Data source and study cohort
We obtained the dataset from the Korean HIRA database between January 1, 2007 and August 31, 2018. The HIRA database includes approximately 59 million individuals and is representative of almost all Koreans insured by the National Health Insurance and National Medicaid. The HIRA database contains data on patient demographics, diagnosis codes, prescription records, medical procedures and services, types of health care facilities, medication use, and dates of treatment. ‘admission.
From the HIRA database, we identified 116,737 enrollees aged 50 or older with either at least one inpatient claim or at least two outpatient claims with ICD-10 spectrum disorder codes bipolar (F30, F31 and F34.0) as their primary or secondary diagnosis between January 1, 2007 and August 31, 2013.
Among them, we identified patients who received at least one valproate prescription between January 1, 2007 and August 31, 2018, but who did not receive lithium as valproate-only users and vice versa for lithium users. alone (N = 31,973 for the VALP group, N = 9,407 for the LITH group). We also found patients who prescribed both valproate and lithium at least once for each between January 1, 2007 and August 31, 2018 (N=17,692 for BOTH groups).
Of these patients, we excluded people who: (1) were prescribed valproate and/or lithium before obtaining a diagnosis code for bipolar spectrum disorder; (2) had a diagnosis code for psychotic disorder, dementia, cerebrovascular disorder, Parkinson’s disorder, epilepsy, or head trauma as the primary diagnosis before being prescribed valproate and/or lithium; or (3) were prescribed acetylcholinesterase inhibitors or memantine before being prescribed valproate and/or lithium. We further excluded people who were prescribed valproate and/or lithium between January 1, 2007 and December 31, 2007 to ensure that there was at least a one-year washout period among patients who were prescribed valproate or lithium before 2008. As we had data up to August 31, 2018, in order to obtain follow-up data of at least 5 years, we excluded patients who were prescribed for first time valproate or lithium after August 31, 2013. (Fig. 1).
After that, we matched age, sex, indexing year, and health insurance type between the VALP group, LITH group, BOTH group, and patients who never received valproate or of lithium between January 1, 2017 and August 31, 2018, as a reference group. (group NONE). To minimize confounding by indication, we also matched antipsychotic use between four groups. Finally, 1164 patients from the VALP group, 621 from the LITH group, 621 from the BOTH group and 2378 from the NONE group are included in the analysis.
We defined the index date as the first day of prescription of valproate or lithium after the diagnosis of bipolar spectrum disorder in the VALP, LITH and BOTH groups. In the NONE group, we defined the index date by adding the interval between the date of diagnosis of bipolar disorder and the index date of the matched VALP, LITH and BOTH groups to the date of diagnosis of bipolar disorder in the NONE group (Fig. 2 ).
This study was approved by the Institutional Review Board of Bundang Hospital, Seoul National University (IRB no. X-1906-546-901) and complied with the ethical guidelines of the Declaration of Helsinki of the World Medical Association. Informed consents were overruled by the Institutional Review Board of Seoul National University Bundang Hospital due to retrospective design.
Exposure to valproate (ATC; N03AG01) or lithium (ATC; N05AN01) was assessed between the index date and the last day of follow-up. To assess the dose-response relationship, we calculated the DDD, CPD of valproate and lithium in days. When calculating the DDD, we have adapted the widely used drug standardization method developed by the World Health Organization21. We divided the level of exposure into tertiles of cumulative DDD and CPD.
The outcome of the study was the incident diagnosis of dementia. We defined incident dementia as having at least one hospitalization claim or at least two outpatient care claims with the ICD-10 dementia codes (F00, F01, F02, F03, G30, G31.0, G31. 8 or G31.9). primary diagnosis at least one year after the index date. We did not include patients whose dementia diagnosis code occurred within one year of the indexing date as incident dementia to avoid protopathic bias due to undiagnosed pre-existing dementia and to treat disease latency for drug-induced dementia. We followed patients from the index date until one of the following conditions was met: (1) diagnosis of dementia; (2) the patient was excluded from the HIRA database for reasons such as death; or (3) the date of August 31, 2018.
We defined the following factors as potential confounders: (1) the existence of comorbidities, assessed as having at least one inpatient or at least two outpatient ICD-10 comorbidity codes (Table 1) before the date of indexing; and (2) use of prescribed co-medications (Table 1) for at least one week prior to the index date. Using comorbidities, we estimated the Charlson Comorbidity Index (CCI) score29.
We compared the risk of dementia in the VALP, LITH, and BOTH groups to that of the NONE group using a Cox multivariate proportional hazard model. In this analysis, we adjusted results for confounders that showed a significant association with dementia risk in univariate Cox proportional hazard models (eTable 1 in Supplement).
We then categorized the DDD and CPD of valproate and lithium into tertiles, estimated the risk of dementia in each tertile using Cox’s multivariate proportional hazard models, and adjusted the results for the aforementioned confounders.
We also performed similar analyzes after changing the definition of incident dementia as a code for patients with dementia in the primary code and prescribing a cognitive enhancer for 7 or more days. All statistical analyzes were performed using SAS Enterprise 7.1 for Windows (SAS Institute Inc., Cary, NC, USA) and considered p values less than 0.05 are statistically significant.
This study was approved by the Institutional Review Board of Bundang Hospital, Seoul National University (IRB no. X-1906-546-901) and complied with the ethical guidelines of the Declaration of Helsinki of the World Medical Association.
Consent to participate
Informed consents were overruled by the Institutional Review Board of Seoul National University Bundang Hospital due to retrospective design.