Authors :
Presenting Author: Ashwin Mahesh, BA – Weill Cornell Medical College
Carson Gundlach, BS – Medical Student, Weill Cornell Medical College; Alexander Zhao, BS – Medical Student, Weill Cornell Medical College; Gabrielle Dykhouse, BS – Medical Student, Weill Cornell Medical College; Natasha Basma, BS MPH – Department of Pediatric Neurology – Weill Cornell Medicine; Zachary Grinspan, MS MD – Director of Pediatric Epilepsy, Interim Chief Child Neurology, Department of Pediatric Neurology, Weill Cornell Medicine
Rationale: For children with epilepsy (CWE), genetic testing yields valuable information for prognosis and treatment. Developing a quality measure to learn if CWE are receiving genetic testing when indicated may help clinicians identify CWE who have not received testing but can benefit from it. One challenge is that genetic information is stored variably across electronic health records (EHRs), impeding feasibility of calculating this measure automatically. Resultantly, developing this quality measure requires constructing computable phenotypes. We aimed to (1) enumerate indications for genetic testing in CWE, (2) develop a computable phenotype to determine if testing was performed, (3) investigate clinical adherence to testing when indicated, (4) describe factors associated with receipt of testing when indicated, and (5) provide recommendations for quality improvement (QI).
Methods: We interviewed clinicians to outline genetic testing indications in CWE. Our data source was two WCM ARCH [ref PMID 34850911] Structured Query Language (SQL) pediatrics databases. We used regular expressions and pattern matching to code eight algorithms (one shown in Figure 1) to build our computable phenotype that identifies patients that received genetic testing. Two gold standard cohorts that received testing were built from lists of children co-managed by genetics and neurology for one cohort and managed by genetics for the other. Sensitivity, positive predictive value (PPV), and F-measure of our genetic testing phenotype were evaluated using these cohorts. We then used ICD codes and EHR data to extract computable phenotypes for each indication and calculated PPV with chart review of simple random samples (n=30) from each phenotype. We described sensitivity and PPV as “excellent” if >= 90%, “very good” if 80-90%, and “fair” if < = 80%. Indication adherence was evaluated and logistic regression and chi square testing were used for demographic analysis.
Results: We interviewed five geneticists and one child neurologist to propose seven genetic testing indications in CWE (Table 1B). Two gold standards (n=35, n=30) were built (Table 1A). Our genetic testing computable phenotype performance was excellent (Table 1A). The PPV of computable phenotypes for indications was excellent for three (1, 2, 5), very good for two (3,4), and fair for two (6,7) as seen in Table 1B. Clinical adherence was roughly half for all indications (Table 1C). Asian and Hispanic CWE were more likely to receive genetic testing when indicated than White CWE (Table 1D). No significant difference was found in likelihood of receiving genetic testing when indicated between male and female sex or public and private insurance (Table 1E/1F).
Conclusions: We showed that (1) composite SQL queries can reliably determine receipt of genetic testing and (2) clinical adherence was roughly half for each genetic testing indication with evidence of racial and ethnic disparities. Ongoing work will refine, operationalize, and implement these measures for QI initiatives.
Funding: N/A