CLN - The Sample

Machine learning and miRNA show promise in diagnosing esophageal cancer

Recent research shows machine learning (ML)-based microRNA (miRNA) classifiers’ potential for diagnosing esophageal cancer (EC) (J Appl Lab Med; doi: 10.1093/jalm/jfae037).

Often diagnosed at advanced stages, EC has high mortality rates, and current tests for the disease have limited efficacy. In response, the researchers wanted to harness miRNA’s potential as a noninvasive diagnostic EC biomarker. The study aimed to determine miRNAs’ diagnostic accuracy and the ability to distinguish miRNAs associated with EC from control miRNAs.

The researchers used ML software Waikato Environment for Knowledge Analysis (WEKA) and a custom-built TensorFlow Keras neural network to analyze complex biomolecular datasets of miRNAs. Three WEKA models — random forest, Hoeffding tree, and naïve Bayes — delivered promising results, with naïve Bayes reaching the highest accuracy of 93.75%. The TensorFlow Keras model achieved an average accuracy of 87.7% over five trials, a Matthews correlation coefficient (MCC) of 0.758, and an area under the curve of 0.910. When tested on independent data, TensorFlow Keras achieved even higher accuracy of 91%, with an MCC of 0.822 and an AUC of 0.896, demonstrating its ability to generalize and accurately identify miRNAs associated with EC in unseen data.

Although their study yielded an ML model that can accurately differentiate miRNAs associated with EC from unrelated miRNAs, the model might not directly apply to clinical settings where EC cases are significantly less prevalent than in the general population, the researchers noted.

They called for further studies to validate their findings and explore the full clinical potential of the approach. The researchers plan to focus on validating the model and enhancing its capabilities by incorporating a broader spectrum of noncoding RNAs and integrating more comprehensive clinical data. The data might include patient demographics and cancer stages. Testing the model in diverse, realistic, and clinical settings with datasets that reflect actual cancer prevalence rates will be crucial, the researchers noted.

Genetic variant common in black individuals contributes to cardiovascular disease

A genetic variant present in 3-4% of self-identified Black individuals in the U.S. increases the risk of heart failure and death (JAMA 2014; doi:10.1001/jama.2024.4467).

Previous research has shown that the amyloidgenic V142I variant of the transthyretin gene increases heart failure and mortality risk. The variant causes misfolding of the blood protein transthyretin, leading to deposits of abnormal amyloid protein in the heart and other parts of the body. In the heart, these deposits cause cardiac amyloidosis — thickening and stiffening of muscle — which can lead to heart failure.

The researchers aimed to better define the natural history of disease in V142I carriers during mid-to-late life, assess variant modifiers, and estimate cardiovascular burden in the U.S. population.

They analyzed data from a total of 23,338 self-reported Black participants initially free of heart failure in four large observational U.S. studies. Of these participants, 3.2% were V142I carriers. Participants had a mean of 15.5 years of follow-up. Outcomes were heart failure hospitalizations and all-cause mortality determined by generating 10-year hazard ratios for each age between 50 and 90. Using actuarial methods, the researchers estimated mean survival by carrier status and applied it to the 2022 U.S. population using U.S. census data.

The researchers found that V142I increased the risk for heart failure hospitalization by age 63 and the risk of death by age 72. The variant’s contribution to heart failure risk increased substantially with age but was not itself increased by other known risk factors such as diabetes and hypertension. The team also showed that female and male carriers of the variant were equally at risk, contrary to some previous studies showing that men were more affected. This suggests that women are likely underdiagnosed, the researchers wrote. They estimated that individual carriers with the V142I variant live 2 to 2.5 years less than expected.

The researchers say their data may guide discussions with patients about genetic screening and inform strategies for early targeted therapies.

Cell-free DNA test proposed for colorectal cancer screening

An experimental cell-free DNA (cfDNA) blood test may be useful for colorectal cancer screening, according to recent research (NEJM 2024; doi: 10.1056/NEJMoa2304714).

Early detection could prevent more than 90% of colorectal cancer-related deaths, but more than one-third of people eligible for screening are not up to date with screening despite multiple available tests. An easy blood test could improve screening adherence, detect disease earlier, and reduce mortality related to colorectal cancer.

The researchers tested a cfDNA screening test in 7,861 average-risk patients ages 45 to 84. Primary outcomes were sensitivity for colorectal cancer and specificity for advanced neoplasia relative to screening colonoscopy. The secondary outcome was sensitivity for precancerous lesions.

A total of 83.1% of participants with colorectal cancer detected by colonoscopy had a positive cfDNA result and 16.9% had a negative result, for a sensitivity of 83.1% (95% CI, 72.2-90.3). Sensitivity for stage 1, 2, or 3 colorectal cancer was 97.5% (95% CI, 75.3 to 94.1). Sensitivity for advanced precancerous lesions was 13.2% (95% CI, 11.3 to 15.3).

A total of 89.6% of the participants without colorectal cancer or advanced precancerous lesions identified on colonoscopy had a negative cfDNA blood-based test, while 10.4% had a positive cfDNA blood-based test. The researchers said these results indicate a specificity for any advanced neoplasia of 89.6% (95% CI, 88.8-90.3). Specificity for negative colonoscopy (no colorectal cancer, advanced precancerous lesions, or nonadvanced precancerous lesions) was 89.9% (95% CI, 89.0-90.7).

In average-risk patients, the cfDNA test had a specificity of 83% for colorectal cancer, 90% for advanced neoplasia, and 13% for advanced precancerous lesions. Blood-based testing offers a colorectal cancer screening alternative to the available stool-based tests and may improve screening participation and early detection of colorectal cancer, the researchers said.

View the full CLN July/August 2024 issue.

Advertisement
Advertisement