Study design

The study employed a concurrent mixed methods design [27], whereby quantitative and qualitative data were collected and analysed within the same timeframe. The quantitative phase employed an analytical cross-sectional design, which involved collection of data using a structured questionnaire. The qualitative phase, on the other hand, employed a descriptive phenomenological design which involved individual interviews with participants.

The use of multiple methods or data sources can enhance the validity of the study findings by identifying biases that may be present in a single method and allowing the researchers to compare quantitative and qualitative results [28]. Moreover, it helps the researchers to gain a comprehensive understanding of the phenomenon [28]. Qualitative methods can provide rich description and context but may not be representative of a larger population, while quantitative methods provide statistical information which can be generalized to a larger population but may lack contextual information [29]. Additionally, quantitative data may be affected by errors in measurement [30] while qualitative data may be influenced by the subjectivity of researchers [29]. By employing multiple methods, the researchers can reduce the impact of these biases.

Study setting

The outpatient diabetes clinics of two district-level hospitals in the Ashanti Region of Ghana were used as the study sites. One of the hospitals, the University Hospital, Kwame Nkrumah University of Science and Technology (also known as KNUST Hospital) is in the Oforikrom Municipal, which is an urban municipality [31]. It is a quasi-government hospital and caters for staff of the Kwame Nkrumah University of Science and Technology (KNUST) and their dependants, students at the university and people from surrounding communities. The hospital has a 125-bed capacity and offers general medical services as well as specialist services.

The other hospital, Ejisu Government Hospital is a public hospital in the Ejisu Municipal, which constitutes both rural and urban localities [31]. It is a 95-bed capacity facility and offers both general and specialist services. Both hospitals hold outpatient clinics once every week for persons with diabetes. Persons who visit the clinic go through blood glucose and blood pressure monitoring, consultations by physicians and dietitians, and receive dietary advice in the form of general health education. Data was collected from persons with T2D who attended the diabetes clinics for one month (between September and October 2022).

The study sought to target both rural and urban population. Hence, the choice of the KNUST Hospital and Ejisu Government Hospital is based on their diverse patient populations. The KNUST Hospital provides services to a mix of students, university staff and locals while the Ejisu Government Hospital serves both rural and urban communities. The presence of patients from different socioeconomic backgrounds makes the two hospitals ideal settings to capture a wide range of perspectives on dietary self-care in diabetes.

Study population

All persons with T2D who attended the outpatient diabetes clinics in the hospitals constituted the population for the study. The total population of persons with T2D who had been diagnosed for at least one year was estimated at 120 and 80 for KNUST Hospital and Ejisu Government Hospital respectively, resulting in a total of 200. The average monthly attendance was used for this estimation, and it was obtained from the outpatient diabetes clinic registers of the two hospitals.

Inclusion and exclusion criteria

Participants who were eligible to be sampled for the study included those who: a) were 18 years and above, b) had a confirmed diagnosis of T2D for at least one year and had sought care from the diabetes clinic at least twice during the last 12 months, and c) could speak, understand, or write either English or Twi or both. Persons who had been diagnosed with gestational diabetes and were critically ill or had significant cognitive impairments excluded from this study.

Sample size determination and sampling for the quantitative phase

The sample size for the quantitative phase was calculated using the Taro Yamane formula: [32]. Where n signifies the sample size, N signifies the population under study, and e signifies the margin of error (0.05). The total population (N) was estimated at 200 using the average monthly attendance at both hospitals and the minimum required sample size obtained was 133. To compensate for registration errors and non-response, a contingency sample of 10% was considered, resulting in an estimated sample size of 146.

Using GPower [33] for a post-hoc power analysis, we determined that a sample size of 146 will yield >99% power at 95% confidence level. This determination was based on an assumed odds ratio of 4.7 between diabetes knowledge and dietary adherence as reported in a previous study in Sudan [34] and a two-tailed z-test.

Consecutive sampling techniques were used in selecting eligible participants for this study phase. Thus, all participants who met the inclusion criteria were invited to participate until the desired sample size was reached.

Sample size determination and sampling for the qualitative phase

Participants for the qualitative aspect of the study were purposively sampled using maximum variation techniques. The researchers selected participants who differ across multiple criteria such as age, sex, place of residence, income, education level, employment, and marital status. The aim of employing this technique was to capture a variety of experiences and viewpoints. These participants were selected from the sample members recruited for the quantitative phase of the study. This is a method of integration known as connecting and it occurs when one dataset links to the other through the sampling frame [35]. Participants were invited to participate in the interview immediately after the quantitative data collection. Thus, participant selection and data collection for both quantitative and qualitative phases were done concurrently.

Data saturation (when new data replicates what has already been gathered) was achieved after conducting 14 interviews [36].

Data collection tools for the quantitative phase

A structured questionnaire consisting of three sections was used for the survey. The first section had sociodemographic and clinical variables including age, sex, marital status, place of residence, educational status, monthly income, and employment status) and clinical variables status, (duration of condition, family history of diabetes, and presence of comorbidities). For the place of residence, the researchers obtained name of the locality of residence of the participants and designated them as rural or urban. All localities within the Oforikrom Municipality were designated as urban. For the Ejisu municipal, all localities apart from Ejisu (Krapa inclusive), Kwamo and Fumesua were designated as rural [37, 38].The second section had the Perceived Dietary Adherence Questionnaire (PDAQ) for the measurement of dietary adherence (outcome variable). PDAQ is a nine-item questionnaire that was developed and validated by Ghada Asaad et al. in 2015 [39]. The response is based on a seven-point Likert scale. Some of the questions on the scale include “On how many of the last 7 days did you eat foods high in sugar, such as cakes, cookies, desserts, candies, etc.?”, and “On how many of the last 7 days did you space carbohydrates evenly?” Higher scores reflect higher adherence except for items 4 and 9, which reflect unhealthy choices (foods high in sugar or fat). To calculate a total PDAQ score, the scores for items 4 and 9 were inverted. The first and second items on the scale were changed to: “On how many of the last 7 days have you followed dietary recommendations given by your healthcare provider?” and, “On how many of the last 7 days did you eat the number of fruit and vegetables you are supposed to eat based on recommendations from your healthcare provider?” respectively. These questions were modified because, in the original questionnaire, they were based on Canada’s Food Guide which may not apply in our setting. Additionally, the food examples given in item 3 and item 8 were modified to suit the context of the study area and study population. Participants were classified as having good dietary adherence if they obtained a score equal to or above the median score. The median score was used in this study because the Shapiro-Wilk test indicated that was not normally distributed (W (142) = 0.636, p<0.001).

The third section had nine questions from the revised version of the General Nutritional Knowledge questionnaire (GNKQ-R) [40] to assess participants’ diabetes-related nutrition knowledge (DRNK). The questions were based on the following areas: number of fruits and vegetables eaten per day, oils to cut down, foods to cut down a lot or increase intake, foods high or low in fibre, foods high or low in starch, foods high or low in added sugar, foods with high glycaemic index, foods likely to raise blood cholesterol and healthy food choices. Some modifications were made to the food choices for a better understanding of participants, considering the local context while maintaining the relevance of each question in the original questionnaire. Each item carried one point for a correct answer. Some questions had subsections and those sections were treated as separate items. The total possible score for each participant ranged from 0 to 27. A score equal to or above the median score was categorized as high DRNK and a score below the median score was categorized as low DRNK. The median score was used in this study because the Shapiro-Wilk test indicated that was not normally distributed (W (142) = 0.910, p<0.001).

Data collection tool for the qualitative phase

Semi-structured interviews were conducted in English and Asante Twi (a Ghanaian vernacular) to explore the facilitators and barriers to dietary adherence. The questions asked included: the factors that enabled them to follow the dietary recommendations given by healthcare professionals and the factors that made it challenging for them to adhere to dietary recommendations received from healthcare professionals. The questions were developed based on the research objectives, a review of relevant literature, and feedback from experts and pre-testing with five persons with T2D at the Kumasi South Hospital in the Asokwa Municipal, Ashanti Region.

Data collection procedure for the quantitative phase

Persons attending the outpatient diabetes clinics in the various hospitals were approached and the objectives of the study and procedures involved were explained to them. Persons who agreed to take part in the study gave written consent and the questionnaires were self-administered or interviewer-administered depending on the participant’s choice. The completion of the questionnaires took about 10–15 minutes.

Data collection procedure for the qualitative phase

Fourteen participants, selected from the participants who took part in the quantitative phase, were purposively sampled for individual face-to-face interviews after they had provided informed written consent. The interviews were audio-recorded and were conducted in English or Asante Twi depending on the participant’s language of preference. The interviews lasted between 15 and 30 minutes and were conducted in a quiet location within the hospital premises to allow participants to freely express their views. Follow-up questions and probes were used to help gain more understanding of participants’ responses. Data saturation was reached after conducting 14 interviews.

Interviews were facilitated by the first, third and fourth authors (either DW, RM or PA) whereby one led the discussions whereas the others took notes during the sessions. RM is a male whereas DW and PA are females; all three researchers have a background in nursing and have received training in the conduct of qualitative interviews and data analysis.

Quantitative data processing and analysis

Data were analysed using Statistical Product and Service Solutions (SPSS) version 29.0 for Windows (IBM SPSS Statistics). There were no missing data in the survey responses as the researchers thoroughly examined all sections to ensure their completeness. Descriptive statistics such as means, medians, standard deviations, range, and percentages were used to describe continuous variables including age and diabetes-related nutrition knowledge. Categorical variables were presented in frequencies and percentages and data on dietary adherence were described using medians and interquartile ranges. To identify the factors associated with dietary adherence, a binary logistic regression analysis was done. The variables entered into the binary logistic regression model were selected based on evidence from previous literature. Previous studies in Ghana and other countries have reported that age, sex, marital status, educational level, employment, monthly income, family history of diabetes, presence of comorbidities, duration of condition and dietary knowledge significantly influence dietary adherence [8, 12–14, 21, 41]. Both univariable and multivariable logistic regression analyses were done and reported using odds ratio and their 95% confidence intervals. The significance level was set at a p-value of ≤ 0.05.

Qualitative data analysis

After the end of each interview, member checking was done to allow participants to make the necessary clarifications or corrections to the main issues raised during the session. Of the 14 interviews, 11 of them were conducted in the local Ghanaian language (Asante Twi) whereas the remaining three were conducted using the English language. The interviews that were conducted in Twi were forward-translated into English and back-translated into the local language to ensure the accuracy of the translation process. The six-step framework for thematic analysis provided by Braun & Clark [42]was applied by the researchers during the analysis. The researchers involved in the data collection (DW, RM, and PA) independently read the transcripts several times for familiarisation, then developed and assigned codes to the data using NVivo software, version 20.0. The assignment of codes was done by identifying and labelling meaningful units of the data based on their relationship to the research questions. All codes were solely derived from the data. The three researchers then met to discuss generated codes and resolve discrepancies until a consensus was reached. Where required, disagreements were resolved through the second, fifth and eleventh authors (AKA-D, PAD and JK-D). The codes were later merged based on patterns and relationships between codes to form larger meaningful units (sub-themes and themes). The themes initially developed in the process of analysis were reviewed and refined by the first, second, third, fourth, fifth and eleventh authors (DW, AKA-D, RM, PA, PAD and JK-D) to ensure their consistency and relevance to research objectives. The final step involved naming the refined sub-themes and themes.

Integration methods

At the methods level, integration was accomplished through connecting (explained in data collection for qualitative phase), while joint display was used at the interpretation and reporting level as described by Fetters et al. [35]. For the joint display, related quantitative and qualitative findings are brought together using visual elements such as figures, tables, matrices or graphs. According to Fetters et al. [35], the coherence of the quantitative and qualitative findings, known as the “fit” of data integration may be interpretated in three ways. The findings of the qualitative and quantitative findings may either validate each other (confirmation), contradict or disagree with each other (discordance) or both datasets may provide insights of the phenomenon of interest by addressing different aspects of the phenomenon (expansion) [35].

Operational definitions

The independent variables were the sociodemographic variables (age, sex, marital status, place of residence, educational status, employment status and monthly income), clinical variables (duration of condition, family history of diabetes and presence of comorbidities) and diabetes related-nutritional knowledge.

Diabetes-related nutritional knowledge refers to participant’s knowledge on recommended number of fruits and vegetables to be consumed daily, oils to cut down, foods to cut down a lot or increase intake, foods high or low in fibre, foods high or low in starch, foods high or low in added sugar, foods with high glycaemic index, foods likely to raise blood cholesterol and healthy food choices.

The dependent variable was dietary adherence. Using the PDAQ, participants were classified as having good dietary adherence if they obtained a score equal to or above the median score (the median score in this study was 37.00).

Quality assurance: Validity, reliability, and trustworthiness

The validity and reliability of the PDAQ and GNKQ-R have been established in previous studies [39, 40]. The PDAQ and GNKQ-R had Cronbach Alpha coefficients of 0.78 and 0.93 respectively in the previous studies. The PDAQ has been used in diabetes management research in Ghana [14] and Ethiopia [12, 13]. In the current study, the face and content validity of the data collection tool was established by a panel of six experts in diabetes management, nutrition, nursing, and instrument development. Following the validity test, both questionnaires were pretested among 20 persons with T2D who were receiving care (outpatient care) at the Kumasi South Hospital in the Asokwa Municipal, Ashanti Region, before the commencement of data collection. The Cronbach Alpha obtained for the PDAQ and GNKQ-R from the pre-test were 0.74 and 0.92 respectively. Following the pre-test, necessary amendments were made to the instructions of the questionnaire, contents, order of the questions, and grammatical issues based on findings from the pre-test.

The content of the interview guide was also evaluated by the six-member panel who reviewed the quantitative tools and pre-tested among 5 participants. The evaluation focused on the following areas: clarity of the content, appropriateness of wording, and terminology for the target population, grouping of questions, relevance of questions to study objectives and number of questions. The interview guide was then modified based on the comments of reviewers and findings from the pre-test.

Trustworthiness was ensured in the present study by applying the principles of credibility, conformability, dependability, and transferability [43, 44]. To ensure credibility, researcher triangulation was done during interviews and analysis as well as respondent validation after interviews. The detailed description of the research methods and findings also contributed to the dependability of the study. Additionally, transferability was established by providing a dense description of the demographic characteristics and geographic boundaries of the population in this study.

Ethical considerations

Before the conduct of the study, permission was sought from the management of the two hospitals. Ethical approval with reference number CHRPE/387/22 (initial approval) and CHRPE/AP/588/22(protocol amendment) was obtained from the Committee on Human Research, Publication and Ethics (CHRPE) of the School of Medicine and Dentistry, Kwame Nkrumah University of Science and Technology (KNUST), Kumasi, Ghana. Ethical principles such as informed consent, confidentiality, autonomy, beneficence, non-maleficence, justice, and voluntary participation were adhered to throughout the study.

Quantitative results

Sociodemographic characteristics and clinical data.

Between September and October 2022, a total of 142 persons participated in the survey out of 146 people who were approached yielding a response rate of 97.3%. Reasons given for non-participation included lack of time (n = 3) and disinterest in the study (n = 1). Most of the participants (80.3%, n = 114) were females, were living in urban areas (84.5%, n = 120), and received a monthly income of less than 500 Ghana Cedis (62%, n = 88). More than half of them were married (58.5%, n = 83) and not employed (57%, n = 82) and about seventy-seven percent (77.5%, n = 110) of them were educated.

Regarding clinical data, a greater number of the participants had comorbidities (83.1%, n = 118). About two-thirds (68.3%, n = 97) of them had a family history of diabetes and sixty-three percent (63.4%, n = 90) of the participants had T2D for at least five years (Table 1).

Download:

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

Table 1. Sociodemographic characteristics and clinical data of participants.

https://doi.org/10.1371/journal.pone.0302914.t001

Dietary adherence.

About half (50.7%) of the participants had good dietary adherence (obtained a score equal to or above the median score).

The highest median (IQR) score [5.00 (3.00)] was recorded for the question on the intake of fish or other foods high in omega 3 and the lowest median (IQR) score was recorded for the question on the intake of foods high in sugar [0.00 (0.00)] and foods high in fat [0.00 (1.00)] (Table 2).

Download:

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

Table 2. Perceived Dietary Adherence Questionnaire (PDAQ) score for participants.

https://doi.org/10.1371/journal.pone.0302914.t002

Association between dietary adherence and sociodemographic data, clinical data, and diabetes-related nutrition knowledge (DRNK).

Table 3 shows the results of the binary logistic regression. The results of the univariable logistic regression indicated that living in an urban area (COR = 3.259, 95% CI: 1.193–8.902, p = 0.021), being educated (COR = 2.377, 95% CI: 1.046–5.400, p = 0.039), receiving a monthly income more than or equal to 500 Ghana Cedis (COR = 2.539, 95% CI: 1.260–5.117, p = 0.009) and increase in DRNK (COR = 1.117, 95% CI: 1.027–1.216, p = 0.010) were associated with good dietary adherence.

Download:

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

Table 3. Association between dietary adherence and socio-demographic factors, clinical factors, and diabetes-related nutrition knowledge.

https://doi.org/10.1371/journal.pone.0302914.t003

In multivariable logistic regression, it was found that for every one-unit increase in DRNK score, the likelihood of a participant having good dietary adherence increased by approximately 10% (AOR = 1.101, 95% CI: 1.004–1.208, p = 0.041) (Table 3). Also, participants who were living in urban areas were about 3 times more likely to demonstrate good dietary adherence (AOR = 3.058, 95% CI: 1.014–9.221, p = 0.047).

Qualitative results

Sociodemographic characteristics of participants.

In all, 14 participants were purposively sampled from both hospitals. About half (57.1%, n = 8) of them were selected from the KNUST Hospital and were within the age range of 51–62 (50.0%, n = 7). Many of them were females (78.6%, n = 11), were married (78.6%, n = 11) and lived in urban areas (71.4%, n = 10). Fifty-seven percent (57.1%, n = 8) of them received a monthly income of less than 500 Cedis and about forty-two percent (42.9%, n = 6) were not formally educated. Nearly thirty-six percent (35.7%, n = 5) of them were not employed (Table 4).

Download:

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

Table 4. Sociodemographic characteristics of participants for the qualitative phase.

https://doi.org/10.1371/journal.pone.0302914.t004

Mixed methods results

Table 5 shows the integration of quantitative and qualitative results using the joint display method [35]. The quantitative results speak to the strength of the association between dietary adherence and independent variables such as place of residence, DRNK, educational status and monthly income while participant quotes from the qualitative data provide insight about the nature of the associations. However, in the case of marital status, the qualitative and quantitative findings are incongruous with each other. While the qualitative findings indicate that spousal support facilitates adherence, the results of the logistic regression indicate that no statistically significant association exists between marital status and dietary adherence.

Download:

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

Table 5. Mixed methods results using joint display.

https://doi.org/10.1371/journal.pone.0302914.t005

Implications for policy and healthcare delivery

The findings from this study imply that dietary adherence among persons with T2D is influenced by diabetes-related nutrition knowledge and multiple factors related to the individual and the environment in which they live and work. Given this, health providers and other stakeholders interested in controlling diabetes should use multidimensional and individualized interventions to improve dietary adherence among persons with T2D. Healthcare professionals should provide continuous dietary education to persons with T2D. Additionally, dietary recommendations should be tailored to individual preferences and dietary habits. The Government of Ghana should expand poverty-reduction social interventions such as the Livelihood Empowerment Against Poverty (LEAP) to assist persons with T2D who face financial challenges. Lastly, policymakers should make efforts to improve access to healthy foods in communities.

Limitations of the study

Patients who had a confirmed diagnosis less than one year, those who could not speak, understand, or write English/ Twi, and those who had sought care from the hospital less than twice in the past 12 months (using the study start date as reference) did not participate in this study. Additionally, critically ill patients and those with significant cognitive impairments were not included in this study. This may be a possible source of bias and may limit the generalization of findings to all persons living with T2D in Ghana. Also, the use of self-reporting in the evaluation of dietary adherence could have resulted in overestimation or underestimation due to social desirability bias. There is a tendency that the hospital environment influenced the participants to provide responses they felt were appropriate to investigators. Future research should focus on assessing dietary adherence outside the clinical setting. Despite these limitations, the findings of this study are relevant to improving dietary adherence among persons with T2D. Healthcare professionals should consider the individual preferences and situations of patients when providing care. Additionally, policymakers and other stakeholders should make efforts to develop policies and interventions that will ensure easy access to healthy foods and support patients facing financial difficulties.