Abstract
Gout-related emergency department (ED) visits and hospitalizations have increased significantly, placing a substantial burden on healthcare systems. This study aimed to characterize gout patients at high risk of hospitalization and develop a web-based prognostic model to predict admission likelihood. Using a retrospective cohort of 1,417 patients from a tertiary hospital in Singapore, a multivariable logistic regression model was developed and validated. The final model, incorporating five clinical variables, achieved an AUROC of 0.84, demonstrating high predictive accuracy. This tool can assist clinicians in identifying high-risk individuals for targeted interventions, potentially reducing unnecessary hospitalizations.
Introduction
Gout is a complex inflammatory condition frequently associated with comorbidities such as hypertension, diabetes, and chronic kidney disease. Recent decades have seen a 50–100% surge in gout-related ED visits and hospitalizations in various regions. Managing this surge requires a shift from reactive treatment to proactive risk stratification. By identifying patients at the highest risk of admission during an ED visit, healthcare providers can prioritize specialist reviews and intensive primary care, optimizing resource allocation and improving patient outcomes.
Methodology and Patient Cohort
The study utilized Electronic Health Records (EHR) from a tertiary hospital in Singapore, focusing on patients presenting to the ED with a gout flare between 2015 and 2017. A minimum one-year look-back period was employed to capture medical history and comorbidities.
The final cohort consisted of 1,417 unique patients, predominantly male (82%) with a median age of 56. Notably, 32.5% of these ED visits resulted in hospitalization. The dataset was split into a derivation set (80%) for model building and a validation set (20%) for performance testing.
Model Development and Key Predictors
A multivariable logistic regression model was developed using the derivation set. While various clinical markers were initially considered—including surrogate markers for flare severity like joint radiographs—the final parsimonious model relied on five key clinical variables:
- Age
- Hypertension
- Chronic Kidney Disease (CKD)
- Previous Gout Hospitalization: The strongest predictor, associated with a nearly fivefold increase in admission odds (OR 4.80).
- Acute Gout Treatment (AGT): Interestingly, a current prescription for standby medications (e.g., colchicine, NSAIDs) was a significant protective factor, suggesting better outpatient management capabilities.
Results and Validation
The model demonstrated excellent performance across both training and test datasets. It achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.84, indicating a strong ability to distinguish between patients requiring admission and those who could be safely discharged.
Calibration plots and a Brier score of 0.15 further confirmed that the model's predicted probabilities closely matched actual clinical outcomes. These findings were translated into a web-based risk estimator tool. In practice, hospitalized patients had a median predicted risk score of 0.60, significantly higher than the 0.14 median score for discharged patients.
Discussion and Conclusion
This study provides a validated, simple-to-use tool for risk-stratifying gout patients in the ED. While limited by its retrospective, single-center design, the model's high accuracy and reliance on non-invasive variables make it a promising candidate for clinical implementation. Future work will involve piloting the tool in primary care and outpatient settings to "flag" high-risk patients for early intervention, potentially preventing the need for ED visits and hospitalizations altogether. By targeting resources toward those at highest risk, healthcare systems can improve the long-term health-related quality of life for patients with gout.
Summary updated