Performance of ‘Triple-D’ and ‘Quadruple-D’ Scores Compared to a Regression-Based Predictive Model for Treatment Outcomes in Extracorporeal Shock Wave Lithotripsy: A 1000-Treatment Series Using the Dornier Compact Delta III Pro
Description
A retrospective cohort study was conducted including 1,000 adult patients undergoing ESWL using the Dornier Compact Delta® III Pro lithotripter at a tertiary hospital in Doha, Qatar, between May 2022 and November 2023. Data on demographics, key stone parameters (size, density, skin-to-stone distance), and treatment details were collected. Key predictors of ESWL treatment failure were evaluated using multivariable logistic regression with internal validation. The predictive performances of the Triple-D, Quadruple-D, and regression-based models were compared using receiver operating characteristic (ROC) analysis, with differences assessed by DeLong’s test. Model calibration and clinical utility were further examined through calibration plots and decision curve analysis (DCA).
Files
Steps to reproduce
Evaluation of Predictive Models for Treatment Outcomes To elucidate predictors of ESWL treatment failure and improve prognostic accuracy, the predictive capabilities of the established ‘Triple-D’ and ‘Quadruple-D’ scoring systems were compared to a bespoke regression-based model developed from this patient cohort. The ‘Triple-D’ score consists of stone Density (HU), stone Diameter (mm), and skin-to-stone Distance (mm). The ‘Quadruple-D’ score adds a fourth component—in this study defined as stone location (renal pelvis versus alternative sites)—to enhance predictive utility (13). The regression model was constructed via multivariable logistic regression with backward elimination and underwent internal validation through k-fold cross-validation to ensure robustness. Statistical Analysis All statistical analyses were conducted using IBM SPSS Statistics version 22 (IBM Corp., Armonk, NY, USA). Continuous variables demonstrating normal distribution are reported as means ± standard deviation (SD), while non-normally distributed data are summarized as medians with interquartile ranges (IQR). Categorical variables are presented as frequencies and percentages. Bivariable analyses employed chi-square or Fisher’s exact tests for categorical data and independent t-tests or Mann–Whitney U tests for continuous variables, as applicable. Variables with a p-value less than 0.20 in bivariable analyses or with established clinical relevance were included in multivariable logistic regression models to identify independent predictors of ESWL failure. Model calibration was evaluated using the Hosmer-Lemeshow goodness-of-fit test (p > 0.05 indicating acceptable fit), and model discrimination was assessed through the area under the receiver operating characteristic curve (AUC). Variables exhibiting multicollinearity (variance inflation factor [VIF] ≥ 5), sparse data (cell counts under 5), or excessive missingness (>15%) were excluded from analyses. Missing data were handled using complete-case analysis due to minimal missingness (<5%). Comparisons of predictive performance among the ‘Triple-D’, ‘Quadruple-D’, and regression-based models were performed using ROC curve analyses, with differences between AUC values assessed via DeLong’s test. Statistical significance was defined as a two-tailed p-value less than 0.05. Decision Curve Analysis (DCA) was conducted to evaluate the clinical utility of these scores by comparing the net benefits across different threshold probabilities (15). Importantly, all analyses conformed to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines to ensure transparent and reproducible reporting of predictive modelling methodologies and results (16).
Institutions
- Qatar University