Scholarlink

Background <p>Steroid-induced diabetes mellitus (SDM) is a significant complication of systemic glucocorticoid (GC) therapy, particularly among dermatology inpatients requiring high-dose or prolonged treatment.</p> Objective <p>Early identification of individuals at increased SDM risk is crucial for prevention and optimized clinical management.</p> Methods <p>We retrospectively analyzed 293 dermatology inpatients receiving systemic GCs (August 2019 to October 2024), among whom 42 developed SDM. Candidate predictors from demographic, clinical, and laboratory variables were filtered by LASSO, followed by multivariable logistic regression. Performance was assessed by ROC, calibration (curve and Hosmer–Lemeshow), and decision curve analysis (DCA), with internal validation via a tenfold cross-validation and bootstrap resampling. A 1:3 age-matched sensitivity analysis (<i>n</i> = 168) used conditional logistic regression with prevalence-recalibrated probabilities for evaluation. This study involved the development of a predictive model with internal validation only; external validation was not performed.</p> Results <p>Four variables—family history of diabetes, immunosuppressant usage, average daily GC dose, and triglyceride level—were retained as independent predictors. The model showed excellent discrimination (AUC = 0.860, 95% CI 0.794–0.926) and good calibration (Hosmer–Lemeshow, <i>p</i> = 0.2904, Brier 0.075, scaled Brier 38.7%); mean cross-validated AUC was 0.839. DCA demonstrated consistent net benefit across wide threshold probabilities (0.05–1.00). Findings were robust in the age-matched subset (AUC = 0.847, 95% CI 0.776–0.918), with weighted Brier 0.082 and a calibration slope of 1.014.</p> Conclusions <p>A concise LASSO-based model incorporating four key predictors accurately predicts SDM among hospitalized dermatology patients. The resulting nomogram facilitates individualized risk assessment, enabling proactive monitoring and tailored therapeutic interventions. However, external validation through prospective, multicenter studies is necessary to evaluate its generalizability before broader clinical implementation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and internal validation of a LASSO-based nomogram for predicting steroid-induced diabetes mellitus in dermatology inpatients

  • 期刊论文

  • 发表日期：

    2025年11月14日

查看全文
  • Shantao Qiu,
  • Huimin Tang,
  • Guan Jiang

摘要

Background

Steroid-induced diabetes mellitus (SDM) is a significant complication of systemic glucocorticoid (GC) therapy, particularly among dermatology inpatients requiring high-dose or prolonged treatment.

Objective

Early identification of individuals at increased SDM risk is crucial for prevention and optimized clinical management.

Methods

We retrospectively analyzed 293 dermatology inpatients receiving systemic GCs (August 2019 to October 2024), among whom 42 developed SDM. Candidate predictors from demographic, clinical, and laboratory variables were filtered by LASSO, followed by multivariable logistic regression. Performance was assessed by ROC, calibration (curve and Hosmer–Lemeshow), and decision curve analysis (DCA), with internal validation via a tenfold cross-validation and bootstrap resampling. A 1:3 age-matched sensitivity analysis (n = 168) used conditional logistic regression with prevalence-recalibrated probabilities for evaluation. This study involved the development of a predictive model with internal validation only; external validation was not performed.

Results

Four variables—family history of diabetes, immunosuppressant usage, average daily GC dose, and triglyceride level—were retained as independent predictors. The model showed excellent discrimination (AUC = 0.860, 95% CI 0.794–0.926) and good calibration (Hosmer–Lemeshow, p = 0.2904, Brier 0.075, scaled Brier 38.7%); mean cross-validated AUC was 0.839. DCA demonstrated consistent net benefit across wide threshold probabilities (0.05–1.00). Findings were robust in the age-matched subset (AUC = 0.847, 95% CI 0.776–0.918), with weighted Brier 0.082 and a calibration slope of 1.014.

Conclusions

A concise LASSO-based model incorporating four key predictors accurately predicts SDM among hospitalized dermatology patients. The resulting nomogram facilitates individualized risk assessment, enabling proactive monitoring and tailored therapeutic interventions. However, external validation through prospective, multicenter studies is necessary to evaluate its generalizability before broader clinical implementation.