(AF) Development and Testing of Diagnostic Algorithms for Identifying Subjects with Acromegaly in Administrative Claims Databases

Publication Number: 1030

Background:

Background: Acromegaly is an endocrine rare disease characterized by an excessive production of growth-hormone and insulin-like growth factor 1, typically resulting from a growth-hormone-secreting pituitary adenoma.

Objectives:

Objectives: To compare and measure accuracy of newly developed and previously developed coding algorithms for the identification of acromegaly using claims databases.

Methods:

Methods: This retrospective, population-based study was conducted in the period January 2015-December 2018, using data from the administrative claims databases of Caserta LHU in Italy. To detect acromegaly cases from the general target population, four different algorithms were developed using a combination of diagnostic codes, surgical procedure codes, co-payment exemption codes, specialist’s visits and pharmacy claims for somatostatin analogues or pegvisomant. For each algorithm, diagnostic accuracy was assessed by measuring sensitivity, specificity, positive predictive value and negative predictive value. The Youden Index was computed to assess discriminatory power. The gold standard was defined as the subgroup of subjects with ≥1 registration in the electronic therapeutic plan (ETP) database, during the study period. From this cohort, true cases were defined as subjects with at least one diagnostic code for acromegaly.

Results:

Results: A total of 84, 92, 116 and 146 individuals were identified as having acromegaly by the four different algorithms, respectively. The estimated percentage of positive cases in Caserta LHU for each algorithm ranged from 8.0 (95% CI: 6.5-9.9) to 13.9 (95%CI: 11.8-16.3) per 100,000 inhabitants. The gold standard cohort comprised 81,388 subjects with ≥1 registration in the ETP database during the study period. Among them, 38 individuals with at least one ICD-9 CM code for acromegaly were identified, yielding an estimated prevalence of 3.6 cases per 100,000 inhabitants. Sensitivity ranged from 71.1% (95%CI: 54.1-84.6%) to 84.2% (95%CI: 68.8-94.0%), while specificity was higher than 99.9% for each algorithm. The algorithm based on the presence of claims suggestive of acromegaly in ≥2 different databases (i.e., hospital discharge records, copayment exemptions registry, pharmacy claims and specialist visits registry) achieved the highest Youden Index (84.2) and the highest positive predictive value (34.8; 95%CI: 28.6–41.6).

Conclusions:

Conclusions: We developed and tested four algorithms to identify acromegaly cases using administrative claims databases with high sensitivity and high Youden Index.