It’s all about how you measure things

On October 17, 2017 a gave I gave an invited talk for researchers in the University of Michigan Department of Family Medicine, Ann Arbor, MI. Below is an approximate slide-by-slide transcript of the talk. Slides (with references) are available here.

Slide 1 – It’s all about how you measure things: the challenged study of food environments - implications for health disparities, research, & policy

I want to thank Caroline Richardson for inviting me here today, everyone taking time out of their busy schedules to meet with me, and Jill Bowdler and Lois Phizacklea for coordinating al the details. I’ve heard great things about Ann Arbor and it nice to finally have the opportunity to be here.  

Most of the work I do is related to food and nutrition, and while I often give talks on clinical nutrition and nutrition myths in clinical practice to audiences like this, Dr. Richardson suggested that your group and this meeting might be better served by a presentation on my primary research.  Given this is a research meeting, I ultimately want to get to my main project underway, which is a K award through NIH. But before I get there, I think I really need set up the rationale for my study so you can understand why I am doing what I am doing—because I seem to work in an area that is not familiar to many people (which unfortunately too often includes funders).  I’m actually going to spend a fair bit of time on background and my preliminary studies as context.  I understand many of you work in related areas of obesity and chronic disease so I hope you’ll at least find this tour of my work interesting if not relevant to your own research.

Slide 2 – Disclosures

I have no real or perceived conflicts of interests or disclosures relevant to this presentation.

Slide 3 – Background: MDs

Who’s a family physician here? The physicians in the room will know that doctors spend a good deal of their time counseling patients, often giving advice and trying to promote healthful behaviors. While detailed dietary guidance has not historically been prominent among the kinds of advice physicians give (for a variety of reasons), doctors sometimes advise patient what to eat ... which can often feel like this. Or maybe a little more like this.

Slides 4-5 – Background: MDs vs. environment

Because when patients leave our offices, they often have to navigate this (and this, and this, and this)[images of fast-food restaurants and other neighborhood food sources]. These are just some photos from Bronx neighborhoods where my patients live. The images demonstrate that food environments can really challenge doctors’ advice for healthful eating and patients’ best intentions to eat healthfully. And I love this quote from Harvard’s David Williams from a research conference I attended a few years back: “What if we treat illness and then send them back to the same conditions that made them sick in the first place?”

Slide 6 – Background: NAM (formerly IOM)

Another quote that is relevant here comes from the National Academy of Medicine (formerly Institute of Medicine), where I attended a conference earlier last month. Since this is a blurry cell-phone photo, let me clarify what the slide says. It says:

“Most chronic diseases and conditions are a normal response by normal people to an abnormal environment.

Slide 7 – Background: photo paints a picture

So let’s talk about environment. Who can tell me what this is a photo of? [display of onions and white potatoes in an improvised plywood box] This is a very poor quality image representing a very poor quality situation. This is the produce aisle, of a grocery store, in an urban, low-income, minority neighborhood like the one in which I practice. In fact, this is an actual photo that I took during residency from grocery store just a couple of blocks from my residency clinic. 

Slide 8 – Background: environ. disparities

Now here’s a map showing where that grocery store is (marked “B” on the map). If we go up the road just a few miles, we get to point “ A” which is a supermarket just a few blocks from where I lived during residency. This is what the produce looked like there [beautiful, varied, colorful assortment of fresh fruits and vegetables].  The point is that environment probably matters. 

Slide 9 – Background: health impact pyramid

Here is a figure from a 2010 paper in AJPH by former commissioner of health for New York City and former Director of the Centers for Disease Control and Prevention, Tom Frieden. What it shows is different interventions that can effect health and their associated levels of effort and impact. What it suggests is that rather than trying to intervene here on behavior at the level of the individual—as in doctors’ offices, counseling and educating patients one-on-one, where doctors and patients may put in a lot of effort but have very little impact—maybe we would do better to address more fundamental drivers.  Intervening on socioeconomic factors is hard, but maybe we might do nearly as well to address environmental factors: changing the environmental context.

Slide 10 - Background: food-environ. studies

Well certainly Tom Frieden and I are not the only ones to have this idea.  There has actually been exploding interest in this area. This is a plot showing papers focusing on food environments that have been published over time. What you can see there has been an exponential increase in articles over about the last 10 years. And its right about then when I got into the game (which is mostly coincident, not causal, to the increase in productivity in the field—but please don’t tell my Chair that.).  Still, being part of this rapid proliferation, I have seen first-hand the problems in the research. Which make me want to ask:

Slide 11 - Background: critical question

 “Why is everybody in such a hurry to get the wrong answers?” That question is actually a quotation attributed to the Dutch computer scientist, Edsgar Dijkstra. But the question is highly relevant to food-environment research.

Slide 12 – Background: apparent controversy

For instance, let’s take and apparent controversy that came up not long after I started doing this work. In April of 2012, Gina Kolata published a piece in the New York Times questioning the pairing of food desserts and obesity (Are you all familiar with the term “food desert”? It is a term that is variable defined but often means neighborhoods devoid of healthy food or specifically, devoid of supermarkets). Kolata’s piece referenced a couple studies looking at food environments vs. dietary intake or body weight that showed either no robust relationship between the food environments and consumption patterns or no differences in body weight by food-outlet exposures.  But later that very month in 2012, a study in the journal Obesity seemed to suggest the exact opposite: that food deserts do matter.  When grocery stores and supermarkets were absent, BMI was higher. The Kolata piece in 2012 prompted me to write a letter to the New York Times, commenting on some of the reasons for this kind of “yes there is an association”/“no there isn’t” argument.  Like all the letters I write to the New York Times, that one was unceremoniously not accepted.  So I decided to submit the same arguments in a letter to the editor at Obesity, regarding the study mentioned here.

Slide 13 – Background: problems for the field

That letter was accepted, but at the blazing speed of academic publishing, didn’t appear in print until almost a year later.  The delay didn't really matter though because the field was still struggling with the same issues.  In fact, I wound up expanding the ideas in that letter to write a full-length review article and commentary about the limitations of food environment research, and the problems were just as bad—if not worse—when it was published 2 years later. In both pieces, I reviewed several problems for reproducibility and, indeed, even validity. I am going to highlight 3 of those limitations today. My research over the last decade has addressed each of these limitations, so what I’d like to do now is give you a high-level whirlwind sampling of some of that work as set up and rationale for my main research.

Slide 14 – Background: problem #1

The first problem relates to unvalidated business lists. If you were going to measure the food environment, or the food sources in a community, how would you do it?  Well, and efficient way would be to use pre-existing lists of businesses (like directories) collected for purposes unrelated to determining what food sources are in an area (e.g. for marketing purposes, licensing purposes, etc.).   The vast majority of food environment studies have done this, including the studies referenced above related to the apparent controversy.   The problem is, most of these lists have not been validated.  So I took one of the business lists used most often and said, let’ validate it.  In other words, let’s see how it performs on the street by comparing what food sources we find on the street to what the list says. 

Slide 15 – Background: observation vs. list

Let’s say this is a given street. We’ve got a pizza place, a deli, Taco Bell, CVS, a bodega, Dollar store, and produce market.  And now we want to compare that to what our business list tells us. If the business list data looks like this, then there is no problem. The business list matches up with our direct observations precisely. Thus we can rely on the list as a valid source of food-environment data. Perfect.

Slide 16 – Background: validate business list

But what if when we look at what the business list shows, it actually looks like this? Well, now you can see that CVS lines up with CVS and NY Deli lines up with NY Deli, so that’s good. But then we’ve got Jim’s Bodega vs. Joes’s Bodega. Is a bodega a bodega? Is Jim’s the same as Joe’s? Hard to say. We’ll give it a cautionary check. Likewise, what about Taco Bell vs. Burger King. Well certainly these are both fast food franchises, and to the extent that any fast food might contribute similarly to the offerings of a food environment, perhaps the distinction is not so important. Again, maybe we’ll give a cautionary check for that match up. But then what 7-eleven vs. a produce market? Clearly these are two very different kinds of businesses. And what about these empty lots where the pizza parlor and dollar store should be?  Or the Starbucks where an empty lot should be? Technically that is correct because Starbucks abhors a vacuum will eventually fill any void but it doesn’t speak favorably to the accuracy of our comparison. In the end, we can see what had been a 100% accurate business list is now looking more like … well lets see, we’ve got 8 lots here and at best we’ve got 4 matches at best or 2 matches at worst so somewhere between 25% and 50% accurate.

Slide 17 – Background: standard 2 x 2 table

Determining how good a business list is really just that simple.  Much like determining the predictive value of a diagnostic test, you just make a 2 x 2 table of the candidate test (in this case the business list) vs. the gold standard (in this case direct observation) and then calculate sensitivity (i.e., the probability that a business on the ground will be in the list) and PPV (i.e., the probability that a business in the list will actually be on the ground). 

Slide 18 – Background: business-list validity

For the business list that we examined on streets in the Bronx, we calculated its performance in terms of sensitivity and PPV. What do you think we found? Now keep in mind, this is a list that has been used in literally dozens and dozens of published studies (including those I referenced earlier in regard to the apparent controversy in 2012 [The An & Dubowitz studies; Lee used D&B]).  Well, the sensitivity was actually 58.1%, and the PPV was 67.3%. Now that’s bad. No question. But those results are actually being generous or “lenient” (that is considering Joe’s bodega to be the same as Jim’s bodega and Taco Bell to be the same as Burger King).  If we are strict in matching between what is in the list and what is on the ground, the sensitivity drops to 39.3% and this PPV to 45.5%.  In other words you would be much better off flipping a coin to determine if a business was present or not than to rely on a business list.  The bottom line is that business lists poorly represent actual food environments and are an inadequate substitution for direct observation.  I published a paper noting that this fact alone could account for much of the confusion in the published literature to date. And others have published similar findings on other pre-existing datasets.  But that is only one of the problems.

Slide 19 – Background: problems for the field

Going back to our list, now we know we probably can’t rely on pre-collected data in conducting research on “food deserts” for instance.  And there may be no substitute for shoe leather when it comes to making assessments.  The next issue relates to what to look at when you actually go out onto the streets to look—or what to count when you are measuring food environments.

Slide 20 – Background: problem #2

The second problem is the limited range of food sources considered (and the poor measurement of these food sources)

Slide 21 - Background: food environ focus

This is a slide from a presentation years back from colleagues at the National Cancer Institute. Even though the figure is a bit dated, I am confident the overall findings still hold. What it shows is that the majority of food-environment studies focus on select food stores or restaurants. Most commonly they focus specifically on supermarkets and/or fast-food outlets—like the papers highlighted in the New York Times piece and in the Obesity article that inspired my letters to the editor. Like those papers, they also tend to categorize supermarket as ‘healthy’ and fast food as ‘unhealthy,’ when supermarkets are often the predominant source of junk foods and fast-food outlets often sell healthful items like salads, fruit, and milk (This fact alone might explain much of the disagreement in the literature.)  In any case, researchers have focused in a limited way on some other sources of food in the environment, but nobody is really talking about this …

Slide 22 - Background: street vendors

I don’t know about Ann Arbor, but people selling food on the street is a big deal in New York City and in other cities around the country and around the world.

Slide 23 - Background: street vendors

My team and I looked at street vending in the Bronx and found close to 400 vendors across the borough. About a quarter of them focused on selling healthier items like fresh produce and water. About three quarters were set up to sell less-healthful items like hot dogs, cheesesteaks, ethnic fast foods, ice cream, chips, and candy. But where and what they sold differed by neighborhood.

Slide 24 - Background: street vendors

For instance, Green carts—mobile vendors permitted to sell only whole, fresh, unprocessed produce—weren’t necessarily selling where they might be most needed in neighborhoods.  First of all, Green Carts are only permitted to sell in the neighborhoods within this dark outline. These are neighborhoods that struggle with access to fresh fruit and vegetables.  The map on the left the shows where Green Carts actually were.  Some were in higher income areas, having no issue with food access. The map on the right shows clustering, which occurred around medical, academic, retail, and transportation centers. The clouds or ‘hotspots’ around these clusters are meant to represent ‘reach’. If you look at the white areas, these are areas outside the distance Green Carts ‘reach’, even being conservative. You can think of these white spaces as a “Green Cart Deserts,” which due to clustering represents about 50% of the space in the target area. But Green carts are just one kind of street vendor.

Slide 25 - Background: street vendors

Other street vendors also have spatial issues, and these differ by weather and season. For instance, vendors of all types seem to spread out across the Bronx in the summer. But vendors seem to consolidate into the Southwest of the borough in the winter, with a predominance of less-healthful vendors collected here. I think the Bronx looks a little bit like and anatomical heart and so I sometime refer to it as the heart of darkness because it is where all the bad stuff happens: unhealthful conditions, poor health outcomes, etc. If it is a heart, then this is the left ventricle, where all the action is; this is the area that is disproportionately affected by all the bad stuff. It also happens to be an area with the highest poverty rates and greatest proportion of Hispanic residents.

Slide 26 - Background: street vendors

This is a busy slide, but I want to go through it quickly because it shows how street vending relates to diet, diet-related health, and demographic characteristics of neighborhoods in statistical terms. More vendors selling ‘less-healthy’ foods in neighborhoods were correlated with lower average fruit-and-vegetable consumption, greater consumption of sugar-sweetened beverages, greater BMI, more diabetes, more dyslipidemia, more hypertension, a greater number of racial and ethnic minorities, lower high-school graduation rates, and greater poverty.  At least those were associations during summer months. In winter months, associations were in the same direction, but lower in magnitude (except for associations with Hispanics and poverty as suggested by the map on the preceding slide).

Slide 27 - Background: street vendors

Here are some papers (on measuring street vendors, on Green Carts, on street vendors by weather and season) that came out of all this work if you are interested in more detail, but the point is that there is definitely food out there beyond supermarkets and fast food outlets. Street vendors in particular probably matter and probably should be measured.

Slide 28 - Background: other food sources

And street vendors are just one other source to consider.

Slide 29 - Background: farmers’ markets

Here is another. Famers’ markets.

Slide 30 - Background: farmers’ markets

As with street vendors, my team and I assessed farmers markets in the Bronx, and here is a map of where markets were relative to nearby stores. While not as prevalent as street vendors, there were a number of farmers’ markets in the Bronx, and they probably should not be excluded from consideration when looking at food sources in neighborhoods.  

Slide 31 - Background: farmers’ markets

One interesting finding for farmers’ markets in the Bronx was that almost 1/3 of the products offered on average were highly refined or processed items. For some famer’s markets in some neighborhoods, these refined or processed items accounted for more than 50% of offerings! And these items were often the most promoted and best-selling products.

Slide 32 - Background: farmers’ markets

Just to give you an idea of what that looks like, here are some photographs. You can see tub after tub of sugary drinks, a sign for pies, brownies, scones, croissants, and cookies, and an assortment of pre-wrapped cream-filled pastries that look more like they came from a factory than a farm, or if ever from a plant more the industrial processing kind than the living botanical kind.

Slide 33 - Background: farmers’ markets

Suffice it to say that farmers’ markets may offer many items less than ideal for good nutrition or health with differences by neighborhood.  

Slide 34 - Background: farmers’ markets

If interested, those findings are published along with comparisons of famers’ markets’ quality, price, and convenience to local stores.

Slide 35 - Background: other food sources

Suffice it to say farmers’ markets may be among food sources beyond food stores and restaurants that probably should be considered when looking at local food environments or making designations like ‘food deserts.’

Slide 36 - Background: ‘other businesses’

And then of course there is this: pharmacies … and dollar stores … and sports stores … and quickie lubes … and laundromats … and all the various other storefront businesses that, while not normally considered ‘food stores’, are very often sources of food and/or drink.  And the items they offer tend to be things like highly-processed, pre-packaged, salty snacks, sweetened drinks, and frozen confections.

Slide 37 - Background: ‘other businesses’

We looked at ‘other businesses’ in two different areas of New York City: the Bronx (“the heart of darkness,” with its lower-income, minority communities) and the Upper East Side (UES) of Manhattan (an affluent and mostly white community). Just for orientation for those unfamiliar with NYC, this map shows the Bronx (where I practice and where most of my patients live) along with the island of Manhattan and surrounding boroughs of Staten Island, Brooklyn and Queens. We also see new Jersey just across the river and part of Westchester County New York to the north. For our assessment of storefront businesses, we didn’t make distinctions between the various Bronx neighborhoods (the differences certainly exist, but are smaller than the differences between any part of the Bronx and the UES).

Slide 38 - Background: ‘other businesses’

These photos represent typical streets in the Bronx and the UES and reveal characteristic differences. There were big differences in retail density, and number and percentage businesses offering food and/or drink between the two areas.  In the Bronx more than 1 in 5 other business offered food or drink. In UES it was less than one in 13 did. Also there were differences in the healthfulness of items offered. The bottom line is that in the Bronx there were many more businesses offering any food or drink, and higher percentages offering only ‘less-healthful’ items without healthier alternatives. 

Slide 39 - Background: ‘other businesses’

Here is an example of an ‘other business’ from the Bronx to give you an idea. This is a lock, safe, and gate retailer and inside the store you see a vending machines full of sugar-sweetened beverages, candy, and a variety of salty refined snack chips.  Perhaps that is a bit unexpected, but not as unexpected as this:

Slide 40 - Background: ‘other businesses’

This is a laundromat—a public place to wash clothes. One might expect the sale of detergent and fabric softener here. One probably wouldn’t expect to find a full sandwich counter as in this example.

Slide 41 - Background: ‘other businesses’

Even more surprising, or really shocking and sad is this example of an actual medical office in the Bronx. You can see a vending machine for sugar-sweetened beverages right in the waiting room!

Slide 42 - Background: ‘other businesses’

More details will be available in two papers that will hopefully be published soon.

Slide 43 - Background: ‘other businesses’

In the interim, it seems clear that foods and beverages are far more prevalent than most prior research has assumed.  And ‘other businesses’, that is storefronts other than food stores and restaurants, are important to consider.

Slide 44 - Background: ‘other businesses’

There are of course even other food sources. Food pantries, for instance, might be an example.  We did a study looking at food pantries in the Bronx and showed (as in the studies of other food sources) substantial provision of mostly ultra-processed unhealthful fare with differences in access and availability by neighborhood.

Slide 45 - Background: ‘other businesses’

And while not directly about food provision, we’ve also looked at food advertising in neighborhoods, which is about what gets promoted. Looking at ads in Bronx subway stations, we showed ads for “less-healthful” items were located disproportionately in neighborhoods home to vulnerable populations facing diet and diet-related-health challenges (and directed specially at children, minorities, and Spanish speakers). This work was particularly gratifying because it has supported and helped move actual policy. The data on alcohol ads in particular has been used by various advocacy groups and non-profits to support an alcohol ban in NYC transit (which as of today, looks very likely to pass)

Slide 46 - Background: ‘other businesses’

All of this work together can be summarized by the opening line from a recent application I submitted for an international food-research award: “Foods and beverages are nearly ubiquitous today, with placement and promotion of cheap, highly palatable, ultra-processed, unhealthful, convenience and impulse items just about everywhere, even in unexpected places and particularly directed at vulnerable groups.”

Slide 47 – Background: problems for the field

Back to our list of problems for the filed, we have now reviewed the research suggesting we can’t rely on unvalidated business lists, and limited ranges of food sources measured badly just won’t do. The final issue I want to address relates to GIS (that is, Geographic Information Systems)

Slide 48 – Background: problem #3

Problem 3 is unrealistic and un-nuanced GIS or, in other words, ways of defining spatial exposure that don’t use common sense. To illustrate, consider this map of the upper part of New York City. Lets say we wanted to measure someone’s food exposure who lived in Harlem. Lets say here. Now, a common way to do this is to pick some arbitrary distance and look at all the food sources within that distance of the person’s home.  So let’s say we pick one mile as the distance and the we draw a one mile buffer around the persons home. Like this. Well that buffer would include the store noted by this marker here, which happens to be a very nice Trader Joe’s … across the Hudson river … in New Jersey. And so even though this supermarket is technically within a mile of the persons home, that person would have to either canoe across a river (both ways, and one way with groceries) or drive over 6 miles to get there, and over 6 miles to get back (going across one of the most congested travel route in the country, on a trip that might take 30 minutes each way in typical traffic, and cost $15 in tolls).

Slide 49 – Background: 3 diff views/same area

How you measure access to food sources matters, and different measures can give you very different (even opposite) results. Take this hypothetical example. In each of these three scenarios, the neighborhood is exactly the same. In each scenario, the orange rectangle represents a persons’ home or school or point of interest. The gray lines are city streets, the blue line is a river, and the double line represents train rails.  The apples represent sources of ‘healthier food like produce markets. The hamburgers represent sources of ‘less-healthy’ food like fast-food outlets. The diamonds represent street vendors selling unhealthful food and the triangles represent ‘other’ storefronts (like pharmacies, gas stations, dollar stores, etc.) also selling unhealthful food. When the diamonds or triangles are open they are uncounted or ignored and when then are filled they are counted or assessed.

What the three scenarios show is three different ways of measuring access. The first method is by arbitrary geo-political boundary or administrative area. Often times these will be census tracts or zip codes. You can see that within the arbitrary administrative area shown here there are two apples (or two sources of ‘healthier’ food), two hamburgers (or two sources of ‘less-healthy’ food) and a bunch of other storefronts and street vendors that are not counted. So the total is 2 ‘healthy’ vs. 2 ‘unhealthy’ (even though two of those food sources are not really accessible, being across train tracks where no roads go). In scenario #2, we see the method used in the Harlem-and-Trader-Joe’s example from the last slide (and incidentally the method used in studies related to the NYT-Obesity controversy noted earlier [An & Dubowitz studies]). We simply draw a circular buffer around the point of interest at some pre-determined distance and we ignore barriers to transit like train tracks or rivers. However, in this case we will respect the administrative boundary here (maybe a county line or state line) since perhaps the people funding our study really only care what’s going on in our area, not what is happening over the border. Thus, we don’t count what’s going on in almost half the exposure circle and we find that in the area that remains, there are 3 apples (or three sources of ‘healthier’ food) and no other counted food sources. Thus the total is 3 ‘healthy’ vs. 0 ‘unhealthy’. In scenario #3, the method also relies on distance, but in this case considers travel paths along an existing street network (which do not cross train tacks or the river). In this case, there are three hamburgers (or three sources of ‘less-healthy’ food) along the potential travel paths. And if we also count the other storefronts and street vendors along those paths, we get a total of 0 ‘healthy’ vs. 8 ‘unhealthy’ food sources. Same neighborhood. Three different methods. Three very different—even opposite—conclusions. Notably, the methods from scenarios #1 and #2 have been used most commonly in published studies.

Slide 50 – Background: problems for the field

So that pretty much wraps up problems for the field that need to be addressed (at least the 3 that I’ve chosen to highlight today). What we need is some science that moves beyond these limitations. That brings me to my K.

Slide 51 – Moving Forward: K23 from NICHD

After some convincing (i.e., 4 submissions over 3 years to 2 institutes, by 1 researcher: me), NIH finally decided to fund a 5-year K23 award. The title of that award is “Local food sources around home and school and adolescent dietary intake” and although it was awarded in May of 2015, it was retroactively made to cycle in Feb so I am a little more than halfway through my 3rd year.

Slide 52 - K23: problems to address

The problems I aim to address in this award are the problem we just reviewed (that is the three I highlighted plus the one I mentioned about supermarket not necessarily being ‘healthy’ and fast food not necessarily being ‘unhealthy’). A fifth problem is considering food sources in isolation (e.g. just supermarkets or just fast-food outlets) without considering combined effects or interactions.

Slide 53 - K23: proposed solutions

The solutions I propose are also what we reviewed or at least what I hinted at in doing the review. I’m going to directly observe, not use pre-existing business lists. I’m going to consider a full range of food sources, not just a limited array of food stores and restaurants. I’m going to consider food sources along walkable paths, not within arbitrary geographic shapes or geopolitical boundaries.  I’m going to look at the foods and beverages actually available and not make distinctions about healthful and unhealthful based on business type. Finally, I am going to assess the effect of any food sources relative to other food sources present.

Slide 54 - K23: specific aims

The specific aims for the project are these:

Aim 1. To conduct a comprehensive assessment of local food sources

(direct observation of all food sources in Bronx neighborhoods)

Aim 2. To generate a novel integrated multi-level database.

(linking data on local food sources from Aim 1 with other data on neighborhoods, on schools, and on patients) 

Aim 3. To assess how local food sources relate to diet and diet-related health

(using spatially-informed multilevel regression, assess how local food sources relate to reported dietary intake and diet-related health measures)

Slide 55 - Aim 1: food-source assessment

To conduct the comprehensive food-source assessment, I delineated ?-mile walking buffers around 10 high schools in the Bronx having school-based health clinics. These schools are in demographically distinct areas of the borough with demographically distinct student bodies.

Slide 56 - Aim 1: food-source assessment

To do assessments, I directed a team to walk each side of each street segment (or section of street between cross streets) in the sample of streets around the 10 schools (n >1,500 street segments). We assessed for any foods or drinks anywhere, from any publically accessible source. We categorized sources as ‘food businesses’ if they were primarily focused on the provision of foods or drinks (e.g., Green carts, delis, restaurants, cafés). We categorized sources as ‘other businesses’ if food or drink provision was not the primary focus (e.g., auto shops, laundromats, salons). Analyses included frequency distributions and chi-squared tests.

Slide 57 - Foods, example items, items not examples

For foods, we assessed for the presence or absence of items from three ‘healthful’ categories (fruits and vegetables, whole grains, nuts) and two ‘less- healthful’ categories (refines sweets and salty/fatty fare).  We had a detailed protocol and the fine print here details what qualified and what didn’t qualify for each. Notably ‘healthful’ categorization was generous, including such items as sweetened trail mixes, sugared nuts, and popcorn, and relatively minor items like vegetable toppings for sandwiches and pizzas (so please keep that in mind when we review preliminary results).

Slide 58 - Drinks, example items, items not examples

Drinks could be ‘healthful’ (milk and water), ‘less-healthful’ (sugar-added drinks and alcohol) or neither (100% juice and ‘diet drinks’) given the scientific controversy around these categories.

Slide 59 - Aim 1: food-source assessment

These are the results so far.  Rather than show an enormous table with values for each of the neighborhoods around all 10 schools, what I’ve done here is just highlight the neighborhoods with the lowest and highest values. So for instance, the school with the fewest businesses around it was school #10 from the map I showed earlier, having only 71 businesses within a 1/2 mile by street network. The school with the most businesses around it was school #4, having 369 businesses within a 1/2 mile by street network (a more than 5-fold difference). Note here that schools #1-6 are in the ‘left ventricle’ of the ‘heart of darkness’ as described earlier, so please keep that in mind as we move forward.

Slide 60 - Aim 1: food-source assessment

As for the percentage of businesses offering food or drink, the lowest percentage was around school #9 at just under a third. The highest percentage was around school #5 at just over half.  In other words, you could walk into any storefront in that neighborhood and have better than a coin-toss chance of finding food or drink (again, ‘left ventricle’ of the Bronx).

Slide 61 - Aim 1: food-source assessment

In most cases when food or beverages were available, there were ‘healthful’—or at least healthier—options like pictured here.

Slide 62 - Aim 1: food-source assessment

However, in all neighborhoods, there were cases in which *only* ‘less-healthful items were available.  That ranged from a very small percentage as in neighborhood around school #9, to a comparatively large percentage (more than 5 times higher) in the neighborhood around school #3 (‘left ventricle’ of the Bronx)

Those are the results for all businesses, which would include supermarket, fast food, and other food stores and restaurants. But what if we subtract out those ‘food businesses’ to focus only on business not primarily focused on food and/or drink provision? 

Slide 63 - Aim 1: food-source assessment

First we notice that the ‘other businesses’ represented anywhere from just over half to just over ? of all businesses around schools. If you consider the reciprocal, what that means is that the proportion of businesses specifically focused on food and/or drink provision around schools was just under a quarter to just under ?.

Slide 64 - Aim 1: food-source assessment

There was also large variation in the proportion of ‘other businesses’ offering food or drink, ranging from just under one tenth to nearly a third. And interestingly, the neighborhood with the fewest absolute number of businesses had the highest proportion of ‘other businesses’ selling food and/or drink.  Examples of ‘other businesses’ that offered food and/or drink appear here: auto shops, clothing stores, department stores, dollar stores, furniture shops, gas stations, gyms, hair salons, laundromats, newsstands, party-supply stores, pharmacies, sports stores, tobacco shops, and toy stores.

Slide 65 - Aim 1: food-source assessment

These photos just give you a sense of what that looked like on the street. You see a hair salon with a cooler full of sugar-added beverages. You see a laundromat with a vending machine selling ultra processed salty and sugary bagged snacks. And you can see a tax/real estate/accounting office that houses a full ‘sweet shop’ full of candy and other treats.

Slide 66 - Aim 1: food-source assessment

Interestingly though, not all items offered by other business were unhealthful.  Some did offered healthful options and this too varied by neighborhood (from about 1 in 20 to about 1 in 4). If you consider just the ‘other businesses’ that offered food or drink, 40% to just over 80% offered some healthful option. These options included fresh fruit, dried fruit, applesauce, salsa, canned beans, granola bars, popcorn, whole-grain crackers, nuts, seeds, trail mixes.

Slide 67 - Aim 1: food-source assessment

What it looked like on the street is this. You can see a barbershop with a little nut dispenser. This is a dollar store with an actually pretty-impressive selection of nuts and whole-grain hot cereal.  This is discount store (specifically a Marshall’s) offering dried fruits and nuts. And this here is a pawn shop with a cart offering of fresh plantains at the entrance.

Slide 68 - Aim 1: food-source assessment

Although healthful options were often available, in some neighborhoods a majority of ‘other businesses’ that offered food and/or drink only offered ‘less-healthful’ options.

Slide 69 - Aim 2: multi-level database

Anyway, there is still some work to do to tease apart those numbers and the differences they represent, but that gives you at least a preliminary picture of results related to Aim 1. For Aim 2, I’ll be linking up findings from food-source assessments with other data on schools, on neighborhoods, and on patients. School data is from the NYC Department of Education and school websites. Neighborhood data is from the U.S. Census bureau and the NYPD.  And student/patient data is from my health system’s electronic medical record (remember that each chosen school hosts a school-based health clinic that serves the majority of student who attend—generally in the 80-90% of the student body). Analyses thus far have been with Spearman correlations.

Slide 70 - Aim 2: multi-level database

Here are some correlation results with regard to school characteristics:

There was less food/drink near schools with higher attendance rates or more Asian/white students; there were more businesses offering *only* ‘less-healthful’ items around schools having more minority students; and there were fewer nuts, whole grains, or diet drinks around schools with more students in poverty

Slide 71 - Aim 2: multi-level database

For correlations with neighborhood characteristics:

More businesses offered *only* ‘less-healthful’ items in neighborhoods with more Hispanic residents; unhealthful-food provision was correlated with the percent of residents who were foreign-born; and greater % of businesses offered *only* ‘less-healthful’ drinks when there were higher % of minority residents.

Slide 72-73 - K23: conclusions/implications

Analyses for Aims 1 and 2 have led to the following thinking thus far:

·     Food sources around schools include businesses well beyond expected ‘food stores’ and restaurants (some quite unexpected)

·     The extent and healthfulness of food-and-drink offerings differ substantially by school and neighborhood characteristics

·     Findings might help physicians to better understand the context of adolescent eating and create opportunities for advocacy.

·     Businesses carrying foods and/or drinks might be persuaded to carry more weight/health-friendly items (as some already do).

Slide 74 - Next step: Aim 3 multi-level models

The next step in all of this work is assessing how food-source presence/ proximity/ density/ relative distribution relate to diet-related health metrics (BMI, blood pressure, lipids, etc.) for adolescent patients, in models that include patient, school, and neighborhood characteristics. I’m not going to go into the details, but here is a sample regression equation.

Slide 75 - Subsequent next steps

This work will be cross-sectional but will lay the foundation for longitudinal study (e.g., how changes in local food environments relate to changes in diet-related health outcomes). The ultimate goal of this research will be to identify targets to test in future R01 intervention trials, to inform policy and improve food environments, diets, and health for adolescents.

Slide 76 - Thanks and acknowledgements

That’s probably enough material for today so I am going to leave it there and end with this slide thanking all the people and organizations who have helped support this work. In particular, I would like to thank all of the students (listed here in the middle) who have done the hard work of conducting these assessments on the ground. My email is listed at the very bottom so if anyone is interested in talking more, please don’t hesitate to reach out. Thank you.