patients who later developed AML and 686 controls and detected putative driver mutations of age-related clonal hematopoiesis in 73.4% of pre-AML patients compared with 36.7% of controls. The researchers detected additional differences in mutation patterns between pre-AML patients and controls, including higher cumulative frequencies and variant allele frequencies in the mutations in pre-AML patients.

A regression model based on those factors predicted the likelihood that patients would develop AML with sensitivity of 41.9% and specificity of 95.7% in an independent cohort of 29 pre-AML patients and 262 controls.

The second method used variations in blood parameters such as red blood cell distribution width and blood cell counts from electronic health records in a database of about 52 million patients, including 875 identified cases of AML. A machine learning model predicted progression to AML with sensitivity of 25.7% and specificity of 98.2% six to 12 months prior to diagnosis in additional patients.

The authors suggested a future model could combine mutation analysis with clinical data to improve accuracy.

User Sign in

Trial Subscription

Get a 4-week free trial subscription to BioCentury Extra

Article Purchase

Purchase this article for limited one-time distribution and website posting

$750 USD