The incident gout cohort was identified from the Chang Gung Research Database, which contains the electronic health records of all patients who visited any one of the following seven Chang Gung Memorial Hospitals: Keelung, Taipei, Linko (headquarters), Taoyuan, Yunlin, Chiayi and Kaohsiung. Chang Gung Memorial Hospital is the biggest private medical center in Taiwan and services approximately 7% of all patient visits in Taiwan. Valid internal patient record linkage is achieved by using unique patient identifiers. To ensure confidentiality, these identifiers are encrypted before the data is released to researchers. This study only used anonymized and nontraceable data, and therefore, the need for patient consent was waived.
Study design
We conducted this cohort study to examine the risk of CKD progression among patients with gout associated with different ULDs. People aged ≥ 18 years who had a diagnosis of gout and at least one prescription of a ULD between 2012 and 2015 were identified, and their electronic health record data were collected from their first visit to Chang Gung Memorial Hospital (as early as January 2001) through December 2016 with information regarding patient demographics, outpatient and inpatient diagnoses, medications, laboratory results, operations and procedures. The patient classification is based on the American Rheumatism Association preliminary classification criteria for acute gout [11].
The prescriptions of ULDs (febuxostat, allopurinol and uricosuric agents including benzbromarone, sulfinpyrazone and probenecid) were identified. The patients were classified according to their first ULD (allopurinol users, febuxostat users and uricosuric agent users). The patients prescribed with two or more ULDs were excluded from the analysis. We identified the initial registration date to the Chang Gung Memorial Hospital and index date. The index date was defined as the first date of ULD exposure. Those with CKD stage 5 at the index date or CKD stage progression to stage 5 any time between the initial registration date and the index date and those without data on serum creatinine level during the study period were also excluded. This is to ensure the CKD progression is related to exposure to ULD. The flow chart of cohort selection is shown in Fig. 1. The patients were followed from the index date and censored at the earliest date of the first progression of CKD by one stage, switch to a different ULD, death or the end date of the study (30 June 2016). We then identified those with CKD progression by one stage and followed and censored them at the earliest date of the first recovery of CKD by one stage, death or the end date of the study (30 June 2016).
The CKD stage of each patient was determined at the first ULD prescription (index date). The CKD stage was defined according to the widely used CKD-EPI equation to estimate glomerular filtration rate (eGFR) from serum creatinine as follows [12]:
$$ \mathrm{GFR}=141\ast \min\ {\left(\mathrm{Scr}/\upkappa, 1\right)}^{\upalpha}\ast \max {\left(\mathrm{Scr}/\upkappa, 1\right)}^{\hbox{-} 1.209}\ast 0{.993}^{\mathrm{Age}}\ast 1.018\left[\mathrm{if}\ \mathrm{female}\right]\ast 1.159\left[\mathrm{if}\kern0.5em \mathrm{black}\right] $$
Scr is serum creatinine (mg/dL), κ is 0.7 for females and 0.9 for males, α is − 0.329 for females and − 0.411 for males, min indicates the minimum of Scr/κ or 1 and max indicates the maximum of Scr/κ or 1.
The five stages of CKD and eGFR for each stage were stage 1 with eGFR > 90 mL/min/1.73m2, stage 2 with an eGFR of 60–89 mL/min/1.73m2, stage 3A with an eGFR of 45–59 mL/min/1.73m2, stage 3B with an eGFR of 30–44 mL/min/1.73m2, stage 4 with an eGFR of 15–29 mL/min/1.73m2 and stage 5 with an eGFR < 15 mL/min/1.73m2 or receiving dialysis or renal transplantation.
Confounding variables
We included confounding covariates that were likely to be associated with the risk of a deterioration in renal function. These included age at the diagnosis of gout, gender, Charlson comorbidity index, other comorbidities and medications. The Charlson comorbidity index summarizes 17 diagnostic categories (myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatologic disease, peptic ulcer disease, mild liver disease, moderate or severe liver disease, diabetes mellitus, diabetes mellitus with chronic complications, renal diseases, any malignancy [including leukaemia and lymphoma], metastatic solid tumours and human immunodeficiency virus infection) to represent health status, and it has been shown to be a useful predictor of mortality [13]. Deyo et al. produced a validated version for use with International Classification of Diseases version 9 ICD-9-based databases [14], and we used this version to calculate the index. Medications considered in this study included anti-hypertensive drugs, NSAIDs, insulin, oral hypoglycaemic agents, lipid-lowering agents, aspirin, glucocorticoid and colchicine. NSAIDs and glucocorticoid are mostly in as-needed dose. Laboratory covariates included serum uric acid levels and baseline eGFR.
Statistical analysis
The incidence of renal function impairment was calculated using the number of people with CKD progression by one stage after the index date as the numerator and the person-days of the gout patients with different stages of CKD at the index date as the denominators.
We conducted three different analyses to assess the association between ULD use and CKD progression, recovery or improvement. The first analysis used propensity score (PS) weighting for multiple ULDs [15] and cause-specific Cox proportional hazards model to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) adjusted for aforementioned covariates [16]. Proportionality assumption was examined by the log-log plot. The original cohort of 5860 people was analysed. The PS weighting was conducted using the TWANG package, considering the aforementioned covariates [17]. Inverse probability of treatment weights of propensity scores was used to balance covariates across the ULD groups. The weights for optimal balance were estimated using generalized boosted models with 5000 regression trees.
The second approach used the Cox proportional hazards model to estimate the association between ULDs and CKD progression in the original cohort of 5860 patients. Among patients with CKD progression, we conducted a separate analysis to estimate the incidence of CKD recovery. Cox proportional hazard model were used to estimate HRs (95% CIs) for CKD recovery adjusted for aforementioned covariates which were identified at the time of CKD progression. All statistical analyses were performed using SAS statistical software, V.9.4.