Untapping the potential of utilizing electronic medical records to identify patients with atopic dermatitis: an algorithm using ICD-10 codes

Patient Phenotyping for Atopic Dermatitis With Transformers and Machine Learning: Algorithm Development and Validation Study

BACKGROUND

Atopic dermatitis (AD) is a chronic skin condition that millions of people around the world live with each day. Performing research into identifying the causes and treatment for this disease has great potential to provide benefits for these individuals. However, AD clinical trial recruitment is not a trivial task due to the variance in diagnostic precision and phenotypic definitions leveraged by different clinicians, as well as the time spent finding, recruiting, and enrolling patients by clinicians to become study participants. Thus, there is a need for automatic and effective patient phenotyping for cohort recruitment.

OBJECTIVE

This study aims to present an approach for identifying patients whose electronic health records suggest that they may have AD.

METHODS

We created a vectorized representation of each patient and trained various supervised machine learning methods to classify when a patient has AD. Each patient is represented by a vector of either probabilities or binary values, where each value indicates whether they meet a different criteria for AD diagnosis.

RESULTS

The most accurate AD classifier performed with a class-balanced accuracy of 0.8036, a precision of 0.8400, and a recall of 0.7500 when using XGBoost (Extreme Gradient Boosting).

CONCLUSIONS

Creating an automated approach for identifying patient cohorts has the potential to accelerate, standardize, and automate the process of patient recruitment for AD studies; therefore, reducing clinician burden and informing the discovery of better treatment options for AD.

Patient Phenotyping for Atopic Dermatitis with Transformers and Machine Learning

Background

Atopic dermatitis (AD) is a chronic skin condition that millions of people around the world live with each day. Performing research studies into identifying the causes and treatment for this disease has great potential to provide benefit for these individuals. However, AD clinical trial recruitment is a non-trivial task due to variance in diagnostic precision and phenotypic definitions leveraged by different clinicians as well as time spent finding, recruiting, and enrolling patients by clinicians to become study subjects. Thus, there is a need for automatic and effective patient phenotyping for cohort recruitment.

Objective

Our study aims to present an approach for identifying patients whose electronic health records suggest that they may have AD.

Methods

We created a vectorized representation of each patient and trained various supervised machine learning methods to classify when a patient has AD. Each patient is represented by a vector of either probabilities or binary values where each value indicates whether they meet a different criteria for AD diagnosis.

Results

The most accurate AD classifier performed with a class-balanced accuracy of 0.8036, a precision of 0.8400, and a recall of 0.7500 when using XGBoost (Extreme Gradient Boosting).

Conclusions

Creating an automated approach for identifying patient cohorts has the potential to accelerate, standardize, and automate the process of patient recruitment for AD studies; therefore, reducing clinician burden and informing knowledge discovery of better treatment options for AD.

Patterns in the Development of Pediatric Allergy

OBJECTIVES

Describe clinical and epidemiologic patterns of pediatric allergy using longitudinal electronic health records (EHRs) from a multistate consortium of US practices.

METHODS

Using the multistate Comparative Effectiveness Research through Collaborative Electronic Reporting EHR database, we defined a cohort of 218 485 children (0-18 years) who were observed for ≥5 years between 1999 and 2020. Children with atopic dermatitis (AD), immunoglobulin E-mediated food allergy (IgE-FA), asthma, allergic rhinitis (AR), and eosinophilic esophagitis (EoE) were identified using a combination of diagnosis codes and medication prescriptions. We determined age at diagnosis, cumulative incidence, and allergic comorbidity.

RESULTS

Allergic disease cumulative (and peak age of) incidence was 10.3% (4 months) for AD, 4.0% (13 months) for IgE-FA, 20.1% (13 months) for asthma, 19.7% (26 months) for AR, and 0.11% (35 months) for EoE. The most diagnosed IgE-FAs were peanut (1.9%), egg (0.8%), and shellfish (0.6%). A total of 13.4% of children had ≥2 allergic conditions, and respiratory allergies (ie, asthma, AR) were commonly comorbid with each other, and with other allergic conditions.

CONCLUSIONS

We detail pediatric allergy patterns using longitudinal, health care provider-based data from EHR systems across multiple US states and varied pediatric practice types. Our results support the population-level allergic march progression and indicate high rates of comorbidity among children with food and respiratory allergies.

Validation of Patient Identification Algorithms for Atopic Dermatitis Using Healthcare Databases

INTRODUCTION

The use of real-world data offers a possibility to perform large-scale epidemiological studies in actual clinical settings. Despite their many advantages, administrative databases were not designed to be used in research, and the validation of diagnoses and treatments in administrative databases is needed. The primary objective of this study was to validate an existing algorithm based on dispensed prescriptions and diagnoses of skin conditions to identify pediatric patients with atopic dermatitis (AD), using a diagnosis of AD in primary care as a gold standard.

METHODS

Retrospective observational data were collected from nation-wide secondary care and pharmacy-dispensed medication databases and two regional primary care databases in Sweden. An existing algorithm and a Modified algorithm, using skin-specific diagnoses from secondary care and/or pharmacy-dispensed prescriptions to identify patients with AD, were assessed. To verify the presence of AD, diagnoses from primary care were used in the base case and complemented with diagnoses from secondary care in a sensitivity analysis.

RESULTS

The sensitivity (30.0%) and positive predictive value (PPV) (40.7%) of the existing algorithm were low in the pediatric patient population when using primary care data only but increased when secondary care visits were also included in the Modified algorithm (sensitivity, 62.1%; PPV, 66.3%). The specificity of the two algorithms was high in both the base case and sensitivity analysis (95.1% and 94.1%). In the adult population, sensitivity and PPV were 20.4% and 8.7%, respectively, and increased to 48.3% and 16.9% when secondary care visits were also included in the Modified algorithm.

CONCLUSION

The Modified algorithm can be used to identify pediatric AD populations using primary and secondary administrative data with acceptable sensitivity and specificity, but further modifications are needed to accurately identify adult patients with AD.