Background:

With advancements in healthcare big data, claims data have become essential for identifying drug-related admissions (DRA), analyzing risk factors, and developing predictive models. However, compared to hospital data, claims data lack certain clinical details, limiting their applicability. To enhance their utility and improve accuracy and reliability in DRA detection, a standardized coding framework based on internationally recognized disease and drug classifications is needed.

Methods:

We utilized national claims data from the Korean Health Insurance Review and Assessment Service (HIRA) to identify DRA based on the adjudication guide of the OPERAM trial, employing ICD-10 and ATC classification codes. ICD-10 codes were applied to identify all diagnosis- and laboratory-based triggers, while triggers unavailable in the database were excluded. To minimize excessive false positives, we added optional conditions to certain triggers based on treatment guidelines and clinical research, particularly regarding appropriate indications and drug selection.

Results:

 We included 23 triggers and implemented 43 coding algorithms. Triggers requiring hospital records, such as potential adverse drug events, vaccine-preventable infections, rapid dose adjustments in delirium and epilepsy or inadequate hypertension treatment were excluded. Moreover, we added three independent DRA triggers: drug-related QT interval prolongation, drug-induced disorders, and drug toxicity. We incorporated disease-specific identification criteria. For instance, uncontrolled non-neurological pain was restricted to severe pain conditions such as cancer pain and postoperative pain that necessitate opioid therapy. Likewise, for hypoglycemia identification, the criteria were expanded to include the administration of glucagon or 50% dextrose in emergency settings, in addition to ICD-10 codes. Meanwhile, we refined the existing tool by restricting the β-blockers listed under \"underuse of β-blockers in heart failure exacerbation\" to only those recommended by clinical guidelines.

Conclusion:

      This study is the first to systematically encode a trigger tool for detecting DRAs. While further validation is needed, this structured rule lays the groundwork for future applications in claims data analysis and the development of large-scale risk prediction models.