Diagnostic triage for ultrasound-detected subclinical insulin injection–related lipohypertrophy in older adults with diabetes: development and internal validation
4 affiliations
- 1Department of Endocrinology, Beijing Hospital
- 2National Center of Gerontology and Institute of Geriatric Medicine, Chinese Academy of Medical Sciences
- 3Department of Ultrasound Medicine, Beijing Hospital
- 4Department of Nursing, Beijing Hospital
Abstract
Background: Subclinical insulin injection–related lipohypertrophy can remain undetected after routine injection-site examination. We developed and internally validated a model to prioritize confirmatory ultrasound in older adults with diabetes after negative visual inspection and palpation.
Methods: This single-center cross-sectional study included 818 adults aged 60 years or older from the endocrinology outpatient clinics and inpatient wards of Beijing Hospital between January 2025 and January 2026; 129 had ultrasound-detected subclinical lipohypertrophy. An 11-predictor ridge logistic model combined sex, age, body mass index, insulin injection duration, daily injection frequency, needle reuse, site rotation, injection area, immediate injection after refrigeration, glycated hemoglobin, and weekly glucose-variability frequency. It was compared with a 9-predictor core model that excluded the two metabolic markers. Internal validation used 2,000 bootstrap resamples and repeated nested 10-fold cross-fitting. Out-of-bag bootstrap validation assessed thresholds targeting 90% and 95% sensitivity.
Results: Subclinical lipohypertrophy was detected in 15.8% (95% CI, 13.4%–18.4%). The core model had an optimism-corrected receiver operating characteristic (ROC) area under the curve of 0.758 (95% interval, 0.720–0.808) and Brier score of 0.117 (0.100–0.130). The extended model had an optimism-corrected ROC area under the curve of 0.774 (95% interval, 0.741–0.823), precision-recall area under the curve of 0.423 (0.353–0.527), and Brier score of 0.113 (0.097–0.126). At the validated thresholds, mean sensitivities were 89.7% and 94.4%, and 337 and 220 ultrasound examinations per 1,000 patients, respectively, were avoided.
Discussion: The model stratified the probability of ultrasound-detected subclinical lipohypertrophy after a negative routine examination. Its high-sensitivity operating points quantify the balance between ultrasound use and missed lesions, providing a basis for evaluating selective ultrasound referral in older adults receiving insulin. The core model warrants evaluation as a lower-burden option. Temporal and multicenter geographic external validation is required before clinical software deployment or widespread protocol adoption.
Related
- Can maternal characteristics improve due date prediction without ultrasound? Development and internal validation of a machine-learning prediction model in 600 000 pregnancies in Sweden
- An interpretable, formally verified point-of-care ultrasound risk equation for difficult videolaryngoscopy: development and internal validation
- Development and validation of a clinical prediction model for diagnosing nephrotic syndrome in adults: a retrospective diagnostic prediction study
- Multimodal integration of carotid plaque ultrasound radiomics and dynamic systemic immune-inflammation trajectories for subacute risk stratification after stroke thrombectomy: development and internal validation of an exploratory prognostic model
- Multicenter development and validation of a serum microRNA-based diagnostic model for breast cancer and its association with tumor burden and biological aggressiveness: an observational case-control study.
BibTeX
@article{yu2026diagnostic,
title = {{Diagnostic triage for ultrasound-detected subclinical insulin injection–related lipohypertrophy in older adults with diabetes: development and internal validation}},
author = {Shuyi Yu and Yunyu Wang and Weihao Wang and Zhiwen Liu and Quanying Wu},
journal = {Frontiers in Endocrinology},
year = {2026},
month = oct,
doi = {10.3389/fendo.2026.1997042},
url = {https://www.frontiersin.org/articles/10.3389/fendo.2026.1997042/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Diagnostic triage for ultrasound-detected subclinical insulin injection–related lipohypertrophy in older adults with diabetes: development and internal validation
AU - Shuyi Yu
AU - Yunyu Wang
AU - Weihao Wang
AU - Zhiwen Liu
AU - Quanying Wu
PY - 2026
DA - 2026/10/21
JO - Frontiers in Endocrinology
DO - 10.3389/fendo.2026.1997042
UR - https://www.frontiersin.org/articles/10.3389/fendo.2026.1997042/full
AB - Background: Subclinical insulin injection–related lipohypertrophy can remain undetected after routine injection-site examination. We developed and internally validated a model to prioritize confirmatory ultrasound in older adults with diabetes after negative visual inspection and palpation. Methods: This single-center cross-sectional study included 818 adults aged 60 years or older from the endocrinology outpatient clinics and inpatient wards of Beijing Hospital between January 2025 and January 2026; 129 had ultrasound-detected subclinical lipohypertrophy. An 11-predictor ridge logistic model combined sex, age, body mass index, insulin injection duration, daily injection frequency, needle reuse, site rotation, injection area, immediate injection after refrigeration, glycated hemoglobin, and weekly glucose-variability frequency. It was compared with a 9-predictor core model that excluded the two metabolic markers. Internal validation used 2,000 bootstrap resamples and repeated nested 10-fold cross-fitting. Out-of-bag bootstrap validation assessed thresholds targeting 90% and 95% sensitivity. Results: Subclinical lipohypertrophy was detected in 15.8% (95% CI, 13.4%–18.4%). The core model had an optimism-corrected receiver operating characteristic (ROC) area under the curve of 0.758 (95% interval, 0.720–0.808) and Brier score of 0.117 (0.100–0.130). The extended model had an optimism-corrected ROC area under the curve of 0.774 (95% interval, 0.741–0.823), precision-recall area under the curve of 0.423 (0.353–0.527), and Brier score of 0.113 (0.097–0.126). At the validated thresholds, mean sensitivities were 89.7% and 94.4%, and 337 and 220 ultrasound examinations per 1,000 patients, respectively, were avoided. Discussion: The model stratified the probability of ultrasound-detected subclinical lipohypertrophy after a negative routine examination. Its high-sensitivity operating points quantify the balance between ultrasound use and missed lesions, providing a basis for evaluating selective ultrasound referral in older adults receiving insulin. The core model warrants evaluation as a lower-burden option. Temporal and multicenter geographic external validation is required before clinical software deployment or widespread protocol adoption.
SN - 1664-2392
ER -