Alkaptonuria Treatment Bettered by Machine Learning Strategy

February 29, 2020
Kevin Kunzmann

An emphasis on collecting and assessing quality-of-life biomarkers shows promise for tailoring treatment of the rare condition.

Ottavia Spiga

A proof-of-principle study showed patients with alkaptonuria (AKU) may benefit from a machine learning application which may inform quality of life-driven treatment strategy.

The trial, led by Ottavia Spiga and Vittoria Cicaloni, of the Department of Information Engineering and Mathematics at the University of Siena, showed the value of data management and analysis for the care of patients with the ultra-rare autosomal recessive condition.

AKU, investigators noted, is estimated to occur in 1 in every 250,000-1,000,000 births. It was described more than a century ago as the first disorder to conform with Mendelian recessive inheritance principles. Patients carry homozygous or compound heterozygous mutations of the HGD gene, which can lead to a deficiency of the enzyme associated with the tyrosine catabolic pathway.

“The major obstacle in carrying out clinical research on AKU is the lack of a standardized methodology to assess disease severity and response to treatment, which is complicated by the fact that AKU symptoms differ from an individual to another and no correlation between specific HGD mutations and disease severity has been observed so far,” they wrote.

As such, Spiga, Cicaloni, and colleagues identified quality of life (QoL) scores as a relaiable metric by which to monitor both patients’ clinical condition and health status. They assessed the implementation of a Machine Learning (ML) approach for the prediction of clinical data-driven QoL scores, in hopes of providing a comprehensive digital tool for assessing the complicated nature of AKU.

The team derived data from 129 patients with AKU taken from the ApreciseKUre database. All patients had been first examined through preliminary statistical analysis to measure linear correlation between 11 QoL scores.

Of the 110 biomarkers within the QoL health status in the database, 6 were identified as having a direct correlation: age; Serum Amyloid A; Chitotriosidase; Advanced Oxidation Protein Products; S-thiolated proteins; and Body Mass Index (BMI). The 6 biomarkers were most correlated with Knee Injury and Osteoarthritis Outcome Score (KOOS) symptoms.

Data showed inflammation, oxidative stress, amyloidosis, and patient lifestyle all correlated with QoL scores for physical status. Investigators observed no correlation between biomarkers and patients’ mental health status.

The team reiterated that robust statistical results for changes and effects of medication or joint replacement were not produced due to the “very limited” rate of clinical information available for patients with the rare chronic disease.

That said, the findings support the use of machine learning and database collection and assessment as means to better assess treatment progression and targeted therapies for AKU. Future studies will focus on collecting more data to make that more feasible.

“This will be an essential point for a typical precision medicine approach, in which each patient is closely monitored over time and several types of information are collected to understand the uniqueness of each individual,” investigators wrote.

Spiga, Cicaloni, and colleagues concluded their observed predictive system will also eventually allow for assessment of AKU disease evolution and progression—with an understanding the QoL scores and biomarkers can dictate treatment selection.

“In summary, this cost-effective computational method will be beneficial in supporting experimental and clinical studies and, at the same time, will help patients by identifying the most promising treatments,” they wrote.

The study, “Machine learning application for development of a data-driven predictive model able to investigate quality of life scores in a rare disease,” was published online in Orphanet Journal of Rare Diseases.