Database—National Health Insurance Research Database
In March 1995, the Taiwan government started the National Health Insurance Program and accordingly enacted an insurance policy, which considerably increased the national insurance coverage from 60% to 92%. By 2016, the coverage reached 99.6% [26]. In 2000, the Taiwan government established the National Health Insurance Research Database (NHIRD), which comprises data on all insurants, and these data are processed to being deidentified and released for research [27]. With certificate of ethical approval, the data can be accessed by way of clinician/institute application to the Bureau of National Health Insurance. The NHIRD provides all data generated during the reimbursement for insurance, including diagnosis sets for outpatients and inpatients, prescription drugs and doses, examinations, procedures, surgeries, payments, resident locations, and income levels. The diagnosis sets consist of three codes for outpatient visits and five codes for hospitalization. These codes are based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).
We conducted the study in accordance with the guidelines of the Declaration of Helsinki. This study was exempted from obtaining informed consent from the participants, because the database was nationwide and the data were deidentified. All information of the insurants was unidentifiable, and this study did not violate their rights or adversely affect their welfare. The research was approved by the Institutional Review Board of Chang Gung Memorial Hospital (IRB number: 201601249B1).
Study group—Registry for catastrophic illness patients
In the Taiwan health insurance system, the category of catastrophic illness patients (CIP) has been instituted for a particular group who benefit from the treatment of relevant catastrophic illnesses. The verification of the eligibility for CIP certification requires strict investigation. For instance, a patient who has relevant diagnosis codes for ESRD and has received renal replacement therapy for at least 3 months can apply for CIP certification. Patients with CIP certification for ESRD pay concessional fees for ESRD-associated therapy, including dialysis. Insurants with CIP certification are categorized in the Registry for Catastrophic Illness Patients (RFCIP) associated with the NHIRD.
Using the RFCIP, we defined a study cohort consisting of patients with ESRD who received dialysis from 1 January 1997 to 31 December 2013 in Taiwan. We used the following ESRD-associated ICD-9 codes, which was defined for RFCIP [27]: 585 (chronic renal failure), 586 (unspecific renal failure), 403.01 (malignant hypertensive renal disease with renal failure), 403.11 (benign hypertensive renal disease with renal failure), 403.91 (unspecified hypertensive renal disease with renal failure), 404.02 (malignant hypertensive heart and renal disease with renal failure), 404.03 (malignant hypertensive heart and renal disease with congestive heart failure and renal failure), 404.12 (benign hypertensive heart and renal disease with renal failure), 404.13 (benign hypertensive heart and renal disease with congestive heart failure and renal failure), 404.92 (unspecified hypertensive heart and renal disease with renal failure), and 404.93 (unspecified hypertensive heart and renal disease with congestive heart failure and renal failure). In total, 157,340 patients were included in the study cohort.
Control group—LHID2000 database
For providing adequate data for research, until 2016, the NHIRD published three databases composed of 1,000,000 insurants representing all insurants in Taiwan, and these databases were named Longitudinal Health Insurance Database 2000 (LHID2000), LHID2005, and LHID2010. For example, LHID2000 consists of 1,000,000 insurants in 2000 who were randomly statistically selected (Oracle’s internal random number generator) from all insurants in Taiwan. No statistically significant differences exist in age, sex, or health care costs between the LHID2000 sample group and all enrollees, according to a National Health Research Institutes report [27], and LHID2000 has been used in several population-based studies [28,29,30]. Therefore, we used LHID2000 to generate a control cohort that was matched with the study cohort according to sex, age, urbanization level, income level, DM, and hypertension and analyzed the effects of DNI on the two cohorts.
Main outcome – Incidence of DNI
A DNI was defined as a severe infection in the deep neck that required hospitalization for intensive care. Therefore, while determining the incidence of DNI in the study and control cohorts, we searched for diagnoses during hospitalization that included the following ICD-9 codes: 528.3 (cellulitis and abscess of oral soft tissues; Ludwig angina), 478.22 (parapharyngeal abscess), 478.24 (retropharyngeal abscess), and 682.11 (cellulitis and abscess of neck) [28]. We excluded patients with a previous inpatient diagnosis of DNI before ESRD dialysis to ensure the validity of results concerning the predisposition of ESRD for DNI; consequently, 503 patients were excluded from the study group. After exclusion, 156,837 individuals with ESRD were enrolled in the study group.
Matching process
For each ESRD case, one control without ESRD (1:1) with matching gender, age, urbanization level, income level, DM and hypertension was randomly selected from the LHID2000 database to form a control group. The index date of the study group was the date of registry in the RFCIP for patients with ESRD, and an index date matching that of patients with ESRD was created for the control group. DNI diagnosed before the index date was excluded. A total of 127,283 controls matching the study patients were included. DNI diagnosis codes were used in the two groups to determine the incidence rate of DNI with a follow-up to the end of 2013. The follow-up period was from the index date to the diagnosis of DNI, and some patients were censored because of death (Fig. 1).
Comorbidities
The following comorbidities were defined using ICD-9-CM codes recorded in the claims data: DM (ICD-9-CM code: 250), hypertension (ICD-9-CM codes: 401–405), liver cirrhosis (CD-9-CM codes: 571.2, 571.5–571.6), systemic autoimmune disease (ICD-9-CM codes: 443.1, 446.0, 446.2, 446.4–446.5, 446.7, 696.0–696.1, 710.0–710.4, 714.0–714.4), chronic obstructive pulmonary disease (COPD) (ICD-9-CM codes: 491,492,496), coronary artery disease (CAD) (ICD-9-CM codes: 410–414), and cerebrovascular accident (CVA) (ICD-9-CM codes: 430–438) [28, 31, 32]. Medical comorbidities were included if they appeared one or more times in the diagnoses of inpatients or three or more times in the diagnoses of outpatients. Comorbidities were included if they occurred within 12 months before the index date.
Therapy classification
We assessed the therapeutic methods used for treating patients with DNIs in the two groups. The treatment methods were divided into three subgroups: antibiotics alone, antibiotics with abscess aspiration or drainage, and surgical intervention. The intervention, including abscess aspiration or drainage and surgical debridement, was identified using the claims records during each hospitalization for DNI treatment. If the patients received surgical intervention, they were included in the surgery subgroup, irrespective of the performance of abscess aspiration or drainage. If the patients received abscess aspiration or drainage without surgery, they were included in the aspiration subgroup. If the patients received only antibiotic treatment during hospitalization, they were included in the antibiotic subgroup.
Prognosis evaluation
For evaluating prognosis, we analyzed the duration of hospitalization, care in intensive care units (ICUs), performance of tracheostomy, and mediastinal complications, which were defined according to the receipt of mediastinal surgery during hospitalization or the following ICD-9-CM codes: 510, 513 and 519.2. The mortality and mediastinitis-related mortality were also investigated in both cohorts. Mortality was defined as death occurring during DNI treatment. The mediastinitis-related mortality was defined as death during DNI treatment accompanied by a diagnosis of mediastinitis.
Statistical analysis
The sociodemographics and incidence rate of DNI were compared between the ESRD and control groups using the Pearson’s chi-squared test. The Fisher exact test was used to compare the rates of tracheostomy, mediastinitis, and mortality. The duration of hospitalization was compared using the Student’s t test. The results of Kaplan–Meier analysis revealed the cumulative incidence of DNI in both groups. Furthermore, the analysis was used to investigate the survival outcomes. The risk of DNI for the study group was estimated using a Cox proportional hazard model. All the analyses were performed using SAS software, version 9.4 (SAS Institute, Cary, NC), and the level of statistical significance was set at p < 0.05.