In a previous study, 79 American Indian women (12) and 20 American Indian men (C.A.T., unpublished data, 2002) were interviewed to identify cultural perceptions of health and diabetes. We used qualitative data from the interviews to create items, or statements, for our assessment tool, which increased its content validity because it included issues more germane to the respondents (26). Using dominant themes and text from the interviews, we created the Keeping the Balance Diabetes Assessment Tool, a four-part questionnaire measuring diabetes knowledge and perceptions of health, diabetes, and the social environment with a focus on diabetes prevention.

We created an initial list of items, or statements, to address four major categories: 1) perceptions of health, 2) perceptions of diabetes, 3) knowledge about the etiology of diabetes, and 4) the role of social interactions in health maintenance. When creating the items for the questionnaire, we attempted to use the original wording of the respondents from the interviews (12) to increase content validity. Each item was a statement about health, diabetes, or the social environment and was measured using a 6-point Likert-type scale (with 1 indicating strongly agree and 6 indicating strongly disagree).

Experts with experience in questionnaire development, American Indian research, or American Indian clinical practice were recruited by personal invitation and through a research and an American Indian health care e-mail Listserv. Eleven experts responded to our request and volunteered to provide online feedback and rewording suggestions so that the items would accurately measure a single concept. The panel of experts reviewed the items for cultural appropriateness, clarity, conciseness, and the ability to measure the intended concept. We used the panel's comments to create the final version of the instrument.

Men and women aged 18 to 65 years who were at least 25% American Indian according to a self-report were eligible to participate. Individuals were not excluded if they had diabetes or other chronic diseases. Key informants at two tribal health clinics identified two powwows in northeast Oklahoma where we could use the assessment tool. After receiving invitations to the powwows, a research team attended them in September 2003 and collected data.

At the first powwow, 81 volunteers completed the assessment tool; 116 volunteers completed the assessment tool at the second powwow. Two participants were excluded because they did not meet study criteria. Ten participants were excluded after providing incomplete responses or abnormally patterned responses (e.g., choosing all A's for every answer). Using SPSS (SPSS Inc, Chicago, Ill), data analysis was performed using 185 (94%) of the 197 completed questionnaires. The Oklahoma State University Institutional Review Board approved the study protocol.

Each volunteer provided signed informed consent before participating. The self-administered questionnaire was presented to eligible volunteers and followed by a brief demographic questionnaire. To increase participation, volunteers who completed both questionnaires received $10.

Items from two scales (31 health perceptions items and 21 diabetes perceptions items) were analyzed using principal axis factor analysis. Factor analysis is used to assess item correlations and identify common relationships among similar items, allowing the items to be categorized into various themes, or factors (27). The resulting factors are named based on the overall theme of their corresponding items. Data were excluded pairwise for items with missing or multiple responses. Data from the social interaction scale are not discussed in this article.

The principal factor analysis for each scale involved a standardized approach, and each scale (health and diabetes) was analyzed independently. The correlation matrix, Kaiser-Meyer-Olkin (KMO) measures of sampling adequacy, and Bartlett's tests of sphericity were evaluated for the factorability of the correlation matrix (i.e., to determine whether the items could indeed be classified into a few categories) (27). KMO values greater than 0.6 indicated that the correlation matrix had sufficient structure to result in a factorable solution. A significant Bartlett's test of sphericity indicated that the correlation matrix was significantly different from an identity matrix (a correlation matrix in which items correlate perfectly with themselves and not at all with other items).

To clarify the factor pattern, a rotation analysis of the factors was performed. The eigenvalues (measures of variance) greater than 1 in conjunction with the scree plot (a plot of the eigenvalues and the factors) were assessed to determine the number of factors to use in the rotation analysis (27). The analysis was performed with an oblique rotation (direct oblimin, Δ = 0), which allows the factors to correlate. The correlations between the linear factors were evaluated and if little correlation resulted, the analysis was repeated using an orthogonal (varimax) rotation. Items accounting for at least 16% of the variance on a factor (with loadings greater than an absolute value of 0.4) were considered to load, or be sufficiently correlated with, a particular factor (27). All items loading on a factor were then used as the basis for naming the factor. Cronbach α was computed for all items loading on each factor. Values higher than an absolute value of 0.7 were considered to have acceptable internal consistency (28).

Evaluation of the correlation matrix indicated relationships among the items. The KMO (0.71) and Bartlett's test of sphericity (P <.001) indicated a factorable correlation matrix. The initial factors extracted from the 31 items measuring the health perceptions indicated a potential for extracting three, four, or five overall factors using the scree plot (Figure) and the eigenvalues (Table 1). We found that five factors best categorized the perceptions of health, which accounted for 33.1% of the total variance. An oblique rotation resulted in little correlation among the factors, so a varimax rotation was performed and interpreted.

Items with factor loadings greater than or equal to an absolute value of 0.4 further clarified the factor's theme. The factor loadings for each of the five factors are presented in Table 2. The five items classified by the first factor (lifestyles) were related to health maintenance behaviors, whereas the second factor (barriers) comprised five obstacles to maintaining good health. Culturally related personal responsibility for health and wellness characterized the third factor (personal responsibility). The fourth factor was characterized by self-care behaviors and their association with health, and the fifth factor (cultural wellness) included items related to the association between health and mental and physical well-being.

Cronbach α internal consistency coefficients (Table 1) were modest for the five factors of health perceptions.

Analysis of the correlation matrix, KMO (0.79), and Bartlett's test of sphericity (P <.001) for the 21 items related to diabetes perceptions suggested that the correlation matrix was factorable. Examination of the initial eigenvalues (Table 1) and the scree plot (Figure) suggested that two factors would best categorize the 21 items. A varimax rotation was performed because the oblique rotation resulted in weak correlations between the two factors. Two factors explained 32% of the variance of the diabetes perceptions. Factor loadings for both factors are provided in Table 2. Nine items related to knowledge of and personal experience with diabetes loaded on the first factor (cognitive). The second factor (affective) comprised eight items related to concerns about and fear of diabetes and its complications. One item, the perception of diabetes as a death sentence, loaded on both factors: cognitive (0.45) and affective (0.44). Because this item related to personal experience but also caused an emotional response, it was allowed to load on both factors.

Cronbach α internal consistency coefficients (Table 1) were desirable for the two factors of diabetes perceptions.

Of the factors identified for the health perceptions items, we discovered that two factors (lifestyle behaviors and self-care) involved the impact of behavior on health, a theme found in previous research studies (15,29). In our previous qualitative interviews, we found that participants believed that overall good health was related to a healthy lifestyle, even though they also felt that developing diabetes was inevitable (12). Responses to the lifestyle questions accounted for the greatest amount of variance (9.5%) in the health perceptions items in the assessment tool, indicating that respondents believed healthy lifestyle behaviors were essential for maintaining good health. In addition, the self-care factor had items related to the importance of taking an active role in health maintenance. Hatton (11) reported similar themes among urban American Indians, who believed that health maintenance was directly related to performing certain self-care tasks. In our study, the barriers factor comprised perceived difficulty in changing lifestyle behaviors and the perceived financial and time investments required for health maintenance, which conflict with the beliefs that self-care behaviors affect health.

Another dominant theme from our previous interviews was that the respondents relied on physical symptoms as indicators of their health status (12). Participants believed that their behaviors did not need to change if they had no physical symptoms of illness. The personal responsibility factor and self-care factor in this study included similar items, such as those measuring the cues to change lifestyle behaviors and variables used to define personal health status. The perception that a person is healthy if the person does not have a disease (a culturally defined definition of wellness) likely affects health promotion and diabetes prevention efforts.

Racial and ethnic minorities are a medically underserved population in the United States (9,36). The differences in health status between American Indians and the general U.S. population are becoming more pronounced as rates of various chronic diseases, including diabetes, begin reaching epidemic levels in American Indians (37). In some southwestern American Indian tribes, as many as 50% of adults older than age 35 years are affected (22). The high diabetes rate and its long-term complications (38-41) likely shape the cognitive and affective factors in the assessment tool items measuring diabetes perceptions.

Strong feelings of hopelessness and fear (affective factors) related to the effects of diabetes were evident during our previous interviews (12) and in research with other minority groups (14-16,42). In fact, Arizona Pima Indians have developed a cultural defense mechanism now known as surrender, which stems from the perceived futility of diabetes prevention (14) and hinders the initiation of diabetes prevention behaviors.

The items that loaded on the cognitive factor likely reflect diabetes knowledge obtained from personal experience. In another study, personal experiences of individual Dakota Sioux Indians collectively shaped an entire tribe's cultural perceptions of health and diabetes (13). Experience has also shaped the attitudes of Mexican Americans about health and illness and has affected their social functioning and physical and mental health (42). Our previous interviews revealed that the personal experiences of American Indians with diabetes shaped their perceptions of diabetes prevention, treatment, and etiology (12).

Even though they recognized the relationship between healthy lifestyles and good health, participants in our study believed that their lifestyle behaviors did not have to change until their diabetes resulted in perceivable signs and symptoms. Likewise, in another study, researchers found that African American women did not consider diabetes to be a serious disease and thought medication alone was the cure (43). These findings were similar to findings from our study; participants believed that diabetes treatment involved only medication if they had no diabetes symptoms.

Items measuring the susceptibility of American Indians to developing diabetes and the fear of diabetes and its long-term complications characterized the affective factor. In a similar study involving Mexican Americans, researchers found that women judged the severity of diabetes by the extent of resulting physiologic damage (42). Diabetes was a frightening disease because of the extensive damage it could cause. Likewise, American Indian women thought of diabetes in terms of its complications, such as kidney failure and blindness (12). African American women thought that a lack of physical symptoms indicated they no longer had diabetes (43), but concerns about diabetes were strongly linked to emotional factors, including fears about diabetes, denial, and concerns about insulin therapy and required lifestyle changes (44).

Our study has several limitations. The sample was derived through nonprobability methods and may have decreased the generalizability of the findings, but these sampling methods are often needed to identify individuals from an at-risk population (11). The smaller sample size could account for the lower internal consistencies for some of the health factors and the loading of the "death sentence" item on the cognitive and affective diabetes factors. In addition, in future versions of the assessment tool, the wording of the items may need to be changed.

As mentioned, the internal consistency coefficients for three of the health perceptions factors (Cronbach α = 0.50–0.55) were lower than desired, possibly because of the partialing of variance — and the factoring of residualized variance — inherent in the factor analysis method or the smaller sample size. Furthermore, the sample size may not have been large enough to account for the variance associated with the health and diabetes perceptions. Methodologically, complications arose as we tried to determine the number of factors to rotate for health perceptions. Despite the lower Cronbach α values, the five factors provided a sound theoretical explanation of health perceptions. Additional testing of the assessment tool in larger samples is needed to determine the relevance of the items and the stability of the factor structure.

Previous studies have identified several health-related beliefs that seem to be common among various tribes: 1) the perception that health is the responsibility of the individual, 2) the perception that health and disease are natural parts of life, and 3) the understanding that spirituality plays a role in health (29). However, identifying differences among tribes and the extent to which individuals subscribe to their tribe's cultural beliefs are important when developing health promotion activities (9,29,30,45). This assessment tool could help clinical and public health professionals obtain data about health and diabetes perceptions among individuals or groups. The resulting data from this study could provide important information for health professionals who are attempting to create culturally appropriate health care counseling and health education programs, which should be customized for specific groups and designed according to the variables that influence the group's behavior (7,46-48).

As mentioned, we found similarities between our research and the findings from other studies, all of which demonstrate relationships among the factors associated with health and diabetes perceptions. Future research must expand the exploration of health and diabetes perceptions so that health professionals can design successful health promotion and diabetes prevention programs.