Original Research

Validating the Adult Primary Care Assessment Tool

Author and Disclosure Information



Table 1 compares the HMO sample with the CHC sample on sociodemographic and health care utilization measures. The HMO sample included predominantly white (81.6%) and higher income subjects (86.8% with annual household income of $25,000 or more). In contrast, the CHC sample included predominantly non-white (83.2%) and lower income subjects (85.9% with an annual household income less than $25,000). Compared with the CHC respondents, HMO subjects had been seeing their regular source of care for a longer time, were more likely to choose their own doctors and visit a specialist, and less likely to have trouble paying for their health care.

Factor Analysis and Construct Validity

In the initial exploratory factor analysis, all 92 applicable questionnaire items measuring the subdomains and domains of primary carefirst contact, longitudinality, comprehensiveness, coordination, family centeredness, community orientation, and cultural competencewere included. Based on the results of the initial factor analysis, 4 criteria were applied to reach the final solution (Table 2; initial factor analyses not shown but available upon request).

Seven common factors were extracted, corresponding to the hypothesized primary care scales: first contactaccessibility, first contactutilization, longitudinalityinterpersonal relationships, comprehensivenessservices available, comprehensivenessservices received, coordination, and community orientation (Table 2). Those extracted factors explained 88.1% of the common variance. Eigenvalues ranged from 16.17 to 1.16. All principal primary care domains were extracted as hypothesized. Only 1 of the 3 derivative features, community orientation, was separately identifiable.

Derivation and Reliability of the Primary Care Scales

Table 3 presents the results of the reliability analyses for both the original items and the final items (based on factor analysis). Item descriptive results (means and standard deviations) are also presented. Scale reliability measures include item-total correlation and alpha coefficient reliability. The distribution of the items varied significantly from a mean of 1.85 (ask about gun safety) to 3.73 (Provider answers questions in ways you understand) on the 4-point Likert-type scale. The distribution tends to skew toward more favorable answers (above 2.5). Apart from the gun safety item, only 2 items fell below a mean of 2 (1.94 for Provider knows neighborhood problems, 1.90 for Provider makes home visits). The first contactutilization and longitudiinalityinterpersonal relationships scales achieved the highest mean scores, whereas scales with lower means were community orientation, first contact-accessibility, and comprehensiveness-services received.

Eighteen of the 92 initial items were deleted on the basis of the criteria imposed for factor analyses. No items were deleted for first contact-utilization, coordination of services, comprehensiveness-services received, and community orientation scales. All items were deleted for family centeredness as were two thirds of the items for first contact-accessibility. Two items (out of 22) were deleted for longitudinality-interpersonal relationships and 3 (out of 24) for comprehensivenessservices available. Items from cultural competence were combined into first contact-accessibility. The revised scales demonstrate internal consistency reliability that was higher than or equal to the original scales, despite the reduction in number of items. Item-total correlations were also high and ranged from 0.34 (If sick, seen same day if office is open) to 0.91 (How to prevent hot water burns and How to prevent falls).

Testing the Likert Scaling Assumptions

Table 4 presents a summary of the results of the tests of Likert scaling assumptions using the revised items. All item-scale correlations well exceeded the accepted minimum (0.30) with the majority greater than 0.50 (Assumption 1). All 7 multi-item scales achieved 100% scaling success, indicating that all items in these scales correlated substantially higher with items in their hypothesized scale than with items in other scales (Assumption 2). Item means within each revised scale generally differed by less than six tenths of a point (except for first contact-accessibility) and item standard deviations within each scale by less than four tenths of a point (Assumption 3). Formal evidence of equal item variance was supported by the equivalence of the intraclass correlation and Scotts homogeneity ratio for each scale. Equal-item scale correlation (Assumption 4) was also observed through the range of item-scale correlations. As shown in column 1 (range of item-scale correlations), the range is relatively narrow (from .17 for coordination of services to .38 for comprehensiveness-services received). Finally, score reliability (Assumption 5) showed that except for first contact-utilization (only 3 items), all alpha levels exceeded .70 and were sufficiently high. Five of the 7 scales had alpha levels above .85.

Descriptive Feature of PCAT-AE

Table 5 displays estimates of central tendency and dispersion of scale score distributions for the 7 primary care scales in this South Carolina sample. Except for community orientation, all primary care scales were negatively skewed, indicating distributions with more positive ratings of primary care. The community orientation scale was positively skewed, indicating distributions with more negative ratings on the community orientation aspect of primary care. The full range of possible scores was observed for all scales except ongoing care.

Next Article: