Development of a Classification Tree to Predict Implant-Based Reconstruction Failure with or without Postmastectomy Radiation Therapy for Breast Cancer.

PURPOSE

The aim of this study was to determine the complications, incidence, and predictors of implant-based reconstruction failure (RF) among patients treated with mastectomy for breast cancer.

METHODS

We retrospectively reviewed 108 patients who underwent mastectomy, tissue expander, and implant-based breast reconstruction with or without radiation therapy (RT) at our institution (2000-2014). Descriptive statistics determined complication incidences, with major complications defined as any complications requiring surgical intervention or inpatient management. Chi square and Fisher's exact tests determined differences in RF incidences, defined as implant loss. Logistic regression analyses identified predictors of RF.

RESULTS

Median follow-up was 42.5 months. Sixty patients (55.6%) experienced major complications. Overall, 27 patients (25%) experienced RF. Incidences of RF were significantly increased in patients who had any major complication (43.3% vs. 2.1%; p < 0.0001), especially infection (61.3% vs. 10.4%; p < 0.0001), delayed wound healing (83.3% vs. 21.7%; p = 0.004), and implant exposure (80.0% vs. 19.4%; p = 0.0002). Receiving RT, but not timing of RT, significantly predicted RF [odds ratio (OR) 4.00, 95% confidence interval (CI) 1.11-14.47; p = 0.03]. On multivariable analysis, infection (OR 7.69, 95% CI 2.12-27.89; p = 0.002) and delayed wound healing (OR 17.86, 95% CI 1.59-200.48; p = 0.02) independently predicted for RF. Our newly developed classification tree, which includes stepwise assessment of major infection, delayed wound healing, implant exposure, age ≥ 50 years, and total number of lymph nodes removed ≥ 10, accurately predicted 74% of RF events and 75% of non-RF events.

CONCLUSIONS

Infection or delayed wound healing requiring surgical intervention or hospitalization and receipt of RT, but not radiation timing, were significant predictors of RF. Our classification tree demonstrated > 70% accuracy for stepwise prediction of RF.