Factors associated with poor glycaemic control among adults with type 2 diabetes mellitus in Albania: a cross-sectional study

Abstract

Background

Poor glycaemic control is common among adults with type 2 diabetes mellitus (T2DM) in low- and middle-income settings in South-Eastern Europe, yet patient-level data on its correlates from Albania are scarce. We examined the demographic, clinical and treatment-related factors associated with poor glycaemic control (HbA1c ≥ 7%) in Albanian adults with T2DM.

Methods

Cross-sectional study of 176 consecutive adults with T2DM attending a hospital-based outpatient diabetes service in northern Albania. Poor glycaemic control was defined a priori as HbA1c ≥ 7% (53 mmol/mol). Sociodemographic characteristics, diabetes duration, insulin therapy and physician-confirmed comorbidities were recorded. Associations were assessed using univariate and multivariable logistic regression, estimating odds ratios (OR) with 95% confidence intervals (CI); comorbidity was modelled both as individual conditions (primary model) and as a comorbidity-burden index (sensitivity model). Model calibration and discrimination were evaluated using the Hosmer–Lemeshow test and the area under the ROC curve.

Results

Mean age was 65 years, 51% were men and mean diabetes duration was 11 years. The prevalence of poor glycaemic control was 58.5% (95% CI 50.9–65.9%), and 54% were receiving insulin. In univariate analyses, poor control was more frequent with insulin therapy (OR 5.64, 95% CI 2.93–10.87), longer diabetes duration (OR 1.42 per 5 years, 1.09–1.85) and ischaemic heart disease (OR 2.10, 1.09–4.08). After mutual adjustment, insulin therapy was the only factor independently associated with poor control (adjusted OR 6.32, 95% CI 2.71–14.71); age, diabetes duration, hypertension, ischaemic heart disease and multimorbidity were not. Results were unchanged in the sensitivity model. The primary model was well calibrated (Hosmer–Lemeshow p = 0.96), with moderate discrimination (area under the ROC curve 0.74).

Conclusions

In this Albanian outpatient population, poor glycaemic control was common and insulin therapy was the only factor independently associated with it. This reflects confounding by indication rather than harm from insulin: insulin is initiated when oral therapy no longer maintains targets, so insulin-treated patients form a clinically identifiable subgroup with more advanced disease who warrant priority for structured treatment intensification and individualised follow-up. Prospective multicentre studies are needed to confirm these associations and their temporal direction.

Trial registration

Not applicable (observational, non-interventional study).

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12902-026-02378-9.

Background

Despite a steadily expanding range of glucose-lowering therapies, a large proportion of adults with type 2 diabetes mellitus (T2DM)- typically 40–60% in international surveys- do not achieve the recommended glycated haemoglobin (HbA1c) target of below 7% (53 mmol/mol) [1–4]. Which patients remain above target, and why, depends on the local mix of treatment regimens, comorbidities and access to care, yet patient-level evidence on the correlates of glycaemic control from South-Eastern Europe is limited. Published Albanian work on T2DM has concentrated on prevalence, hospitalisation and mortality, with little characterisation of the clinical and treatment-related factors associated with control at the individual level [5, 6].

Two features of routine practice make such an analysis informative. First, longer diabetes duration and insulin therapy are consistently associated with higher HbA1c, but the insulin–HbA1c relationship largely reflects confounding by indication — insulin is reserved for patients whose disease is no longer controlled on oral agents — rather than a harmful effect of insulin itself [7–10]. Second, multimorbidity is an increasingly cited candidate correlate whose independent contribution, once age, duration and treatment intensity are accounted for, remains uncertain [11–13]. Against this background, we conducted a cross-sectional study of consecutive adults with T2DM attending an outpatient diabetes service in Albania, to estimate the prevalence of poor glycaemic control (HbA1c ≥ 7%) and to identify the demographic, clinical and treatment-related factors associated with it using univariate and multivariable logistic regression.

Methods

Study design and setting

This was a single-centre, cross-sectional observational study conducted in the diabetes outpatient service of the endocrinology unit at the Regional Hospital of Shkodra (a publicly funded hospital) in northern Albania. Recruitment took place over a six-month period among consecutive adults attending routine follow-up consultations. The study followed the principles of the Declaration of Helsinki and the STROBE reporting recommendations for observational studies [14]. According to the International Diabetes Federation Diabetes Atlas (2024), approximately 261,400 adults are living with diabetes in Albania, corresponding to a national prevalence of about 12.4%.

Participants

Eligible individuals were adults aged ≥ 18 years with a physician-confirmed diagnosis of T2DM of at least six months’ duration. Exclusion criteria were type 1 or other specific types of diabetes; gestational diabetes; severe cognitive impairment, active psychosis or any other condition precluding informed consent or self-completion of a structured questionnaire; active malignancy under treatment; and end-stage organ failure judged to dominate the clinical picture independently of diabetes (chronic dialysis, decompensated cirrhosis, NYHA class IV heart failure).

Of 320 consecutive patients screened, 144 were excluded (80 were ineligible, 40 met exclusion criteria, and 24 declined participation), leaving 176 participants; the screening and inclusion process is summarised in Fig. 1. No participant had missing data on the outcome or on the covariates analysed.

Fig. 1.

Participant flow diagram showing screening, exclusions and the analysed sample (N = 176)

Variables and measurements

Outcome: glycaemic control. HbA1c was taken from the most recent routine measurement within three months of enrolment, performed by high-performance liquid chromatography standardised to the National Glycohaemoglobin Standardization Program (NGSP); values are reported in NGSP (%) units with International Federation of Clinical Chemistry (IFCC, mmol/mol) equivalents. The outcome, poor glycaemic control, was defined a priori as HbA1c ≥ 7.0% (53 mmol/mol) — the threshold most widely applied in routine practice in our setting and consistent with general guidance for the majority of non-pregnant adults with T2DM [2, 15]. We applied this single cut-off uniformly; we recognise that less stringent, individualised targets are appropriate for some patients, such as older adults or those with established cardiovascular or renal disease, and we return to this in the limitations. HbA1c was also retained as a continuous variable for description.

Predictor variables. Age (years) and sex were recorded by structured interview and verified against the medical record. Educational attainment was categorised as primary, secondary or higher. Diabetes duration was derived from the year of diagnosis and analysed both continuously (per 5-year increment) and in three categories (≤ 5, 6–10, > 10 years). Current insulin therapy (yes/no) was taken from the treatment plan documented on the day of enrolment. Comorbidities were ascertained as physician-documented diagnoses in the medical record and defined as follows: hypertension [documented diagnosis and/or use of antihypertensive medication]; ischaemic heart disease [documented myocardial infarction, angina or coronary revascularisation]; chronic kidney disease [documented diagnosis and/or eGFR < 60 mL/min/1.73 m²]; diabetic retinopathy or other eye disease [ophthalmologist-documented diagnosis]; chronic lung disease [documented chronic obstructive pulmonary disease or asthma]; and dyslipidaemia [documented diagnosis and/or use of lipid-lowering therapy] [add supporting references]. The number of comorbidities (0–6) was summed, and a three-level comorbidity-burden variable was derived: none (0), single condition (1) and multimorbidity (≥ 2) [11].

Sample size. At the observed prevalence of poor control (≈ 58%), 176 participants provided ≥ 90% power (α = 0.05) to detect an odds ratio of 2.0 for a binary predictor of 30–70% prevalence. The primary multivariable model carried eight predictor degrees of freedom, giving an events-per-variable ratio of approximately 13 (103/8), within the range conventionally considered adequate for stable estimation [16].

Statistical analysis

Continuous variables were summarised as mean (standard deviation) and categorical variables as counts (percentages); characteristics are presented overall and by glycaemic-control status for description only, without between-group significance testing, because all eligible attenders were included rather than sampled from a defined source population. The prevalence of poor glycaemic control was reported with an exact binomial 95% confidence interval (CI).

Associations with poor glycaemic control were quantified by logistic regression as odds ratios (OR) with 95% CIs. Continuous predictors were rescaled (age per 10 years, diabetes duration per 5 years). The primary multivariable model (Model A) included a pre-specified set of covariates selected on clinical grounds rather than by univariate screening: age, sex, education, diabetes duration, insulin therapy, hypertension and ischaemic heart disease [17]. Because the comorbidity-burden variable is composed in part of hypertension and ischaemic heart disease, entering it alongside those conditions would adjust the burden index for its own components; the two were therefore not included simultaneously. Instead, a sensitivity model (Model B) replaced the individual conditions with the comorbidity-burden variable while retaining all other covariates. Multicollinearity was assessed using variance inflation factors (VIF > 5 considered problematic). Model calibration was assessed with the Hosmer–Lemeshow goodness-of-fit test and discrimination with the area under the ROC curve; Cox–Snell and Nagelkerke pseudo-R² values were also reported. As effect estimates are presented with 95% CIs, p-values are reported as secondary. A two-sided α of 0.05 defined statistical significance. Analyses were conducted in Stata 18 (StataCorp, College Station, TX, USA).

Results

Participant characteristics and prevalence of poor glycaemic control

A total of 176 adults with T2DM were analysed. Mean age was 65 years (SD 14), 90 (51%) were men, and mean diabetes duration was 11 years (SD 6). Ninety-five participants (54%) were receiving insulin.

Hypertension was the most common comorbidity (104; 59%), followed by ischaemic heart disease (60; 34%), chronic kidney disease (35; 20%), dyslipidaemia (25; 14%), eye disease (19; 11%) and chronic lung disease (18; 10%); 77 participants (44%) had two or more comorbidities. The prevalence of poor glycaemic control (HbA1c ≥ 7%) was 58.5% (103/176; 95% CI 50.9–65.9%). Mean HbA1c was 6.2% (44 mmol/mol) in the controlled group and 8.8% (73 mmol/mol) in the poorly controlled group. Participant characteristics overall and by glycaemic-control status are presented in Table 1.

Table 1.

Characteristics of the study population overall and by glycaemic control status (N = 176)

Variable	Total (N = 176)	HbA1c < 7% (n = 73, 41.5%)	HbA1c ≥ 7% (n = 103, 58.5%)
Age, years, mean (SD)	65.1 (14.3)	62.9 (13.5)	66.7 (14.7)
Male sex, n (%)	90 (51.1)	42 (57.5)	48 (46.6)
Education, n (%)
Primary	46 (26.1)	19 (26.0)	27 (26.2)
Secondary	84 (47.7)	30 (41.1)	54 (52.4)
Higher	46 (26.1)	24 (32.9)	22 (21.4)
Diabetes duration, years, mean (SD)	10.8 (6.0)	9.4 (5.9)	11.8 (5.9)
Diabetes duration category, n (%)
≤5 years	64 (36.4)	38 (52.1)	26 (25.2)
6–10 years	35 (19.9)	10 (13.7)	25 (24.3)
>10 years	77 (43.8)	25 (34.2)	52 (50.5)
HbA1c, %, mean (SD)	7.74 (1.79)	6.18 (0.52)	8.84 (1.54)
HbA1c, mmol/mol, mean (SD)	61.1 (19.6)	44.1 (5.7)	73.1 (16.9)
Insulin therapy, n (%)	95 (54.0)	22 (30.1)	73 (70.9)
Hypertension, n (%)	104 (59.1)	37 (50.7)	67 (65.0)
Ischaemic heart disease, n (%)	60 (34.1)	18 (24.7)	42 (40.8)
Chronic kidney disease, n (%)	35 (19.9)	14 (19.2)	21 (20.4)
Eye disease, n (%)	19 (10.8)	10 (13.7)	9 (8.7)
Chronic lung disease, n (%)	18 (10.2)	6 (8.2)	12 (11.7)
Dyslipidaemia, n (%)	25 (14.2)	12 (16.4)	13 (12.6)
Comorbidity burden, n (%)
None	31 (17.6)	15 (20.5)	16 (15.5)
Single condition	68 (38.6)	34 (46.6)	34 (33.0)
Multimorbidity (≥ 2)	77 (43.8)	24 (32.9)	53 (51.5)

Notes: Values are mean (standard deviation, SD) or number (percentage); percentages are column percentages within each group. Groups were defined by the glycated haemoglobin (HbA1c) threshold of 7% (53 mmol/mol). NGSP, National Glycohaemoglobin Standardization Program; IFCC, International Federation of Clinical Chemistry. Consistent with the cross-sectional design and the inclusion of all eligible attenders, characteristics are summarised descriptively without between-group significance testing; associations were formally assessed by logistic regression (Table 2)

Factors associated with poor glycaemic control

In univariate analyses (Table 2), poor glycaemic control was associated with insulin therapy (OR 5.64, 95% CI 2.93–10.87), longer diabetes duration (OR 1.42 per 5 years, 1.09–1.85) and ischaemic heart disease (OR 2.10, 1.09–4.08); hypertension showed a borderline association (OR 1.81, 0.98–3.34). Age, sex, education, chronic kidney disease, eye disease, lung disease, dyslipidaemia and comorbidity burden were not associated with the outcome. Insulin use was markedly more frequent among poorly controlled than controlled participants (71% versus 30%).

Table 2.

Logistic regression analyses of factors associated with poor glycaemic control (HbA1c ≥ 7%)

Predictor	Unadjusted OR (95% CI)	p	Model A – adjusted OR (95% CI)	p	Model B – adjusted OR (95% CI)	p
Age (per 10 years)	1.21 (0.98–1.49)	0.082	0.96 (0.68–1.37)	0.836	1.00 (0.71–1.40)	0.979
Male sex (vs. female)	0.64 (0.35–1.18)	0.154	0.68 (0.35–1.34)	0.268	0.66 (0.34–1.29)	0.223
Education (vs. primary)
Secondary	1.27 (0.61–2.65)	0.530	1.66 (0.69–4.00)	0.254	1.79 (0.74–4.34)	0.194
Higher	0.65 (0.28–1.47)	0.297	0.98 (0.31–3.11)	0.969	1.04 (0.32–3.31)	0.951
Diabetes duration (per 5 years)	1.42 (1.09–1.85)	0.010	0.90 (0.60–1.35)	0.615	0.93 (0.62–1.40)	0.724
Insulin therapy (yes vs. no)	5.64 (2.93–10.87)	< 0.001	6.32 (2.71–14.71)	< 0.001	5.99 (2.62–13.69)	< 0.001
Hypertension (yes vs. no)	1.81 (0.98–3.34)	0.057	1.34 (0.63–2.88)	0.448	—	—
Ischaemic heart disease (yes vs. no)	2.10 (1.09–4.08)	0.028	0.96 (0.42–2.20)	0.921	—	—
Chronic kidney disease (yes vs. no)	1.08 (0.51–2.30)	0.843	—	—	—	—
Eye disease (yes vs. no)	0.60 (0.23–1.57)	0.300	—	—	—	—
Chronic lung disease (yes vs. no)	1.47 (0.53–4.12)	0.461	—	—	—	—
Dyslipidaemia (yes vs. no)	0.73 (0.31–1.72)	0.476	—	—	—	—
Comorbidity burden (vs. none)
Single condition	0.94 (0.40–2.19)	0.882	—	—	0.75 (0.28–1.98)	0.557
Multimorbidity (≥ 2)	2.07 (0.88–4.86)	0.095	—	—	0.91 (0.28–2.93)	0.875

Notes: OR, odds ratio; CI, confidence interval. Bold indicates p < 0.05. Model A (primary) is adjusted for age, sex, education, diabetes duration, insulin therapy, hypertension and ischaemic heart disease. Model B (sensitivity) replaces the individual conditions with the comorbidity-burden variable to avoid adjusting the burden index for its own components; all other covariates are retained. Chronic kidney disease, eye disease, chronic lung disease and dyslipidaemia were assessed univariately only. “—” indicates the variable was not entered in that model. Continuous predictors are scaled per 10 years (age) and per 5 years (diabetes duration). Model A: Hosmer–Lemeshow χ² = 2.55, df = 8, p = 0.96; area under the ROC curve 0.74; Nagelkerke R² 0.23; all variance inflation factors < 2.5 (mean 1.70). Model B: Hosmer–Lemeshow χ² = 6.16, df = 8, p = 0.63; area under the ROC curve 0.74

In the primary multivariable model (Model A), insulin therapy was the only factor independently associated with poor glycaemic control (adjusted OR 6.32, 95% CI 2.71–14.71). After adjustment, the univariate associations of diabetes duration (adjusted OR 0.90, 0.60–1.35) and ischaemic heart disease (0.96, 0.42–2.20) were no longer present, and age, sex, education and hypertension were likewise not independently associated. The model was well calibrated (Hosmer–Lemeshow χ² [8] = 2.55, p = 0.96) with moderate discrimination (area under the ROC curve 0.74; Fig. 2) and a Nagelkerke R² of 0.23; all variance inflation factors were below 2.5 (mean 1.70), indicating no important collinearity.

Fig. 2.

Receiver operating characteristic (ROC) curve for the primary multivariable model (Model A); area under the curve 0.74

In the sensitivity model (Model B), in which the comorbidity-burden variable replaced the individual conditions, the findings were unchanged: insulin therapy remained the only independent correlate (adjusted OR 5.99, 95% CI 2.62–13.69), while neither a single comorbidity (0.75, 0.28–1.98) nor multimorbidity (0.91, 0.28–2.93) was associated with the outcome.

Model performance was similar (Hosmer–Lemeshow p = 0.63; area under the ROC curve 0.74). The attenuation of the multimorbidity association on adjustment therefore reflected its overlap with insulin therapy and age, rather than adjustment of the burden index for its own components.

Discussion

Principal findings

In this cross-sectional study of 176 Albanian adults with type 2 diabetes, poor glycaemic control was common, affecting 58.5% of participants. After adjustment, insulin therapy was the only factor independently associated with poor control; age, diabetes duration, education, hypertension, ischaemic heart disease and comorbidity burden — several of which were associated in unadjusted analyses — showed no independent association. The result was unchanged whether comorbidity was modelled as individual conditions or as an aggregate burden index.

Interpretation

The prevalence we observed sits within the 40–60% range reported in international surveys of adults with type 2 diabetes who do not reach an HbA1c target below 7% [2–4], suggesting that the treatment gap in this setting is comparable to that documented elsewhere rather than exceptional.

The central finding — that insulin therapy is the strongest and only independent correlate of poor control — is best understood as confounding by indication rather than a harmful effect of insulin. In routine care, insulin is introduced when oral agents no longer maintain glycaemic targets, so insulin-treated patients are, by definition, those whose disease has already proved more difficult to control [7–10]. The strong cross-sectional association between insulin use and elevated HbA1c therefore marks the more advanced end of the disease spectrum; it does not imply that insulin worsens control. This reading is reinforced by the behaviour of the other variables: diabetes duration and ischaemic heart disease, both associated with poor control in unadjusted analysis, lost that association once insulin therapy was included. Because longer duration and established cardiovascular disease cluster in the same patients who progress to insulin, insulin therapy captures the shared underlying severity, and the apparent contributions of duration and heart disease are not independent of it.

Multimorbidity behaved similarly. Although a burden of two or more conditions was numerically more frequent among poorly controlled participants, it was not independently associated with the outcome in either the individual-condition model or the burden-based sensitivity model. The attenuation on adjustment thus reflects overlap with insulin therapy and age, rather than an artefact of adjusting a composite index for its own components — a possibility we excluded by design through the two-model approach. Our data therefore do not support comorbidity burden as an independent correlate of glycaemic control in this population, a question on which previous reports have been mixed [11–13].

Clinical implications

These findings point to insulin-treated patients as a readily identifiable, high-priority group for structured review, rather than as evidence against insulin. In a resource-constrained setting, insulin status is a simple flag that concentrates poor control and can be used pragmatically to target treatment intensification, adherence support and individualised follow-up at those least likely to be at target. The absence of independent associations for the demographic and comorbidity variables suggests that, once treatment stage is known, these characteristics add little to identifying patients in need of closer attention.

Strengths and limitations

Strengths include consecutive enrolment with complete data on the outcome and all covariates, a pre-specified multivariable model selected on clinical rather than statistical grounds, and a sensitivity analysis confirming the robustness of the principal finding. To our knowledge, this is among the first patient-level analyses of the correlates of glycaemic control in adults with type 2 diabetes from Albania, addressing a gap in evidence from South-Eastern Europe [5, 6].

Several limitations temper interpretation. First, the cross-sectional design precludes any inference about temporality or causation; the insulin–HbA1c association in particular cannot establish direction and is consistent with confounding by indication. Second, the study was conducted at a single [centre] and recruited a modest sample, which limits generalisability and statistical power; the wide confidence intervals around several estimates mean that smaller independent effects cannot be excluded, and the findings should be confirmed in larger, multicentre studies. Third, we applied a single HbA1c threshold of 7% uniformly, whereas contemporary guidance recommends individualised targets that are less stringent for older adults and those with established cardiovascular or renal disease; some participants classified as poorly controlled may in fact have been at an appropriate individualised target, which would tend to overstate the prevalence of true undertreatment. Fourth, comorbidities were ascertained from medical records and may be incompletely documented, and we did not capture factors such as treatment adherence, diet, physical activity, body mass index, socioeconomic status or specific drug regimens, so residual confounding is likely. Finally, recruitment from a single [region] may limit applicability to other Albanian or regional settings.

Conclusions

In this study of Albanian adults with type 2 diabetes, poor glycaemic control was common, affecting close to three in five participants. Insulin therapy was the only factor independently associated with poor control; demographic characteristics, diabetes duration and comorbidity burden carried no independent association once treatment stage was accounted for. This association is most plausibly explained by confounding by indication - insulin marks the more advanced, harder-to-control end of the disease spectrum - rather than by any adverse effect of insulin itself. Practically, insulin-treated patients represent a simple, readily identifiable group in whom treatment intensification, adherence support and individualised follow-up can be concentrated, which is particularly useful in resource-constrained settings. As among the first patient-level evidence on glycaemic control from Albania, these findings address a regional gap, but their cross-sectional nature means prospective, multicentre studies are needed to confirm the associations and clarify their direction.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

aOR

Adjusted odds ratio

AUC

Area under the receiver operating characteristic curve

CI

Confidence interval

EASD

European association for the study of diabetes

HbA1c

Glycated haemoglobin

HPLC

High-performance liquid chromatography

IFCC

International federation of clinical chemistry

NGSP

National glycohaemoglobin standardization program

OR

Odds ratio

SD

Standard deviation

STROBE

Strengthening the reporting of observational studies in epidemiology

T2DM

Type 2 diabetes mellitus

VIF

Variance inflation factor

Author contributions

[Z.SH] designed the study, supervised data collection and drafted the manuscript. [E.H] contributed to participant recruitment, data curation, drafted the manuscript and critical revision. [E.D] performed the statistical analysis and contributed to interpretation.All authors read and approved the final manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Data availability

The de-identified dataset analysed during the current study is available from the corresponding author on reasonable request, subject to local data-protection regulations and approval from the institutional ethics committee.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the ethical standards of the University of Shkodra “Luigj Gurakuqi” and with the principles of the 1964 Declaration of Helsinki and its later amendments. Ethical approval was obtained from the University Ethics Committee of the University of Shkodra “Luigj Gurakuqi” (Approval No. 2025/AEI-18). Participants were fully informed about the objectives of the study and were assured of anonymity and confidentiality. They were informed of their right to refuse participation or to withdraw at any stage without any consequence for their ongoing clinical care, and written informed consent was obtained from every participant prior to enrolment. No personally identifiable information was retained in the analytic dataset, and data were stored securely for research purposes only.

Consent for publication

Not applicable. The manuscript does not contain any individual-level data (text, images, videos or details) by which a participant could be identified.

Competing interests

The authors declare no competing interests.