Internal validation supports six-month response model for infantile hemangioma

A prediction model developed for children with infantile hemangioma who had been selected to receive oral propranolol combined with intralesional polidocanol showed strong internal discrimination and calibration in a cohort of 675 children. The model estimates each child’s probability of an effective response at six months using 10 baseline factors, but prospective external validation is needed before routine clinical use.

Overall, 560 of the 675 children, or 83.0%, had an effective response at six months. The participants were aged 24 months or younger when combined treatment began and were treated during the study period of 2012–2025. Because every child in the development cohort had already been selected for combination therapy, the results do not compare propranolol with polidocanol or establish which treatment a child should receive.

Researchers addressed missing baseline data with 20 multiply imputed datasets. They used LASSO in repeated analyses across those imputations, including a 200-resample stability comparison, to evaluate candidate predictors. A final 10-predictor specification was then locked before a separate internal validation using 1,000 conditional bootstrap resamples. This separation meant that the final model was not altered after the validation assessment was initiated.

The locked Stable10 model incorporated lesion location, premature birth, age at the start of combined therapy, lesion depth, lesion type, baseline echocardiogram pattern, treatment pathway, maximum lesion diameter, lesion thickness and blood-flow grade. For the pooled, shrunken coefficient formula used in the accompanying calculator, the apparent area under the receiver operating characteristic curve, or AUC, was 0.920. The bootstrap optimism-corrected AUC was 0.896, with a conditional location-shifted 95% bootstrap interval of 0.886–0.926.

Measures of calibration also indicated agreement between model predictions and observed outcomes. The corrected Brier score was 0.095, with a 95% interval of 0.075–0.102. The calibration intercept was 0.023, with a 95% interval of −0.028 to 0.064, while the calibration slope was 0.949, with a 95% interval of 0.914–1.018. The model was designed to estimate a continuous probability of response at six months rather than classify children only into fixed response categories.

In a separate analysis of additional-injection burden across the full cohort of 675 children, investigators used robust Poisson regression. Surface elevation was associated with a higher rate of additional injections during the first six months. The adjusted incidence rate ratio was 1.27, with a 95% confidence interval of 1.01–1.58 and P = 0.039. The authors described this finding as exploratory; it is an association within the treated cohort and does not show that surface elevation caused additional injections or that changing treatment would reduce them.

The results support the model’s internal performance in the population used to develop it, but they do not establish how the calculator would perform in children treated at other hospitals, in different populations or under different treatment pathways. Internal validation evaluates whether the model retains its performance when tested within the development data; external validation is required to assess performance in independent settings and populations.

For now, the calculator should be regarded as a research tool rather than a routine clinical decision aid. The study’s central result is that a baseline model built from 10 child and lesion characteristics can estimate six-month response with good discrimination and calibration among children receiving the selected combination therapy. Whether those estimates remain accurate outside the development cohort remains an open question that requires prospective external evaluation.

Subject of Research: Internal validation of a six-month treatment-response prediction model for children with infantile hemangioma

Article Title: Development and internal validation of a prediction model for 6-month response in children with infantile hemangioma selected for propranolol and polidocanol injection

Article References: Liu, X., Sun, Y., Wang, Y., Zhang, Y., Mu, W., & Li, A. (2026). Development and internal validation of a prediction model for 6-month response in children with infantile hemangioma selected for propranolol and polidocanol injection. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07874-4

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07874-4

Keywords: infantile hemangioma, prediction model, internal validation, propranolol, polidocanol, treatment response, pediatrics, lesion characteristics, injection burden, Development, internal, validation