Section

6

Using Secondary Data to Assess Obesity

The methods in Section 5 focus on how to conduct primary data collection for research. Many researchers and practitioners use secondary data to conduct research and evaluation.

In some cases, researchers may have a specific body composition outcome measure in mind and have to look for an existing data set that includes this measure. Alternatively, researchers may have identified a data set of interest with available measures of body composition and need to determine if they can 1) adequately answer their research question with these data and 2) how to analyze and interpret the available measure.

Secondary data can include complex surveys and surveillance systems (e.g., National Health and Nutrition Examination Survey [NHANES]), electronic health records, and data captured from wearable devices. When using secondary data, a research team still needs to consider a method’s validity in estimating body fat; precision, reliability or reproducibility; sensitivity to change over time or with intervention; ability to predict health risks/outcomes (e.g., clinical validity); and the availability of reference ranges or norms for the study population (having data from a standard or reference population for comparisons).

When selecting secondary data to create a comparison group, additional considerations arise. It is important that the two groups are comparable on as many attributes as possible—this includes the outcome assessment methodology (e.g., self-reported vs. objectively measured BMI, cut-points used), sampling approach to ensure representativeness of the populations of interest, timing of data collection, and inclusion of socio-demographic characteristics that may confound results. The NCCOR Catalogue of Surveillance Systems can be used to identify and compare surveillance systems that include weight-related outcomes. The webpage for each surveillance system provides user manuals, which may include information about sampling plans, anthropometric procedures, questionnaires, and analytic guidance. Most nationally representative and state-representative data sets are not over-sampled, which precludes in-depth analyses of specific demographic or health-status subgroups, including those at increased risk of obesity.¹¹³ BMI is the most commonly used measure used to define obesity in surveillance systems.  If using secondary data to evaluate change over time, sample size is another relevant consideration.

Sometimes, published values (e.g., from an NHANES surveillance summary) can offer a useful point of comparison. Often, researchers will want to conduct their own analyses using secondary data. CDC provides SAS code to calculate BMI z-scores and percentiles and to identify biologically implausible values.¹¹⁴ Additionally, the World Health Organization (WHO) provides code for researchers to calculate BMI z-scores and percentiles using WHO growth charts for children younger than 2 years of age. Given that body composition varies by age, gender, and race/ethnicity, it is important that existing data include such demographic variables without systematic missing data. Some data sets may include indicators of pubertal status.

Researchers may find it difficult to identify a data set that is comparable across all these criteria and will need to be transparent about any shortcomings or limitations in their approach. For example, a strength of NHANES is that it includes objectively measured height and weight, a detailed dietary record (24-hour recall), and other health measures. If researchers are looking to evaluate the impact of a national policy on high school students’ BMI and need to be confident that participants are enrolled in grades 9–12, they might prefer a data set such as the Youth Risk Behavior Surveillance System (YRBS). This system uses a self-reported measure of height and weight, which introduces bias, but it is specific to secondary students.

Back to top