Abstracts

Rationale: Infantile spasms (IS) are seizures typically characterized by a sudden, rapid contraction of the trunk and limbs and last 5 to 10 seconds, with varied intensity. IS is a relatively rare disease (estimated incidence: 0.25-0.42/1000 live births/y).1 There is often a substantial delay in diagnosis and treatment of IS. A median delay in treatment of 24.5 days due to misdiagnosis (mistaken for colic or reflux) has been reported.2 Delayed diagnosis and treatment of IS can lead to long-term neurobehavioral problems.3 Predictive factors that can identify undiagnosed infants and/or enable faster diagnosis of IS represent a critical medical need. This study developed predictive models and identified potential IS patients within a population-based claims database. Methods: The project utilized clinical expertise and analyses of the Symphony Health Integrated Dataverse (IDV) database to identify triggers for early identification of IS patients. The IDV database captured de-identified patient-level medical and pharmacy claims from more than 12,000 US health plans, 1.8 million prescribers, and 280 million active patients, with almost 14 years of history as of 2018. The IDV database identified 10,837,709 patients aged <2 years with any claims activity between May 2017 and April 2018. Among these patients, we searched the International Classification of Diseases (ICD)-10 and ICD-9 procedure codes and drug codes generated by the IS experts. We evaluated combinations of these codes to identify those most likely to predict a subsequent diagnosis of IS (ICD-9: 345.60, 345.61, ICD-10: G40.821-G40.824; ICD-10 codes implemented starting Oct 1, 2015). Using input from IS medical experts, we determined clinical, electrographic, radiologic, procedural, and medication variables that may predict development of IS. Deductive models were used to identify combinations of variables before IS diagnosis and during the IS treatment pathway that would best predict IS. Results: Among 10,837,709 patients, 557 had >=2 symptoms pertaining to IS (e.g., seizures, developmental delay, lack of eye contact, lack of muscle tone) and a moderate/high severity emergency department (ED) visit. In this group, 304 patients (55%) had an IS diagnosis within a median of 0.8 months of the triggering event. The most notable of the deductive combinations (>=2 symptoms + ED visit of moderate/high severity) focused on patients prior to an IS diagnosis (Table). Conclusions: This deductive approach identified >=2 symptoms (e.g., developmental delays, convulsions, dysphagia, cerebral palsy) and a moderate/high severity ED visit as strong predictors for IS patient identification in real-world patient care scenarios. Limitations include use of ICD-10 for evaluation of IS, which may fail to identify all IS cases. Similarly, the deductive rules relied on the accuracy of coding for diagnoses and procedures in the ED. To raise awareness and enable simpler application into rules-based electronic medical records, these results support leveraging deductive rules on the basis of standardized ICD-10 diagnoses, procedures, and claims. Further research is needed to examine the impact of optimizing early diagnosis and treatment to improve patient health outcomes.  Funding: The sponsor (Mallinckrodt Pharmaceuticals, ARC, LLC) funded the study and editorial services (provided by MedLogix Communications, LLC).