Highlight
• A prognostic model derived from 173,812 UK new users of allopurinol predicts 100‑day risk of severe cutaneous adverse reactions (SCAR) with good discrimination (Harrell’s C 0.82 development, 0.79 validation).
• Major independent predictors: increasing age, chronic kidney disease (CKD) stage (dose‑response across stages 3–5), initial allopurinol dose ≥300 mg, and South/other Asian ethnicity.
• Absolute risks are very low (≈0.04% in both development and validation cohorts), but model may guide shared decision‑making, consider alternative urate‑lowering strategies, or targeted genetic testing in higher‑risk patients.
Background
Allopurinol is the most widely prescribed urate‑lowering drug worldwide and a well‑recognized cause of severe cutaneous adverse reactions (SCAR), including Stevens‑Johnson syndrome (SJS), toxic epidermal necrolysis (TEN), and drug reaction with eosinophilia and systemic symptoms (DRESS). Although SCARs are rare, they carry high morbidity and mortality. Clinicians face a practical challenge when initiating allopurinol: balancing gout management benefits against a small but potentially catastrophic early risk of SCAR.
Risk stratification at the point of prescribing could support decisions about initial dose, alternative agents (for example, febuxostat or uricosurics), and targeted pharmacogenetic testing (HLA‑B*5801) where appropriate. Cipolletta and colleagues developed and externally validated a prognostic model to estimate 100‑day risk of allopurinol‑induced SCAR using linked UK primary care, hospitalisation, and mortality data (CPRD Aurum for development; CPRD GOLD for validation), aiming to provide an evidence base for risk‑tailored prescribing.
Study design
This was a retrospective new‑user cohort study including adults (≥18 years) in England newly prescribed allopurinol between Jan 1, 2001 and Mar 29, 2021. Patients were followed for 100 days after initiation for SCAR recorded in hospitalisation or mortality records. The development cohort included 173,812 patients from CPRD Aurum; the validation cohort included 41,610 patients from CPRD GOLD.
Candidate predictors were chosen a priori, reflecting demographics, comorbidity and prescribing factors: age, sex, ethnicity, CKD stage, initial allopurinol dose, ischaemic heart disease, and heart failure. The authors used multivariable Cox regression, pseudo‑values, penalisation to reduce overfitting, and then externally validated the model. Performance metrics included explained variation (Royston & Sauerbrei’s R2D), discrimination (Harrell’s C), calibration slope, and assessment of clinical utility across a prespecified absolute risk interval (0.0001 to 0.003; i.e., 0.01%–0.3%).
Key findings
Cohort characteristics: In the development sample (n=173,812), mean age was 63.9 years (SD 15.0); 74.3% were male and 88.8% White. In the validation sample (n=41,610), mean age 64.4 years (SD 14.9); 74.0% male and 89.5% White. Observed 100‑day SCAR events were 63 (0.04%) in the development cohort and 16 (0.04%) in the validation cohort — consistent and very low absolute event rates.
Independent predictors (multivariable-adjusted hazard ratios [95% CI]):
- Age: HR 1.03 per year (1.01–1.06).
- Chronic kidney disease: stage 3 HR 2.24 (1.20–4.17); stage 4 HR 6.65 (2.90–15.23); stage 5 HR 18.85 (6.32–56.19).
- Initial allopurinol dose ≥300 mg: HR 5.99 (CI as reported in the paper contains a typographical issue — see text below).
- South Asian ethnicity: HR 5.35 (2.37–12.07); other Asian ethnicity: HR 5.63 (1.34–23.61).
Model performance:
- Development (optimism‑adjusted): R2D = 0.50; Harrell’s C = 0.82.
- External validation: calibration slope 0.93 (95% CI 0.18–1.68); R2D = 0.44 (95% CI 0.20–0.62); Harrell’s C = 0.79 (95% CI 0.71–0.88).
- Clinical utility: authors report decision‑curve analyses showing net benefit across the prespecified absolute risk range (0.0001–0.003).
Interpretation of absolute risk: Observed 100‑day SCAR incidence was about 0.04% (≈4 per 10,000; ~40 per 100,000) in both cohorts. The model is therefore estimating risk for an uncommon but serious event; factors that multiply relative risk can still yield small absolute increases in risk. For example, a baseline 0.04% risk multiplied by an HR of ~6 would predict an absolute risk approaching 0.2% over 100 days in a high‑risk subgroup — clinically meaningful for shared decision‑making.
Note on reported precision: the paper reports an adjusted HR of 5.99 for initial dose ≥300 mg with a confidence interval printed as [3.56–0.08], which appears to be a typographical error in the reported manuscript. The magnitude of association (HR ≈6) is large and consistent with prior observations that higher starting doses increase early SCAR risk, but the exact CI should be checked in the final published erratum or data supplement before clinical application.
Expert commentary and clinical context
Biological plausibility
The associations make biological and pharmacologic sense. Allopurinol is metabolised to oxypurinol, renally cleared; impaired renal function leads to accumulation and higher exposure, plausibly raising immunologic risk. Ethnic differences likely reflect population allele frequencies of genetic risk factors such as HLA‑B*5801, which is more prevalent in some Asian populations and has been strongly associated with allopurinol‑related SCAR in prior genetic studies. Higher initial dosing may create greater early exposure and immune activation.
How this model complements existing strategies
Clinical pharmacogenetics guidelines and expert consensus in some jurisdictions recommend targeted HLA‑B*5801 testing in populations at higher genetic risk (for example, people of Han Chinese, Thai, or Korean ancestry and other groups with elevated allele frequency). This prognostic model uses routinely available clinical data (age, CKD stage, ethnicity, dose) and could operationalise a pragmatic, rapid risk estimate at prescribing without immediate genetic testing. In higher predicted risk patients, clinicians could consider targeted HLA‑B*5801 testing, a lower starting allopurinol dose, alternative urate‑lowering therapy, or closer early monitoring.
Limitations and caveats
Several important limitations affect interpretation and generalisability:
- Low absolute event rate and wide confidence intervals: Despite large cohorts, the rarity of SCAR produced imprecise effect estimates for some predictors (notably later CKD stages and non‑White subgroups), and calibration uncertainty at the extreme ends of risk.
- Outcome ascertainment and misclassification: SCAR events were identified from hospitalisation and mortality records using diagnostic coding. Some events may be missed (primary care‑managed milder reactions) or misclassified. The consequences are unpredictable but could bias estimates if misclassification is differential by predictor.
- Unmeasured confounding — particularly HLA‑B*5801 status: The model did not include genetic markers. HLA‑B*5801 is a strong risk factor; absent genotype data, ethnicity acts as a partial proxy but is imperfect and risks ecological inference.
- External validity beyond the UK: Both development and validation cohorts were UK‑based and predominately White; performance in populations with different ethnic mixes, healthcare systems, or prescribing patterns remains to be established.
- Short time horizon: The model predicts 100‑day risk only. Some/allopurinol reactions may occur later; this model is not designed for long‑term risk prediction.
Clinical implications and recommended use
This prognostic model offers a pragmatic tool to quantify near‑term (100‑day) SCAR risk at the point of allopurinol initiation. Potential clinical applications include:
- Shared decision‑making: present individualized absolute risk estimates to patients when discussing starting allopurinol and alternatives.
- Targeted interventions: in patients with elevated predicted risk, consider lower initial dosing, early safety monitoring (education about prodromal symptoms, earlier review), targeted HLA‑B*5801 testing where available and appropriate, or selection of alternative urate‑lowering drugs.
- Electronic health record integration: embedding the model as a decision support alert in primary care prescribing systems could flag high‑risk patients and prompt risk mitigation steps.
Importantly, because absolute risks are low for most patients, the majority will still have low predicted risk and may be started safely on allopurinol with usual precautions (start low, titrate, patient education). The model is best viewed as an adjunct to clinical judgement rather than a binary rule.
Research and implementation priorities
Next steps to translate this work into practice include:
- Prospective impact studies embedding the model in clinical workflows to assess whether use reduces incidence of SCAR, influences prescribing behaviour, and is acceptable to clinicians and patients.
- Cost‑effectiveness analyses comparing strategies: model‑guided selective HLA‑B*5801 testing versus universal testing versus no testing.
- Validation in non‑UK settings and ethnically diverse cohorts, and exploration of adding HLA genotype where available to improve discrimination and clinical utility.
- Clarification of optimal decision thresholds for different contexts (primary care versus specialist care), informed by patient preferences and risk tolerance.
Conclusion
Cipolletta et al. present a rigorously developed and externally validated prognostic model for the 100‑day risk of allopurinol‑induced SCAR, using large linked UK datasets. Predictive performance was good (Harrell’s C ≈0.8) and the model highlights clinically plausible, actionable predictors — most notably CKD stage, higher starting dose, and Asian ethnicity. Given the low absolute event rate, the tool is most useful for identifying a minority of patients with materially higher near‑term risk in whom clinicians might consider alternative management or targeted genetic testing. Before routine deployment, prospective implementation work, additional validation in diverse populations, and integration with shared decision‑making tools are needed.
Funding
The study was funded by the University of Nottingham.
Reference
Cipolletta E, Nakafero G, Rozza D, Mahil SK, Smith CH, Riley RD, Abhishek A. Development and validation of a prognostic model for predicting the risk of allopurinol-induced severe cutaneous adverse reactions: a retrospective new-user cohort study using linked primary care, hospitalisation, and mortality data. Lancet Rheumatol. 2025 Dec;7(12):e840-e850. doi: 10.1016/S2665-9913(25)00165-1. Epub 2025 Sep 4. PMID: 40915297.
Author note
Patients with lived experience of gout or allopurinol were not involved in the study design but will be engaged in dissemination of results, per the original report.
