EN | UA
EN | UA

Help Support

Back

Machine learning offers hope for early detection of primary immunodeficiency diseases

Primary immunodeficiency disease (PIDD) Primary immunodeficiency disease (PIDD)
Primary immunodeficiency disease (PIDD) Primary immunodeficiency disease (PIDD)

What's new?

An innovative predictive model leveraging clinical data has the capacity to facilitate the early identification of people with primary immunodeficiency diseases by analyzing their prior symptomatic treatment history.

The findings of a recent study emphasized the utilization of indicators related to past symptom treatment to create a predictive model for primary immunodeficiency disease (PIDD) early detection. PIDD encompass a range of immune-linked disorders, and the current average time to diagnosis often spans between 6 to 9 years. Detecting and treating PIDD at an earlier stage has been linked to more favorable patient outcomes.

This study aimed to create a machine learning model utilizing electronic health record data, specifically focusing on prior symptomatic treatments, to forecast the presence of PIDD.

Researchers performed a retrospective analysis of patients with PIDD based on inclusion criteria that considered PIDD-linked diagnoses, medications specific to immunodeficiency, and low immunoglobulin levels. A control group of patients with asthma, matched by age, gender, and race, was also established. The key goal was to identify PIDD diagnoses.

Various factors, including comorbidities, laboratory test results, medications, and radiological orders, were used as features, all of which were examined before the formal diagnosis and indicated symptom-linked treatment. Systematic application of the features was done to random forest classifiers, elastic net, and logistic regression. These models were trained with the aid of a nested cross-validation strategy to enhance their performance. The study cohort encompassed 6422 patients, among whom 247 (4%) received a PIDD diagnosis.

The logistic regression model, when considering comorbidities alone, illustrated a notable ability to distinguish between those with PIDD and those with asthma (c-statistic: 0.62 [0.58-0.65]). However, incorporating additional data such as laboratory results, medications, and radiological orders considerably enhanced discrimination (c-statistic: 0.70 vs. 0.62), along with enhancing specificity and sensitivity. The application of more advanced machine learning models did not yield further improvements in performance.

Thus, the researchers successfully developed a predictive model for early PIDD diagnosis utilizing historical data tied to symptomatic care. Implementing this approach holds the potential to address a crucial need by diminishing the time required to diagnose PIDD, ultimately leading to improved prognosis and outcomes for individuals battling immunodeficiency disorders.

This model could represent a crucial initial phase in creating an electronic health record alert for healthcare professionals, prompting them to contemplate a PIDD diagnosis. This, in turn, could expedite the administration of immunoglobulin replacement therapy, potentially minimizing the likelihood of mortality and morbidity.

Source:

The Journal of Allergy and Clinical Immunology: In Practice

Article:

Early Diagnosis of Primary Immunodeficiency Disease Using Clinical Data and Machine Learning

Authors:

Anoop Mayampurath et al.

Comments (0)

You want to delete this comment? Please mention comment Invalid Text Content Text Content cannot me more than 1000 Something Went Wrong Cancel Confirm Confirm Delete Hide Replies View Replies View Replies en ru ua
Try: