March 28, 2024

Research published in european heart journal We are investigating the correlation between the built environment and cardiovascular disease.

Kevin McConway, Professor Emeritus of Applied Statistics, The Open University of Japan stated the following.

“This study is an interesting start along a new path of investigating locational variation in the risk of coronary heart disease (CHD, or heart attack, angina, and related conditions).” Despite the epic scale of data analysis, it is still not possible to clearly establish how useful these new tools will ultimately be, or for what specific purposes they will serve. And so far, we know little new about what actually causes people to develop his CHD or what we might do to cause it. do to reduce risk. This line of research could develop into something very useful, but for now I think it’s too early to say what direction things will go.

“The obvious problem is that we need information about what causes the disease and possible interventions to reduce risk. However, this study is observational. and operates entirely at the level of regions and their populations, rather than at the level of individual people. Both of these characteristics make it difficult to say anything about what causes it at the individual level. .

“There will be many differences between people living in a neighborhood that will look different on Google Street View, apart from what is visible visually. These other differences may explain the actual difference in CHD risk. It could be the cause, not the difference in neighborhood appearance at all. That’s a problem common to all observational studies. Studies may show correlation, but correlation doesn’t necessarily mean causation. Not.

“For example, both the new research paper and its accompanying editorial note that pollution, particularly air pollution, has previously been shown to be correlated with increased CHD risk. does it play a role? cause While CHD is not always clear-cut, it is interesting that the new study does not use any data on air pollution levels in the area of ​​interest. Air pollution was not included in the “traditional” demographic and socio-economic factors that the researchers used in their statistical models to compare with those obtained from Google Photos, making it a social determinant of health. It is not included in the three composite indicators. Already used. Presumably, higher pollution levels have some effect on how an area looks on Google Street View, and could be indirectly incorporated into statistical models, but at least modeling pollution measures more directly would have been interesting to include.

“I’m not here to comment on how strong a role pollution levels play in understanding the causes of CHD. However, there may be an important role, and this study explores that. In any case, contamination is just one example of a factor that was not directly considered.

“At least it is possible to measure air pollution and use those measurements for future research. But there may be many other possible aspects of cause and effect at play. Researchers found that having buildings and roads in poor condition in an area correlated with a higher risk of CHD, while having more trees and more houses in good condition correlated with higher risk of CHD, as seen on Google Street View. found that being in the home was associated with a lower risk of CHD. However, this does not directly imply the condition of the home. cause The risks of CHD vary.that maybe Alternatively, people who are at increased risk of CHD may happen to live in homes or roads that are in poor condition for reasons other than the condition of the buildings or roads. It’s not difficult to imagine how that could happen. This study doesn’t even tell us if that’s happening.

“In their research paper, the researchers clearly point out the issue of causation: “…it is important to note that these correlations do not establish causation.” However, in press The release doesn’t mention this at all (and it should have).

“Because this study is cross-sectional, meaning it looked at CHD risk and Google Street View imagery at roughly the same time, we can’t directly see how changes in what you see on Google Street View are related. We can’t even look into it.” As CHD risk changes. This is just another reason why we cannot directly know whether or how improvements in the urban environment will change residents’ risk of CHD. And, as Dr. Kela points out in the linked editorial, “Most large-scale structural changes [in the built environment] Occurs in the context of gentrification and simply displaces high-risk individuals [in terms of CHD risk] Rather than improving health in the original environment. ”

“The problem with using data about neighborhoods or groups of people rather than individuals is that neither the local conditions seen in Google Street View nor the CHD risk for individuals are exactly the same across a census tract of perhaps 4,000 people. The study found a correlation between the average characteristics of a census tract on Google Street View and the average CHD risk within that tract. People most likely to be diagnosed with the disease were found in areas within the census tract where homes that appeared to be in worse condition were found, or had other common characteristics among the 4,096 characteristics in the photos surveyed. The researchers’ model was used. It is possible that this may be the case, but it is still possible that this is not the case. This study does not imply that I don’t understand.

“This problem of assuming that correlations measured between averages for groups of people are reflected in correlations about individuals is a well-known fallacy that has been addressed in many contexts. If you really want to know what’s going on, you need to get data on individual people.

“I think Rohan Khera’s linked editorial is very good. It shows some enthusiasm about the future prospects of this kind of approach, but it’s clear that things are not that far along yet.” He also notes that adding data based on Google Photos increased the level of correlation with CHD risk above that based on demographics and socio-economic characteristics alone, but that It noted that the increase was “rather modest.” In short, a significant portion of the association between what Google’s photos revealed and his risk of CHD is due to correlations between the photo information and previously known demographic and social issues. That means there is a possibility. Using the photos added something, but not much. Perhaps some of that addition was already tied to factors such as pollution or climate that were not explicitly included in the new model at all.

“Dr. Kela also pointed out that the method of validating statistical models based on Google Street View, while a very common and useful method in machine learning, did not go beyond using data from the same census tract. However, there is no guarantee that the same model, or even the same approach, will work in other parts of the United States, such as small cities and rural areas, let alone in completely different places in other parts of the world (including the United States). Not at all (in the UK). And he argues that people who live in areas where they haven’t even explicitly consented to allowing anyone to view or reuse their images may want their images of their neighbors to be used for these purposes. It properly draws attention to ethical issues, including rights.”

‘“Artificial Intelligence-Based Built Environment and Coronary Artery Disease Prevalence Assessment from Google Street View.” Written by Zhou Chen other. Published in european heart journal in Thursday, March 28, 2024, 12:05 a.m. GMT.

Doi: https://doi.org/10.1093/eurheartj/ehae158

declared profits

Professor McConway: I am a board member of SMC and a member of its advisory board. My quotes above are from my position as an independent professional statistician.