Complication Burden and Demographic Trends in Autologous Breast Reconstruction Outcomes: Insights from the NIH All of Us Program.

Abstract

Background: While the risks and benefits of autologous breast reconstruction have been widely examined, nationally representative, longitudinal data on complication burden, flap utilization trends, and patient-level risk factors remain limited. The NIH All of Us Research Program provides an opportunity to address these gaps using a diverse, population-scale cohort.

Methods: We identified 260 patients who underwent autologous breast reconstruction using current procedural terminology codes within the All of Us Registered Tier Dataset (1995-2025). Complications were tracked at 30 days and 1 year postoperatively. Logistic, multivariate regressions, and Kaplan-Meier analyses evaluated predictors and timing of complications. Unsupervised machine learning via k-means clustering was utilized to uncover phenotypic subgroups by age and body mass index (BMI).

Results: Deep inferior epigastric perforator flap utilization increased over time, particularly among younger patients. Complication rates did not significantly differ across flap types. BMI > 32.7 kg/m² was associated with increased 30-day complications, while age and race were not independent predictors. Chronic pain and persistent postoperative pain were the most common 1-year complications. Flap failure occurred in fewer than 2% of cases. Clustering revealed three patient subgroups with distinct complication profiles; older patients and those with higher BMI experienced greater morbidity but maintained high flap success rates.

Conclusion: Autologous breast reconstruction is broadly effective across diverse patient populations. Complication risk is more strongly influenced by BMI than by age or race. Chronic pain emerged as a common long-term morbidity, underscoring the need for improved detection and management efforts. The diversity, depth, and follow-up available through All of Us enable nuanced insights into reconstructive outcomes not possible with traditional datasets.