Development of a Breast Reconstruction Risk Assessment (BRA) Score: An Individualized Risk Calculator for Complications Using the NSQIP and TOPS Databases
John Y. Kim, MD1, Alexei Mlodinow, BA1, Nima Khavanin, BS1, Karol A. Gutowski, MD2, Keith M. Hume, MA3, Christopher J. Simmons, BS3, Michael Weiss, MPH4, Robert X. Murphy, Jr., MD, MS4.
1Northwestern Feinberg School of Medicine, Chicago, IL, USA, 2Private Practice, Northbrook, IL, USA, 3American Society of Plastic Surgeons, Chicago, IL, USA, 4Lehigh Valley Health Network, Allentown, PA, USA.
Purpose: With over 90,000 prosthetic and autologous breast reconstructions each year, many studies have evaluated population level measures of post-operative risk. However, there is a paucity of data providing an objective measure of individual risk for complication following immediate breast reconstruction. The recent advent of robust, prospectively planned and standardized databases provides the data needed to achieve adequate statistical power to develop such models. Two such databases are the National Surgical Quality Improvement Program (NSQIP) and Tracking Operations and Outcomes for Plastics Surgeons (TOPS) databases. This study aims to use these to develop a Breast Reconstruction Risk Assessment (BRA score) calculator for post-operative medical and surgical complications.
Methods: The TOPS (2008-2011) and NSQIP (2005-2011) participant use files were queried for immediate breast reconstructions. Regression analysis was used to develop predictive models for each of seven outcomes of interest: seroma, dehiscence, surgical site infection (SSI), flap failure (autologous cohort only), explantation (tissue expander cohort only), reoperation, and medical complications. Hosmer-Lemeshow p-value, c-statistic, and Brier-score were computed to assess model performance. Bootstrap analysis validated the models. The models served as the basis for a risk calculator available at BRAScore.org.
Results: A total of 16,069 patients from NSQIP and 5,180 from TOPS were included in the analysis. The seven regression models developed between the two data sets were based on a variety of preoperative variables including procedure type, age, BMI, smoking status, diabetes, ASA class, chemotherapy, and bleeding disorders. The characteristics of all models are displayed in Table 1, along with the baseline risk of each complication assessed. Additionally, Figures 1 and 2 present a series of histograms representing the distribution of individual surgical and medical complications, respectively, for the cohorts from which each model was developed. The calculator for medical complications is currently available at BRAScore.org, and similar calculators for surgical complications will soon follow.
Conclusions: The BRA score risk calculator for medical and surgical complications mitigates the potentially inaccurate extrapolation of population-based measures of average risk to individual patients. The models demonstrate acceptable performance, and when applied to our study cohort reveals a positively skewed distribution of individual risk. We developed an easy-to-use online interface_accessible by patients and surgeons alike_to more objectively assess individual risk for complications and inform surgical decision making in a more patient-centric fashion.
Table 1 - Benchmark Complication Rates and Regression Model Characteristics
|Complication||H-L Statistic||c-Statistic||Optimism Corrected |
|Brier Score||Baseline Incidence|
Figure 1 - Distribution of Predicted Probabilities for Each Surgical Complication
Vertical lines represent the mean probabilities
Figure 2 - Distribution of Predicted Probabilities of Medical Complications
Vertical line represents the mean probability
Back to Program