Study setting and participants
A cross-sectional observational study was conducted using all ED visit records of the Korea Health Insurance Review Agency (HIRA). HIRA is a government organization that reviews and assesses medical claims for the National Health Insurance and maintains a database of medical records for the entire Korean population. HIRA provided identifiable information that was specific to each individual. These identifiable information mbers anonymously corresponded to record files, which were utilized to provide a comprehensive picture of health care use and patient diseases. HIRA records showed that a total of 10,599,311 patients visited EDs in Korea from 2016 to 2017. It should be noted that HIRA data does not include patients who receive car insurance or worker’s compensation. Due to this, the number of ED patients accounted for in this study was less than the actual number of ED patients.
Definition Of Less-frequent, Frequent, And Highly Frequent ED Users
To measure ED visit frequency, each patient’s most recent ED visit during the 2017 fiscal year was used as a reference point. In order to create each patient’s total ED visit frequency, the ED visits that preceded the most recent visit were counted, beginning from one calendar year (365 days) before the reference point. “Frequent ED users” were defined as patients who visited an ED more than 7–17 times per calendar year and “highly frequent ED users” were defined as patients who visited an ED eighteen or more times during the same period.15 Patients who were under 18 years old were excluded in this study, since their diagnoses contributed to less than 100 patients.
Selection Of Risk Factors
Risk factors associated with ED use were chosen according to a literature review and grouped by patient demographic, illness and disposition. Demographic factors (patient age, sex, insurance status) were also included and classified according to the most recent ED visit. Patient diagnoses were based on the 10th revision of International Statistical Classification of Disease (ICD-10) code algorithms. In the ICD-10, patient death was not classified as a separate group, so it was included in the discharge group.
Grouping And Ranking Of Patient Diagnoses
Patients were labeled under chronic disease if they had any of the following ICD-10 codes: C22, R18, K70, 74 (for liver diseases); J44, 45 (for lung diseases); I10, 20 (for cardiovascular diseases) or E11, 14, 87, N18 (for renal and endocrine diseases). Similarly, patients were labeled under mental health disorders if they had codes F10, 32, 41 or G47; under trauma-related injury if they had codes S01 (for trauma to the head and face); S22, 33 (for trauma to the thorax and abdomen) and S43, 61, 80, 83, 91, 93, T14 (for trauma to extremities). Patients were categorized as having joint pain-related disorders if emergency physicians reported any of the following ICD-10 codes as the main diagnosis of the ED visit; M10, 13, 17, 19, 25, 75 (for arthritis); M79, 89 (for musculoskeletal disease) and G43, 44, 47, 56, M47, 48, 50, 51, 54 (for neuropathy). Relatively simple diseases were also grouped based on their code and diagnosis: J00, 02, 03, 06, 20, 30, 40 (for upper respiratory infection); K52, 59 (for acute gastroenteritis); L23, 50 (for urticaria/dermatitis) and Z48 (for a follow-up visits including wound dressing changes). Each disease was ranked by prevalence to observe the characteristic diagnoses of each ED frequency group (less frequent, frequent, and highly frequent ED user group).
Primary Data Analysis
To identify the risk factor of frequent and highly frequent ED visits, two separate logistic regression models were constructed. The first model compared frequent ED users with less frequent ED users, while the second model compared highly frequent ED users with frequent ED users. For logistic regression analyses, patient-level demographic factors such as age, gender, type of insurance and disease categories such as chronic disease, pain-related disease, mental disorder, trauma-related disease, and minor cases were included. Standardized ß values were used to rank risk factor importance; this process can directly translate the size of these values by converting the unit ß values into standardized units. Assuming equal variance, the standardized odds ratio is affected by the effect and sample size. The methods created by Agresti were used for the standardized odds ratios. All analyses were performed with SPSS (version 18.0; IBM SPSS Statistics, Chicago, IL).