Study context, design, and population

This research was conducted at the oncology centers of two tertiary public hospitals, Tikur Anbessa Specialized Hospital (TASH) and Saint Paul’s Hospital Millennium Medical College (SPHMMC) in Addis Ababa, Ethiopia. These hospitals are among the top referral units in the country, providing specialized and comprehensive healthcare services along with medical training. Established in 1969 by the German Evangelicals, SPHMMC is currently under the administration of the federal Ministry of Health of Ethiopia (FMOH). In addition to its academic functions, it offers a diverse range of healthcare services with 392 beds, serving an annual average of 200,000 patients and a catchment population of over 5 million [33]. The hospital, currently, provides all chemotherapy services except radiotherapy to patients diagnosed with cervical cancer. TASH is likely the oldest and largest referral hospital in the country, boasting 700 beds and being owned by Addis Ababa University. In addition to providing training across various health disciplines, the hospital offers healthcare services to patients referred from all parts of the country. The hospital’s dedicated oncology unit, situated in the 5^th police station, provides both inpatient and outpatient oncology services, covering radiotherapy, chemotherapy, surgery, and chemoradiotherapy [34]. A cross-sectional design was implemented between 1^st January and 30^th March 2023 to validate the 30-item Amharic version of perceived access to health-care services in patients with cervical cancer. Efficient and swift, this design was deemed suitable for our study, aiming to gauge the validity of constructs related to access to healthcare services. The assessment occurs at a specific point in time, capturing a snapshot of the data from patients. All patients with a confirmed diagnosis of cervical carcinoma, undergoing follow-up during the study period, and consenting to participate, were consecutively enrolled. Although not a strictly random selection technique, consecutive sampling was chosen in this study because the participants shared homogeneity in terms of diagnosis, the patient population’s number and flow were limited and stable, ensuring temporal representation during the three-month data collection period. This technique also reduces selection bias, provided that the primary referral settings were simultaneously included. As studies suggested that a participant to item ratio of 5:1 or that a sample size of at least 200 would be adequate for factorial validity [35, 36], a total sample of 308 patients was recruited and assessed.

Validity and reliability of the instrument

Ensuring the validity and reliability of any quantitative research tool is a crucial step to assure the internal and external validity of findings. This is especially pertinent in cases like ours, where the tool predominantly incorporates a hierarchical and abstract psychometric component. This, in turn, plays a significant role in generating robust and accurate results that can be replicated [37]. The validity and reliability of the instrument used in data collection have been emphasized, employing a variety of approaches sequentially to ensure its accuracy and consistency. Presently, the concept of access to healthcare remains intricate, and the proposed domains and items may not comprehensively capture its complexities. Various additional factors may still influence, potentially resulting in insufficient scoring or factor loading. In some cases, an item might show reliability without indicating validity, posing a challenge when trying to generalize results. Despite this, the steps highlighted in the guidelines by Tsang et al. [38], and Yusoff and colleagues [36] were adhered to during the adaptation and validation process of the questionnaire. In general, the entire process of instrument development has undergone a sequence of five key steps, which include instrument design and adaptation, translation, validity testing of the draft instrument, piloting, and psychometric evaluation of the construct. The specific method, whether qualitative or quantitative, to be employed, along with the detailed steps and necessary inputs during each stage, are described below.

Instrument design and adaptation

The psychometric domains and constructs for assessing patients’ perceived access to healthcare services have been adapted from a recently developed and validated questionnaire by Hoseini and colleagues [8]. The tool has demonstrated satisfactory fit in the Persian context, where the health system and economic conditions are quite similar to ours. In fact, some sections of the tool were designed and recommended by previous researchers for assessing access to healthcare [1, 4]. The earlier instrument, which focused on primary healthcare services, has been modified to suit the current study settings, particularly secondary healthcare services for patients with cancer. Consequently, the constructs were adjusted to be applicable to any health facility, not restricted to health centers. The items assessing availability were entirely revised to inquire about the availability of medicines, an adequate number of staff, and the skills of health personnel at the settings. The first domain, originally named ’accessibility,’ has been changed to ’reachability’ to better reflect geographic access to services. Additionally, the first item, ’the services I request are given at the health center,’ was removed as it was considered redundant with specific inquiries under the ’availability’ domain. Furthermore, an item in the ’acceptability’ domain has been revised to account for religious diversity in the current setting, in addition to the cultural differences mentioned in the earlier version. As details on the contents of sections 1 and 2 of the instrument have been provided in previous sections, the focus of validation and subsequent descriptions will be on the adapted psychometric instrument. To navigate cultural nuances during the translation and validation processes, various measures were implemented. These included engaging experts in public health and research with a background in tool translation, consulting with senior experts familiar with the local health system, ensuring that all experts involved in the tool content validation were informed about the study objectives, incorporating back-translation, and conducting pilot testing of the instrument.

Forward translation to Amharic version

Two translators took part in the translation process of the initially adapted psychometric instrument into Amharic. In an effort to minimize disparities in content interpretation, the principal investigator (PI), possessing a master’s degree in pharmacoepidemiology and considerable experience in translating research instruments, undertook the initial translation of the original content. Subsequently, a public health specialist, well-versed in tool translation and qualitative research, and without prior knowledge of the study’s objective, conducted the second translation. Any discrepancies arising between the two translations were addressed through discussion or by seeking the input of a third translator.

Backward translation

In this phase, the Amharic version of the instrument underwent back-translation into English to ensure the consistency of items with the adapted version and to identify any unforeseen deviations. Two public health specialists, well-versed in research methodology and unfamiliar with the study objectives and the intended message of the entire instrument, participated in the back-translation process.

Content validity

Validity, by definition, refers to the extent to which collected data accurately represents the phenomenon under investigation or "measures what is intended to be measured" [39]. Both qualitative and quantitative methods of content validity have been implemented, as they were employed by a panel of experts in the primary psychometric evaluation of the original tool [8]. The purpose of this assessment was to evaluate the instrument, in a multidisciplinary approach, for potential use in the Ethiopian context. It is recommended that the content validation process includes four essential steps: domain definition, domain representation, domain relevance, and appropriateness of the test construction procedure [40]. As a qualitative method for ensuring validity, a purposively selected group of health professionals, including physicians, nurses, pharmacists, researchers, and public health experts, were invited to assess the domain content, wording, appropriate placement of items, and relevance for measuring perceived access to healthcare services, specifically in patients with cervical cancer. The selection of the expert panel was specifically guided by the contents of the tool from the perspectives of the mentioned professions, some of whom had experience in cancer care and research. Subsequently, individual information was collected and incorporated into the tool before proceeding to quantitative evaluation. The feedback received from the panel played a crucial role in refining and enhancing the final instrument. A systematic process was followed to incorporate the feedback, which included thorough reviews of each input, prioritization based on relevance and potential impact, discussions with panel experts, iterative revisions, and validation.

The quantitative process of content validation in this study has undergone the necessary steps outlined elsewhere, including domain definition and representation. The approach was carried out using both a face-to-face (by panel of experts) and a non-face-to face approach (using an electronic assessment form) [36] and was centered on assessing the essence, relevance, and appropriateness of the scale.

The evaluation criterion for item relevance was structured using a 3-level Likert scale: "item is essential," "item is useful but not essential," and "item is not essential." Subsequently, content validity ratio (CVR) and content validity index (CVI) values were calculated. The formula for content validity ratio was CVR = (Ne—N/2) / (N/2), where Ne is the number of panelists indicating that the item is "essential," and N is the total number of panelists, as documented earlier by Lawshe [41]. The decision of whether to include an item in the scale depended on the number of experts and the corresponding minimum ratio recommended in Lawshe’s table. Panelists or experts were furnished with an evaluation format to assess the level of agreement they could report based on the essence of the items in question.

Similarly, we employed a widely recognized method of content validity known as the content validity index (CVI). The CVI assesses the relevance and clarity of items. According to Davis [42], a 4-point ordinal scale (1 [not relevant], 2 [somewhat relevant], 3 [quite relevant], and 4 [highly relevant]) was incorporated to evaluate both dimensions.

Utilizing the scale provided in the table, two measures were computed: the item-level content validity index (I-CVI) and the scale-level content validity index (S-CVI). The I-CVI indicates the proportion of agreement on the relevance of each item, ranging between 0 and 1, while the S-CVI is defined as "the proportion of total items judged content valid." An I-CVI of 80 percent or higher is generally considered "appropriate" [43]. Some studies suggest that 78% (0.78) is an acceptable lower threshold for I-CVI, while S-CVI is considered excellent when it achieves a score of at least 80% [44].

Face-validity

Face validity was assessed to determine if the instrument ultimately gains acceptability based on its face value by the respondents. This aspect of validity addresses whether an instrument appears valid to subjects, patients, and/or other participants [38]. One can address this question from the participants’ perspective by assessing their agreement with the items and their wording in the instrument to fulfill the research objectives. Another definition of face validity is the extent to which a test seems to measure what it asserts to measure. It provides a broad evaluation as a quick assessment of what the test aims to measure [45]. Face validity can be assessed using a 5-point Likert questionnaire with options such as ’very important,’ ’important,’ ’relatively important,’ ’slightly important,’ and ’unimportant.’ An item impact score is calculated by multiplying the frequency and importance. Frequency is determined by the percentage of participants with a score of 4 or 5, while importance is assessed by averaging the sum of scores for each item. Ultimately, the item impact score is obtained by multiplying these two results. The general guideline is that if the item impact of an item is equal to or greater than 1.5 (mean frequency of 50% and an importance mean of 3 on the 5-point Likert scale), it is advisable to retain it in the instrument; otherwise, it may be eliminated [46].

Piloting the instrument

A sample size of 30 participants was employed in this test, a size deemed suitable for identifying any issues (such as ambiguities or misunderstandings) in the instrument, as per documented acceptability [47]. In this phase, the demographic makeup closely mirrored that of the broader study population with respect to variables like diagnosis, study setting, and source population. However, it’s crucial to emphasize that the pilot sample was separate and not included as part of the official study population. The primary objective of the pilot testing was to verify the face validity and internal consistency of the items and assess the duration required for an interviewer to complete the questionnaire. According to Yusoff [36], acquiring a standardized method for distributing the tool and conducting interviews, particularly for interviewer-administered questionnaires, was emphasized. Additionally, improvements were made to the flow, process of items, administration, and logistical aspects of the fieldwork during this phase. Accordingly, altering the sequence some items, determining the estimated time that an interview would take, and the management of the data collection process through the KoboCollect application were among the areas revised during this time. The rationales behind the modifications were to ensure logical flow of items, improve responders’ trust, and ensuring the quality of the data collected. Finally, at this phase, a preliminary internal consistency test was performed to make sure if items are eligible for further development. This was done so because Cronbach’s alpha is not an absolute measure of consistency under all circumstances [48], and variability across different samples is always inevitable-hence the need to test reliability of the measure for both the pilot and the construct validation. While the use of Cronbach’s alpha was meant to check and improve the internal consistency of items per each domain, the confirmatory factor analysis was to assess the underlying construct validity of the instrument. In the context of our study, the goal was to measure specific constructs related to access to healthcare, and the CFA allows us to confirm that the observed variables (survey items) align with the hypothesized latent constructs, both at the first and the second order levels.

Initial reliability analysis of factor constructs

Reliability assesses the internal consistency of items within a scale, indicating whether the items in the instrument are conceptually compatible. In this study, the internal consistency of item constructs was evaluated using Cronbach’s alpha. While a sample size of 20 to 30 patients is considered sufficient for evaluating the reliability of a survey instrument, as recommended elsewhere [49], we used the entire sample of respondents recruited as part of the factor analysis above, which also enhances the precision of estimates. In doing so, an alpha value of at least 0.7 was deemed acceptable [50].

Confirmatory factor analysis

Construct validity is a crucial concept in assessing a questionnaire designed to measure an unobservable construct, often referred to as latent variables. Evaluating the construct validity of a new instrument involves estimating its association with other measures within the same domain, typically utilizing a different scale. This correlation may manifest as positive, negative, or null [48]. As such, this study assessed four types of construct validity: convergent validity (positive correlation), divergent (discriminant) validity (poor correlation), known-group validity, and factorial validity.

Derived from content analysis, factorial validity was utilized in this study as the suitable method for evaluating the validity of item constructs within the domains of access to healthcare services. This choice was made due to the presence of 30 items distributed across six latent domains associated with the concept of perceived access to healthcare. In the analysis of factorial validity, it is expected that several items intended to measure a specific dimension within a construct of interest would exhibit higher interrelatedness compared to those measuring other dimensions [7, 48]. This implies that factor analysis unveils relationships among items, between items and constructs, and among different constructs [51].

The interpretation of results from the confirmatory factor analysis (CFA) was as follows: Model fit was considered good if the Tucker Lewis Index (TLI), Normed-Fit Index (NFI), non-normed-fit index (NNFI), and goodness of fit (GFI) were ≥ 0.95 or acceptable if ≥ 0.90. Comparative fit index (CFI) and adjusted goodness of fit (AGFI) were considered acceptable if ≥ 0.90. For the root mean square error of approximation (RMSEA), a fit was considered good if ≤ 0.05 and moderately acceptable if between 0.05 and 0.08. The ratio of χ2 to degrees of freedom (χ2/df) was deemed the best fit when below 2 and moderate fit when it ranged between 2 and 3 [52, 53]. To distinguish the discriminant validity across included latent variables in the model, Henseler’s later heterotrait-monotrait (HTMT2) ratio method [54] was utilized.

Data collection process and quality assurance

After translation and initial content and face validation, data for the study were gathered using an interviewer-administered structured instrument created in the KoboCollect mobile-based application. Subsequently, the data were managed through the KoboToolbox server [34]. Five data collectors, consisting of three individuals with master’s degrees and two general practitioners, were engaged in the data collection process. A one-day training session was conducted for both data collectors and supervisors, encompassing a comprehensive orientation to research ethics, study objectives, and familiarity with the instrument. This thorough training aims to equip data collectors and supervisors with the essential skills, understanding, and cultural awareness needed for consistent and accurate data collection, thereby enhancing the overall quality and reliability of the study findings. In addition to pretesting and training, continuous supervision of the data collection process and adherence to the study protocol were implemented to ensure quality.

Data analysis

Data was cleaned and coded in MS-Excel for windows, R programming [55], and Amos software v.26 for analysis [56]. Descriptive statistics were utilized to illustrate participants’ profiles, while structural equation modeling (SEM) was employed to present model fitness indices, a standardized correlation matrix, and the structure of the construct. Given the violation of normality assumptions, measurements were compared against bootstrapped estimates with confidence intervals and standard errors of the sample. Known group validity was confirmed using Student’s t-test (for two-grouped variables) and one-way analysis of variance (one-way ANOVA) for variables with more than two categories. The estimates in the t-test were based on the second row (assuming unequal variances), while the post-hoc pairwise comparison in ANOVA utilized Dunnett’s T3 calculation [57]. The computation of divergent validity was estimated the methods proposed by Henseler et al. [54]. The application of the mentioned models and specified analysis methods was considered suitable for the nature of the collected data, aiming to generate quantitative findings relevant to the research question objectives.

Ethics consideration

The study received ethics approval from Saint Paul’s Hospital Millennium Medical College (SPHMMC) (Ref.no: PM23/284), and a support letter was issued to Tikur Anbessa Specialized Hospital (TASH) (Ref.no: PM23/295) as part of the mini-grant offer obtained from Saint Paul Institute for Reproductive Health and Rights (SPIRHR). Verbal informed consent was obtained from all patients participating in this study. The informed consent process entails a comprehensive explanation of the study objectives, procedures, and potential risks and benefits to participants. Verbal consent requests are presented to participants, and any questions they may have are addressed before securing their voluntary agreement to participate. To uphold confidentiality and privacy, participant identities are anonymized, data is securely stored, and access is restricted to authorized personnel. These measures are implemented to safeguard sensitive information and maintain ethical standards throughout the study.

Characteristics of participants

A total of 308 participants took part in the current survey. Among them, 202 (65.6%) were referred patients from outside of Addis Ababa. The mean (SD) age of the patients was 50 (±11) years, ranging from 24 to 83. Over half (57.8%) were married, and approximately half (51%) had basic literacy skills. Additionally, more than two-thirds (69.5%) reported having a family size larger than 5. The median monthly income for households was reported to be 4000 ETB, with a mean (SD) of 4539 (±3325) ETB. About 55.5% of the respondents surpassed the extreme poverty line according to the World Bank’s income classification for low-income countries. Among the respondents, 138 (44.8%) were housewives, and nearly three-quarters (74%) had coverage under community-based health insurance (CBHI). Examining clinical profiles, 128 (41.6%) were in stage II of the disease, followed by 111 (36%) in stage III. The majority (64%) had no history of any chronic comorbidity (see Table 2). In broad terms, the demographic profile of participants indicates that a significant portion of oncology center attendees were referred from various regions of the country, were married, had family sizes exceeding 5, earned monthly incomes above the extreme poverty line, possessed coverage under CBHI, and had a history of comorbidities. Consequently, the validated instrument is expected to be applicable and relevant to patients with similar characteristics throughout the country.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of study participants among patients visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.t002

Initial validity and reliability assessment

In this phase, a series of qualitative and quantitative validity assessments were conducted. Seven experts with diverse specializations, including clinical (MD), public health research (MPH), qualitative methods (MPH), pharmacy service (BPharm), Microbiology (MSc), pharmacoepidemiology (MSc), and lab quality assurance (MSc), were invited to evaluate the content and face validity of the 30-item adapted instrument. Two experts were involved in both forward and backward translation processes. After modifying two components (changing the factor name from ’access’ to ’reachability’ or referring to geographic access and altering ’screening for cervical cancer’ to ’breast or colon cancer’), quantitative measures were derived from expert opinions. The first adjustment aimed to prevent confusion. In the earlier version, the term ’access’ referred specifically to geographic access, causing confusion with the broader concept of access covering all constructs under investigation. In the second modification, we opted to replace "screening for patient breast or colon cancers" instead of cervix. This change was made to align with the study population, which consists of patients already diagnosed with cervical cancer. The Content Validity Ratio (CVR) calculated ranged from 0.5 to 1, while the Item-Level Content Validity Index (I-CVI) values varied from 0.75 to 1. The Scale-Level Content Validity Ratio (S-CVR) was 0.91. Additionally, face validity scores ranged from 2.4 to 4.8. The obtained content validity ratio (CVR) and item-level content validity index (I-CVI) values suggest strong content validity for the instrument. Higher values indicate a consensus among experts on the necessity and relevance of the items, affirming the quality of the instrument in assessing the intended constructs related to healthcare access in the study context. Following content and face validity assessments, the instrument was distributed to a sample of 30 participants to measure its reliability. The Cronbach’s alpha for the entire scale was 0.72. Higher alpha values, typically greater than 0.7, reflect stronger internal consistency, reinforcing the instrument’s reliability in consistently measuring the factors under investigation.

Internal consistency of factors

As illustrated in Table 3 below, an assessment of factor-level reliability was conducted to evaluate the internal consistency of item constructs within each of the six domains. If the removal of an item resulted in an improvement in factor-level consistency, that specific item was excluded. However, all factors were required to exhibit a Cronbach’s alpha of at least 0.7 to qualify for subsequent factor analysis. Additionally, a factor loading of at least 50% in the structure matrix was expected. Despite the factor ’reachability’ achieving an acceptable Cronbach’s alpha value of 0.98 with two items, it was excluded from the final second-order structure matrix due to its very low loading (0.12).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Factor-level reliability test assessment of indicators on perceived access to healthcare among patients visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.t003

Convergent validity

The assessment of convergence among the items within each factor and the latent variables explaining perceived access to healthcare services involved evaluating composite reliability (CR) and average variance extracted (AVE). Upon examination of these measures, it was observed that all factors met the recommended minimum threshold of 0.6 for CR, signifying a high level of reliability for the scale. This measure aligns with Cronbach’s alpha values, suggesting good construct reliability. Similarly, the AVE for each scale was computed, and the findings indicated that the average reliability of each factor exceeded the recommended threshold of 0.5, affirming model fit. In conclusion, both CR and AVE measures support the notion that the construct fits the data well. Additionally, the higher CR compared to AVE suggests strong reliability. The latent variables of the first-order scale were treated as inputs for the second-order model (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Convergent validity of latent factors in perceived access to health-care services in patients with cervical cancer visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.t004

Divergent validity

The instrument’s divergent validity was assessed using the updated criterion of the hetero-trait to mono-trait (HTMT2) ratio, as proposed by Henseler et al. (48). This method involves calculating the ratio of the sum of intra-factor level correlations to inter-factor level correlations to gauge the discriminant ability of a latent variable in comparison to other latent variables. Among the ten pairs of factors examined, eight demonstrated an acceptable level of discriminant validity (below 0.90), while two pairs, specifically acceptability vs. awareness and acceptability vs. availability, exhibited a slightly higher value (0.91) compared to the recommended threshold mentioned earlier (Table 5). While this ratio helps determine the extent to which different latent variables discriminate from each other, the implications of the two factor pairs, acceptability vs. awareness and acceptability vs. availability, showing higher values (0.91), suggest a potential overlap or shared variance between these constructs. This may indicate some degree of interrelatedness or conceptual similarity between acceptability and awareness, as well as acceptability and availability.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Divergent validity of latent factors in perceived access to health-care services in patients with cervical cancer visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.t005

Known-group validity

The instrument’s validity was assessed by examining its ability to discriminate between various groups within the same sample, anticipating differences in their perceptions of access to healthcare services. These groups included income category, health insurance coverage, social support presence, and occupation. The construct effectively differentiated between groups based on average monthly income, with those above the extreme poverty line scoring significantly higher than those below the line (t = -5.95; df = 187.8; p<0.001). Similarly, individuals with current community-based health insurance (CBHI) coverage exhibited a relatively low mean aggregate score, indicating a poorer perception of access to healthcare services (t = -5.65 df = 200.7; p<0.001). Moreover, respondents reporting any form of social support had a significantly higher mean aggregate score for perceived access to healthcare services (t = 3.39; df = 186.4; p = 0.001). Furthermore, a post-hoc analysis using one-way ANOVA with Dunnett’s T3 comparison method revealed variations in perceived access to healthcare services among different occupation groups. Specifically, government employees had a significantly higher mean aggregate score compared to housewives (mean difference = 7.10; 95% CI = 1.85–12.27; p = 0.003) or those with no job at all (mean difference = 12.72; 95% CI = 5.16–20.18; p<0.001). The findings from the known group validity analysis indicate that the instrument effectively distinguishes between groups based on income, health insurance status, social support, and occupation. Higher mean scores among individuals with income above the extreme poverty line, those without current community-based health insurance (CBHI) coverage, and those reporting any form of social support suggest a more favorable perception of access to healthcare services in these groups. Additionally, the instrument highlights significant variations in perceived access to healthcare services among different occupation groups, with government employees scoring notably higher compared to housewives or those without employment. Overall, these results demonstrate the instrument’s ability to discriminate between diverse groups based on their characteristics and perceptions of healthcare access.

Model fitness indices

The overall fit of the construct was evaluated using various parameters, as outlined in Table 6. Despite the chi-square statistics for goodness of fit not rejecting the null hypothesis of good fit (pCMIN<0.001), the ratio of minimum discrepancy to the degree of freedom fell within the range of 2<CMIN/df<3, indicating a well-fitting model. The normed fit index (NFI), a size-insensitive measure of fitness, yielded a favorable value of 0.93, suggesting that the present model was a good fit. An alternative adjustment to the NFI using the incremental fit index (IFI) and comparative fit index (CFI), both considering sample size and degrees of freedom, resulted in higher values (0.95), indicating a well-fitted model. Additionally, the Tucker-Lewis Index (TLI), accounting for degrees of freedom, indicated a better fit. The standardized root mean square of residuals (SRMR), measuring the difference between observed and implied correlation matrices, was notably lower at 0.04, well below the recommended ceiling of 0.08. Lastly, the standardized root mean square error of approximation (RMSEA) value was quite low at 0.07, suggesting a moderate to acceptable fit of the data to the model (see Table 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Measures of fit indices of latent constructs in perceived access to health-care services among patients with cervical cancer visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.t006

Observing these results, the model fitness indices collectively indicate a good fit of the data. While the chi-square statistic did not reject the null hypothesis of a good fit, other indices such as NFI, IFI, CFI, TLI, SRMR, and RMSEA consistently support the conclusion of a well-fitted model. These indices consider factors such as size, sample characteristics, and degrees of freedom, providing a comprehensive evaluation of the overall fit of the model.

The second-order structure of the instrument, comprising 19 items and 5 latent domains, along with selected model fit indices, is illustrated in the figure below (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Structural equation modeling of the construct structure in perceived access to healthcare services among patients with cervical cancer visiting oncology centers in Addis Ababa, Ethiopia.

https://doi.org/10.1371/journal.pone.0300815.g001