Study design, period, and setting

A multicenter, cross-sectional study was conducted in the Amhara region's comprehensive specialized referral hospitals (CSRHs) from March 24, 2025, to May 5, 2025. The Amhara region is the second-largest region in Ethiopia, located in the northwestern part of the country. According to the 2007 census, the estimated total population of the Amhara region was 17,221,976 (male = 8,641,580 & female = 8,580,396) [41].

The region is served by eight comprehensive specialized referral hospitals, two general hospitals, 73 primary hospitals, 847 health centers, and 3,342 health posts. The eight public comprehensive specialized referral hospitals are the University of Gondar Comprehensive Specialized Referral Hospital (UGCSRH), Bahir Dar Felege Hiwot Comprehensive Specialized Referral Hospital (BFCSRH), Bahir Dar Tibebe Gion Comprehensive Specialized Referral Hospital (BTCSRH), Debere Markos Comprehensive Specialized Referral Hospital (DMCSRH), Debere Tabor Comprehensive Specialized Referral Hospital (DTCSRH), Debere Bihrhan Comprehensive Specialized Referral Hospital (DBCSRH), Dessie Comprehensive Specialized Referral Hospital (DCSRH), and Woldia Comprehensive Specialized Referral Hospital (WCSRH). Based on the information from the regional health bureau and the hospital health management information system (HMIS) report, the annual average number of outpatient follow-ups for T2D was 1,297 [42].

Source and study population

All adults with T2D who had outpatient follow-up care at CSRHs in the Amhara region, Ethiopia, constituted the source population, whereas those who attended during the data collection period in the randomly selected CSRHs were the study population.

Inclusion and exclusion criteria

All individuals aged 18 years or older diagnosed with T2D who had at least 6 months of outpatient follow-up care and who were attending during the data collection period were included in the study.

Sample size determination

The sample size for confirmatory factor analysis (CFA) or structural equation modeling (SEM) was determined using the recommended parameter-to-sample ratio, which ranges from 1:5–1:20 [43]. Based on this guideline, we used the 1:20 parameter-to-sample ratio, resulting in a total sample size of 520.

Sampling procedure

Out of eight comprehensive specialized referral hospitals, three hospitals were randomly selected. These were Debre Markos Comprehensive Specialized Referral Hospital (DMSCRH), Bahir Dar Felege Hiwot Comprehensive Specialized Referral Hospital (BFCSRH), and Woldia Comprehensive Specialized Referral Hospital (WCSRH). The sample size for each CSRH was determined using the proportional allocation formula where ni represents the sample size for a single hospital, n is the total sample size, Ni is the total population of a single hospital, and N is the total population across all eight hospitals. Participants were selected using a systematic random sampling technique. The sampling interval was calculated using the formula (K = N/n), where N represents the total population and n is the total sample size. Therefore, participants were selected at every K = 2 interval based on their order of arrival at the hospital during their follow-up period.

Data collection tool and procedure

The Patient Assessment Chronic Illness Care (PACIC-5As) measure is the most widely used tool designed to assess the implementation of chronic illness care from the patient’s perspective [31,44]. The tool aligns with the concept of the Chronic Care Model (CCM) [30]. The PACIC tool was originally developed by Glasgow et al. (2005) [45] and subsequently expanded into the PACIC-5As by incorporating items that included concepts from the behavioral change model (behavioral counseling intervention). The tool demonstrated an internal consistency of 0.97 [32].

The 5As model is an evidence-based approach to behavioral change employed to improve patients’ self-management [46]. The PACIC-5As comprises 26 items, integrating the original PACIC instrument with the “5As” behavioral change model, which is recommended by the United States Preventive Services Task Force (USPSTF) [46]. The 5As model includes five subscales: Assess, Advise, Agree, Assist, and Arrange. “Assess” includes items 1, 11, 15, 20, and 21, which evaluates behavioral health risks, symptoms, or current behaviors. “Advise” contains items 4, 6, 9, 19, and 24, which provide clear, specific, and personalized behavioral counseling. “Agree” consists of items 2, 3, 7, 8, and 25, which focus on collaboratively selecting appropriate treatment goals and reaching a shared agreement. “Assist” includes items 10, 12, 13, 14, and 26, which helps or supports patients in making lifestyle changes. “Arrange” comprises items 16, 17, 18, 22, and 23, which are designed to assess ongoing assistance or follow-up care [46].

The items are scored on a 5-point Likert scale, where 1 represents “almost never” and 5 represents “almost always.” The overall PACIC-5As score (summary score) was obtained by averaging all items, excluding item 5 (1−4, 6−26). Each subscale score of the PACIC-5As was calculated by averaging the items corresponding to each subscale [45]. Higher PACIC-5As scores indicate better perceived quality of diabetes care [33,45,47].

Data were collected using Kobo Toolbox software version 4.4. Following the translation and validation process, the questionnaire was created and prepared in both English and Amharic languages within the software. The software is suitable for collecting data both online and offline. Subsequently, data were gathered through face-to-face interviews.

Prior to conducting the translation and validation process, permission was obtained from the tool developer, Glasgow et al. (2005) [45]. In this study, the translation, cultural adaptation, and validation process adhered to the guideline for the translation, adaptation, and validation of instruments for use in cross-cultural health care research [48]. This guideline consists of a series of stages, ranging from translation to psychometric evaluation. We followed this guideline and conducted the translation and adaptation process as follows:

Stage I: Forward translation: The forward translation was carried out by two bilingual translators whose native language is Amharic, the target language. Each translator independently translated the English PACIC-5As tool into Amharic. The first translator (FT-1) is an assistant professor of medical nursing at the University of Gondar, familiar with healthcare terminology and the instrument's content. The second translator (FT-2) is an assistant professor of teaching English as a foreign language (TEFL) at the University of Gondar. He possesses expertise in language and is familiar with the culture being studied, but he is not well-versed in healthcare terminology and the instrument's content. Both translators provided written reports highlighting challenging phrases and explaining their translation choices. This process included translating all components of the instrument, including item content, response options, and instructions.

Stage II: Synthesis of the translation: At this stage, the two translators (FT-1 and FT-2) compared and synthesized their translations through a committee approach. Any discrepancies or ambiguities in wording, sentences, and meanings between their translations were discussed. After discussion, both translators confirmed content equivalence and reached a consensus. The principal investigator also actively participated in the discussions, clarifying issues and supporting both translators in reaching an agreement. Finally, the preliminary translated Amharic version of the PACIC-5As-ET instrument (FT-12) was produced.

Stage III: Back translation: After completing the forward translation, two bilingual back translators, Back Translator 1 (BT-1) and Back Translator 2 (BT-2), independently translated the Amharic version of the PACIC-5As-ET tool back into English. BT-1 was an assistant professor in the Department of Medical Nursing at the University of Gondar and an expert in healthcare terminology and the conceptual content area of the instrument. BT-2 was an assistant professor of Teaching English as a Foreign Language (TEFL) at the University of Gondar. Whereas he understood the cultural and grammatical nuances of English, however, his familiarity with medical terminology and the instrument being studied was limited.

During the translation process, the back translators (BT-1 and BT-2) were unaware of the original instrument and did not have access to it (blind back translation). After completing the back translation, both translators convened to discuss and reconcile each question with the original English version. Any issues, including differences in translation, lack of clarity, or ambiguities in words and sentences, were thoroughly identified and resolved through mutual agreement. Additionally, the principal investigator actively participated in the discussions and solved the problem together.

Stage IV: Expert committee: At this stage, an expert panel discussion was conducted to ensure the cross-cultural equivalence of the newly translated PACIC-5As-ET instrument and to evaluate its content validity. A total of ten experts participated in the panel discussion, including one PhD holder in public health (a member of the research team), one PhD in nursing, and experts with MSc degrees in nutrition, midwifery, surgical nursing, and pediatric nursing. Both forward and back translators also participated in the panel discussion. All the experts had over ten years of experience in their respective fields.

The expert committee evaluated and reviewed the instructions, items, and response formats for any similarities or differences in meanings, wording, or grammatical structures between the forward translations (FT-1, FT-2, and FT-12) and the back translations (BT-1 and BT-2). All translations were then compared with the original questionnaire. The committee discussed any grammatical errors, ambiguities, discrepancies in meaning, or challenges in language construction and then reached a consensus. At this stage, initial conceptual, content, and semantic equivalence were established. Accordingly, the experts independently evaluated and rated the content of the pre-final version of the PACIC-5As-ET in terms of its relevance for measuring the perceived quality of care among patients with T2D. Finally, the committee approved the pre-final Amharic version of the PACIC-5As-ET tool for use in Ethiopia.

Stage V: Pilot testing: In this stage, the pre-final version of the newly translated and adapted PACIC-5As-ET measure was evaluated by participants. Approximately 20 patients with T2D were included in the pretest at the University of Gondar Comprehensive Specialized Referral Hospital.

Stage VI: After completing the content and face validity assessments, the psychometric properties of the newly translated and adapted PACIC-5As-ET tool were evaluated among 520 patients with T2D. This analysis included the assessment of the tool’s reliability, validity, model fit, and other psychometric properties.

Data management and analysis

Following data collection, the data were reviewed, cleaned, and exported to Statistical Package for the Social Sciences (SPSS) version 25 for analysis. Additionally, Analysis of Moment Structures (AMOS) version 26 software was used to conduct confirmatory factor analysis (CFA). The results are presented using descriptive statistics, including mean (±SD), frequencies with percentages, figures, and tables.

Content validity

After the panel discussion, all expert committee members independently evaluated and rated the pre-final version of the PACIC-5As-ET tool. The committee assessed the relevance of the tool using a four-point Likert scale: 1 (not relevant), 2 (somewhat relevant), 3 (quite relevant), and 4 (highly relevant). Items rated as 1 (not relevant) or 2 (somewhat relevant) were considered for revision. The scores were then dichotomized into relevant and not relevant categories. Ratings of 3 (quite relevant) and 4 (highly relevant) were categorized as relevant and scored as one, while ratings of 1 (not relevant) and 2 (somewhat relevant) were categorized as not relevant and scored as zero. The Content Validity Index (CVI) was then calculated at both the item level (I-CVI) and scale level (S-CVI) [49]. An item-level content validity index (I-CVI) of ≥ 0.78 was considered the minimum acceptable threshold, while a scale-level content validity index average (S-CVI/Ave) of ≥ 0.90 indicated acceptable overall content validity [48].

Furthermore, Gwet’s kappa coefficient was computed to assess the level of inter-rater agreement while accounting for chance agreement. Gwet’s kappa provides a more stable and reliable estimate of agreement; Cohen’s and Fleiss's kappa, in contrast, can yield negative values and tend to underestimate agreement in the case of imbalanced responses [50]. A kappa value < 0 indicates poor agreement; 0 to 0.20 indicates slight agreement; 0.21 to 0.40 indicates fair agreement; 0.41 to 0.60 indicates moderate agreement; 0.61 to 0.80 indicates substantial agreement; and 0.81 to 1 indicates almost perfect agreement [51,52].

Face validity

The pre-final Amharic version of the PACIC-5As-ET tool was evaluated by twenty study participants at the University of Gondar Comprehensive Specialized Hospital. The participants were asked to review the questionnaire’s instructions and response categories, as well as the clarity and ease of understanding of each item. Any instruction, response format, or item identified as unclear by 20% or more of the participants was revised; those not meeting this threshold remained unchanged [48]. Both content and face validity confirmed the appropriateness and relevance of the tool among patients with T2D.

Ceiling and floor effect

The ceiling effect occurs when a measurement instrument is not sensitive enough to detect higher levels of a variable, resulting in scores due to the clustering at the upper end. Conversely, the floor effect occurs when scores aggregate at the lower end of the scale, making it difficult to detect lower levels of a variable [53]. Both floor and ceiling effects indicate a limited or incomplete distribution of responses [54]. A floor or ceiling effect was identified if ≥ 15% of participants scored at the minimum (total score = 0) or maximum (total score = 100) levels, respectively [55–57]. In the present study, we identified ceiling and floor effects at the overall scale, subscale, and item levels.

Reliability test

The reliability (internal consistency) of the instrument was assessed using Cronbach’s alpha coefficient and composite reliability (CR). Alpha measures the degree to which items produce consistency scores and should be interpreted as the proportion of measurement variance attributable to true score variance among individuals [58]. Internal consistency should range from 0 to 1, with values closer to 1 indicating higher internal consistency. A Cronbach’s alpha coefficient of ≥ 0.90 was interpreted as excellent, 0.80–0.90 as good, 0.70–0.80 as acceptable, 0.60–0.70 as questionable or needing improvement, 0.50–0.60 as poor, and < 0.50 as unacceptable [59,60].

Composite reliability was computed after factor analysis using standardized factor loadings of the constructs with the formula CR = (∑λ)² / [(∑λ)² + ∑(1-λ²)], where (∑λ)² represents the squared sum of factor loadings, and ∑(1-λ²) denotes the sum of error variance of the squared loadings [61,62]. According to Fornell and Larcker (1981), a CR value greater than 0.70 is considered acceptable, while values as low as 0.60 may also be deemed adequate [63].

Inter-item and item-total correlation

The strength of relationships between items was evaluated using inter-item and item-total correlations. A two-tailed Pearson correlation was employed to examine the strength of association between items. An inter-item correlation of ≥ 0.30 was considered acceptable, and values below 0.30 indicated a weak correlation [64]. An item-total correlation (corrected item-total correlation) value of ≥ 0.20 was deemed acceptable, while a value below 0.20 was considered less acceptable [64,65].

Test-retest reliability

Test-retest reliability reflects the consistency of scores obtained from the same respondents at two different times [66]. The test-retest reliability indicates the stability of the results over time [67]. A total of 40 participants with T2D participated in the test-retest assessment [68], which was conducted two weeks after the initial test [69,70]. Test-retest reliability was evaluated using the Intraclass Correlation Coefficient (ICC) based on a two-way mixed-effects model. ICC values less than 0.5 indicate poor reliability; 0.5–0.75 indicate moderate reliability; 0.75–0.90 indicate good reliability; and > 0.90 is interpreted as excellent test-retest reliability [71].

Confirmatory factor analysis (CFA)

In this study, second-order (hierarchical) confirmatory factor analyses were conducted. Second-order factors should account for a significant portion of the variance in both the lower-order factors and the observed (manifest) variables, demonstrating their relevance and explanatory power across all levels of the model [62]. Prior to conducting the CFA, we checked the normality assumption using skewness and kurtosis. For normally distributed data, skewness and kurtosis values should generally fall within ±2, although kurtosis values up to ±7 may be acceptable [72,73].

In this confirmatory factor analysis, we evaluated the factor structures of PACIC-5As and its subscales, as originally proposed by Glasgow et al. (2005) [32], to determine whether the observed items adequately represented the underlying constructs and supported the proposed theoretical model [74].

Model fit and factor loadings were also assessed as part of this CFA. The model fit of the proposed PACIC-5As instrument and its constructs was evaluated using the Generalized Least Squares (GLS) method. In factor analysis, GLS can be used as the primary alternative fitting function to maximum likelihood, particularly when the observed variables are correlated and underlying assumptions of multivariate normality are violated [75,76]. GLS addresses the issue of estimating correlated errors, which is important for accurate model fitting and interpretation [77]. Similarly, GLS allows for more flexible assumptions about error variances and is a more powerful tool than other estimators for obtaining more efficient and unbiased estimates in the presence of heteroskedasticity [76–79].

To assess the model fit, various fit indices were evaluated. The absolute fit indices included the chi-square test (χ²), the chi-square test to degree of freedom ratio (χ²/df), the Root Mean Square Error of Approximation (RMSEA), the Root Mean Residual (RMR), the Standardized Root Mean Square Residual (SRMR), the Goodness of Fit Index (GFI), the Adjusted Goodness of Fit Index (AGFI), and the Parsimony Goodness of Fit Index (PGFI). The relative (incremental) fit indices, including the Comparative Fit Index (CFI), the Parsimony Comparative Fit Index (PCFI), the Incremental Fit Index (IFI), the Normed Fit Index (NFI), the Parsimony Normed Fit Index (PNFI), the Relative Fit Index (RFI), and the Tucker-Lewis Index (TLI), were used to assess model fitness.

Model fitness was evaluated using the recommended thresholds; however, the thresholds for fit indices can vary based on model complexity and research context [80]. An acceptable model fit indicated by the non-significant chi-square test (p-value > 0.05) or a ratio of chi-square to degree of freedom (χ²/df) ≤ 3, which is considered an acceptable fit [81]. RMSEA values less than 0.05 are considered an acceptable fit, and values up to 0.08 or 0.1 are also considered acceptable, while other studies suggest that a value close to zero indicates an acceptable fit [80]. For RMR and SRMR, a value ≤ 0.08 is deemed an acceptable fit, although other studies suggest that values closer to zero are considered acceptable [80]. For the other fit indices, such as GFI, AGFI, IFI, CFI, NFI, and PRATIO, values ≥ 0.90 are generally considered an acceptable fit; also, other studies suggest that values closer to 1 are considered acceptable [80,81]. In contrast, the smallest values are preferred for the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) [80,82]. There are no universal criteria or rules for the number of fit indices required to confirm good model fit. However, a model is generally considered to have an acceptable fit when at least two fit indices meet the recommended threshold values [83]. It is also possible for a model to adequately fit the data despite certain fit indices indicating a poor fit [81]. Likewise, standardized factor loading coefficients were evaluated at both the item and construct levels, with values greater than or equal to 0.40 considered acceptable [61,84].

Moreover, bootstrapping and modification indices were applied to enhance the accuracy and robustness of statistical estimates as well as to improve model fit. Bootstrapping is particularly useful when dealing with small sample sizes or non-normally distributed data [85]. Conversely, modification indices help reduce error terms and refine model fitness by adding covariances within a construct’s indicators or relationship paths between constructs, thereby adjusting the chi-square test value [78,85].

Construct validity

After assessing the model fit, construct validity was evaluated. Construct validity refers to the degree to which a measurement tool accurately measures the theoretical construct it intends to measure [31]. In this study, both convergent and discriminant validity were assessed. Convergent validity was evaluated by calculating the average variance extracted (AVE) for the first-order model (between the subscales and observed variables), also known as the measurement model, and the second-order model (between the overall scale and subscales), also known as the structural model.

The AVE was computed using the formula: AVE = ∑λ² / [∑λ² + ∑(1-λ²)], where ∑λ² is the sum of the standardized squared loadings of the items, and ∑(1-λ²) is the sum of the error variance of the squared loading of the items [61,62]. According to Fornell and Larcker (1981) and other reporting guidelines, the AVE of the construct should be greater than or equal to 0.5 for observed variables [62,63]. However, if the composite reliability (CR) of the construct is above 0.6, an AVE as low as 0.4 is considered acceptable. Discriminant validity between each scale was evaluated using the AVEs values and the correlations between the constructs. Discriminant validity is considered high when the square root of the AVEs exceeds the inter-construct correlations [86,87].

Data quality assurance

The tool was properly translated, culturally adapted, and validated in accordance with standardized guidelines [48]. Furthermore, we documented and reported all steps of the translation and validation process, including decisions made by the expert committee. Three data collectors with Master of Science (MSc) degrees in nursing and three MSc-level supervisors were recruited for the data collection process. All data collectors were recruited from the chronic care department within the hospital and had prior experience in data collection. However, to minimize potential social desirability bias, each data collector was recruited from a different chronic care unit (HIV, TB, and cancer care unit), excluding the diabetes care unit.

Data were collected using the Kobo Toolbox software. Prior to data collection, a four-day training session was conducted for the data collectors on the study objectives, data collection procedures, and the use of the KoboTool software. Before the main data collection began, 10 sample questionnaires were collected to identify and correct any errors in the questionnaire and to assess the functionality of the Kobo software.

During the data collection period, data collectors submitted their completed tasks to the principal investigator daily and reported any issues that required corrections to their supervisors by phone or in person. The principal investigator addressed any challenges or incidents that arose during the data collection period. Every day the supervisors monitored the daily activities of the data collectors and submitted a daily progress report to the principal investigator. The data collection process and reporting were completed within the planned timetable. All data were accurately recorded and documented, ensuring transparency and supporting the study’s reliability. The findings were reported according to the STROBE guideline for observational studies [88].

Ethical consideration

This study was approved by the Institutional Research Ethics Review Committee (IRERC) of the University of Gondar College of Medicine and Health Sciences (CMHS) (IRERC/65/102025). All methods and procedures were carried out in accordance with the ethical principles outlined in the Declaration of Helsinki [89]. Permission and support letters were secured from the Amhara Public Health Institution (APHI). Additional authorization was also obtained from the chronic care center coordinator of each hospital. The study participant was informed about the purpose, methods, expected benefits, and risks of the study. Before data collection, written informed consent was obtained from each participant. Participants were informed that they had the right to decline participation in the study, and they had also been assured that refusal to participate would not affect their usual healthcare service in the hospital. Participants were informed that their confidentiality would be protected and that no personal information would be disclosed to third parties. To ensure confidentiality, codes were used to identify participants. For those who could not read or write, a thumbprint was used in place of the participant's signature.

Translation and cultural adaptation

The PACIC-5A measurement tool was accurately translated by forward and backward translators. Subsequently, the expert committee thoroughly reviewed and evaluated each translation, reaching a consensus that the tool was relevant for measuring perceived quality of care. No items were removed from the original questionnaire.

Two forward translators, FT-1 and FT-2, independently translated the original English version of the PACIC-5As into Amharic. After completing their independent translations, the two translators met to discuss and synthesize their version into a single Amharic translation. They agreed that most items were accurately translated, with only minor grammatical errors. Items 4, 7, and 17 showed consistent translation by both translators. After correcting these grammatical errors, they produced a preliminary Amharic version called FT-12.

Next, FT-12 was translated back into English by two independent back translators (BT-1 and BT-2). Both translators completed their translations independently, after which they met to verify the items and content equivalence. The committee then reviewed the instructions, items, and response formats of all translations. They compared the back translations (BT-1 and BT-2) and forward translations (FT-1, FT-2, and FT-12) with each other and with the original English version.

During the panel discussion, BT-1 and BT-2 were thoroughly evaluated and compared to identify issues related to sentence structure, clarity, and meaning, with the goal of determining which version most accurately reflected the original intent of the tool. Items 2, 3, 4, 5, 6, 8, 9, 10, 12, 13, 16, 17, 21, 23, 24, 25, and 26 from BT-1 were selected as the most consistent with the original English version. In contrast, items 1, 7, 11, 14, 15, 18, 19, 20, and 22 were taken from BT-2. Minor grammatical corrections were made to the selected items. Finally, the committee confirmed the content equivalence of the tool and produced the final English version of the PACIC-5As tool (S1 Table), which aligns with the original version.

The committee also reviewed the Amharic versions (FT-1, FT-2, and FT-12) in comparison to the original English language version. They reviewed and corrected grammatical errors and selected culturally appropriate wording, meanings, and interpretations of the Amharic translations that aligned with the original English version. Finally, the committee established the initial conceptual, content, and semantic equivalence of the tool and approved the pre-final Amharic version of the PACIC-5As-ET tool for use in Ethiopia (S2-S3 Table).

The pre-final Amharic version (PACIC-5As-ET) was evaluated with twenty patients living with type 2 diabetes at the University of Gondar CSRH. The participants were asked to evaluate and provide feedback on the newly translated tool, focusing on its clarity, sentence construction, response format, meaning, and ease of understanding. The majority of participants (19/20) agreed that the instructions, response format, and items of the tool were clear and easy to understand. However, one participant (5%) commented that item 6, “Have you been guided to understand how the actions you have taken to care for yourself have contributed to your current condition?” was unclear, and another participant noted that item 10, “Have you been advised to go to a specific group or specialist who can provide support or assistance to improve or cope with your condition?” was unclear. Therefore, less than 20% of the participants reported the items were unclear; as a result, the items were not re-evaluated.

Content validity

The content validity index was evaluated and rated by ten committee members. The content validity was computed at both the item and scale levels. The item-level content validity index (I-CVI) was ≥ 80 (80% to 100%), and the average or scale-level content validity index (S-CVI/Ave) was 97.69%; all were found to be acceptable. Moreover, the inter-rater agreement (Gwet’s kappa value) was 95% (95% CI: 0.91–0.99) with a p-value of < 0.001, indicating perfect agreement. The content validity assessment revealed that the measurement tool was relevant, valid, and accurately represented the intended construct.

Face validity

In addition to the expert evaluation, the face validity of the newly translated and validated Amharic version of the PACIC-5As-ET measure was assessed by 20 study participants at the University of Gondar Comprehensive Specialized Hospital. We received valuable feedback from the participants regarding sentence construction, relevance, and clarity. Based on their input, almost all individual items were found to be correct and easily understood by the participants. Overall, the face validity assessment indicated that the measurement tool was clear, relevant, valid, and accepted by the majority of the study participants with T2D.

Socio-demographic characteristics of the study participants

A total of 520 study participants were enrolled, of whom 517 were included, with a response rate of 99.4%. The mean age of the respondents was 54.34 (±11.51) years. More than half of the participants were male, 285 (55.1%), and 384 (74.5%) identified as Orthodox Christians, while 497 (96.1%) belonged to the Amhara ethnic group. The majority of participants, 410 (79.3%), resided in urban areas, and most respondents, 361 (69.8%), were married. Regarding educational status, the largest proportion, 161 (31.1%), had attained a college education or higher. Among occupations, the most common was housewives, comprising 114 (22.1%) individuals. Most participants, 389 (75.2%), had been diagnosed with diabetes for a period ranging from 6 months to 5 years (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Sociodemographic characteristics of patients with T2D attending in the Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t001

Descriptive statistics of the PACIC-5As-ET measure

In this study, the overall mean summary score of PACIC-5As was 2.68 (± 0.62). The mean subscale scores were 2.79 (± 0.69) for Assess, 2.73 (± 0.71) for Advise, 2.63 (± 0.74) for Agree, 2.72 (± 0.69) for Assist, and 2.52 (± 0.76) for Arrange. Within the Assess subscale, item 21 had the highest mean score (3.57 ± 0.81), while item 11 had the lowest (2.25 ± 1.11). In the Advise subscale, item 6 had the highest mean (3.09 ± 0.79), whereas item 9 had the lowest (1.85 ± 1.05). For the Agree subscale, the highest score was recorded for item 3 (3.48 ± 0.88), and the lowest for item 25 (1.99 ± 1.26). In the Assist subscale, item 12 had the highest mean score (3.26 ± 0.97), while item 26 had the lowest (2.07 ± 1.16). In the Arrange subscale, the highest mean was observed for item 16 (3.21 ± 0.76) and the lowest for item 18 (2.02 ± 1.10) (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Mean and standard deviation of the PACIC-5As-ET measure among patients with T2D attending in the Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t002

Ceiling and floor effect

The overall PACIC-5As-ET scale did not demonstrate floor or ceiling effects; only 0.2% of participants achieved the minimum and maximum scores. Similarly, the subscales also showed no floor effects, ranging from 0.2% for Arrange to 0.8% for Assist, and no ceiling effects, ranging from 0.2% for Agree to 1.4% for Arrange. The results indicated that the measurement tool possesses sufficient sensitivity to detect variations at both the lower and upper ends of the score distribution, demonstrating good content validity and reflecting a better distribution of responses [57]. Although fewer items showed a floor effect, none exhibited a ceiling effect (all < 15%), implying that the measure is sensitive to detecting high levels of variation, but not low levels.

Reliability assessment

Internal consistency of the PACIC-5As-ET and its constructs.

The Cronbach's alpha for the overall PACIC-5As-ET was 0.93, indicating excellent internal consistency, while the subscale alpha values were 0.74 for Assess, 0.72 for Advise, 0.76 for Agree, 0.71 for Assist, and 0.82 for Arrange, indicating acceptable reliability. The composite reliability of the overall PACIC-5As-ET was 0.93, demonstrating excellent internal consistency, while the subscales fell within a good to acceptable range: 0.76 for Assess, 0.73 for Advise, 0.79 for Agree, 0.75 for Assist, and 0.83 for Arrange. The subscale reliability tests also confirmed that the items were consistent and suitable for measuring the underlying constructs (Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Cronbach’s alpha, composite reliability, average variance extracted, and test-retest reliability of the PACIC-5As-ET and its subscales among patients with T2D attending in the Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t003

Inter-item and item-total correlation

The study demonstrated that the inter-item correlation ranged from 0.06 (between items 19 and 22) to 0.69 (between items 18 and 23). Based on expert recommendations, the weakly correlated items 19 and 22 were retained in the questionnaire. The corrected item-total correlation ranged from 0.49 to 0.69 for Assess, 0.31 to 0.66 for Advise, 0.41 to 0.72 for Agree, 0.39 to 0.66 for Assist, and 0.53 to 0.66 for Arrange. These correlations demonstrated that almost all items were reliable and suitable for measuring the underlying constructs. Additionally, the ‘Cronbach’s alpha if item deleted’ values for each item indicated good internal consistency, showing that no single item significantly affected the reliability of the scale (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Item-total statistics of the PACIC-5As-ET scales among patients with T2D attending in Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t004

Correlation between constructs

The strength of the association between constructs was computed using Pearson correlation analysis. All constructs demonstrated a strong, statistically significant correlation (p < 0.001, two-tailed). Among the subscales, the strongest correlation was found between Assess and Agree (r = 0.79), while the weakest correlation was between Arrange and Advise (r = 0.43). When comparing the correlations between the overall PACIC-5As-ET score and its subscales, the strongest correlation was observed with Agree (r = 0.92), and the weakest was with Advise (r = 0.79) (Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Correlation between constructs among patients with T2D attending CSRHs in the Amhara region, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t005

Test-retest reliability

The average test-retest value of the overall PACIC-5As-ET was 0.94, with subscale ICC values ranging from 0.66 for “Agree” to 0.87 for “Assist.” All results demonstrated moderate to excellent internal consistency, indicating the stability of the measurement tool over time (Table 3).

Confirmatory factor analysis

A confirmatory factor analysis (CFA) was conducted to assess the model fit and the factor structure of the hypothesized model. Before conducting the CFA, the normality assumption was assessed using skewness and kurtosis. The univariate normality assessment revealed skewness values ranging from −0.31 to 1.07 and kurtosis values ranging from 1.10 to 0.16, suggesting a normal distribution. The multivariate normality assessment indicated that a skewness value of 58 suggested a departure from normality.

Although various fit indices were used to assess the model's fitness, the chi-square (χ²) statistic was 781.7 with a p-value of < 0.001, indicating that the model did not fit well. However, the χ²/df ratio was 2.79, which meets the recommended threshold of ≤ 3, suggesting an acceptable fit. The values of RMSEA and SRMR were 0.06 and 0.08, respectively, both of which satisfy the acceptable threshold of ≤ 0.08. The RMR was 0.12, not satisfying the recommended threshold of ≤ 0.08. The GFI and PRATIO were 0.89, close to the recommended threshold of ≥ 0.90 or near 1, indicating an acceptable fit. Furthermore, the AGFI and PGFI were 0.86 and 0.73, respectively; both values were close to 1 but slightly below the recommended threshold of ≥ 0.90. The IFI (0.57), NFI (0.30), PNFI (0.28), RFI (0.20), CFI (0.40), PCFI (0.35), and TLI (0.32) were all below the recommended threshold of ≥ 0.90. Additionally, the AIC value (860) was lower compared to those of the independence and null models. Overall, the absolute fit indices were generally within the recommended range, whereas incremental fit indices were low; therefore, support for the five-factor structure is limited and should be interpreted with caution (Table 6).

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Model fit indices of the hypothesized model among patients with T2D attending in the Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t006

To enhance the model fit, error covariances such as e1-e10, e3-e21, e2-e4, and e1-e16 were adjusted based on modification indices. As a result, some of the fit indices improved: the χ²/df ratio decreased from 3.01 to 2.79, the GFI increased from 0.87 to 0.89, the AGFI improved from 0.85 to 0.86, and the RMSEA decreased from 0.062 to 0.06; the remaining values remained unchanged.

The standardized regression weights revealed that the factor loadings ranged from 0.34 (Q-19) to 0.84 (Q18, 23, and 25) for the measurement model and from 0.95 (Arrange) to 0.99 (Agree) for the structural model. In the “Assess scale,” the highest loading was 0.79 (Q1), and the lowest loading was 0.52 (Q21). In the “Advise scale,” the highest loading was 0.78 (Q9), and the lowest was 0.34 (Q19). In the “Agree scale,” the highest loading was 0.84 (Q 25), and the lowest loading was 0.43 (Q 3). In the “Assist scale,” the highest loading was 0.77 (Q 10), and the lowest loading was 0.48 (Q 26). In the “Arrange scale,” the highest loading was 0.84 (Q 18 and 23), and the lowest loading was 0.56 (Q 16).

For the “Advise Scale” variables, item 19 of the newly validated and translated question (Question 19: “Have you been told that being treated by other healthcare professionals like ophthalmologists, surgeons, and others helped you to improve your healthcare or treatment?”) had a low loading (0.34). Before deleting the item, the expert committee evaluated its significance and decided to retain it in the questionnaire. Overall, almost all of the factor loadings exceeded the recommended threshold of 0.4, suggesting that the items represent their underlying constructs (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Second order confirmatory factor analysis among patients with T2D attending in Amhara region CSRHs, 2025 (n  = 517).

https://doi.org/10.1371/journal.pone.0329197.g001

Construct validity

Convergent and discriminant validities were assessed based on the Fornell-Larcker criterion. The convergent validity of the first-order constructs (measurement model) was 0.50 for Assess, 0.47 for Advise, 0.50 for Agree, 0.52 for Assist, and 0.59 for Arrange. All subscales showed adequate convergent validity. The convergent validity of the second-order construct (overall PACIC-5As) was 0.93, indicating strong convergent validity (Table 3). While the second-order construct demonstrated high discriminant validity, some of the first-order constructs exhibited lower discriminant validity (Table 7).

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Discriminant validity of the PACIC-5As-ET and its subscales among patients with T2D attending Amhara region CSRHs, 2025 (n = 517).

https://doi.org/10.1371/journal.pone.0329197.t007

Implications of the study

The cross-cultural adaptation and psychometric evaluation of the PACIC-5As is important for assessing the perceived quality of care among patients with T2D. This study enables healthcare professionals and policymakers to identify gaps and enhance the quality of diabetes care. It also supports researchers in evaluating patient-centered care and provides insights to relevant stakeholders for informed decision-making. Moreover, this study promotes patient-centered care practices, enhances self-management support, and supports evidence-based decision-making to improve health outcomes for patients with T2D.

Strengths and limitations of the study

In this study, we employed various statistical tests to assess the reliability and validity of the tool, including a test-retest assessment. We also utilized a maximum sample size (sample-to-parameter ratio), which may enhance the external validity of the study. However, the participants’ responses were clustered at the lower end of the scale, which may limit the tool's sensitivity to detect variations at that level. Although the incremental fit indices were low, the factorial validity should be interpreted with caution. In addition, the quality of care provided to diabetes patients may vary across healthcare settings, and studies conducted only in tertiary healthcare settings may limit generalizability to other healthcare settings, like primary and secondary healthcare settings.