Developing AI-measures of Pedestrian Environment Features for Physical Activity and Cancer Prevention in Rural Communities

7. Project Summary/Abstract Residents of rural areas often exhibit lower rates of physical activity (PA), correlating to elevated cancer incidence and mortality rates compared to urban dwellers. The lack of PA resources significantly contributes to the lower rates of PA observed among racially and ethnically diverse and lower-income rural populations. A major obstacle to addressing urban-rural PA and cancer disparities is an insufficient understanding of the neighborhood environment—specifically, the pedestrian environment features that inhibit or promote PA— which can be cost-effectively modified. Existing, publicly available pedestrian environment measures assess macroscale walkability features (e.g., land use mix, street intersection density) that are costly and infeasible to improve in rural areas. While smaller-scale research has identified more affordable, microscale Pedestrian Environment Features (PEFs) (e.g., sidewalks, crosswalks, lighting), person-led, microscale audits of PEFs show limited feasibility across expansive rural geographies. Machine learning algorithms have been developed using data from urban and suburban areas to audit microscale PEFs, but these can introduce bias when scaled up for use across expansive rural areas to study their relationship with PA. Addressing urban-rural cancer disparities necessitates assessing the association between microscale PEFs and PA in both urban and rural areas of the US. Therefore, we propose three specific aims: 1) further validate existing machine learning algorithms to assess 9 microscale PEFs (sidewalks, sidewalk buffers, curb ramps, zebra and line crosswalks, walk signals, bike symbols, benches, and lighting) for rural areas, 2) test the relationship between rural microscale PEFs and middle to older age adult PA, and 3) identify disparities in microscale PEFs by income levels, racial and ethnic composition, & geographic location across the US. We will retrain and leverage existing deep learning classifiers developed as part of preliminary work funded by the National Cancer Institute to assess urban and suburban microscale PEFs, to create classifiers that generalize and perform well in rural areas. We will validate these microscale PEF classifiers with human virtual audits and examine their relationship with PA among middle and older age adults, given this age group is at high risk for physical inactivity. We hypothesize that greater rural PEFs will be associated with greater minutes per week of total PA and walking, as measured by the International Physical Activity Questionnaire, after adjusting for covariates. Finally, we will explore income, racial, ethnic, and regional disparities in rural microscale PEFs across the US where policy or environmental intervention may be necessary. This project aims to validate a new scalable tool for identifying rural PEFs and uncovering potential environmental and health-related disparities in diverse rural locales across the US. Results will inform larger-scale research that uses AI-measured PEF assessments to address physical inactivity and cancer-related health disparities. In turn, existing cancer prevention and PA promotion initiatives can intervene on lower-cost and modifiable neighborhood environment features.