Using Machine Learning to Identify Factors Associated with Mammography Adherence in Korean American Immigrant Women

Background and Purpose: The incidence of breast cancer is rising among Korean American women (Tuan et al., 2020). However, these women report lower rates of breast cancer screening than other racial and ethnic groups (Lee et al., 2018). To promote breast cancer screening in this population, researchers have identified various factors associated with mammogram use such as sociodemographic characteristics, accessibility to health care, and cultural health beliefs (Oh et al., 2017). The purpose of this study is to expand this research by using machine learning to model additional factors associated with mammography uptake in Korean American immigrant women based on American Cancer Society (ACS) screening guidelines.

Methods: A cross-sectional survey was administered to 538 Korean immigrant women in North Carolina in 2019. The study participants were recruited using a study flyer and snowball sampling strategies at various community-based sites (e.g., churches and grocery stores). The survey was either self-administered or conducted face-to-face by the researcher. The participants’ mean age was 57.4 years old (SD=8.3) and their average length of time in the United States was 25.2 years (SD=11.9). About 89% completed undergraduate or graduate education, and 40% reported their income was less than $50,000 per year. A machine learning decision tree model, using chi-square automatic interaction detection (CHAID), was performed to identify factors associated with adherence to ACS mammography screening guidelines.

Results: Approximately 91% of the participants had a mammogram at some point in their lifetime, and among them, 65% adhered to ACS guidelines. The most significant factor associated with adherence to ACS guidelines was having a regular medical check-up. Other significant factors included: primary care physician’s recommendation, social support, alcohol use, health information from Korean newspapers and magazines, mammogram frequency, family history of cancer, the country of their first mammogram, follow-up test experience after mammogram, and having heard about another woman’s mammogram experience. The overall accuracy of correct predictions for the machine learning model was .84.

Conclusions and Implications: The study findings highlight the importance of access to routine primary care services and screening referrals for Korean American immigrant women. In addition, education about risk factors for breast cancer (alcohol use, family cancer history) and the need for follow-up tests after receiving abnormal screening results are warranted. One mechanism to promote breast cancer screening in this population is to use Korean newspapers and magazine to provide public health messaging about the primacy and benefits of having a mammogram. Interestingly, women who had their first mammogram in Korea tended to adhere to ACS screening guidelines as compared to those who had their first mammogram in the US. Future research needs to explore the motivational factors associated with breast cancer screening in Korean American immigrant women.