|
This epidemic increase in obesity prevalence is sure to have an effect on the US health care system, diverting resources to management of chronic conditions related to obesity: diabetes, cardiovascular disease, and certain cancers. The increased morbidity and mortality, loss of productivity, and decreased quality of life will be staggering.
This increase in obesity rates is not limited to the US. Many other countries in the world are also experiencing a similar growth in obesity. The Organization for Economic Co-operation and Development (OECD, http://www.oecd.org) publishes obesity statistics for developed countries. Figure 1.4 shows the obesity rates for OECD members across different years. Although many countries are well below a threshold of concern for population-level obesity, all OECD member states exhibit increasing obesity levels over the last few decades.
click on the image to enlarge it
Source: OECD Health Data 2009.
Figure 1.4: OECD obesity
Obesity has high prevalence in both developed and developing countries. Data from the UN and WHO show several additional countries with alarmingly high rates of overweight and obesity (Figure 1.5). Furthermore, for many countries these obesity rates are higher for women than for men.
click on either image to enlarge it
 
Figure 1.5: Worldwide obesity rates, age 15+, most recent estimates from UN/WHO
If obesity rates are also increasing over time for those countries for which longitudinal data are not available, this points to a global crisis of staggering proportions.
2: What do we know about the causes of obesity?
Not much is definitively known about the underlying etiologies of the increase in obesity over the last several decades. New associative evidence is uncovered frequently, yet because of the lack of establishing specific causative factors, few effective interventions have been developed, and obesity continues to increase. Therefore, it is necessary to include the caveat that much of what is presented here is based on a combination of current scientific evidence and my own theoretical understanding. More than likely, the propensity for humans to gain weight under sufficient environmental conditions stems from our basic biological/evolutionary makeup: we have evolved to eat good food and to be lazy. In the face of uncertain food supplies, our ancestors focused on efficiency and parsimony. Prior to the development of effective food preservation and storage technology, in times of plenty, we gorged; in times of famine, we starved, and the traits that led to survival have persisted. This is why the most calorically dense foods—fats and sugars—taste so good to us. Expending calories for non-essential activities could lead to dangerous caloric deficit at a time of crisis.
But why the sudden rise in obesity? The simple equation is that in order to maintain homeostasis of weight, the number of calories consumed must equal the number of calories expended. An increase in caloric consumption with no change in caloric expenditure will result in weight gain. A decrease in caloric expenditure with no change in caloric consumption will also result in weight gain. Of course, an increase in caloric consumption coupled with a decrease in caloric expenditure will also result in weight gain. Although this is a simple equation and more than likely known to most people, the actual process of either maintaining or losing weight is not easy. Our current belief is that the reason it is difficult to maintain or lose weight is not because of intrinsic personal factors per se (e do not expect that genetics or will power has changed in the last few decades), but rather that the main reason for the rise in obesity is due to changes in the environments in which individuals live, work, and play.
2.1: Changes in physical activity
Without a doubt, in most areas in the modern industrialized world, fundamental changes in daily physical activity are responsible for decreased energy expenditure.
2.1.1: Labor saving devices
A substantial portion of the calories we used to expend on a daily basis has been assumed by the work of machines. We have essentially substituted the solar energy we used to expend performing many daily tasks for fossil fuel energy to power a growing number of labor saving devices. Before the invention of washing machines, dishwashers, and vacuum cleaners, we swept with a broom, and washed dishes and clothes by hand (Figure 2.1). Now, not satisfied with electric vacuum cleaners, we can buy a Roomba and let the device itself clean our messes, with no work at all on our part.
Figure 2.1: Old technology and new
It is surely a bit naïve to suggest that life was any better before the development of many of the technologies that have improved our quality of life. Not only have some of these devices given us more time to pursue other endeavors, devices such as the electric dishwasher are “greener” with respect to generation of both heated and waste water. Nevertheless, it should be acknowledged that these changes in lifestyle have had an effect on the amount of calories expended on a daily basis.
2.1.2: Transportation
Other modern conveniences related to transportation have also had an effect on daily caloric expenditure. The streetcar required walking from home to the main arterial streets on which rails were located. The streetcar became replaced in many cities by buses, which could service larger areas of town, requiring relatively less walking getting to bus stops versus trolley lines (Figure 2.2, left).
Once the transition from streetcar to bus had occurred, the next logical step was to replace the bus with an even more convenient form of personal transportation: the car. Although the car allowed unrivaled personal freedom to get from place to place, replacement of walking with driving has certainly had an effect on energy balance. Futuristic visions have predicted even more convenient (and outlandish) solutions for personal transport (Figure 2.2, right).
Figure 2.2: Streetcars, buses, cars: to infinity and beyond
2.1.3: Changes in employment patterns
Along with different ways of getting to work, our work habits themselves have changed dramatically over the last century. More people worldwide are moving from rural areas to urbanized areas (Figure 2.3). As the number of urban dwellers increases, fewer people will work as physical and manual laborers; we should expect larger portions of the population employed in service industries that require less caloric expenditure.
Figure 2.3: Urban and rural populations of the world, 1950-2050
2.1.4: Decreased use of the outdoors
Another reason for decreased physical activity, especially among children in developed areas, is the decreased use of the outdoors. One of the major drivers of the reduction in outdoor time is the increased perception of danger. As individuals become more frightened of the outside world (e.g., random violence, stranger danger, road rage) we can expect more time spent at home, work, or other controlled indoor locations, which will in turn lead to reduced caloric expenditure. This is exemplified by the amount of space a set of 8 year old children in England used in their typical weekly routines (Figure 2.4). Whereas the great-grandfather used to roam over a large area of Sheffield, the great-grandson is allowed to walk only to the end of his street. For a typical child living in these controlled conditions, physical activity must be structured into the daily life patterns of school, extracurricular sports, or other exercise or recreation.
Figure 2.4: Decreased realm of physical activity over 4 generations
2.2: Dietary changes
So on the one hand our lives have changed in a way that demands less physical activity, on the other, our diets have changed in a way that allows excessive caloric consumption, as well as diets conducive to weight gain.
2.2.1: Portion sizes
One simple reason for increased caloric intake is simply an increase in portion sizes. Only a few decades ago, typical portion sizes for foods commonly consumed outside the house were much smaller than they are today. Consider a few common food items in Figure 2.5 below. Simply purchasing and consuming the quantity of the same item in its modern size means a substantial increase in the number of calories consumed. With larger portion sizes being frequently the only option available, the burden is completely on the consumer to exert self-control in consumption.
Figure 2.5: Portion sizes, 1976 to 1996
2.2.2: Preparations
In addition the change in portion size, food preparations have changed, and the most nutritious—and least obesogenic—foods have become the most costly on a calorie-by-calorie basis. Those foods that are most nutritious, such as fresh fruits and vegetables, are sometimes orders of magnitude more costly than the most calorically dense foods, including starches, fats, and sugars (Figure 2.6).
Figure 2.6: Cost and energy density of selected foods (Drewnowski, A. and N. Darmon, 2005.
For people living on limited budgets or fixed incomes, the clear and rational choice for getting the most calories at the least cost means purchasing those foods that are the least nutritious on a caloric basis.
3: Current research on obesity: what are the limitations?
A tremendous amount of recent research has attempted to tease apart some of the causes of obesity. However, nearly all of these studies have been severely limited by one or more fundamental shortcomings in experimental design.
Cross sectional study design. Many studies have a cross-sectional design. That is, studies have investigated a population at a specific point in time. Without knowing about a population’s life history, it is only possible to characterize that population; it is not possible to determine whether any particular factor or set of factors has actually led to its condition. For example, it has been shown that residents of areas with fewer supermarkets had lower prevalence of obesity and overweight (Morland, Diez Roux et al. 2006). Also, areas with more people of color were shown to have fewer food choices (Baker, Schootman et al. 2006) fewer supermarkets and more grocery, liquor, and convenience stores (Block, Scribner et al. 2004). These relationships are fairly clear, but because of cross sectional design, they leave several questions unanswered:
- (How) have residents’ BMI changed over time?
- (How) have the distributions of supermarkets changed over time?
Problems with cross sectional design can only be overcome by the use of longitudinal data with sufficient control over other factors. Randomized clinical trials are considered the “gold standard” of longitudinal studies. However, the suggestion of randomizing individuals to live in particular locations, or to be subjected to particular standards of living should be abhorrent to any sensible researcher, much less to even be considered by an institutional review board.
Lack of generalizability. Many studies have examined the relationship among environment, obesity, food access, and/or socioeconomic position (SEP) for a particular location, e.g., Detroit (Zenk, Schulz et al. 2005), Los Angeles (Lewis, Sloane et al. 2005), New Orleans (Block, Scribner et al. 2004). Again, the relationships are clear, but the results may not be generalizable to other cities or countries. One of the problems with studies that focus on a particular place is that not all places are alike. Each place has a unique history, a unique population, and unique environmental characteristics that limit the ability to extrapolate beyond the study area.
By performing identically designed studies, attempting to match control conditions, and measuring all important variables, some amount of generalizability can be obtained. This is one reason why studies should be documented thoroughly, so that studies can be replicated with the same methodologies. Applied social health sciences lag behind other, more well-established sciences (e.g., chemistry, physics) in this regard.
Aggregation. Most studies investigating obesity and its correlates have used aggregate data. Popular units of analysis include ZIP code areas and census tracts. Typically a large number of these spatial units are measured for several variables, and a statistical model is developed based on these variables. Frequently significant relationships are found that indicate associations between environment, behavior, health outcomes, and/or SEP. There are two main problems with the use of spatially aggregated data.
- The ecological fallacy occurs when ascribing the characteristics of a population to its individuals. Suppose a statistical model has predicted that obesity is strongly associated with wealth (poorer people having higher BMI). Now imagine a census tract in which 25% of the residents are extremely poor and 75% are very wealthy, but the only measured variable is median household income. The “average” individual in this area would be considered relatively wealthy. Applying the statistical model to this population without knowing about the population’s inherent variation in wealth would yield questionable results.
- Zonation problems occur when areal units are constructed for one purpose but analyzed for another. For example, ZIP code areas in the US are developed solely for the efficient distribution of mail. The population within a given ZIP code area may have large heterogeneity with respect to variables of interest, such as health status, race, SEP, medical coverage.
- Boundary problems may occur when areal units are arbitrarily defined. For example, it has been shown that obesity risk was related to the presence of supermarkets in ZIP code tabulation areas (ZCTAs) (Lopez 2007). One of the problems with this type of measurement is that the boundaries of census units typically are created along major arterials; these are also the places where supermarkets tend to be located. Whether a particular store, fast food restaurant, or set of convenience stores fall in one zone or another will change the effective presence and density of these establishments within each zone. However, a resident living near this major arterial will most likely be unaware of and unaffected by the existence of such a boundary. Simply changing the boundaries can result in wildly different results (Fotheringham and Wong 1991).
Measuring and modeling at the individual level will be important in determining the relationship of health outcomes, personal, and environmental characteristics. An active area of development in statistical technique is focused on leveraging individual and aggregate data within the same modeling framework.
Self-report. Much of what is currently known about obesity and its covariates is from self-reported data. For example, the maps of obesity rates in the US come from the CDC’s Behavioral Risk Factor Surveillance System (BRFSS), which relies on self-reported data. It is well known that people tend to overestimate their height, underestimate their weight, and otherwise deviate from “truth” when completing surveys. On the bright side, answers tend to be biased toward what is socially desirable, which means that there is systematic, rather than random error in these measurements. Under conditions of systematic error, and given sufficient information about the pattern in the difference between self-reported and “true” data, it is theoretically possible to develop valid estimates of true measurements.
The only way to get around the known limitations of self-report is to obtain objective measures. For example, instead of asking people to report their height and weight, they can be directly measured, as in the National Health and Nutrition Examination Survey (NHANES). NHANES visits different locales with large cargo trucks filled with sampling equipment for taking measures of dozens of subjective and objective measures, e.g., psychological profile tests, blood chemistry, cardiovascular health. Obtaining a large data set of these measurements is costly, time-consuming, difficult, potentially dangerous with respect to protection of personal information, and subject to sampling bias.
4: Selected obesity research at the University of Washington
Our interdisciplinary research team at the University of Washington (UW) is comprised of students, staff, and faculty in Urban Design and Planning, Public Health / Epidemiology, and Computer Science & Engineering. In this section I briefly describe four ongoing projects designed to address several of the limitations in current obesity research.
4.1: BEST MoveS: The Built Environment Space-Time Movement Study
Many studies of environment and behavior have basic limitations in either spatial or temporal data. In BEST MoveS, we are using a novel device (the Multi-Sensor Platform, or MSP) that measures several environmental variables as well as geospatial position. The device is unobtrusive, can be worn on the belt (Figure 4.1), and provides continuous data at 1 second intervals. In addition to the MSP, subjects carried a cell phone that was customized with the MyExperience survey tool (Froehlich, Chen et al. 2007) for obtaining hourly surveys on activity type, location, duration, and purpose.
Figure 4.1: the MSP, as worn by a subject
Figure 4.2: The customized cell phone interface
Over 50 subjects were enrolled in the study; each participant wore the MSP for an entire week during normal waking hours. A raw GPS trace from a single subject’s data collection over a week is shown in Figure 4.3. These data will be used to determine the overall spatial realm of activity, as well as to identify locations of greatest use (i.e., dwell time).
Figure 4.3: Raw GPS data for one subject over 1 week
A subset of this trace is shown in Figure 4.4. The area in the red rectangle (left) is shown at larger scale (right), with three survey responses. This allows us to sample the location, duration, and purpose of different activities as they occur, rather than in diaries that are typically filled out by subjects at the end of the day. Although this does not prvode continuous diary (log) data, we expect that the individual diary entries will be more accurate, because they are recorded at the time at which the sample was taken.
Figure 4.4: Part of a GPS trace showing survey responses
The multivariate data set from the MSP (Figure 4.5) is processed using a mixed Markov model with Decision Stumps (Lester, Choudhury et al. 2005) for probabilistically classifying data into different activity types.
Figure 4.5: Raw MSP data
This results in a probabilistically classified map representing spatially and temporally continuous data (Figure 4.6). These classified activity traces will be used with GIS overlay analysis to determine the land uses that are most strongly associated with particular types of activities in different individuals.
Figure 4.6: Classified activity traces
This study addresses several fundamental limitations of existing research. First, the unit of measure and analysis is the individual. Second, it measures individuals continuously as they move through space and time, rather than as a single time step. Third, it associates individualized space-time measures with particular spatial locations, rather than associating personal characteristics with properties of a single location (such as home or work). The major limitations of this study are incomplete data coverage (due to battery charge duration, data collection periods did not cover subjects’ entire waking hours) small sample size (with only 50 subjects it will be quite unlikely to produce generalizable results), and potential subject bias (this was a convenience sample of UW students, staff, and faculty).
4.2: BALANCE: Bioengineering Approaches for Lifestyle Activity and Nutrition Continuous Engagement
Although the simple equation for weight homeostasis is Econsumed = Eexpended, one of the most challenging tasks for anyone attempting to maintain or lose weight is actually knowing how many calories they have expended or consumed. For the BALANCE project we have developed a prototype device combination that includes the MSP and a Windows Mobile Smartphone. Previous controlled experimental lab and field work with the MSP, along with a mobile VO2 max caloric expenditure meter, allows us to estimate caloric expenditure in real time. The Smartphone is equipped with a customized database of common food items, which allows users to enter the types and amounts of foods they consume, at the time of consumption. The food database includes fields for calories by item, which allows a near-real-time estimate of caloric consumption (Figure 4.7). The Smartphone contains a Bluetooth module that is used to communicate with the MSP.
Figure 4.7: Food database interface
With the combination of MSP-generated caloric expenditure estimate and the database-driven caloric intake estimate, we produce an “energy gauge” that gives users feedback about their instantaneous energy balance (Figure 4.8). This tool allows individuals to keep track of their behavior, as well as to make changes to their behavior for improving their energy balance. We believe that a tool like this, which provides more fine-grained feedback for caloric consumption and expenditure, will be a great improvement over such other common feedback tools (such as the bathroom scale). In addition to the food database interface and energy gauge, we have built a map interface based on ArcPad (ESRI, Redlands, CA), for displaying the food and fitness facilities located near the user’s current position. The map interface is intended to display locations that will allow users to make sensible choices for eating and exercise. A publication detailing our work on this project is forthcoming (Duncan, Bruemmer et al. in press).
Figure 4.8: The BANANCE energy gauge
Figure 4.9: the BALANCE map interface
One of the strengths of the BALANCE study is its focus on individual-level feedback and intervention. The system we developed has not been used in any clinical trials to date, but one of the reasons we expect it to be useful is that the system will allow us to track—at very fine spatial and temporal scales—the behaviors and choices individuals make on a daily basis. Although public health is concerned with the overall health of populations, many of the important changes in behavior that will drive population-level health improvement need to be ultimately implemented and managed at an individual level.
4.3: TRAC: Travel Assessment and Community
Many built environment features and characteristics are believed to be related to physical activity. Greater neighborhood “walkability” has been associated with higher residential density, presence of utilitarian destinations, availability of transit, and greater land use mix (Cervero and Kockelman 1997; Lee and Moudon 2006; Moudon, Lee et al. 2007). Although associations are strong and clear, some questions persist due to the lack of longitudinal evidence. If people who live in these environments walk more, is it because the use of these facilities stimulates (or requires) walking (e.g., to and from transit stops)? Or is it because people who like to walk preferentially live near these types of features? This analytical problem is known as self selection.
It is both impractical and unethical to randomize subjects to different living environments. However, sometimes planned environmental changes set the stage for natural experiments. In the TRAC study, we are taking advantage of the introduction of a new light rail system to measure the effects of changes in the transit system on overall physical activity, and specifically, transit-related walking.
The study has enrolled some 600 subjects in the first phase of measurement. Subjects wear belt-mounted Actigraph accelerometers (which differentiate no activity from moderate-to-vigorous physical activity) and belt-mounted GPS receivers that include data collectors. The accelerometers and GPS units are configured to record overall activity and location at 30 second intervals. In addition to these automated/objective measurement devices, subjects record their daily activity patterns (e.g., commuting, shopping, recreation) in a travel diary (developed by the Puget Sound Regional Council for its regional travel survey).
The main aim is to measure individuals’ overall physical activity as well as their movements over a one week period, both before and after introduction of the light rail system. Our study employs a longitudinal case/control panel design. We have identified potential cases as those people who reside within walking distance of proposed light rail stops. Controls are identified as those who live beyond walking distance of light rail stops. We have also used a spatial sampling design (Lee, Moudon et al. 2006) that uses census data to match demographic characteristics of the pool of potential cases with the pool of potential controls. We will perform a follow-up measurement phase shortly after the light rail system is completed. We also expect land use changes to occur at a temporal lag that is greater than the individual response to existence of the light rail system. Because we expect these land use changes to have an additional effect on neighborhood-level walking, we will perform a secondary follow-up measurement phase 2 years after completion of the light rail system in the study area.
4.4: SOS: The Seattle Obesity Study
Many important studies have shown relationships among the “food environment,” SEP, and health outcomes. As noted, however, many of these studies have been hampered by fundamental limitations of study design. Probably the two most important shortcomings are cross-sectional design and the use of ecologic (aggregate), rather than individual data. Another important, and frequently overlooked limitation, is the assumption that physical access, whether measured by presence of supermarkets within a ZCTA (e.g., Lopez 2007) or the kernel density estimate of healthy food outlets (e.g., Rundle, Neckerman et al. 2009), equals use of those same facilities. In the SOS we have several aims, the most important of which address the problems of aggregate vs. individual data and the assumption that physical access equals use. In this study, we sampled 2000 households in the greater Seattle area to determine demographic characteristics, food habits, and use of the food environment. Households were randomly selected from a group of ZIP code areas (Figure 4.10, left), stratified for oversampling to obtain sufficient sample size in underrepresented racial/ethnic groups (the Seattle area is predominantly White).
We conducted a 25 minute telephone interview, which included questions about basic demographics (e.g., count of household members, family income, race, education, car ownership), health status (e.g., diabetes, heart disease, perception of health status, height, weight), food usage (e.g., types and amounts of different foods consumed, shopping frequency, amount of money spent on food), and detailed spatial information (home and work address, as well as the location of the primary and secondary food stores and most frequented fast food restaurant [Figure 4.10, center and right]). We also administered a food frequency questionnaire (FFQ) to estimate diet quality for one member of each household. Along with the FFQ, we surveyed the leading local supermarkets with the standard USDA market basket, from which we calculated the standardized cost of the diet for each FFQ.
Figure 4.10: SOS samples: ZIP code sampling areas (left), primary food stores (center), fast food restaurants (right)
The basic survey data (telephone, FFQ) will allow us to investigate individual level relationships among demographics, health outcomes, food preference and use, diet cost, and diet quality. The major strengths of the SOS is the collection of individual data on home and work location, as well as the name and location of the food stores that are actually used by study participants. We use advanced GIS methods for characterizing the local neighborhood around each respondent’s home and work location, and network analysis to estimate the travel distance and time between home, work, and the food stores and restaurants identified in the survey.
- Although we are in the preliminary stages of analysis, some interesting general findings have already been uncovered: The mean network distance from home to the primary food store is much greater than the distance to the closest supermarket (Figure 4.11). This suggests that people are willing to travel further to shop at their selected food store, rather than shopping at the supermarket closest to home.
Figure 4.11: Closest supermarket vs. primary used food store
- One quarter of our subjects live within 1 mile of the primary store at which they shop. One quarter of our subjects live within 0.6 miles of their closest supermarket. This indicates that for many residents of the Seattle area, walking to shop at a grocery store or supermarket is a realistic option.
- The mean network distance from home to the closest fast food restaurant is substantially lower than the distance to the fast food restaurant patronized by our respondents. This suggests that people may take advantage of fast food restaurants as a matter of convenience when traveling between home, work, and other locations. It also suggests that measures to restrict land use permits for fast food restaurants in particular locales may be of limited effectiveness at curbing consumption of fast foods.
- Clear socioeconomic patterns are displayed for diet and food shopping choices. More affluent and well educated residents tend to shop at particular stores, with less affluent and educated residents shopping elsewhere (although the price of many foods does not vary from store to store).
This approach to data collection: obtaining large samples of individual data on location, health outcomes, food consumption, shopping choices, and actual utilization of environment should shed light on frequently assumed—but rarely empirically investigated relationships governing many of the factors purported to be responsible for the recent obesity epidemic.
5: Acknowledgments
This work has been funded by the National Center for Research Resources under award 5P20RR020774-03 (BEST MoveS), the University of Washington Royalty Research Fund (BEST MoveS), the National Institute on Aging and the National Heart, Lung, and Blood Institute under award 5R21AG032232-03 (BALANCE), the National Heart, Lung, and Blood Institute under award 5R01HL091881-02 (TRAC), National Institute of Diabetes and Digestive and Kidney Diseases under award 5R01DK076608-02 (SOS).
Team members include: Anju Aggarwal, Adrienne Andrew, Shirley Beresford, Gaetano Borriello, Barbara Bruemmer, Tamara Denning, Adam Drewnowski, Glen Duncan, Carl Hartung, Deonna Hughes, Jonathan Lester, Anne Moudon, and Laura Streichert.
|
button (top left)