Today’s post covers some work I did a few years ago to visualize the diffusion of digital mammography in New York State. I obtained Medicare claims data that let me distinguish between mammography types based on billing codes. I then divided the state into small grid cells and used a spatial filtering approach to find the proportion of mammograms that were digital as opposed to film-based for each cell. I computed a 12-month moving average, produced 92 maps, and stitched them together. (You can watch the full animation sequence here.
The first map, covering all of 2004, reflects digital mammography technology in wide use at only a single site – the community hospital in the small town of Gloversville. It’s not typical for a place like this to be the earliest adopter, but that’s what happened here. Indeed, the hospital is quite proud of this fact. The last map, from 2012, reveals that the entire state had made the conversion, with inner-city Buffalo and Jamestown, in the southwestern corner of the state, the last holdouts. In between, there was a period of several years when your chances of getting a digital or film mammogram, and thus your risk of getting a true positive or false positive, depended on where you lived. The image above captures what was roughly the 50-50 point. It correlates somewhat with where where is more wealth and a higher density of medical providers, but with plenty of exceptions. This exercise could be repeated for any medical procedure, though I am not aware that anyone else has done so. If I had the time and resources, I would pool this method over large number of procedures, which would allow for the creation of an “early adopter index” for hospitals. For more detail, see: Boscoe FP, Zhang X. Visualizing the diffusion of digital mammography in New York State. Cancer Epidemiology Biomarkers and Prevention 2017, 26(4): 490-494.