Prevalence, Risk Factors, and Outcomes of Gout Flare in Patients Hospitalized for PCR-Confirmed COVID-19: A Multicenter Retrospective Cohort Study

METHODS

Study design and population. This retrospective cohort study included adults with comorbid gout who were hospitalized for COVID-19 confirmed on PCR between March 1, 2020, and December 31, 2021, in 3 hospitals in the Bangkok metropolitan area: Thammasat University Hospital, Thammasat Field Hospital, and Vajira Hospital.

Included study participants met all the following inclusion criteria: (1) age ≥ 18 years, (2) hospital admission with COVID-19 (confirmed on PCR on the first day of admission) as the primary admission diagnosis, and (3) having gout as a comorbid disease at the time of hospital admission. Comorbid gout was defined as having received a gout diagnosis by a doctor before the current admission, according to previous outpatient records, previous hospital discharge notes, or referral letters. Patients were excluded from analysis if they were hospitalized with gout as the primary admission diagnosis or if diagnosed with gout for the first time during the current hospital admission.

Patient identification. Eligible patients were identified from the hospitals’ electronic databases in 2 steps. The first step aimed to identify patients hospitalized for COVID-19. Patients who received International Classification of Diseases, 10th Revision (ICD-10) code for COVID-19 (U071) as the primary diagnosis during the study period were extracted from the hospital databases. Electronic records for Thammasat University Hospital and the field hospital were kept in the same database managed by Thammasat University, whereas electronic records from Vajira Hospital were maintained by Navamindradhiraj University. The second step was to identify patients with comorbid gout among the previously identified COVID-19 cohort. Comorbid gout was identified if the participants received discharge ICD-10 codes for gout (M10) as comorbid diagnosis or if there was mention of any gout key words (gout, tophus, podagra, allopurinol, febuxostat, or colchicine) in the participant’s electronic discharge letter. The identification of gout keywords was done manually by research assistants who had been trained to look for gout keywords in hospital discharge letters and had successfully performed similar manual search in another gout study.³ Manual screening of discharge letters was necessary because some participants with inactive gout might not have received ICD-10 code for gout at their hospital discharge but gout would still have been mentioned as comorbidity in the letters.

Patient ascertainment. Discharge letters, hospital records, and referral letters of all participants identified from the initial 2-step screening were subsequently reviewed in detail by the investigators (KJ and PS) to ensure that the participants met all inclusion criteria and none of the exclusion criteria. During this process, the investigators made sure that the definition of comorbid gout was followed (physician-diagnosed gout). Patients who were initially identified by the key words “podagra,” “colchicine,” “allopurinol,” or “febuxostat” in the discharge letters would be included in the analysis only if they received explicit gout diagnosis by a doctor in the same discharge letter, or in the hospital record or referral letter associated with the same hospital event.

Variables and their definitions. Gout flare is defined as an episode of joint pain and swelling that developed during hospital stay and was judged to be gout by the attending doctor based on hospital notes, or an episode of joint pain that satisfied the definition of gout flare by Gaffo et al.¹³ Duration of gout flare episode was defined as the number of days between the onset of acute joint pain and swelling and the resolution of gout flare or date of hospital discharge.

Other variables were divided into 5 domains: demographics (age, sex), admission data (length of stay, comorbidities), COVID-19 data (pneumonia, oxygen support, invasive ventilation, COVID-19 treatments), gout history (tophus, preadmission gout medication, in-hospital adjustment of gout medications, SU level), and gout flare episodes (joint number and location, episode duration, treatment). Comorbid diseases included obesity, smoking status, and comorbid conditions from the Charlson Comorbidity Index (CCI), which has been associated with poor outcomes in patients hospitalized for COVID-19.¹⁴ Table 1 lists all the variables collected for this study and their descriptions.

Databases and data collection. Electronic and scanned copies of hospital notes were available for review at each participating hospital. SU measurement was performed in-house by the hospital where participants were admitted, and the results were available in the same database as the electronic hospital notes. Data were extracted by the investigators (KJ and PS) and recorded in a predetermined case record form. If a participant had more than 1 COVID-19 admission during the study period, only the first admission was recorded. If a participant developed gout flare more than once during the same hospital event, only data prior to the first gout flare episode were recorded.

Statistical analysis. Characteristics of the cohort and prevalence of gout flare were described with descriptive statistics. Categorical data were expressed as frequency and percentage, and continuous data were expressed as mean (SD) or median (IQR). Comparison between patients who developed gout flare (flare group) and patients who did not develop flare (nonflare group) were made using the chi-square test for categorical variables and the Mann-Whitney U test for continuous variables.

Multivariate logistic regression model. Factors associated with the development of inpatient gout flare (dependent variable) were explored using multivariate logistic regression. Fourteen exposure variables were considered for the logistic regression, including traditional predictors of gout flare (GOUT-36 prediction rule, no ULT, no prophylaxis, tophus, SU > 6 mg/dL, ULT adjustment, and prophylaxis stopped/decreased), COVID-19-related variables (COVID-19 pneumonia, invasive ventilation, corticosteroids, favipiravir), demographics (age and sex), and CCI. These variables were first screened by least absolute shrinkage and selection operator (LASSO). LASSO is a type of penalized regression procedure augmented by 10-fold cross validation that applies penalty factor (λ) to the variables’ regression coefficients. Variables that hold little to no discriminatory power would have their regression coefficients shrunk to 0 and removed.¹⁵ LASSO was used for variable selection over the traditional univariate logistic regression because LASSO was more robust in handling a dataset with a small number of outcome events as well as a dataset whose variables showed some degree of correlation (multicollinearity).^15,16 Variables selected by LASSO were subsequently fitted into a logistic regression model. Regression coefficients, odds ratios (ORs) and their 95% CIs, and P values were reported.

Two performance indicators were used to assess the logistic regression model: C-statistics and the Hosmer-Lemeshow test. C statistics indicate the ability to distinguish patients with flare from patients without flare (discrimination). The C statistics range between 0.5 and 1.0, with the latter indicating perfect discrimination. The Hosmer-Lemeshow test indicates how well the predicted event rates match the observed event rates (goodness of fit). A Hosmer-Lemeshow test with a nonsignificant P value (≥ 0.05) suggests that there is no evidence of poor model fit in the dataset.

Multivariate linear regression model. We explored the association between gout flare and the length of hospital stay using multivariate linear regression analysis. The primary outcome (dependent variable) was hospital length of stay (in days) and the exposure variable was inpatient gout flare (yes or no), adjusting for age, sex, CCI, presence of COVID-19 pneumonia, and invasive ventilation. Regression coefficients, their 95% CIs, and P values were presented. We also reported the coefficient of determination (R²) of the linear regression model, which indicated the degree of variance in the hospital length of stay that could be explained by the model.

All statistical analyses were performed using IBM SPSS Statistics software (version 25), except the LASSO which was performed in RStudio (version 2022.02.3+492; RStudio Team).

Sample size estimation. Initial database inquiries estimated the number of patients hospitalized for PCR-confirmed COVID-19 at 8000 patients in 3 participating hospitals. Since the prevalence of gout in the general population ranges between 1% and 4%, we anticipated that at least 80 patients (1%) hospitalized for COVID-19 would have comorbid gout.¹ Assuming that 34% of hospitalized patients with comorbid gout developed gout flare, we anticipated gout flare in 27 patients in our cohort.³ The event number would be sufficient to support the final logistic regression model that included no more than 5 exposure variables (5 events for each variable).¹⁷

Ethics approvals. The study protocol was reviewed and approved by the Human Research Ethics Committee of Thammasat University (reference no. MTU-EC-IM-1-321/64) and the Institutional Review Board of the Faculty of Medicine Vajira Hospital (reference no. 265/64E) in compliance with the Declaration of Helsinki. This was a retrospective, chart review cohort study for which informed consent was not required.