In the world of rare diseases, finding the right diagnosis can be a harrowing and drawn-out journey. Patients often endure multiple misdiagnoses and seek opinions from numerous healthcare professionals before finally receiving an accurate diagnosis. This lengthy diagnostic process can take a toll on patients and may result in delayed treatment and irreversible damage.
However, recent research presented at the National Society of Genetic Counselors Annual Meeting by the ThinkGenetic clinical team and its collaborators offers a promising solution to this challenge. The study, “Utilizing AI and Electronic Health Records for Enhanced Patient Discovery in Rare Diseases,” highlights the potential of artificial intelligence (AI) and electronic health records (EHRs) in identifying individuals at risk for rare diseases.
The Challenge of Rare Disease Diagnosis
Rare diseases, such as Aromatic L-amino acid decarboxylase deficiency (AADCd), pose unique diagnostic challenges. These conditions often manifest with diverse and overlapping symptoms, leading to delayed and sometimes inaccurate diagnoses. Patients can experience significant delays in receiving the correct diagnosis, particularly in cases where clinical indicators resemble more common ailments.
The Hypothesis: AI-Powered Patient Identification
This study hypothesized that AI could be harnessed to identify individuals at risk for rare diseases by analyzing EHR data and subsequent review by genetic counselors.
To implement this hypothesis, an automated prediction scoring system (APSS) was developed, integrating structured and unstructured data from EHRs. Structured data included ICD10 codes, procedures, lab results, and imaging results, while unstructured data relied on Natural Language Processing (NLP) to extract pertinent information from medical notes. A genetic counselor validated the relevance of the NLP-identified data points.
The APSS’s effectiveness was validated using known cases of AADCd from medical literature and real-life controls. The system was tested on extensive de-identified EHR data from various healthcare systems, representing a diverse patient population.
The Results: Identifying At-Risk Patients
The study yielded promising results. Across diverse patient populations and healthcare systems, the APSS successfully identified individuals at risk for AADCd. Seven patients were categorized as high-risk with an additional 32 patients falling into the possible risk category. Genetic counselors subsequently reviewed these cases for accuracy.
Furthermore, the study revealed a more diverse racial and ethnic background among at-risk patients, in contrast to traditional referral patterns. This indicates that AI-driven approaches can help mitigate unconscious provider referral biases related to patients’ demographics.
Rare Disease Patient Discovery
One of the most significant challenges in rare disease diagnosis is ensuring healthcare providers consider these conditions during their assessments. This study demonstrates that the development, validation, and implementation of APSS can assist healthcare providers in identifying patients at risk for rare genetic conditions like AADCd.
The implications of this research extend beyond diagnosis. By identifying at-risk individuals early in their healthcare journey, providers can engage in more informed discussions with patients and their families, potentially leading to earlier referrals for genetics evaluations and testing.
This study highlights the immense potential of AI and EHR data in reshaping the landscape of rare disease patient discovery. By fusing clinical expertise with advanced technology like NLP and AI, we can make significant strides in early diagnosis and treatment for individuals with rare conditions, offering them a brighter and healthier future.
For pharmaceutical companies interested in enhancing their rare disease patient-finding initiatives, this research showcases how AI can aid in identifying potential patients, facilitating early interventions, and improving patient outcomes.
