Predictive Utility of Alternate Measures of Physical Activity and Diet for Overweight and Obesity in Low-Income Minority Women

Abstract

Purpose

The purpose is to compare the predictive utility of alternate measures of diet and physical activity for overweight and obesity among low-income minority women.

Design

Cross-sectional analysis of baseline data from a cohort study.

Setting

Three public housing developments in South Los Angeles.

Subjects: Adult women (N = 425).

Measures

Primary outcome—weight status (normal BMI, overweight, or obese). Primary predictors— diet: 24-hour dietary recalls (Healthy Eating Index), dietary screener (intake of specific food groups), and single-item survey question (diet quality); physical activity: accelerometry (minutes/day of moderate-to-vigorous activity), short recall questionnaire (minutes/week of moderate and vigorous activity), and single-item questions (days per week did exercise; self-assessment of overall activity level).

Analysis

Multinomial logistic regression models, controlling for socio-demographic covariates. Models are built up starting with least resource-intensive measures of diet and physical activity (single items) and sequentially adding more resource-intensive measures. Model performance is assessed via information-based model selection indices.

Results

Adjusted relative risk for obesity for single-item measures ranged from .61 to .64 for diet (P < .01) and from .80 to .81 for physical activity (P <.05). The added value of resource-intensive measures was negligible for physical activity and at best small for diet.

Conclusion

Single-item questions for diet and physical activity can provide valuable information about risk for overweight and obesity in low-income minority women when more resource-intensive assessments are infeasible.

Keywords

obesity body mass index (BMI)dietary screener physical activity measurement instrument validation

Introduction

Overweight and obesity are a major public health concern and place a disproportionate and growing burden on low-income and Hispanic and non-Hispanic Black women.^1,2 Addressing these disparities requires developing a better understanding of diet and activity behaviors and their association with overweight and obesity in low-income, minority populations.

Conceptually, there are several aspects of diet and activity that may be of interest such as total energy intake, diet-quality, -variety, or -patterns, total energy expenditure, time spent doing vigorous and moderate activity, or fitness for physical activity. A wide range of assessments have been used in population studies to measure these aspects of diet and physical activity.^3,4 These assessments are primarily based on self-reports and range from single-item measures, to short and long recalls over a pre-specified period, to detailed diaries and logs of diet and activity. A central question is which conventional measures of diet and physical activity are most informative, more specifically, most predictive of overweight or obesity, in at-risk populations.

The validity of these alternate assessments has been studied extensively.^4,5 The emerging consensus is that there is no single best measure that optimizes both validity and cost or feasibility in population-based studies, nor is universally applicable in different contexts (e.g., across regions within states or across countries) or across populations (age, race/ethnicity, income, literacy, etc.). Experts have suggested administering multiple measures in a study to maximize the strengths of each measure when the budget allows.^6,7 However, there are two key gaps in the literature that we seek to address in the present study. First, the bulk of the validation literature is conducted in Caucasian samples,⁵ so more evidence is needed in minority populations, which bear the larger burden of overweight and obesity. Second, the diet and physical activity validation literatures have been largely siloed, despite both contributing to overweight and obesity. Although there is not yet a consensus causal model of how physical activity interacts with food intake to influence body composition status, both are critical components of the overall energy expenditure⁸ and the joint improvement on both aspects was shown to induce larger long-term weight loss than focusing on one aspect alone.⁹ Our study intends to build upon the limited literature^10-13 by systematically examining the predictive utility of diet and physical activity measures for body composition simultaneously.

In this paper, we compare the performance of alternate measures of diet and physical activity for predicting overweight and obesity in a sample of low-income minority women. Our goal is to assess whether measures based on approaches that are more time and resources-intensive (i.e., dietary recalls and accelerometry) add value over easy-to-administer single-item measures in terms of their predictive utility for overweight and obesity.

Methods

Study Design and Sample

Our data come from the Watts Neighborhood Health Study, an ongoing longitudinal cohort study of low-income urban public housing residents located in South Los Angeles. Briefly, the study is designed to examine the impact of a community redevelopment on residents’ obesity and related behaviors. The study recruited adult residents from three public housing developments in the Watts neighborhood over a 2-year period (2018–2019). A total of 868 adults were recruited and two waves of baseline data were collected during this period, prior to the beginning of the redevelopment. Adults who were recruited in year 1 were invited to participate in both rounds of baseline data collection (N = 494), whereas those recruited in year 2 only participated in one round. Over the two waves of baseline data collection, we administered a few different diet and physical activity assessments allowing us to study how they correlated with each other and compare their performance in terms of how well they predicted objectively measured overweight and obesity status in this high-risk population.

This study uses data on adult women who participated in both waves 1 and 2 (n = 425). There were 45 men who participated in both waves, but were excluded from the analyses due to small sample size. The higher participation of women is not unusual and is in line with the demographic composition of public housing developments; about 56% of housing units at the three study sites are female-headed with children.¹⁴

In both waves, respondents participated in a Computer-Assisted Personal Interview (CAPI) and body composition measurements by trained study staff.

Measures

Body Mass Index, Overweight, and Obesity

Trained study staff measured participant’s height and weight using a standardized protocol. Height was measured using a stadiometer (Charder HM200P Portstad Portable Stadiometer, Charder), rounded to the nearest .1 cm. Weight was measured using a Tanita UM-081 digital scale, recorded to the nearest .1 kg. Prior to each wave of data collection, accuracy of each scale was assessed by study staff using a standardized 5lb weight. All participant measurements were taken at least twice, and a third measurement was taken if the two measurements differed by a pre-determined amount (>0.5 cm for height and >0.2 kg for weight). The average of the two closest measurements was used as the final measure. BMI was computed as the ratio of the measured weight [kg] to height [m]-squared. We constructed indicators of normal weight, overweight, and obese, where normal weight is defined as BMI <25, overweight is defined as BMI between 25 and <30, and obesity is defined as BMI at 30 or higher.

Diet and Physical Activity

We first administered the assessments that are considered to have higher validity—24-hour dietary recalls for diet, and accelerometry for physical activity. However, these assessments presented several challenges in this population, including logistics, compliance, mistrust, and respondent burden. Therefore, shorter and easier assessments were added thereafter, creating the opportunity to compare the predictive utility of the different measures in this population, with respect to the outcomes of interest— overweight and obesity. These measures are described in detail below.

Three different dietary assessments were administered at baseline, including 24-hour dietary recalls, a brief dietary screener, and a single-item question for self-reported overall diet quality.

Dietary Recalls: Two interviewer-assisted 24-hour dietary recalls were conducted using National Cancer NCI’s ASA24® 24-hour recall tool.¹⁵ The tool guides respondents through multiple steps of recalls including meal-based list, gap review, detailed pass, forgotten foods, and a final review. It also asks respondents if the recall captures usual intake, more-than-usual intake, or less-than-usual intake. We used data from the ASA24® to create the Healthy Eating Index (HEI),¹⁶ a measure (0–100) of diet quality to assess how well food intake aligns with key recommendations of the Dietary guidelines for Americans. We also computed average cups of fruit and vegetables per day, average ounces of whole grains per day, and average cups of dairy per day from the food recall data for sensitivity analyses.

Dietary Screener: Adult participants completed the National Cancer Institute’s (NCI’s) Dietary Screener Questionnaire (DSQ) to assess consumption frequency of key food groups.¹⁷ Respondents were asked how often each item was consumed in the past 30 days (never, 1 time, 2–3 times, 1 time per week, 2 times per week, 3–4 times per week, 5–6 times per week, 1 time per day, and 2 or more times per day). Using the data from the DSQ, we constructed measures of predicted cups of vegetables including legumes but excluding french fries per day, predicted ounces of whole grains per day, predicted cups of dairy per day following data processing and scoring procedures developed for the National Health and Nutrition Examination Survey.¹⁸ In separate analyses not shown here, we also included predicted teaspoons per day of added sugars from sugar-sweetened drinks but it was excluded from the present analyses as higher reported sugar intake was associated with lower relative risk of overweight and obesity, suggesting either underreporting of or reduced intake of sugar-sweetened drinks among those who have higher body fat composition.

Single-item Diet Quality Question: Participants were also asked a single-item validated^19,20 self-reported measure of overall diet quality: “In general, how healthy is your diet? Would you say it is… Excellent (5), Very good (4), Good (3), Fair (2), Poor (1).”^19,20

Assessment of physical activity (PA) included accelerometer-assessed moderate-to-vigorous physical activity (MVPA), a short recall of leisure time physical activity, and two single-items for self-reported overall physical activity.

Accelerometer-Assessed Physical Activity: Participants were also asked to wear an Actigraph, Inc. wGT3X-BT accelerometer on their waist for 7 days to collect continuous physical activity data. Moderate-to-vigorous physical activity (MVPA) was defined as at least 2020 counts per minute, consistent with other studies.²¹ Total minutes of MVPA during each valid day was computed.²¹ We calculated the average minutes per day of MVPA for up to 7 valid days of accelerometer data among participants with at least 2 days of valid data.^22-24

Short Recall of Leisure Time Physical Activity: Adults reported days per week and minutes per day of moderate and vigorous activity outside of work for the past week, similar to the National Health and Nutrition Examination Survey.^25,26 Specifically, adults were asked, “During the last week, on how many days did you do vigorous physical activities for 10 minutes or more at a time, like circuit weight training, sports (basketball, soccer, baseball), Zumba, heavy housework, running, when you were NOT at work?” If participants did vigorous activity on one or more days, they were then asked, “How much time did you usually spend on 1 of those days doing vigorous physical activities when you were not at work?” Similar questions were asked for moderate physical activity, such as carrying light loads, bicycling, gardening, putting groceries away, skateboarding, moderate housework, or fast walking. Vigorous and moderate physical activity measures were each capped at 2940 min per week to avoid distortion of the results due to implausibly large values. Adults also reported time spent watching TV and playing video games. We constructed a measure of total weekly time spent on these sedentary activities, capped at 3360 minutes per week. Time spent on these three types of activities were used as separate predictors in the analysis, allowing a more nuanced model specification.

Single-Item Physical Activity Questions: We developed two single-item questions about physical activity to address two concerns with the other self-reported measure. First, given that the moderate and vigorous activity self reports had to be capped for some respondents, it suggested that the distinction between moderate and vigorous physical activity was not consistently clear to respondents despite providing visual examples and textual explanations of each type. And second, the sample varied considerably in terms of age, disability/health status, and education levels, so a question that was simple to interpret for respondents of varying backgrounds was needed. The first question was a global measure of physical activity level that asked adults to rate their level of daily physical activity on a scale of 1 to 10 (On a scale of 1 to 10, where 1 means you spend most of your day sitting or lying down and 10 means you spend most of your day moving around on your feet, how physically active are you? 1 means spend most of day sitting or lying down; 10 means spend most of day moving around on feet). The second question asked on how many days they were physically active for at least 20 minutes at a stretch during the past week.²⁷

Covariates

Adults reported the following covariates: age (18–34, 35–54, and 55 or older), ethnicity (Hispanic and non-Hispanic), education level (less than high school, high school graduate, more than high school, and some college/technical school or higher), household income ($9,999 or less, $10,000–$19,999, and $20,000 or more), coupled (married or living as married), and working on a paid or unpaid basis.

Analysis

The analyses were restricted to adult women who completed both waves of the surveys (N = 425), excluding 24 cases with missing variables other than the accelerometer data. For a total of 76 cases that were missing accelerometer data, we imputed their MVPA minutes per day using regression-based multiple imputation with other PA variables as well as dietary and demographic variables, and averaged across the 20 imputed sessions.²⁸ We examined the summary statistics describing the variable distributions, including correlation among dietary and PA measures with Pearson and Spearman’s rank correlations.

The primary analyses rely on the multinomial logistic regression predicting the probability of being in one of three BMI categories (normal weight, overweight, and obese) using the dietary or PA measures. For each domain (i.e., diet and PA), we ordered the measures by the easiness and cost-efficiency to administer the assessment. For diet, the order was single item, then dietary screener (DSQ), and lastly 24-hour dietary recall (ASA24®). For PA measures, the order was single items, short recall of leisure time physical activity, and accelerometer-assessed MVPA. The model was built up by adding one type of measure at a time.

Selection of the most parsimonious model, or essentially, the most predictive and efficient sets of variables, was determined by several criteria that strive to balance the statistical significance, interpretability and practical value. We considered a costlier assessment to have added value if (a) the derived measures significantly predicted overweight and/or obesity status with clinically meaningful effect size, (b) the model-data fit was significantly poorer when the derived measures were omitted based on the likelihood ratio test, in consideration of other information-based model selection indices such as Akaike’s information criterion (AIC) and Bayesian information criterion (BIC),²⁹ or (c) inclusion of such derived measures meaningfully increased the overall predictive power of the model according to McFadden’s R², a popular pseudo-R² statistic.³⁰ Such a measure selection process was performed on dietary and PA measures separately, and the final selected measures were used jointly in the same model to predict the likelihood of normal weight, overweight, or obese status.

Statistical sensitivity analyses were also conducted to investigate the robustness of the findings. Specifically, we replicated the above analyses restricting to the subsample that had complete accelerometer data, as a check for the robustness due to multiple imputation. In addition, we performed the model building using (a) alternative ASA24®-derived measures (average cups of fruit and vegetables per day, average ounces of whole grains per day, and average cups of dairy per day, adding quadratic term of HEI), (b) the sample with 2 completed ASA24® recalls, or the sample with only recalls that represent usual intake, and (c) the sample with at least 4 days of valid accelerometer data (to increase reliability.³¹ Furthermore, as an alternative to using the predictors as continuous variables, we also conducted the analyses after categorizing the variables, with thresholds recommended in the literature or, if not available, by splitting the variable at the observed median. Finally, we replicated the measure selection process when severe obesity, with BMI≥40, was considered as an additional outcome category, separated from non-severe obesity (i.e., 40>BMI≥30).

Analyses were performed using Stata, version 16. All models adjusted for age, race/ethnicity, education, household income, married or living as married, and working status.

Results

Table 1 includes the socio-demographic characteristics of the sample as well as the summary statistics of the dietary and PA measures. More than half (60.9%) were Hispanic, with the rest almost all identifying as non-Hispanic Black, except 6 who identified as non-Hispanic and non-black. Three quarters (74.8%) were ≥ 35 years old. Close to half (47.3%) held a degree less than high school, while half (50.1%) had annual household income <$10,000. More than a quarter (29.4%) were married or living as married, and less than a third (31.3%) worked on a paid or unpaid basis. Among them, 65.2% were obese and another 21.4% were overweight.

Table 1.

Sample Descriptive Statistics (N = 425).

	% or Mean (SD)
Hispanic	60.9%
Age
18–34	25.2%
35–54	52.9%
55+	21.9%
Education
Less than high school	47.3%
High school	32.5%
More than high school	20.2%
Household income
$9,999 or less	50.1%
$10,000–$19,999	31.5%
$20,000 or more	18.4%
Married or living as married	29.4%
Working on paid or unpaid basis	31.3%
Weight status
Normal weight	13.4%
Overweight	21.4%
Obese	65.2%
Diet
Overall diet quality (rating 1–5)	2.35 (1.08)
DSQ fruits and vegetables (cups per day)	2.52 (.82)
DSQ whole grains (oz per day)	.91 (.52)
DSQ dairy (cups per day)	1.48 (.53)
ASA24® fruits and vegetables (cups per day)	2.29 (1.56)
ASA24® whole grains (oz per day)	.45 (.75)
ASA24® dairy (cups per day)	.92 (.90)
ASA24® healthy eating index	51.01 (12.39)
Physical activity
Overall physical activity (rating 1–10)	8.05 (2.28)
Number of days physically active for at least 20 minutes in past week	4.60 (2.24)
Leisure time vigorous activity (minutes per week)	201.98 (412.01)
Leisure time moderate activity (minutes per week)	534.25 (646.63)
Sedentary activity (minutes per week)	877.71 (901.46)
Accelerometer moderate-to-vigorous physical activity (MVPA) (minutes per day)	28.22 (21.48)

DSQ: Dietary Screener Questionnaire.

ASA24®: Automated Self-Administered 24-hour Dietary Assessment Tool.

Table 2 shows the correlations across the dietary and PA measures, with the Pearson correlations below the diagonal line and the Spearman’s rank correlations above the diagonal. The correlations across different dietary measures based on the same assessment (i.e., DSQ or ASA24 ®) were low to moderate (.04 to .53 for Pearson’s correlation; −.06 to .54 for Spearman’s correlation), whereas the correlations across assessments were generally low (.00 to .25 for Pearson’s correlation; −.01 to .23 for Spearman’s correlation). For the PA measures, the correlations within the same assessment (i.e., single item, short recall of leisure time physical activity, and accelerometer) and across assessments were mostly low (−.01 to .23 for Pearson’s correlation; −.02 to .26 for Spearman’s correlation).

Table 2.

Correlation of Dietary and Physical Activity Measures (Lower Triangle: Person’s Correlation; Upper Triangle: Spearman’s Correlation).

		(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)
(1)	Overall diet quality	1.00	.12	.05	−.05	.13	.07	.06	.23	.20	.18	.09	.01	−.07	−.03
(2)	DSQ fruit and vegetables	.09	1.00	.31	.17	.23	.10	−.04	.18	.17	.14	.18	.11	.01	.00
(3)	DSQ whole grain	.03	.38	1.00	.39	.12	.13	.05	.12	.05	−.04	.13	.11	.04	−.01
(4)	DSQ dairy	−.02	.21	.47	1.00	−.01	.07	.24	−.08	.06	−.02	.19	.06	.08	.06
(5)	ASA24® fruit and vegetables	.06	.25	.14	−.02	1.00	.10	.08	.54	.01	−.04	−.03	−.02	−.14	.02
(6)	ASA24® whole grain	.01	.03	.09	.00	.10	1.00	.13	.40	.02	−.04	.08	.09	−.06	−.03
(7)	ASA24® dairy	−.01	−.03	.05	.27	.07	.04	1.00	−.06	.07	.00	.18	.10	.05	.05
(8)	ASA24® healthy eating index	.18	.21	.15	−.13	.53	.37	−.06	1.00	.01	.03	−.09	−.01	−.23	.05
(9)	Overall physical activity	.23	.19	.10	.06	.02	−.03	.07	.00	1.00	.26	.07	−.02	−.08	.22
(10)	Number of days physically active	.20	.10	−.01	−.02	−.06	−.02	−.02	.01	.23	1.00	.17	.11	−.04	.09
(11)	Leisure time vigorous activity	.00	.17	.22	.23	.03	.01	.15	−.07	.08	.13	1.00	.09	.07	.06
(12)	Leisure time moderate activity	−.01	.10	.12	.03	−.05	.10	.00	.00	.05	.08	−.01	1.00	.07	.08
(13)	Sedentary activity	−.07	.00	.04	.14	−.08	−.07	.01	−.20	−.09	−.04	.13	.05	1.00	−.02
(14)	Accelerometer MVPA	−.03	.02	.01	.06	.04	−.01	.02	.05	.17	.08	.02	.04	−.06	1.00

DSQ: Dietary Screener Questionnaire; ASA24®: Automated Self-Administered 24-hour Dietary Assessment Tool; MVPA: moderate-to-vigorous physical activity.

Table 3 reports the adjusted relative risks (ARRs) for the dietary measures. Better diet quality overall (single item) reduced the risk of obesity (ARR ranged from .61 to .64 across models, P < .01). Higher dairy intake in the DSQ reduced the risks of being obese (for each additional cup per day, ARR = .51, 95% CI = .28, .91 in Model 2 and slightly higher in Model 3) or being overweight (ARR = .47, 95% CI = .23, .97 in Model 2 and slightly higher in Model 3). Higher ASA24® HEI score was associated with a slightly higher risk of being overweight (for each additional 10 points on HEI, a unit size slightly smaller than the observed SD of 12 points, ARR = 1.04, 95% CI = 1.01, 1.08).

Table 3.

Adjusted Relative Risk of Dietary Measures Predicting Obesity Groups.

	Model 1		Model 2		Model 3
	Relative Risk	95% CI	Relative Risk	95% CI	Relative Risk	95% CI
Overweight
Overall diet quality (rating 1–5)	.90	[.66,1.22]	.89	[.65,1.22]	.83	[.61,1.15]
DSQ fruit and vegetables (cups per day)			.80	[.49,1.30]	.73	[.45,1.19]
DSQ whole grains (oz per day)			1.86	[.84,4.13]	1.50	[.66,3.42]
DSQ dairy (cups per day)			.47*	[.23,0.97]	.60	[.28,1.27]
ASA24® healthy eating index (per 10 pts)					1.04*	[1.01,1.08]
Obese
Overall diet quality (rating 1–5)	.64**	[.49,0.84]	.63**	[.48,0.83]	.61***	[.46,0.81]
DSQ fruit and vegetables (cups per day)			1.11	[.74,1.65]	1.05	[.70,1.58]
DSQ whole grains (oz per day)			1.25	[.62,2.49]	1.10	[.54,2.23]
DSQ dairy (cups per day)			.51*	[.28,0.91]	.58	[.32,1.06]
ASA24® healthy eating index (per 10 pts)					1.26	[.94,1.69]
Likelihood ratio test (df)			9.98 (6)		6.41* (2)
AIC	733.88		735.90		733.49
BIC	823.03		849.36		855.05
Pseudo R²	.076		.089		.098

* P < .05, ^** P < .01, ^*** P < .001.

DSQ: Dietary Screener Questionnaire; ASA24®: Automated Self-Administered 24-hour Dietary Assessment Tool.

Model 1 includes single-item diet quality; Model 2 adds dietary screener variables; Model 3 adds Health Eating Index from the ASA24® dietary recall.

All models include age, ethnicity, education level, household income, coupled, and working on paid or unpaid basis.

Models were then assessed and compared by whether the additional predictors improved the model-data fit or enhanced the explanatory power of the outcome variability. The likelihood ratio test suggests that adding DSQ variables beyond the single-item overall rating of dietary quality (Model 2 vs 1) did not significantly improve the model-data fit (χ²⁶ = 9.98, P > .05), and the information criterion, AIC and BIC, also increased, implying that the little gain in model-data fit did not sufficiently compensate for the increase in model complexity, favoring the simpler model with overall dietary quality only. However, the pseudo R² increased from .076 to .089 after including the DSQ variables, suggesting that the DSQ variables helped to explain more variability in obesity status. When assessing the added value of ASA24® HEI (Model 3 vs 2), both the likelihood ratio test (χ²² = 6.41, P < .05) and AIC (decreased from 735.9 to 733.5) favored inclusion of HEI, but the BIC increased (from 849.4 to 855.1). The pseudo R² also increased with HEI. Although the model comparisons suggest that the addition of the ASA24® HEI resulted in marginal improvement in model-data fit, the coefficient estimate on the ASA24® HEI was counter-intuitive, with higher index of diet quality associated with higher risk of overweight and obesity. Therefore, Model 2 was favored, which included the single item for self-reported overall diet quality and measures derived from the DSQ.

Table 4 presents the results for the PA measures. Higher rating of overall physical activity was associated with lower risk of obesity (ARR ranged from .80 to .81, P < .05), while higher self-reported number of days per week of physical activity was associated with lower risk of overweight (with one more day, ARR ranged from .80 to .81, P < .05). Neither recall of leisure time physical activity (with additional 30 minutes per week, ARR ranged from .99 to 1.00, P > .05) nor accelerometer-assessed MVPA (with additional 10 minutes per day, ARR ranged from 1.04 to 1.08, P > .05) was predictive of obesity status. The model comparisons (Model 2 vs 1 and 3 vs 2) suggested that inclusion of recall of leisure time physical activity or accelerometer-assessed MVPA did not meaningfully improve model-data fit (non-significance by likelihood ratio test and the increase in both AIC and BIC) or enhance the overall explanatory power (pseudo R² increased trivially from .080 in Model 1 to .085 in Model 2 and .087 in Model 3). These results favored the simplest model using only the two single-item questions.

Table 4.

Adjusted Relative Risk of Physical Activity Measures Predicting Obesity Groups.

	Model 1		Model 2		Model 3
	Relative Risk	95% CI	Relative Risk	95% CI	Relative Risk	95% CI
Overweight
Overall physical activity (rating 1–10)	.92	[.76,1.12]	.92	[.76,1.12]	.92	[.75,1.12]
Number of days physically active for at least 20 minutes in past week	.83^*	[.69,0.99]	.82^*	[.68,0.98]	.82^*	[.68,0.98]
Leisure time vigorous activity (per 30 min)			1.00	[.97,1.02]	1.00	[.97,1.02]
Leisure time moderate activity (per 30 min)			1.00	[.98,1.02]	1.00	[.98,1.02]
Sedentary activity (per 30 min)			.99	[.98,1.00]	.99	[.98,1.00]
Accelerometer MVPA (per 10 min)					1.04	[.86,1.26]
Obese
Overall physical activity (rating 1–10)	.81^*	[.68,0.96]	.81^*	[.69,0.96]	.80^*	[.67,0.95]
Number of days physically active for at least 20 minutes in past week	.87	[.74,1.01]	.86	[.73,1.01]	.86	[.73,1.01]
Leisure time vigorous activity (per 30 min)			.99	[.97,1.01]	.99	[.97,1.01]
Leisure time moderate activity (per 30 min)			1.00	[.99,1.02]	1.00	[.99,1.02]
Sedentary activity (per 30 min)			.99	[.98,1.00]	.99	[.98,1.00]
Accelerometer MVPA (per 10 min)					1.08	[.91,1.28]
Likelihood ratio test (df)			3.63 (6)		1.05 (2)
AIC	734.66		743.03		745.98
BIC	831.91		864.59		875.64
pseudo R²	.08		.085		.087

* P < .05, ^** P < .01, ^*** P < .001.

MVPA: moderate-to-vigorous physical activity.

Model 1 includes single-item physical activity items; Model 2 adds short recall of leisure time physical activity items; Model 3 adds accelerometer-assessed physical activity.

All models include age, ethnicity, education level, household income, coupled, and working on paid or unpaid basis.

Because both the dietary and the PA measures were associated with obesity status and inclusion of one domain in the model might alter the results of the other, we then repeated the above model building and selection procedure starting with the selected set of variables from the other domain in the model. In other words, we conducted the dietary model build and selection after controlling for the two single-item PA questions and socio-demographics, and the PA model build and selection after controlling for the single-item dietary question and DSQ measures in addition to the socio-demographics. The results were similar to those described in Tables 3 and 4, and are available upon request. Results from the final joint model that included the preferred diet and PA measures are reported in Table 5 and were similar to those from separate models for diet and PA reported in Tables 3 and 4, respectively.

Table 5.

Adjusted relative risk of selected diet and physical activity measures predicting obesity groups.

	Relative Risk	95% CI
Overweight
Overall diet quality (rating 1–5)	.97	[.70,1.35]
DSQ fruit and vegetables (cups per day)	.85	[.52,1.38]
DSQ whole grains (oz per day)	1.84	[.83,4.08]
DSQ dairy (cups per day)	.46^*	[.22,0.96]
Overall physical activity (rating 1–10)	.95	[.78,1.16]
Number of days physically active for at least 20 minutes in past week	.82^*	[.68,0.98]
Obese
Overall diet quality (rating 1–5)	.70^*	[.52,0.93]
DSQ fruit and vegetables (cups per day)	1.22	[.81,1.84]
DSQ whole grains (oz per day)	1.20	[.60,2.41]
DSQ dairy (cups per day)	.51^*	[.28,0.93]
Overall physical activity (rating 1–10)	.84^*	[.71,0.99]
Number of days physically active for at least 20 minutes in past week	.87	[.74,1.03]
AIC	729.36
BIC	859.03
pseudo R²	.109

* P < .05, ^** P < .01, ^*** P < .001.

DSQ: Dietary Screener Questionnaire.

Model includes age, ethnicity, education level, household income, coupled, and working on paid or unpaid basis.

Sensitivity Analyses

The main results presented above were robust across several sensitivity analyses. Analyses restricting to the subsample of participants with only recalls representing usual intake (Supplementary Appendix Table 1) and only participants with two ASA24® recalls yielded similar results. In addition, analyses restricting the subsample to participants with at least 4 days of valid accelerometer MVPA data (Supplementary Appendix Table 2), restricting to participants with observed accelerometer MVPA, and alternative configurations for diet and PA variables data yielded similar estimates. Models predicting a 4-category body composition status, with severe obesity (n = 75) as a separate category (Appendix Tables 3 and 4), yielded similar findings.

Discussion

This study compared the predictive utility of alternate measures of diet and PA with respect to overweight and obesity among low-income minority women. The easiest to collect were single-item questions included in surveys, followed by brief dietary screeners and physical activity questions, also fielded in surveys. These measures have often been found to have lower validity.^4,5 In contrast, the most resource-intensive (for researchers) and burdensome (for participants) assessments were the 24-hour dietary recalls and accelerometry (over a 7-day period), which have been found to have higher validity.^4,5

Our results indicate that single-item measures of diet and PA can be very informative about the risk for overweight and obesity in this population. The added value of more resource-intensive measures based on 24-hour dietary recalls and accelerometry was negligible for PA and at best small for diet.

The predictive utility of self-reported single-item measures of diet quality and physical activity has been demonstrated in prior studies. Loftfield et al. assessed a single-item measure of diet quality, similar to that used in the present study, in a large community sample of adults in New York City and found it to be highly predictive of blood pressure and BMI.¹⁹ A single-item measure of physical activity, the Saltin-Grimby Physical Activity Level Scale, has been assessed in several studies conducted in European samples.^32-34 This measure asks respondents to self-rate their level of leisure time physical activity and has been shown to be predictive of cardiovascular risk factors, including BMI. Our contribution is to assess how well short, simple, and easy-to-administer measures predict obesity among low-income minority women. Moreover, we simultaneously assess the predictive utility of diet and PA, in contrast to prior work that either examines diet or PA.

There are several possible reasons that might explain the relatively higher utility of single-item measures of diet and PA relative to the more intensive measures in our study. First, dietary recalls and accelerometry assess short-term diet and PA compared to short questionnaires or single-item questions that can capture long-term or “usual” diet and may, therefore, correlate better with obesity. For example, the substantial day-to-day variation inherent in 24-hour dietary recalls generates within-person random error that leads to attenuation in the diet-disease relationship.³⁵ Therefore, it is recommended that multiple recalls, often more than two non-consecutive recalls, are needed to minimize measurement error. However, because the respondents felt most challenged by the lengthy time to complete and the requirement of multiple recalls, our study was only able to administer at most 2 recalls and only a portion of respondents completed both. Second, our sample consists entirely of women, and women have been shown to be more likely than men to misreport energy intake when using multiple pass recalls, as in the ASA-24.⁵ Furthermore, misreporting is also higher among overweight and obese individuals,⁵ which is another potential factor given the high rates of both in our sample. Comparison of our findings to a NHANES-based analysis for low-income adult women also shows that our sample reported lower overall calorie intake than NHANES (an average of 1543 total calories/day in our sample compared to about 1700 total calories/day in NHANES), which may suggest under-reporting in our ASA-24 recalls. Third, the high participant burden from dietary recalls and accelerometry can induce reactivity, especially when these assessments are announced in advance, as was the case in our study. Respondents may try to reduce the burden by “simplifying the reporting process (e.g., consuming single foods rather than combination foods) or to comply with socially desirable norms (i.e., to appear to have a healthy diet by reporting intake as per recommended in dietary guidelines)”.⁵ Finally, much of the validation work for 24-hour dietary recalls and accelerometry-based PA measures is based on Caucasian samples, with limited evidence in low-income, minority women.

Our study has some limitations. First, the analysis was based on cross-sectional data, in which PA and diet measures were administered in the same wave as the outcome BMI measures. As a result, there is the possibility of reverse causation, where obesity status influenced the self-reported healthiness of diet and level of physical activity. As we collect more waves of data on this cohort, we will be able to prospectively examine the predictive utility of our baseline diet and PA measures on overweight. Second, our sample comprises of adult women in 3 public housing developments in Los Angeles, which may limit its generalizability. However, our sample characteristics were comparable to those of public housing residents in LA County (77.4% Hispanic, 19.34% African American or Black; 49% 18–40 years old, 33% 41–60 years old, 19% 61 years old or older; average household income $23,063).¹⁴ Moreover, the high rates of overweight and obesity in our sample are comparable to those found in other studies of public housing sites^36-38 and make this a particularly important group to study. Third, while BMI is a useful measure for public health surveillance purposes and is cost-effective for studies with large and/or geographically dispersed samples, it is an imperfect measure of adiposity.³⁹ Four, although our study robustly found that single-item measures have good predictive utility, these measures were developed for our study specifically and their psychometric properties need to be tested. Since multi-item instruments usually have better psychometric properties, a deeper understanding of why multi-item instruments did not perform better than single-item questions in this case is needed, for example, to what extent it is attributable to education attainment, and in particular, respondents’ numeracy or social desirability. Last, we considered physical activity and dietary measures as linear predictors of body composition in our analysis, explaining only a small proportion of BMI variance. Although not unique to our study,⁴⁰ this suggests that model misspecification may be a potential concern. Expansion of theoretical models to describe more accurately the ways in which dietary choices and physical activity measures are associated with body composition is critical for future work.

In conclusion, single-item questions that assess diet quality and physical activity level can provide valuable information about risk for overweight and obesity in low-income minority women when more resource-intensive assessments are infeasible.

So What?

• What is already known on this topic?

There are no single best measures of diet or physical activity that optimize both validity and cost or feasibility in population-based studies, or universally applicable in different contexts or across populations.

• What does this article add?

An assessment of whether diet and physical activity measures based on approaches that are more time and resources-intensive (i.e., dietary recalls and accelerometry) add value over easy-to-administer single-item measures in terms of their predictive utility for overweight and obesity in low-income minority women.

• What are the implications for health promotion practice or research?

Supplemental Material

sj-pdf-1-ahp-10.1177_08901171211069992 – Supplemental Material for Predictive Utility of Alternate Measures of Physical Activity and Diet for Overweight and Obesity in Low-Income Minority Women

Supplemental Material, sj-pdf-1-ahp-10.1177_08901171211069992 for Predictive Utility of Alternate Measures of Physical Activity and Diet for Overweight and Obesity in Low-Income Minority Women by Ying Liu, Victoria Shier, Sara King and Ashlesha Datar in American Journal of Health Promotion

Footnotes

Acknowledgments

This research was supported by grants from the National Cancer Institute (R01CA228058) and the Eunice Kennedy Shriver National Center for Child Health and Human Development (R01HD096293). All opinions are those of the authors and not of the funding agency. We acknowledge the generous support of the Housing Authority of the City of Los Angeles (HACLA) for conducting this study. The authors thank the Community Coaches and residents at the study sites for their participation and support for this study.

Author Contributions

Ying Liu contributed to the concept of the work, analysis of data, interpretation of data, drafting the article, and article revisions. Victoria Shier contributed to the concept of the work, interpretation of data, drafting the article, and article revisions. Sara Ellen King contributed to cleaning the data, interpretation of data, and drafting the article. Ashlesha Datar contributed to the concept of the work, analysis of data, interpretation of data, drafting the article, and article revisions.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Cancer Institute and Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Ethical approval

The study was approved by the University of Southern California’s Institutional Review Board (UP-17-00842).

ORCID iD

Ashlesha Datar

Supplemental Material

Supplemental material for this article is available online.

References

Hales

Carroll

Fryar

Ogden

. Prevalence of Obesity and Severe Obesity Among Adults, United States 2017-2018. National Center for Health Statistics; 2020.

Ogden

Fakhouri

Carroll

Hales

Fryar

, et al. Prevalence of Obesity Among Adults, by Household Income and Education. United States: Centers for Disease Control and Prevention; 2017:1369-1373. 2011–2014.

Shim

Kim

. Dietary assessment methods in epidemiologic studies. Epidemiology and Health. 2014;36.

Dowd

Szeklicki

Minetto

, et al. A systematic literature review of reviews on techniques for physical activity measurement in adults: a DEDIPAC study. Int J Behav Nutr Phys Activ. 2018;15:15.

Burrows

Rollo

Collins

. Validity of dietary assessment methods when compared to the method of doubly labeled water: a systematic review in adults. Front Endocrinol. 2019;10:850.

Subar

Freedman

Tooze

, et al. Addressing current criticism regarding the value of self-report dietary data. J Nutr. 2015;145(12):2639-2645.

Carroll

Midthune

Subar

, et al. Taking advantage of the strengths of 2 different dietary assessment instruments to improve intake estimates for nutritional epidemiology. Am J Epidemiol. 2012;175(4):340-347.

Westerterp

. Control of energy expenditure in humans. Eur J Clin Nutr. 2017;71(3):340-344.

Johns

Hartmann-Boyce

Jebb

Aveyard

Group

BWMR

. Diet or exercise interventions vs combined behavioral weight management programs: a systematic review and meta-analysis of direct comparisons. J Acad Nutr Diet. 2014;114(10):1557-1568.

10.

Van Gemert

Peeters

May

, et al. Effect of diet with or without exercise on abdominal fat in postmenopausal women - a randomised trial. BMC Public Health. 2019;19(1):174-179.

11.

Kim

Hwang

J-Y

Kim

Park

Kwon

. Diet quality, physical activity, and their association with metabolic syndrome in Korean adults. Nutrition. 2019;59:138-144.

12.

Salas-Salvadó

Díaz-López

Ruiz-Canela

, et al. Effect of a lifestyle intervention program with energy-restricted Mediterranean diet and exercise on weight loss and cardiovascular risk factors: one-year results of the PREDIMED-Plus trial. Diabetes Care. 2019;42(5):777-788.

13.

Epstein

Wing

Koeske

Valoski

. Effects of diet plus exercise on weight change in parents and children. J Consult Clin Psychol. 1984;52(3):429-437.

14.

Statistics & Demographic Overview . Housing Authority of the City of Los. Angeles Public Housing Sites; 2018.

15.

Subar

Kirkpatrick

Mittl

, et al. The automated self-administered 24-hour dietary recall (asa24): a resource for researchers, clinicians, and educators from the national cancer institute. J Acad Nutr Diet. 2012;112(8):1134-1137.

16.

Krebs-Smith

Pannucci

Subar

, et al. Update of the healthy eating index: HEI-2015 (vol 118, pg 1591, 2018). J Acad Nutr Diet. 2019;119(10):1759.

17.

Thompson

Midthune

Kahle

Dodd

. Development and evaluation of the national cancer institute's dietary screener questionnaire scoring algorithms. J Nutr. 2017;147(6):1226-1233.

18.

Data Processing & Scoring Procedures Using Current Methods. National Cancer Institute. Available at: https://epi.grants.cancer.gov/nhanes/dietscreen/scoring/current/

19.

Loftfield

Immerwahr

Eisenhower

. Construct Validity of a Single-Item, Self-Rated Question of Diet Quality. J Nutr Educ Behav. 2015;47(2):181-187.

20.

Chasens

Dinardo

Imes

Morris

Braxter

Yang

. Citizenship and health insurance status predict glycemic management: NHANES data 2007-2016. Prev Med. 2020;139:106180.

21.

Troiano

Berrigan

Dodd

Mâsse

Tilert

McDowell

. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc. 2008;40(1):181-188.

22.

Dunton

Liao

Almanza

, et al. Joint physical activity and sedentary behavior in parent-child Pairs. Med Sci Sports Exerc. 2012;44(8):1473-1480.

23.

Rich

Geraci

Griffiths

Sera

Dezateux

Cortina-Borja

. Quality control methods in accelerometer data processing: defining minimum wear time. PLoS One. 2013;8(6):e67206.

24.

van Schooten

Rispens

Elders

PJM

Lips

van Dieën

Pijnappels

. assessing physical activity in older adults: required days of trunk accelerometer measurements for reliable estimation. J Aging Phys Activ. 2015;23(1):9-17.

25.

Gao

Zhu

. Differential item functioning analysis of the 2003-04 NHANES physical activity questionnaire. Res Q Exerc Sport. 2011;82(3):381-390.

26.

Schuna

Johnson

Tudor-Locke

. Adult self-reported and objectively monitored physical activity and sedentary behavior: NHANES 2005-2006. Int J Behav Nutr Phys Activ. 2013;10:126.

27.

Jackson

Cunningham

. The stability of children's weight status over time, and the role of television, physical activity, and diet. Prev Med. 2017;100:229-234.

28.

Pedersen

Mikkelsen

Cronin-Fenton

, et al. Missing data and multiple imputation in clinical epidemiological research. Clin Epidemiol. 2017;9:157-166.

29.

Kuha

. AIC and BIC: Comparisons of assumptions and performance (vol 33, pg 188, 2004). Socio Methods Res. 2005;33(3):417.

30.

Williams

. Scalar Measures of Fit: Pseudo R2 and Information Measures (AIC & BIC). University of Notre Dame; 2020. Available at: https://www3.nd.edu/∼rwilliam/stats3/L05.pdf

31.

Matthews

Ainsworth

Thompson

Bassett

. Sources of variance in daily physical activity levels as measured by an accelerometer. Med Sci Sports Exerc. 2002;34(8):1376-1381.

32.

Grimby

Börjesson

Jonsdottir

Schnohr

Thelle

Saltin

. The “saltin-grimby physical activity level scale” and its application to health research. Scand J Med Sci Sports. 2015;25:119-125.

33.

Aires

Selmer

Thelle

. The validity of self-reported leisure time physical activity, and its relationship to serum cholesterol, blood pressure and body mass index. A population based study of 332,182 men and women aged 40-42 years. Eur J Epidemiol. 2003;18(6):479-485.

34.

Rödjer

Jonsdottir

Rosengren

, et al. Self-reported leisure time physical activity: a useful assessment tool in everyday health care. BMC Publ Health. 2012;12:693.

35.

Dietary Assessment Primer. Usual Dietary Intake. National Institutes of Health , National Cancer Institute. https://dietassessmentprimer.cancer.gov/. Accessed January 5, 2021.

36.

Gudzune

Peyton

Pollack

, et al. Overweight/obesity among social network members has an inverse relationship with Baltimore public housing residents' BMI. Preventive medicine reports. 2019;14:100809.

37.

Meza

BPL

Pollack

Levine

Latkin

Clark

Gudzune

. You are what you think your social network eats: public housing, social networks, and fast-food consumption. J Health Care Poor Underserved. 2020;31(4):1712-1726.

38.

Quintiliani

DeBiasse

Branco

Bhosrekar

Rorie

J-AL

Bowen

. Enhancing physical and social environments to reduce obesity among public housing residents: rationale, trial design, and baseline data for the healthy families study. Contemp Clin Trials. 2014;39(2):201-210.

39.

Gallagher

Visser

. The first two authors reply. Am J Epidemiol. 1997;145(1):82-83.

40.

Goni

García-Granero

Milagro

Cuervo

Martínez

. Phenotype and genotype predictors of BMI variability among European adults. Nutr Diabetes. 2018;8(1):27-28.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.08 MB