Anthropometric estimates can predict satisfaction with breast in a population of asymptomatic women.

Background: Several authors hypothesized that normative values of breast related quality of life in asymptomatic populations can be helpful to better understand changes induced by surgery. Breast related quality of life can be associated to breast anthropometry. This study was designed to explore this hypothesis, find relevant correlations and, using machine learning techniques, predict values of satisfaction with breast from easy body measurements.

Methods: Asymptomatic women undergoing routine clinical examination for breast cancer prevention were interviewed using the BREAST_Q V1 Breast Conserving Surgery Pre-op. Descriptive statistics was performed to describe the characteristics of the population. The Pearson correlation test defined correlation between relevant anthropometric variables and scores in each domain of the BREAST_Q. Regression analysis was employed to assess variation in the "Satisfaction with breast" domain when looking at the mirror dressed or undressed. Three machine learning algorithms were tested to predict scores in the "Satisfaction with breast domain" given body mass index and nipple to sternal notch distance.

Results: One-hundred and twenty-five women underwent clinical examination and assessment of anthropometry. The reply rate to the BREAST_Q ranged from 99.2 to 88% depending on the domains. The "satisfaction with breast" domain was negatively associated either to BMI [r_Pearson = -0.28, CI (-0.41, -0.15) p < 0.005] and Age [r_Pearson = -0.15, CI (-0.29, -6.52e-03) p = 0.04]. The N_SN distance was also negatively associated to this domain with the following values for the right [r_Pearson = -0.34, CI (-0.45, -0.21) p < 0.000] and left side [r_Pearson = -0.31, CI (-0.43, -0.17) p < 0.000]. Linear regression analysis was performed on questions 1 and 4 of the "Satisfaction with Breast" domain revealing a steeper decrease for women with higher BMI values looking in the mirror undressed (Adjusted R-squared BMI: Dressed - 0.03329/Undressed - 0.08186). The combination of two parameters (BMI and N_SN distance) generated the following accuracy values respectively for three machine learning algorithms: MAP (Accuracy = 0.37, 95% CI: (0.2939, 0.4485)); Naïve Bayes (Accuracy = 0.70, 95% CI: (0.6292, 0.7755); SVM (Accuracy = 0.63, 95% CI: (0.5515, 0.7061)).

Conclusions: This study generates normative scores for a Mediterranean population of asymptomatic women and demonstrates relevant associations between anthropometry and breast related quality of life. Machine learning techniques may predict scores of the "satisfaction with breast" domain of the Breast_Q using body mass index and nipple to sternal notch estimates as input. However, the algorithm seems to fail in approximately one third of the sample probably because is not able to capture many aspects of personal life. Much larger sample and more qualitative research is required before establishing any direct association between body estimates and quality of life. Clinical implications are given.