Available online 13 September 2022
HighlightsWhat is already known about this topic?
Significant delays in diagnosing primary immunodeficiency diseases (PIDD) contributes to high morbidity and mortality. Models that utilize electronic health record data to diagnosis PIDD currently use known risk factors as predictors.
What does this article add to our knowledge?
We highlight how features indicative of prior treatment of symptoms can be used to develop a prediction model for early diagnosis of PIDD.
How does this study impact current management guidelines?
Early diagnosis of PIDD using our approach could lead to initiation of treatment and improved outcomes in immunodeficient patients.
AbstractBackgroundPrimary immunodeficiency diseases (PIDD) are a group of immune-related disorders that have a current median delay of diagnosis between six and nine years. Early diagnosis and treatment of PIDD has been associated with improved patient outcomes.
ObjectiveTo develop a machine learning model utilizing elements within the electronic health record data that are related to prior symptomatic treatment to predict PIDD.
MethodsWe conducted a retrospective study of PIDD patients identified using inclusion criteria of PIDD-related diagnoses, immunodeficiency-specific medications, and low immunoglobulin levels. We constructed a control group of age, sex, and race-matched asthma patients. Primary outcome was the diagnosis of PIDD. We considered comorbidities, laboratory tests, medications, and radiological orders as features, all prior to diagnosis and indicative of symptom-related treatment. Features were presented sequentially to logistic regression, elastic net, and random forest classifiers, which were trained using a nested cross-validation approach.
ResultsOur cohort consisted of 6,422 patients, of which 247 (4%) were diagnosed with PIDD. Our logistic regression model with comorbidities demonstrated good discrimination between PIDD and asthma patients (c-statistic 0.62 [0.58-0.65]). Adding laboratory results, medications, and radiological orders improved discrimination (c-statistic 0.70 vs. 0.62 P < 0.001), sensitivity, and specificity. Extending to the advanced machine learning models did not improve performance.
ConclusionsWe developed a prediction model for early diagnosis of PIDD using historical data that is related to symptomatic care, which has potential to fill an important need in reducing the time to diagnose PIDD, leading to better outcomes for immunodeficient patients.
Keywordsprimary immunodeficiency disease (PIDD)
immunodeficiency
common variable immunodeficiency (CVID)
specific antibody deficiency
machine learning
electronic health record (EHR)
List of abbreviationsPIDDPrimary Immunodeficiency Disease
CVIDCommon Variable immunodeficiency
EHRElectronic Health Record
ICDInternational Classification of Disease
CRICenter for Research Informatics
AUCArea Under the receiver operating characteristic Curve
© 2022 Published by Elsevier Inc. on behalf of the American Academy of Allergy, Asthma & Immunology
留言 (0)