Data sources
This retrospective population‐based cohort study used the Longitudinal Health Insurance Database 2005 (LHID 2005) released by the Taiwan National Health Research Institutes (NHRI) for research purposes. The National Health Insurance (NHI) Program implemented on March 1, 1995, covers more than 99% of Taiwan's population of 23.74 million population. The LHID 2005 consists of a random sample of 1 million, and includes demographic data of enrollees; service records and expenditure claims from outpatient, inpatient, and ambulatory care; and data associated with contracted pharmacies for reimbursement purposes. The International Classification of Disease, 9th Revision, Clinical Modification (ICD-9-CM) codes was used to identify diseases in this study. The accuracy of diagnoses in the NHIRD has been verified in previous articles [22-24]. This study was approved by the Institutional Review Board of the Ditmanson Medical Foundation, Chia-Yi Christian Hospital, Taiwan (CYCH-IRB No: 2018078).
Study population
We conducted a retrospective cohort study covering the period from January 1, 2000 to December 31, 2013. We selected subjects 20 years of age or over with a first diagnosis of IDA (ICD-9-CM: 280) from the LHID 2005 between January 1, 2000 and December 31, 2012. Excluded were patients diagnosed with anxiety disorders, depression, psychotic disorders, bipolar disorders, sleep disorders, RLS, or dementia before 2000, or before their first visit for IDA. In order to increase the validity of IDA diagnoses, this study only included cases that had at least two diagnoses of IDA in their medical claims prior to their index date as IDA cases. Supplementary iron medication data were also collected. Detailed iron supplementation classification is shown in Supplementary Data, Table S1. Patients who had taken any oral iron between the investigation follow-up periods were defined as iron users; the remaining subjects were defined as iron non-users. Individuals with missing data and those who were diagnosed without blood tests were excluded. Moreover, on the basis of the clinical guidelines and health insurance regulations of the NHI, patients suspected of having IDA might receive a diagnosis of unspecified anemia (ICD-9-CM: 285) on their first visit. However, in order to confirm the diagnosis of IDA, patients underwent laboratory testing for decreased serum iron and ferritin, and increased total iron binding capacity. We retrieved the non-IDA subjects for the comparison cohort from the remaining insured people among the LHID 2005. Individuals in the comparison cohort were individually matched with those in the IDA cohort at a 2:1 ratio based on age, sex, index year, and the year of IDA diagnosis. Individuals with a diagnosis of psychiatric disorders prior to the index date were excluded. A total of 38,794 non-IDA subjects were included in this study.
Main outcome
Patients in both the IDA and non-IDA groups were followed up from the index date until the end of December 31, 2013, or until one of the following events occurred: diagnosis with psychiatric disorder, including anxiety disorders (ICD-9-CM: 300), depression (ICD-9-CM: 296.2-296.3, 300.4 and 311), psychotic disorders (ICD-9-CM: 295 and 297-298), bipolar disorders (ICD-9-CM: 296.0, 296.4-296.8), sleep disorders (ICD-9-CM: 307.4 and 780.5), RLS (ICD-9-CM: 333.90 and 333.99), and dementia (ICD-9-CM: 290, 294.1, and 331.0), withdrawal from the NHI program, or death, whichever came first. Moreover, anxiety disorders, depression, psychotic disorders, bipolar disorders, sleep disorders, RLS, and dementia were also extracted as outcome variables of interest, separately. As with the main outcome, all of the subjects were also followed until withdrawal from insurance, occurrence of events, or until December 31, 2013.
Baseline characteristics and comorbidities
The general characteristics of individuals were age, gender, and insurable salary (in New Taiwan Dollars [NT$]; <19,100, 19,100 – 41,999, ≥ 42,000). The present study used the urbanization stratification of Taiwan townships developed at Taiwan's NHRIs. This index was derived from a cluster analysis of five indicators: population density, percentage of population with college or greater educational level, percentage of population aged 65 years or over, percentage of population working in agriculture, and density of physicians per 100,000 people. The 368 townships in Taiwan were classified into seven levels of urbanization except for the isolated isles in Kinmen and Lienchiang counties. We further classified the urbanization levels as urban (levels 1 and 2), suburban (levels 3 and 4), rural (levels 5–7) and the isolated isles as remote areas [25]. The covariates of comorbidities that were selected in this study included hypertension (ICD-9-CM: 401 – 405), diabetes mellitus (DM, ICD-9-CM: 250), dyslipidemia (ICD-9-CM: 272), hyperthyroidism (ICD-9-CM: 242), hypothyroidism (ICD-9-CM: 244), chronic obstructive pulmonary disease (COPD, ICD-9-CM: 490 – 496), stroke (ICD-9-CM: 430 – 438), coronary artery disease (CAD, ICD-9-CM: 410 – 414), chronic kidney disease (CKD, ICD-9-CM: 585) and liver cirrhosis (ICD-9-CM: 571.2, 571.5, and 571.6).
Statistical analysis
Demographic characteristics were expressed using means and standard deviations for continuous variables, presented as numbers and percentages for categorical variables. The differences in continuous variables were estimated using t-tests, and differences between categorical variables were analyzed using the chi-square test or Fisher exact test, as appropriate. The incidence rate was calculated as the number of first diagnoses of psychiatric disorders per 1,000 person-years. Univariate and multivariate Cox proportional hazards models were used to calculate hazard ratios (HRs) and 95% confidence interval (CI) for developing outcomes (including overall events and dementia, anxiety disorders, depression, bipolar disorders, sleep disorders, RLS, and psychotic disorders, respectively). Multivariate Cox proportional hazards models were used to explore the associations between IDA and risk of psychiatric disorders, controlling for age, gender, and medical comorbidities. The Kaplan–Meier method and log-rank test were used to estimate the cumulative risks of psychiatric disorders between the IDA and non-IDA groups. A 2-tailed p < 0.05 was considered significant. The SPSS for Windows version 21.0 (IBM, Armonk, NY, USA) was used for the statistical analysis of the results. Statistical graphs were plotted with R version 3.5.1, with the KMsurv, survfit and survival packages. Stata statistical software (version 15; StataCorp, College Station, TX, USA) was used to calculate the power. The power for survival data calculation was estimated with the Stata command stpower log-rank and the set up conditions were a sample of at least 19,397 patients, an effect size of 1.50 (expressed as an HR), and an α of 0.05 with a 2-sided test. The statistical power was estimated to be more than 99% and would be able to detect any significant difference in the two groups.