Behavioural and social determinants of type 2 diabetes self-care adherence in a low-resource setting in South Ethiopia: a cross-sectional study using integrated health belief model and health-related quality-of-life frameworks

Abstract

Objectives

To investigate the behavioural and social determinants of type 2 diabetes mellitus (T2DM) self-care adherence in South Ethiopia using the integrated health belief model (HBM) and health-related quality-of-life (HRQoL) frameworks.

Design

A cross-sectional study.

Setting

Three public hospitals in South Ethiopia: Wolaita Sodo University Comprehensive Hospital, Humbo Primary Hospital and Boditi Primary Hospital.

Participants

404 systematically sampled adults aged 18–60 years with a confirmed diagnosis of T2DM who had been attending follow-up clinics for at least 12 months. Exclusion criteria included newly diagnosed T2DM, pregnancy, severe comorbidities or critical illness and unwillingness to participate.

Primary and secondary outcome measures

The primary outcome was adherence to diabetes self-care, assessed using the Summary of Diabetes Self-Care Activities scale across five domains: diet, physical activity, medication intake, blood glucose monitoring and foot care. Good adherence was defined as engagement in recommended behaviours on ≥50% of days per week. Secondary outcomes included socio-demographic factors, clinical variables, HBM constructs (perceived susceptibility, severity, benefits, barriers, cues to action and self-efficacy) and HRQoL domains (physical, psychological, social and environmental).

Results

Of 404 participants, 58.4% demonstrated good adherence. In multivariable analysis, insulin-only treatment (AOR=3.0; p<0.001), having comorbidities (AOR=2.02; p=0.007) and poor glycaemic control (AOR=3.6; p=0.003) were positively associated with adherence. Factors associated with poor adherence included low income (AOR=0.18; p=0.002), living alone (AOR=0.19; p=0.012), low self-efficacy (AOR=0.18; p<0.001) and poor psychological health (AOR=0.53; p=0.024).

Conclusion

The findings challenge the direct application of standard behavioural models in low-resource settings. Structured factors, such as poverty, can overwhelm psychological mechanisms. Effective interventions must integrate economic support with psychological care to improve self-care adherence.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study used a large, systematically sampled cohort of 404 participants from three public hospitals in a low-resource setting.

• The integration of the health belief model and health-related quality-of-life frameworks provides a comprehensive theoretical approach.

• Validated instruments (Summary of Diabetes Self-care Activities, WHO Quality of Life–based tools) with acceptable internal consistency (Cronbach’s alpha 0.756) were employed.

• The cross-sectional design precludes causal inference about the relationships between determinants and adherence.

• The reliance on self-reported data may introduce recall and social desirability bias, and facility-based sampling limits generalisability to non-healthcare-seeking populations and receiving care exclusively at primary health centres or private facilities.

Introduction

Type 2 diabetes mellitus (T2DM) accounts for >90% of global diabetes cases and poses a major public health challenge, particularly in low-income and middle-income countries (LMICs) where healthcare systems are often under-resourced.^{1 2} In sub-Saharan Africa (SSA), rising prevalence driven by urbanisation, dietary transitions and sedentary lifestyles is compounded by limited access to structured care and education.^{3 4} Effective T2DM management requires not only pharmacological treatment but also sustained adherence to self-care behaviours, including dietary modification, physical activity, medication compliance, glucose monitoring and foot care.^{5 6} However, studies across SSA consistently report poor adherence to these practices, leading to inadequate glycaemic control and increased complications.^{7 8}

In Ethiopia, the burden of T2DM increased rapidly, yet structured education and individual self-care support remain limited.^{9 10} Existing Ethiopian studies have primarily examined clinical and socio-demographic correlates of adherence, with less emphasis on psychological and behavioural factors such as health beliefs, self-efficacy and quality of life (QoL).^{11 12} This gap is critical because behavioural and psychological determinants significantly shape chronic diseases self-management, especially in low-resource contexts where health literacy and social support are variable.¹³

To address this gap, the Health Belief Model (HBM) was selected as the primary behavioural framework in this study. The HBM is one of the most widely used and empirically supported models for understanding why individuals engage in or avoid health behaviours, including medication adherence.¹⁴ It was specifically chosen over other models (eg, Theory of Planned Behaviour and Social Cognitive Theory) because it directly operationalises key perceptual factors that are modifiable through education and counselling, factors that are particularly relevant in low-resource settings where patients may have limited prior knowledge about diabetes.

The HBM proposes that adherence to health behaviour (such as taking diabetes medication or following a diet) is determined by six core constructs: perceived susceptibility (patient’s belief about likelihood of developing diabetes complications), perceived severity (patient’s belief about the seriousness of diabetes and its consequences), perceived benefits (the patient’s belief that taking recommended actions, such as medication adherence and blood glucose monitoring, will be effective), perceived barriers (the patient’s assessment of obstacles to performing the behaviour, such as medication costs, side effects or lack of transport), cues to action (internal or external triggers that prompt the behaviour) and self-efficacy.¹⁵

In this study, each HBM construct was applied to T2DM self-care by developing specific questionnaire items that measured, for example, perceived susceptibility (“How likely do you think you are to develop foot ulcers or lose vision because of your diabetes?”), perceived benefits (“How helpful is taking your diabetes medication every day for controlling your blood sugar?’), perceived barriers (“How difficult is it for you to afford your diabetes medication or to remember to take it?’) and self-efficacy (“How confident are you that you can follow your diet even when you are busy or stressed?”). By quantifying the constructs, this study aimed to identify which specific beliefs were most strongly associated with adherence or non-adherence in this population, thereby informing targeted behavioural interventions.

HBM alone, however, may not capture the broader impact of living with diabetes on a patient’s overall well-being and functional capacity. Therefore, the health-related QoL (HRQoL) framework, which assesses physical, psychological, social and environmental, was also incorporated.^{16 17} The HRQoL framework complements HBM by measuring the actual lived experience of diabetes, for instance, whether poor physical health or psychological distress (depression and anxiety) undermines a patient’s ability to act on their positive health benefits. The combined use of HBM and HRQoL remains rare in SSA, and their integration offers a more comprehensive understanding of adherence determinants.

Simply applying these models in new geographical settings is insufficient for generating novel insights. The significant novelty of this study lies in its critical investigation of these established frameworks within a context of profound resource scarcity. In settings like South Ethiopia, where patients face intense structural constraints such as food insecurity, poverty and limited healthcare access,^{17 18} the conventional pathways proposed by HBM and HRQoL models may be fundamentally altered or overwhelmed. For instance, the perceived benefits of healthy diet may be rendered irrelevant by food insecurity, or high self-efficacy may be eroded by persistent financial barriers.

Therefore, this study moves beyond simple application to a critical test of the boundaries and applicability of these behavioural models under conditions of extreme socio-economic constraints. Specifically, the aim of this study was to (a) critically investigate the behavioural and social determinants of T2DM self-care adherence in South Ethiopia using an integrated HBM and HRQoL framework, (b) examine the applicability of these models in a context of profound resource scarcity and identify the predominant factors influencing adherence.

Methods and materials

Study design and setting

A facility-based cross-sectional analytic study was conducted from 1 November to 30 December 2024, across three public hospitals in Wolaita Zone in South Ethiopia: Wolaita Sodo University Comprehensive Specialised Hospital, Humbo Primary Hospital and Boditi Primary Hospital in Wolaita Zone. These hospitals serve both urban and rural populations and are primary sites for chronic disease management and follow-up, including T2DM.

Study population and eligibility

The study population comprised adults aged 18–60 years with a confirmed diagnosis of T2DM who had been receiving follow-up at diabetes clinics for at least 12 months. Patients with newly diagnosed T2DM, pregnant women, those with severe comorbidities or critical illness and individuals unwilling to participate were excluded.

Sample size and sampling technique

The sample size was determined using a single population proportion formula. Owing to the absence of previous local studies on self-care adherence among patients with T2DM in the study area, a 50% prevalence of poor self-care practices was assumed, which maximises the required sample size and provides adequate power to detect associations. With a 95% confidence level and 5% margin of error, the calculation yielded a sample size of 384. An additional 5% was added to account for potential non-response, resulting in a final sample size of 404.

Operational definition

Diabetes self-care adherence

Diabetes self-care adherence was measured using the Summary of Diabetes Self-care Activities (SDSCA) scale.¹⁹ The SDSCA assesses the frequency (in days per week) of engagement in key self-care behaviours such as diet, exercise, medication use, blood glucose testing and foot care over the previous 7 days. The threshold of ≥3 days/week (≥50) for ‘good adherence’ was selected based on established clinical practices where engagement in self-care behaviours at least half of the days per week is considered to confer therapeutic benefit.^{20 21} This threshold aligns with previous studies using the SDSCA scale in similar populations, which have employed comparable cutpoints to distinguish adherent from non-adherent patients^{17 22} and provides a clinically meaningful distinction for analysing determinants of adherence.

Health Belief Model constructs

HBM constructs were measured using a structured questionnaire adapted from validated tools and literature on diabetes self-care.^{19 21} Items were scored on a 5-point Likert-scale and categorised into three levels as low (1.0–2.0), moderate (2.1–3.0) and high (3.1–5.0).

Health-related quality of life

HRQoL was measured across physical, psychological, environmental and social domains using WHO Quality of Life (WHOQoL)-based tools especially in LMICs.²³ Scores <3 were categorised as low and ≥3 as good.

Variables

Dependent variable

Adherence to diabetes self-care (based on SDSCA scores, adherent vs non-adherent).

Independent variables

Socio-demographic: age, sex, marital status, education, occupation, monthly income, residence, healthcare access and satisfaction with healthcare service and family structure. While a dedicated, validated socio-economic scale was not employed, these multiple indicators provide a comprehensive assessment of participants’ economic circumstances.

Clinical: duration of diabetes, presence of comorbidities (hypertension, etc), treatment regimen (oral hypoglycaemic agents, insulin and combination therapy).

Behavioural and psychosocial: HBM constructs (perceived susceptibility, severity, benefits, barriers, cues to action and self-efficacy).

HRQoL: This domain includes physical health, social health and psychological health.

Quality assurance

The SDSCA scale has been previously validated in multiple settings and populations, demonstrating good psychometric properties with Cronbach’s alpha values ranging from 0.70 to 0.85 across different self-care domains. The HBM questionnaire was adapted from validated tools used in previous diabetes research in LMICs, with content validity assessed by three public health experts. The WHOQoL-based questionnaire for HRQoL assessment has been validated cross-culturally including in African settings. Internal consistency of the combined questionnaire in our study population was confirmed with Cronbach’s alpha of 0.756, indicating acceptable reliability. A pilot study was conducted prior to the main data collection. The questionnaire was pretested on 30 patients with T2DM (approximately 7% of the final sample size) from non-study sites. The pilot testing assessed face validity, clarity of questions, cultural appropriateness and average completion time. Minor wording adjustments were made based on pilot feedback to improve comprehensibility. Pilot participants were excluded from the main study sample.

Data collection

Participants recruitment was conducted at each hospital’s diabetes follow-up clinic. On each data collection day, the sampling interval (k=4) was applied to the daily patient list. When a patient was selected, a trained data collector approached the patient after their clinical consultation. Data collectors were three final-year public health students who received 1-day training on the study objectives, ethical considerations, interviewing techniques and standardised data collection procedures. Patients were invited to a private room within the clinic to complete the interviewer-administered questionnaire. For patients who have difficulty reading or understanding certain questions, data collectors read the questions aloud and provide clarification as needed, without leading or influencing responses. Each data collection session lasted approximately 25–35 min. Supervisors (two M.Sc.-level public health professionals) were present at each site to address any logistical issues and ensure quality control. Water and a comfortable waiting area were provided for participants before and after the interview.

All questionnaires were administered in the local language, Amharic (the most widely spoken language in South Ethiopia), to ensure comprehensibility for all participants. The original English versions of SDSCA, HBM questionnaire and WHOQoL-based tools were translated into Amharic by a professional translator fluent in both languages and then back-translated into English by an independent translator to verify accuracy. Discrepancies were resolved through discussion with the research team.

Clinical variables including duration of diabetes, type of diabetes (T2DM confirmation) and treatment regimen and presence of comorbidities were abstracted from participants’ medical records at the respective hospitals by trained data collectors, not self-reported. This approach was taken specifically to enhance data validity and reduce recall bias. For the few cases in which medical records were incomplete (n=12, 3.0%), patients’ self-reports were supplemented with physician verification when possible.

Data analysis

Data were entered and analysed using SPSS. Descriptive statistics such as means, SD, frequencies and percentages were computed. Variables with p<0.25 in the univariate logistic regression were considered candidates for multivariable logistic regression analysis. This threshold was chosen based on the recommendations by Hosmer and Lemeshow and Bursac et al to avoid excluding variables that may become significant after adjusting for confounders.^{24 25} These variables were entered into a multivariable binary logistic regression model to identify independent predictors of self-care adherence. Statistical significance was set at p<0.05, and adjusted ORs (AORs) with 95% CIs were reported as justified in epidemiological and clinical studies.

Multicollinearity was assessed using variance inflation factors (VIFs) and the correlation matrix. No predictor variable exceeded the VIF threshold of 10 or had a correlation coefficient >0.80, suggesting that multicollinearity was not a concern. Residual analysis showed that predicted values ranged from 0.31 to 1.08, and standard residuals fell to ±3, indicating the absence of major outliers. The model showed strong discriminative ability with an area under the receiver operating characteristics (ROC) curve (AUC) >0.90, indicating excellent capability to distinguish between adherent and non-adherent patients.

STROBE compliance

This manuscript was prepared following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies.²⁶ A STROBE flow diagram is provided as online supplemental figure S1, and a completed STROBE checklist is included in the online supplemental file 1.

Results

Participant characteristics

Of 404 eligible patients approached during the study period, all 404 (100%) agreed to participate and completed the full questionnaire, yielding a 100% response and completion rate. A STROBE flow diagram summarising participant selection is provided in the online supplemental figure S1. No eligible patients refused participation.

The sample was balanced by sex (51.75% male). Most participants were aged >40 years (76%), married (80.7%) and urban residents (79.0%). Approximately 30% earned <3000 Ethiopian Birr (ETB) per month, whereas 9.4% earned >10 000 ETB. Nearly half (53%) reported food insecurity, and 10% lived alone.

A large proportion (74.8%) reported comorbidities and complications, with neuropathy (30.0%), visual problems (29.5%) and hypertension (29.0%) being the most common. Kidney and heart diseases were each reported by 10.5% of participants. Only 16.1% of participants achieved their haemoglobin A1c (HbA1c)<7% (table 1).

Table 1. Socio-demographic characteristics of participants of Wolaita Zone, South Ethiopia.

Variable	Category	Frequency (%)
Sex	Male	209 (51.7%)
	Female	195 (48.3%)
Age (years)	<40	97 (24%)
	40–50	151 (37.4%)
	>50	156 (38.5%)
Marital status	Single	30 (7.4%)
	Married	326 (80.7%)
	Divorced	24 (5.9%)
	Widowed	24 (5.9%)
Educational level	No formal education	58 (14.4%)
	Primary education	93 (23.0%)
	Secondary education	97 (24%)
	Diploma	52 (15.3%)
	Degree and above	94 (23.3%)
Occupation	Housewife	92 (22.8%)
	Merchant	101 (25.0%)
	Government employed	110 (27.2%)
	NGO workers	45 (11.1%)
	Student	22 (5.4%)
	Farmer	15 (3.7%)
	Other	22 (5.4)
Residence status	Urban	319 (79.0%)
	Rural	85 (21.0%)
Food security	Secured	190 (47.0%)
	Not secured	214 (53.0%)
Family structure	Living alone	41 (10.0%)
	With nuclear family	341 (84.6%)
	With extended family	22 (5.4%)
Monthly income, ETB	Low (<3000)	121 (30.0%)
	Middle (3000–5000)	108 (26.7%)
	Upper middle (5000–10000)	137 (33.9%)
	High (>10 000)	38 (9.4%)
Distance to health facility	Yes	304 (75%)
	No	100 (25%)
Satisfaction with healthcare services	Satisfied	226 (55.9%)
	Not satisfied	178 (44.1%)
Years lived with diabetes, years	<5	172 (42.6%)
	5–10	164 (40.6%)
	>10	68 (16.8%)
Blood pressure status	Normal (<120/80)	127 (31.4%)
	Elevated (120–129/80)	14 (3.5%)
	Hypertension stage 1 (130–139/80-89)	139 (33.9%)
	Hypertension stage 2 and higher (>140/90)	126 (31.2%)
Type of medication patients take	OHAs	257 (63.6%)
	Insulin	93 (23.0%)
	Both	54 (13.4%)
Comorbidities	Yes	302 (74.8%)
	No	102 (25.2%)
Comorbidities and complications	Hypertension	90 (29.0%)
	Neuropathy	91 (30.0%)
	Visual problem	89 (29.5%)
	Kidney disease	32 (10.5%)
	Heart disease	32 (10.5%)
FBS (mg/dL)	Controlled (≤130)	38 (9.4%)
	Moderate (131–180)	196 (48.5%)
	Poor (>180)	170 (42.1%)
Glycaemic control (HbA1c %)	Good (<7%)	65 (16.1%)
	Moderate (7.0%–8.0%)	167 (41.3%)
	Poor (>8.0%)	172 (42.6%)

Note: Participants with <1 year duration were excluded per inclusion criteria.

ETB, Ethiopian birr; HbA1c, haemoglobin A1c; OHAs, oral hypoglycemic agents.

Adherence to self-care practices

The self-care practices assessed included five domains: diet adherence, physical activity, medication intake, foot care and blood glucose monitoring (table 2). Among these, medication intake had the highest rate of adherence, with 283 participants (70.0%) reporting high adherence (>5 days per week). Foot care and physical activity also showed relatively favourable patterns, with 192 (47.5%) and 196 (48.5%) reporting high adherence, respectively.

Table 2. Self-reported frequencies of weekly diabetes self-care practices among participants of Wolaita Zone, South Ethiopia.

Variable	Category	Frequency (%)
Diet adherence, days	Poor adherence (0–2)	301 (74.5%)
	Moderate adherence (3–4)	40 (10.0%)
	High adherence (≥5)	63 (15.5%)
Physical activity, days	Poor adherence (0–2)	116 (28.7%)
	Moderate adherence (3–4)	92 (22.8%)
	High adherence (≥5)	196 (48.5%)
Blood glucose test, days	Poor adherence (0–2)	290 (71.8%)
	Moderate and high adherence (≥3)	114 (28.2%)
Medication intake, days	Poor adherence (0–2)	83 (20.5%)
	Moderate adherence (3–4)	38 (9.4%)
	High adherence (≥5)	283 (70%)
Foot care practice, days	Poor adherence (0–2)	128 (31.7%)
	Moderate adherence (3–4)	84 (20.8%)
	High adherence (≥5)	192 (47.5%)
Overall adherence	Poor (<50%)	168 (41.6%)
	Good (≥50%)	236 (58.4%)

In contrast, dietary adherence was notably poor, with only 63 participants (15.5%) reporting high adherence and 301 (74.5%) reporting poor adherence (0–2 days per week). Blood glucose monitoring was the most neglected domain, with 290 participants (71.8%) reporting poor adherence and only 114 (28.2%) testing three or more times per week. These findings reveal specific behavioural targets for intervention, particularly diet and self-blood glucose monitoring, which are critical to maintaining glycaemic control and preventing complications.

Overall, using the predefined threshold of ≥3 days/week (≥ 50%) adherence to each domain of all practices, 236 participants (58.4%) demonstrated good adherence, while 168 (41.6%) had poor adherence. The substantial proportion of poor adherence indicates a concerning gap that signals increased risks for poor clinical outcomes.

Behavioural beliefs of participants

Based on key constructs of the HBM, the findings reveal prevailing beliefs and attitudes that significantly influence self-care practices and health-seeking behaviours.

Perceived susceptibility

A strong majority of participants (74.0%) reported high perceived susceptibility to the complications of diabetes, indicating a broad awareness of personal risk. A smaller proportion (21.0%) expressed moderate susceptibility.

Perceived severity

More than half (59.4%) of the participants had a high perception of the seriousness of diabetes and its consequences. Moderate severity was noted by 27.2%, and 13.0% perceived the conditions as less severe. This reflects good awareness of diabetes-related health risks.

Perceived benefits

A considerable proportion (71.3%) recognised the high benefits of engaging in appropriate diabetes management behaviours such as medication adherence, dietary modification, physical activity, foot care and glucose monitoring. This suggests a positive orientation toward proactive health behaviours.

Perceived barriers

Despite the high perceived benefits, 64.4% reported high perceived barriers to effective diabetes self-care. These barriers could include financial constraints, limited access to services, cultural factors and personal obstacles. Moderate barriers were reported by 33.4%, indicating the majority of patients experiencing challenges to effective self-care.

Cues to action

More than half (57.7%) of the participants experienced high cues to action, such as reminders from healthcare providers, family support or peers encouraging them to manage their diabetes. Moderate cues were reported by 34.7%.

Self-efficacy

Confidence in one’s ability to manage diabetes varied across the sample. While 39.6% participants demonstrated high self-efficacy, the largest proportion (42.8%) had a moderate level in managing their condition (table 3).

Table 3. Behavioural beliefs of participants of Wolaita Zone, South Ethiopia.

Variable	Category	Frequency (%)
Perceived susceptibility	Low	20 (5.0%)
	Moderate	85 (21.0%)
	High	299 (74.0%)
Perceived severity	Low	54 (13.0%)
	Moderate	110 (27.2%)
	High	240 (59.4%)
Perceived benefit	Low and moderate	116 (28.7%)
	High	288 (71.3%)
Perceived barriers	Low and moderate	144 (35.6%)
	High	260 (64.4%)
Perceived cues to action	Low	31 (7.7%)
	Moderate	140 (34.7%)
	High	233 (57.7%)
Self-efficacy	Low	71 (17.6%)
	Moderate	173 (42.8%)
	High	160 (39.6%)

Health-related quality of life

Four domains of HRQoL were also assessed: physical, environmental, social and psychological health.

Physical health

Approximately half of participants 48.8% reported low physical health, indicating many patients experience functional limitations such as pain, reduced mobility or dependence on medical care.

Environmental health

A majority (56.2%) of the participants rated their environmental health as poor, suggesting barriers related to accessing resources, financial challenges and living conditions that may hinder effective diabetes self-care.

Social health

Approximately 43.6% of participants reported low social health, indicating that a significant number of patients with T2DM may suffer from social isolation, weak interpersonal relationships and lack of support from friends, family members and community.

Psychological health

This domain was most affected; 70.3% reported poor psychological health, including frequent negative feelings such as depression, anxiety, despair and feelings of inadequacy, which may severely impair self-care behaviours (table 4).

Table 4. Health-related quality-of-life domains of participants in Wolaita Zone, South Ethiopia.

Variable	Category	Frequency (%)
Physical health domain	Low	197 (48.8%)
	Good	207 (51.2%)
Environmental health domain	Low	227 (56.2%)
	Good	177 (43.8%)
Social health domain	Low	176 (43.6%)
	Good	228 (56.4%)
Psychological health domain	Low	284 (70.3%)
	Good	120 (29.7%)

Factors associated with diabetes self-care adherence among participants

Socio-demographic and clinical predictors of self-care adherence

In multivariable logistic regression analysis, several socio-demographic factors were independently associated with adherence to self-care practices among individuals with diabetes. The model included monthly income, family structure, food security status, distance from home to the health facility and satisfaction with healthcare services. Monthly income was a significant predictor of adherence to diabetes self-care practices. Compared with the highest income group, the lowest income group had 82.0% lower odds of adherence (AOR=0.18, 95% CI 0.06 to 0.53, p=0.002). The middle-income group hanond 67.3% lower odds of adhering to self-care (AOR=0.33, 95% CI 0.12 to 0.88, p=0.028). This finding suggests that financial constraints may limit access to healthy food, medication and follow-up care.

Individuals living alone were 81.0% less likely to be adherent than those living with family members (AOR=0.19, 95% CI 0.05 to 0.69, p=0.012), indicating that social support from family members may facilitate better adherence to self-care practices. Participants dissatisfied with healthcare services had a 43% lower adherence (AOR=0.57, 95% CI 0.36 to 0.91, p=0.018), highlighting the importance of quality of care and healthcare provider–patient interaction.

Regarding the relationship between clinical factors and diabetes self-care adherence, the model included variables such as duration of diabetes, comorbidity status, medication regimen, fasting blood sugar and HbA1c. Patients who had lived with diabetes <5 years were 66.0% less likely to be adherent (AOR=0.34, 95% CI 0.16 to –0.70, p=0.003). This suggests that longer disease duration allows more time for patients to learn and adapt to self-care routines. Compared with patients on both oral hypoglycaemic agents (OHAs) and insulin, those on insulin were three times more likely to be adherent (AOR=3.0, 95% CI 1.70 to 5.3, p<0.001). This may reflect greater disease severity perception among insulin users.

Participants with poor glycaemic control (HbA1>8%) had 3.6 times higher odds of adherence (AOR=3.6, 95% CI 1.57 to 8.44, p=0.003). Similarly, participants with moderate glycaemic control were 3.2 times more likely to be adherent (AOR=3.2, 95% CI 1.1 to 9.23, p=0.034). Interestingly, this poor glycaemic control was associated with increased adherence, possibly indicating reactive behavioural changes due to disease progression or clinical advice. Participants with comorbid conditions were two times as likely to be adherent (AOR=2.02, 95% CI 1.20 to 3.38, p=0.007), possibly due to more healthcare contact (table 5).

Table 5. Socio-demographic and clinical predictors of diabetes self-care adherence of Wolaita Zone, South Ethiopia.

Variable	Category	Adherence (%)	Non-adherence (%)	AOR (95% CI	P value
Monthly income, ETB	Lowest (<3000)	23 (28.4)	58 (71.6)	0.18 (0.06 to 0.52)	0.002*
	Middle (3000–5000)	42 (44.2)	53 (55.8)	0.33 (0.12 to 0.88)	0.028*
	Upper middle (5000–10000)	51 (59.3)	35 (40.7)	0.70 (0.28 to 1.73)	0.45
	High (>10 000)	120 (70.2)	22 (29.8)	1
Family structure	Living alone	12 (28.6)	30 (71.4)	0.19 (0.05 to 0.69)	0.012*
	Nuclear family	88 (56.8)	67 (43.2)	0.97 (0.46 to 2.01)	0.928
	Extended family	135 (67.5)	65 (32.5)	1
Satisfaction with healthcare services	Satisfied (yes)	169 (67.6)	81 (32.4)	1
	Not satisfied	67 (47.2)	87 (52.8)	0.57 (0.36 to 0.90)	0.018*
Distance to health facility	Near (<1 hour)	165 (62.5)	99 (37.5)	1
	Far (>1 hour)	71 (52.6)	64 (47.4)	0.62 (0.36 to 1.04)	0.071
Food security	Secured	132 (60.3)	87 (39.7)	1
	Not secured	104 (56.5%)	80 (43.5)	1.44 (0.77 to 2.71)	0.249
Years lived with diabetes, years	<5	30 (32.3)	63 (67.7)	0.34 (0.16 to 0.07)	0.003*
	5–10	61 (46.2)	71 (53.8)	0.36 (0.18 to 0.74)	0.006*
	>10	145 (75.5)	34 (24.5)	1
Medication regimen patients take	OHAs	180 (59.0)	75 (41.0)	1.10 (0.55 to 2.02)	0.777
	Insulin	87 (80.6)	21 (19.4)	3.0 (1.70 to 5.30)	0.001*
	Both	41 (45.1)	50 (54.9)	1
Comorbidities	Yes	127 (70.2)	54 (29.8)	2.02 (1.20 to 3.38)	0.007*
	No	109 (48.2)	117 (51.8)	1
HbA1c (%)	<7% (good	40 (66.7)	20 (33.3)	1
	7%–8% (moderate)	48 (77.4)	14 (22.6)	3.20 (1.1 to 9.23)	0.034
	>8.0% (poor)	62 (83.8)	12 (16.2)	3.60 (1.57 to 8.44)	0.003*

^*Statistically significant at p<0.05.

FBS, fasting blood sugar; HbA1c, haemoglobin A1c; OHAs, oral hypoglycaemic agents.

Behavioural beliefs and health-related quality of life predictors

Multivariate analysis showed that several HBM constructs and HRQoL domains were significantly associated with diabetes self-care adherence (table 6).

Table 6. Behavioural beliefs and health-related quality-of-life predictors and diabetes self-care adherence in Wolaita Zone, South Ethiopia.

Variable	Category	Adherence (%)	Non-adherence (%)	AOR (95% CI	P value
Perceive susceptibility	Low	12 (3.0)	8 (2.0)	5.10 (1.2 to 23.1)	0.035*
	Moderate	50 (12.4)	35 (8.7)	0.83 (0.39 to 1.75)	0.620
	High	174 (43.1)	125 (30.9)	1
Perceived severity	Low	29 (7.2)	25 (6.2)	2.11 (0.83 to 5.34)	0.117
	Moderate	65 (16.1)	45 (11.1)	0.85 (0.42 to 1.74)	0.663
	High	180 (44.6)	60 (14.9)	1
Perceived benefit	Low and moderate	61 (15.1)	55 (13.6)	0.08 (0.04 to 0.18)	0.001*
	High	175 (43.3)	113 (28.0)	1
Perceived cues to action	Low	14 (3.5)	17 (4.2)	0.37 (0.08 to 1.57)	0.183
	Moderate	102 (25.2)	38 (9.4)	0.23 (0.12 to 0.41)	0.001*
	High	107 (42.1)	63 (15.6)	1
Self-efficacy	Low	24 (5.9)	47 (11.6)	0.18 (0.07 to 0.46)	0.001*
	Moderate	95 (23.5)	78 (19.4)	0.27 (0.15 to 0.59)	0.001*
	High	117 (29.1)	43 (10.5)	1
HRQoL-Physical health	Low	82 (50.9)	79 (49.1)	0.51 (0.28 to 0.92)	0.027*
	Good	154 (64.4)	85 (35.6)	1
HRQoL-Social health	Low	74 (40.0)	111 (60.0)	0.29 (0.17 to 0.50)	0.001*
	Good	162 (71.7)	64 (28.3)	1
HRQoL-Psychological health	Low	78 (50.0)	78 (50.0)	0.53 (0.31 to 0.92)	0.024*
	Good	158 (66.1)	81 (33.9)	1

^*Statistically significant at p<0.05.

AOR, adjusted OR; HRQoL, health-related quality of life.

Perceived susceptibility

Participants with low perceived susceptibility had significantly higher odds of adherence than those with high perceived susceptibility (AOR=5.10, 95% CI 1.12 to 23.10, p=0.035). This may suggest reverse causation; adherent individuals may feel less vulnerable due to their active self-care.

Perceived benefits

Low or moderate perceived benefits were strongly associated with non-adherence (AOR=0.08, 95% CI 0.04 to 0.18, p<0.001).

Cues to action

Participants with moderate cues to action were 78.0% less likely to be adherent than with those with high cues to action (AOR=0.23, 95% CI 0.12 to 0.41, p<001). This underscores the importance of frequent reminders, health education and peer support.

Self-efficacy

Both low self-efficacy (AOR=0.18, 95% CI 0.07 to 0.46, p<0.001) and moderate (AOR=0.27, 95% CI 0.15 to 0.59, p<0.001) self-efficacy levels were associated with substantially lower adherence than high self-efficacy, making it the strongest behavioural predictor. This underscores the importance of frequent reminders, health education and peer support. Perceived barriers and perceived severity were not significant in the adjusted model (table 6).

Health-related quality-of-life domains

Low social health (AOR=0.29, 95% CI 0.17 to 0.50, p<0.001), low physical health (AOR=0.51, 95% CI 0.28 to 0.92, p=0.027) and low psychological health (AOR=0.53, 95% CI 0.31 to 0.92, p=0.024) were all significantly associated with poor adherence. Low or moderate self-efficacy and low perceived benefits were the strongest behavioural predictors, whereas low social and psychological health domains were key quality-of-life correlates (table 6).

Discussion

This study examined the determinants of diabetes self-care adherence by integrating the HBM and HRQoL framework among 404 adults with T2DM in resource-limited settings in South Ethiopia. The main findings reveal that adherence is suboptimal, with only 58.4% of participants demonstrating good adherence. More importantly, our results show that low income (AOR=0.18), living alone (AOR=0.19), low self-efficacy (AOR=0.18) and poor psychological health (AOR=0.53) were the strongest independent predictors of poor adherence. Conversely, insulin-only treatment (AOR=3.0), having comorbidities (AOR=2.0) and poor glycaemic control (AOR=3.6) were associated with higher adherence, suggesting reactive behavioural patterns.

The potent influence of economic hardship and social isolation in this study underscores the primary limitation of applying behavioural models without contextual adaptation. The strong association between low income and poor adherence, even after adjusting for psychological factors, indicates that financial barriers can act as a primary, overwhelming determinant.^{13 21} This finding challenges models that predominantly emphasise individual cognition by suggesting that in settings of extreme scarcity, structural factors may establish the boundary conditions within which psychological beliefs can operate. Similarly, the significantly lower adherence among individuals living alone highlights that in low-resource settings, social support is not merely a facilitator but may be a necessary prerequisite for sustaining self-care behaviours, providing both practical assistance and emotional reinforcement that compensates for systemic gaps.^{27 28}

Our findings also confirm established clinical predictors, such as diabetes duration, insulin-only therapy and presence of comorbidities, which align with studies conducted in different parts of the world.^{29 30} This may be because with longer duration of diabetes, patients are more likely to develop coping mechanisms, and disease experience, perceived severity and healthcare engagement may encourage better self-care. Interestingly, poor glycaemic control was associated with increased adherence. This may reflect a delayed response to deteriorating health, in which patients begin adhering to recommended behaviours only after experiencing complications or repeated clinical warnings.

Our analysis of HBM constructs yielded nuanced findings that contribute to theory refinement. The emergence of self-efficacy as the strongest behavioural predictor aligns with Bandura’s¹⁵ theory and underscores its universal importance. However, the fact that both low and moderate self-efficacy were strongly associated with non-adherence suggests that a threshold effect may be at play in challenging environments; a moderate level of confidence, which might be sufficient in a high-support setting, is inadequate when faced with profound structural barriers. Furthermore, the counter-intuitive finding that poor glycaemic control was associated with higher adherence is a pivotal result. Contrary to results of some previous studies in Ethiopia,³¹ this may indicate a ‘reactive adherence’ pattern, in which the experience of complications or strong clinical warnings (heightened perceived severity) finally triggers behavioural change. This suggests that the motivational pathways of HBM may be delayed or activated only by tangible health threats in this population rather than by preventive education alone. This presents a crucial boundary condition for the HBM: the model’s predictive power may depend on the stage of disease and the patient’s direct experience with negative consequences, especially in contexts where preventive healthcare is less accessible.

The profound impact of poor quality of life, particularly in the social and psychological domains, further elucidates the mechanism through which context affects behaviour. The high prevalence of poor psychological health (70.3%) indicates that depression, anxiety and despair are normative experiences that severely impair the cognitive and motivational resources required for daily self-management.^{32 33} This finding moves beyond merely correlating HRQoL with adherence; it suggests that the high burden of psychological distress is a central pathway through which the challenges of living with diabetes in low-resource settings translate into poor self-care. Therefore, psychological health is not just an outcome but a critical mediator of adherence.

This study did not collect data on body mass index, which is an important clinical parameter associated with diabetes self-care. Additionally, while the presence of comorbidities (including neuropathy, visual problems, kidney disease and heart diseases) was evaluated, complications were not systematically classified as microvascular versus macrovascular, which may have provided more nuanced insights. Future studies should include these clinical measurements.

Theoretical implications: refining behavioural models in low-resource settings

From a theoretical perspective, the proposed integrated HBM–HRQoL approach demonstrates that in low-resource settings, the relationship between beliefs, quality of life and behaviour is not linear but interactive and constrained by structure. The model requires refinement to account for the fact that structural deprivation can (a) overwhelm perceived benefits and barriers, (b) erode self-efficacy and (c) dictate that perceived severity only becomes a potent motivator after complications occur.

Practical implications and recommendations

Practically, these findings demand a shift from generic patient education to integrated, multi-level intervention. Recommendations must be specific and mechanistic:

1. Economic integration: diabetes programmes should explore integrating economic empowerment components such as micro-savings for medical supplies to directly alleviate the structural barriers that undermine psychological constructs like self-efficacy.

2. Systematic psychosocial support: routine screening and management of psychological distress must be embedded within diabetes care. Task-shifting to trained community health workers could make this feasible.

3. Staged messaging: health education should be staged; for newly diagnosed patients, focus on building self-efficacy and emphasising benefits; for patients with complications, leveraging their experienced severity to reinforce adherence.

4. Family-centred care: interventions must actively engage family members as co-managers of care, providing training and support to enhance the practical and emotional backing they offer.

Conclusion

This study demonstrates that diabetes self-care adherence in South Ethiopia is influenced by a complex interplay in which socio-economic constraints dictate the operational boundaries of psychological and behavioural determinants. The findings argue for a reconceptualisation of behavioural models to better account for structural deprivation. Moving forward, interventions that synergistically address economic barriers, psychological distress and social support are not just beneficial but essential to improving diabetes outcomes in Ethiopia and similar LMICs

Data availability statement

Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.