Development and testing of diagnostic algorithms to identify patients with acromegaly in claims databases in southern Italy


Information source

This retrospective Italian population-based study was conducted between January 2015 and December 2018, using data extracted from the fully anonymized administrative record linkage databases of the Caserta Local Health Unit (LHU), with an average of 1,060,904 inhabitants. Additionally, data from fully anonymized health administrative databases of the region of Sicily for the period January 2011 to December 2018, with an average population of 5,031,655, were used as a test dataset to assess performance of the proposed algorithms and to support estimates of acromegaly cases identified at Caserta LHU. Both databases contain demographic and medical data collected through services provided by the Italian National Health Service (NHS). They include information on the demographics of residents in each catchment area, outpatient pharmacy claims, hospital discharges, co-payment exemptions, referrals for outpatient diagnostic tests, and the hospital visit database. specialists. The content and context of these NHS complaints databases have been described in detail elsewhere.19. The Caserta LHU claims databases also include the Electronic Therapy Plans (ETP) database. In Italy, the ETPs are completed directly by specialists, who provide information on the exact name of the brand, the number of packaging dispensed and the indication for use. Dispensed drugs were coded using the Anatomical Therapeutic Clinical Classification (ATC) system and the Italian Marketing Authorization Code (AIC), while comorbidities and indications for use were coded by the ninth revision of the International Classification of Diseases – Clinical Modification (ICD-9 CM).

Definition of the algorithm

To detect the presence of acromegaly in the target population, four different algorithms were proposed and developed based on a systematic review of published articles on the epidemiology of acromegaly using algorithms based on complaint databases.2. Specifically, each algorithm was developed using a combination of ICD-9 CM diagnosis codes related to acromegaly (253.0), ICD-9 CM surgical procedure codes (07.6x and 92.3x) , co-payment exemption codes (001 and 253.0), specialist visits, laboratory examinations and pharmaceutical claims for somatostatin analogues (ATC: H01CB02, H01CB03, H01CB05) or pegvisomant (ATC: H01AX01). “Algorithm 1” was developed by Caputo et al. in 20184“Algorithm 2” consisted of a revision of “Algorithm 1”, which included other codes related to acromegaly, while “Algorithm 3” and “Algorithm 4” were developed through different combinations of acromegaly related codes used in publications. articles3,4,21,22,23. For each algorithm, as proposed by Caputo et al.4, pharmaceutical claims for drugs approved for the treatment of acromegaly were not considered if (i) patients had received fewer than three separate prescriptions for drugs for the treatment of acromegaly (occasional users of drugs) during the observation period; (ii) the drugs were not extended release formulations (LAR); (iii) patients taking octreotide or lanreotide have had at least one hospitalization with a diagnosis which, as reported in the summary of product characteristics, is one of the indications for use of these medicinal products [i.e., malignant neoplasms (ICD-9 CM: 140–209, 230–239), liver disorders (ICD-9 CM: 570–573), gastrointestinal bleeding (ICD-9 CM: 578), esophageal varices (ICD-9 CM: 42), Cushing’s disease (ICD-9 CM: 255; 255.0)]; (iv) the patients had a user fee exemption code for Cushing’s disease (code: 032). Inclusion and exclusion criteria are shown in Table 1.

Table 1 Inclusion and exclusion criteria for each proposed algorithm for identifying cases of acromegaly. *Italian coding system.

Definition of the reference cohort

At Caserta LHU, index cases were defined as subjects who had at least one ETP database record with at least one ICD-9 CM code for acromegaly during the study period. Non-index cases were defined as all remaining subjects (i.e. registered at Caserta LHU without an ICD-9 CM code for acromegaly in the ETP database).

In the region of Sicily, index cases were defined as patients with a confirmed diagnosis of acromegaly in the endocrinology unit of the University Hospital of Messina, a province located in the northeastern part of Sicily. , during the study period. Data for non-cases were not available.

statistical analyzes

For each proposed algorithm, the precision was assessed by the sensitivity (SE) and specificity (SP) measures, as well as their 95% confidence intervals (CI) calculated using the Clopper-Pearson exact method. for a binomial proportion. Additionally, Youden’s index (i.e. a summary statistic that balances both SE and SP)24 was calculated and the algorithm that achieved the highest accuracy (i.e. Youden’s index) was considered the preferred one over the others. In addition, the positive predictive value (PPV) and negative predictive value (NPV) along with their 95% CIs were also estimated to assess the accuracy of the algorithm. To visually assess the number of overlaps of acromegaly cases detected by each coding algorithm against different data sources, Venn diagrams were produced.

The four proposed algorithms were also applied to patients diagnosed with acromegaly in the Endocrinology Unit of the University Hospital of Messina by probabilistic record linkage and the proportion of true positives among these cases was compared to the SE obtained for each algorithm. Additionally, to provide an estimate of the percentage of acromegaly cases in the Sicilian population, only the algorithm with the highest discriminating power was performed in the administrative data of the Region of Sicily (test dataset).

Finally, for the algorithm with the highest Youden index, a network plot was produced to show the number of acromegaly patients identified in Caserta LHU and the region of Sicily, respectively, to further substantiate the chronological occurrence. any acromegaly-specific code identified in each complaint database. In particular, for each claims database, all possible routes by which subjects have been identified over time have been represented. Each lane consists of the chronological sequence by which each acromegaly-specific code occurred for each patient. This can be seen as a proxy for the patient journey25.

All statistical analyzes were performed using R Foundation for Statistical Computing (R Development Core Team 2008, Vienna, Austria, version: 4.0.3, packages: caret and PropCI).

Ethics approval

Analyzes were conducted in accordance with institutional and national research committee ethical standards and the Declaration of Helsinki of 1964 and subsequent amendments. This study was approved by the ethics committee of Azienda Ospedaliera Universitaria Integrata of Verona (Protocol number 55986, September 27, 2021).

Consent to participate

Informed consent was obtained from all individual patients of the Endocrinology Unit of the University Hospital of Messina included in the study.


Comments are closed.