Introduction/aims: Administrative health data has been increasingly used to study the epidemiology of myasthenia gravis (MG) but a case ascertainment algorithm is lacking. We aimed to develop a valid algorithm for identifying MG patients in the older population with Medicare coverage.
Methods: Local older patients (age ≥65) who received healthcare at the Cleveland Clinic and possessed Medicare coverage in 2014 and 2015 were selected. Potential MG patients were identified by using a combination of ICD9 or ICD10 codes for MG and MG-related text-word search. Diagnosis was categorized as "definite MG", "possible MG" or "non-MG" after review of clinical summaries by 5 neuromuscular specialists. Performances of various algorithms were tested by use of the definite MG cohort as a reference standard, and calculation of sensitivity, specificity, and predictive values.
Results: A total of 118 988 local older patients with Medicare coverage were identified. Usage of MG ICD codes and text-word search resulted in 125 patients with definite and 67 with possible MG. A total of 45 algorithms involving ICD usage, medication prescription, and specialty visit were tested. The best performing algorithm was identified as 2 office visits using MG ICD codes separated by at least 4 weeks or 1 hospital discharge and 1 office visit each using MG ICD codes separated by at least 4 weeks within the two-year period, resulting in a sensitivity and positive predictive value of 80% for identifying definite MG patients.
Discussion: Algorithms using ICD codes can reliably identify patients with MG with a high degree of accuracy.
Keywords: algorithm; incidence; myasthenia gravis; prevalence; sensitivity; specificity.
© 2022 Wiley Periodicals LLC.