In this cross-sectional study, a subsample of participants at high risk for diabetes in the North Carolina Health Project (NCHP) cohort was mailed a survey about physical activity, supports, and barriers. NCHP participants were recruited from the North Carolina Family Medicine Research Network (NCFMRN) (15). The purpose of the NCHP was to develop a representative cohort of adult primary care patients for use in multiple projects. Details of the development of the cohort are described elsewhere (15). This study was reviewed and approved by the University of North Carolina Biomedical Institutional Review Board.

The NCHP cohort was established in spring and summer 2001. A total of 4876 individuals aged 18 years and older from the NCFMRN family practice sites completed a baseline  enrollment questionnaire that included questions about health-related quality of life (16,17), comorbid conditions, and risk factors for diabetes (18). In each practice office, all adults seen during a 1-month enrollment period were offered the opportunity to participate; the enrollment rate was 62%. The mean age of the NCHP cohort at baseline was 47.7 years; 70.8% were female, and 19.7% were African American (15).

Participants in the NCHP cohort were determined to be at risk for diabetes according to their original responses to the American Diabetes Association (ADA) diabetes risk test questions (18). These questions included age, body mass index (BMI), family history of diabetes, history of gestational diabetes, and exercise ("I get little or no exercise in a typical day"). A total score of 10 points or higher placed a person at high risk. Sensitivities of the diabetes risk test have been described as 69% to 78%, and specificities for prediabetes or diabetes are 51% to 54% (19). In the original NCHP cohort, 45% of the adults reported performing little or no exercise in a typical day; 15% reported having diabetes; and an additional 36% scored 10 or higher (high risk) on the ADA test. In comparison, 6.5% of people in North Carolina reported having diabetes in the Behavioral Risk Factor Surveillance System (BRFSS) (20). 

Of the 4876 people who completed the NCHP baseline questionnaire, 4139 (85%) consented to receive additional surveys. The two groups, those who consented to receive further surveys and those who did not, differed in age (mean age 47.7 years for those who consented compared with 43.7 years for those who did not, P < .001), BMI (mean BMI 29.5 kg/m² for those who consented compared with 28.1 kg/m² for those who did not, P < .001), and having less than a high school education (20.5% for those who consented compared with 13.5% for those who did not, P < .001), but they did not differ by sex or race. Of the 4876 baseline enrollees, there were 1077 people at risk for diabetes who 1) did not currently have diabetes, 2) were white or African American, 3) consented to receive surveys, and 4) had a valid mailing address. This group included 204 white men, 627 white women, 193 African American women, and 53 African American men. In April 2003, we mailed a self-report questionnaire assessing physical activity supports and barriers to a random stratified sample of 522 adults of the 1077 adults identified by the 2001 enrollment questionnaire as being at high risk for diabetes. These 1077 adults were stratified by race and sex to assure adequate representation in our sample. The initial mailing was accompanied by a self-addressed, stamped return envelope, and a postcard reminder was mailed 2 weeks later. Nonrespondents were mailed a second questionnaire 1 month later. Attempts were made to telephone individuals who did not respond, and they were asked to complete the survey over the telephone. 

In addition to baseline demographic data obtained from the original NCHP cohort survey, the survey designed for this study included physical activity items adapted from the BRFSS (converted from telephone to written survey) (21). The physical activity outcome for this study, specified as an objective in Healthy People 2010 (22), is 150 minutes or more of moderate to vigorous activity per week (calculated by the responses to the BRFSS activity questions). A checklist of physical activities respondents enjoyed (walking, running, biking, stretching, water aerobics, gardening, tennis, swimming, lifting weights, yoga, hiking, other) and whether or not their health provider talked with them about physical activity in the past year were also included in the survey. Questions also addressed stage of change for physical activity (23), whether they had to cut back on activity because of health in the past 6 months, and self-efficacy using the Self-Efficacy for Exercise (SEE) scale (24). We also included items on influences on physical activity in an assessment instrument, the Influences on Physical Activity Instrument (IPAI).

The IPAI was included as part of the mailed survey to assess physical activity barriers and supports based on factors reported in the literature to be associated with activity (7-12,25,26) as well as questions included in the "Starting the Conversation" tool for physical activity (used by many doctors offices in North Carolina) (27). The IPAI consisted of three separate domains, each representing a different type of influence — individual, support, and environment. (The IPAI and information about its scoring is available in the Appendix.) The IPAI is written for a fourth- to sixth-grade reading level. Comprehension was assessed qualitatively through written and oral feedback by 15 white and African American adults (six were older than 65 years, and nine were younger than 65 years) at the University of North Carolina at Chapel Hill Family Practice before survey administration.

All statistical analyses were performed using Stata (StataCorp, College Station, Tex). Univariate and bivariate analyses of the continuous and categorical variables were performed for missing data, sparse numbers, extreme values, and linearity. Bivariate analyses evaluated the unadjusted relationship between the covariates and the physical activity outcome of 150 minutes or more of moderate to vigorous activity per week.

Exploratory principal components factor analysis (PCFA) were conducted on each of the three domains of the IPAI (individual, support, and environmental influences) to clarify the factor structure of each domain. Kaiser's eigenvalue greater-than-1 rule and Cattell's scree test were used to determine the number of factors to extract (28,29). An eigenvalue equal to 1.0 meant that a component accounted for the average variance of one item (28). A scree test included a plot of the eigenvalues on the y axis and consecutive factors on the x axis, and the last factor retained was before the downward curve where the slope levels off (29). If more than one factor was present, then the factors were rotated using an oblique rotation to determine if each item would load on a single factor. The Cronbach α reliability coefficient was calculated after the PCFA for each factor present.

Predictive logistic regression models were used for the dichotomous outcome, reaching 150 minutes or more of moderate to vigorous activity per week. Independent variables, including the IPAI and SEE scale, were assessed as continuous variables in the model. Education was assessed in three categories: some high school or less, high school graduate, and some college or more, with some high school or less as the reference category. All other variables were dichotomized; age was dichotomized into younger than 55 years and 55 years and older after testing for the linearity assumption. Obesity was defined as having a BMI of 30 kg/m² or higher. Interaction terms, sex with age and race with all other covariates, were evaluated during exploratory analyses to help identify which risk factors may be effect modifiers. A backwards stepwise variable reduction was used to identify important variables. Interaction terms, race with age and race with working status, were significant in the exploratory analyses but were dropped from the final model by a P value of .05 for the −2 Log likelihood ratio test. The covariates including age, sex, race, and work status were also dropped from the logistic model using the −2 Log likelihood ratio test. For the final model, a population average effects model was used to adjust for any intraclass correlation due to the cluster sampling design. Clusters were the practice sites in this analysis.

Exploratory PCFA of the individual influences domain of the IPAI identified three factors that explained 60% of the total variance among items (Figure 1). After an oblique rotation, the factor loadings made the interpretation of the three factors clearer. Four items (Appendix) loaded on factor 1 (items a, c, d, and g) and appeared to represent the view that physical activity is a low priority; two items loaded on factor 2 (items e and f) and appeared to represent weight control; and three items (items h, j, and k) loaded on factor 3 and appeared to represent injury concerns. Items b and i were dropped because they had loadings greater than 0.3 on more than one factor. For each item retained, the primary loading was greater than 0.58, and the secondary and tertiary loadings were never greater than 0.25. Thus, three scales comprise the individual influences domain of the IPAI.

An average score was calculated for each of these three scales based on the factor structure. The items loading on factor 1 (low priority) had a Cronbach α of 0.68 and a mean score of 2.31; factor 2 (weight control) items had a Cronbach α of 0.67 and a mean score of 2.28; and factor 3 (injury concerns) items had a Cronbach α of 0.63 and a mean score of 2.22.

Exploratory PCFA of the support influences domain identified one factor that explained 60% of the total variance among the items (Figure 2). Item p was dropped because it loaded weakly on one factor, and when two factors were forced, it loaded ambiguously. The factor loading for items l to o were all greater than 0.72. An average score was calculated for the support influences scale. The scale comprising the support items l to o had a Cronbach α of 0.77 and a mean of 2.28. 

The exploratory PCFA on the environmental influences identified two factors that explained 65% of the total variance among the items (Figure 3). The factor loadings became clearer after an oblique rotation. Three items (q, r, and s) appeared to represent having a place for activity (factor 1). Two items (t and u) appeared to represent having time for activity (factor 2). For each item on the two environmental influences scales, the primary loading was greater than 0.27, and the secondary loadings were never greater than 0.17. Factor 1 (having a place for physical activity) had a Cronbach α of 0.60 and a mean score of 2.13, and factor 2 (having time for physical activity) had a Cronbach α of 0.53 and a mean score of 2.23. 

Interfactor correlations and a higher order factor analysis suggested that a global summary score based on the six scales was not the best way to represent IPAI. Therefore, the average score for the items loading on each factor was calculated and used in the modeling rather than a single summary score based on combining the six scales of the IPAI. 

There were no significant differences between the characteristics of those who were physically active and those who were not, except for level of education (Table 2). A higher proportion of respondents with a college education or high school diploma (65.9%) reported meeting Healthy People 2010 activity levels than of those with less than a high school diploma (30.9%), P < .001. SEE was higher among active individuals than inactive individuals (5.6 for active compared with 4.9 for inactive, P = .007), as were scores on several questions about influences on activity. Fewer physically active individuals reported that activity was a low priority (P < .001), disagreed that physical activity influences their weight (P = .02), and worried about injury (P < .001). Additionally, the active group less often had difficulty finding a place for activity (P < .001) and making time for activity (P < .001). Interestingly, health care provider counseling about physical activity was similar between groups. Social support was also similar.

In the multivariate model (Table 3), higher education remained a significant predicator of physical activity (OR, 1.72; 95% CI, 1.08–2.75). Participants who reported that physical activity was a low priority were less likely to be physically active than participants who reported that physical activity was a high priority (OR, 0.45; 95% CI, 0.23–0.89). Participants who reported concerns about injury were also less likely to be physically active (OR, 0.42; 95% CI, 0.25–0.69). Additionally, those who had difficulty finding time for activity were less likely to be physically active (OR, 0.38; 95% CI, 0.17–0.87). Beliefs about weight being affected by activity, having support for activity, having a place for activity, and self-efficacy were not significant independent predictors of physical activity in this sample.