Researchers Derive Streamlined Quality of Life Scale for Mexican Diabetes Patients

A psychometric evaluation conducted in a public primary care setting in Mexico City has demonstrated that the established five-domain structure of the Diabetes-39 instrument fails to capture distinct dimensions of health-related quality of life in Mexican adults living with type 2 diabetes, leading investigators to develop an abbreviated, unidimensional 23-item version. Published in Frontiers in Clinical Diabetes and Healthcare, the study evaluated the structural validity, internal consistency, and construct validity of the original instrument alongside previously proposed configurations. The researchers discovered that the multi-factor representations of the scale exhibited substantial inter-factor overlap, indicating that the assessed health aspects function as a single underlying construct. By systematically removing redundant and low-performing items, the team derived a shortened tool designated as the D-23, which demonstrated satisfactory fit indices and robust statistical reliability when tracking the perceived impact of diabetes on daily life.

Health-related quality of life has gained increasing recognition across clinical disciplines as a meaningful patient-reported outcome that complements conventional metabolic indicators such as glycated hemoglobin and capillary blood glucose. In primary care environments where routine clinical consultations are typically brief, evaluating patient perceptions of their general well-being provides critical insight into how chronic disease management alters daily functioning. In Mexico, where national health data recorded a diabetes prevalence of 17.0 percent between 2021 and 2024, approximately one in three affected adults attains recommended glycemic control targets. Because chronic self-management behaviors can place significant psychological demands on patients, systematically assessing health perceptions remains vital. However, previous psychometric investigations of the Diabetes-39 in Mexico were restricted to exploratory factor analysis and convergent validity assessments, leaving confirmatory structural evidence absent and leaving clinicians without a streamlined, empirically verified instrument tailored for routine consultations.

To investigate the internal structure of the instrument, investigators employed a nonexperimental, instrumental, cross-sectional design involving 201 adult outpatients receiving care at an Instituto Mexicano del Seguro Social clinic situated in the Gustavo A. Madero district of Mexico City. The sample comprised 115 women, representing 57.2 percent of the cohort, and 86 men, representing 42.8 percent, with a mean participant age of 65.12 years. Enrolled individuals presented with a mean diabetes duration of 13.4 years, ranging from 1 to 35 years, alongside a mean preprandial capillary glucose level of 158.84 milligrams per deciliter. Sociodemographic profiles indicated that 65.7 percent of the participants had attained a primary school education, 23.4 percent had completed middle school, 10.0 percent had finished high school, and 1.0 percent held a bachelor’s degree. Comorbid medical conditions were widespread across the patient group, with hypertension identified in 61.7 percent of individuals and obesity diagnosed in 25.4 percent.

Data collection took place between April and July 2021 amid institutional public health containment measures associated with the COVID-19 pandemic. To adhere to healthcare facility access protocols, medical interns received formal training from the principal investigator to administer the assessment battery as structured in-person interviews during medical visits. The interview battery integrated a sociodemographic and clinical questionnaire, the original 39-item Diabetes-39, the 21-item Therapeutic Adherence Scale, and the 24-item World Health Organization Quality of Life brief version, known as the WHOQOL-BREF. Eligibility criteria required participants to be at least 18 years old, possess a verified diagnosis of type 2 diabetes of at least six months, and complete at least 90 percent of the measurement protocol. Patients with type 1 diabetes and individuals with documented psychological or psychiatric disorders in their clinical files were systematically excluded to prevent psychiatric symptoms from confounding self-reported diabetes impacts.

Statistical modeling was carried out using maximum likelihood confirmatory factor analysis to compare competing configurations of the instrument. The researchers evaluated four sequential models: Model 1, representing the original 39-item structure spanning five domains comprising Energy and Mobility, Diabetes Control, Social Burden, Anxiety and Worry, and Sexual Functioning; Model 2, testing the 21-item short form derived from Brazilian data nested within the original five domains; Model 3, specifying a unidimensional model comprising all 39 items; and Model 4, evaluating a newly refined unidimensional short form. Confirmatory testing revealed that neither multi-factor specification achieved acceptable goodness-of-fit benchmarks. Model 1 yielded a chi-square of 2019.511 across 692 degrees of freedom, producing a comparative fit index of .881, a Tucker-Lewis index of .872, and a root mean square error of approximation of .098. Model 2 similarly fell short of standard criteria, recording a comparative fit index of .941 and an error of approximation of .093.

Crucially, the five-factor models exhibited severe inter-factor correlations ranging between .85 and .97. Evaluation of internal discriminant validity using the Fornell-Larcker criterion demonstrated that four of the five factors had average variance extracted values lower than their highest squared inter-factor correlation, which reached .94. These statistical associations indicated that the five theoretical domains proposed in the original instrument were not empirically distinguishable within the Mexican patient sample. The unidimensional 39-item model also performed poorly, producing a comparative fit index of .873 and an approximation error of .100. Consequently, investigators initiated a sequential, data-driven item reduction procedure from the unidimensional model, removing items exhibiting standardized residual covariances greater than 2.58 in absolute value across multiple items, standardized factor loadings below .50, or conceptual redundancy and lack of disease specificity.

Through this iterative reduction process, 16 individual items were systematically eliminated, yielding the 23-item D-23 instrument. The refined model incorporated seven residual covariances specified exclusively between items originating from identical baseline domains, reflecting shared wording nuances and localized domain variance within Diabetes Control, Anxiety and Worry, Sexual Functioning, and Energy and Mobility. Model 4 achieved satisfactory fit parameters, documented by a chi-square of 416.061 with 223 degrees of freedom, a normed chi-square of 1.866, a standardized root mean square residual of .0238, an error of approximation of .066 with a 90 percent confidence interval between .056 and .076, a comparative fit index of .968, and a Tucker-Lewis index of .963. Standardized factor loadings across the 23 items ranged from .69 to .92, establishing an average variance extracted of .74. When residual covariances were fixed to zero in sensitivity testing, the normed chi-square remained at 2.540 and the standardized residual was .0279, demonstrating that the structural model did not depend entirely on post hoc modifications.

To rigorously examine dimensionality and establish empirical support for score calculation, the investigators evaluated a bifactor model and a second-order factor model. The bifactor configuration failed to achieve computational convergence, reflecting empirical underidentification that occurs when secondary group factors account for minimal variance independent of a general factor. Conversely, the second-order model converged successfully, showing that all five original domains loaded strongly on an overarching health-related quality of life construct, with loadings ranging from .937 for Sexual Functioning to .999 for Social Burden. A Schmid-Leiman transformation revealed that item loadings on the general construct remained high at .69 to .92, while specific domain loadings collapsed to values between .04 and .33. The general factor captured 97.5 percent of the common variance, yielding an explained common variance of .975, a percentage of uncontaminated correlations of .779, a construct replicability coefficient of .987, and a hierarchical omega of .982, demonstrating that the measure is essentially unidimensional and supports only a single total score.

Assessment of internal consistency revealed exceptionally high reliability parameters for the D-23 total scale, with Cronbach’s alpha reaching .985, accompanied by a 95 percent confidence interval spanning .982 to .988, and McDonald’s omega reaching .985. Construct validity testing aligned with the study’s predefined directional hypotheses. Total D-23 scores exhibited a moderate negative correlation with the overall WHOQOL-BREF quality of life score, recorded at r = -.542, and correlated negatively with all individual domains of the generic scale at values ranging between -.399 and -.545. A weaker negative association emerged with the total Therapeutic Adherence Scale at r = -.237, reflecting adherence as an external self-management behavioral factor rather than an internal health state. Within adherence subscales, D-23 scores correlated moderately with self-efficacy at r = -.377, while displaying non-significant associations with medication and dietary control at r = -.046 and behavioral medical follow-up at r = -.088.

Known-groups validity analyses further supported the clinical utility of the shortened measure. Total scores are interpreted such that higher numerical values indicate a greater perceived negative impact of diabetes on personal quality of life, spanning a theoretical score range from 23 to 161. Nonparametric Mann-Whitney testing revealed that patients presenting with uncontrolled glycemia, defined as preprandial glucose exceeding 130 milligrams per deciliter, reported significantly higher burden scores than patients with controlled glycemia between 70 and 130 milligrams per deciliter. This comparison produced a test statistic of U = 2029 with p less than .001 and an effect size of r = .34, which remained statistically significant following Holm-Bonferroni correction. Occupational categories evaluated via Kruskal-Wallis analysis also displayed statistical differences with H equal to 12.810 and an epsilon-squared effect size of .064, although researchers advised cautious interpretation because a single individual represented the professional employment category.

The authors outlined several methodological constraints that frame the interpretation and practical implementation of their findings. Primarily, the D-23 was derived and evaluated within the same clinical sample through data-driven item reduction and freed residual covariances without split-sample or external confirmation. The investigation was conducted at a single urban facility during severe COVID-19 pandemic conditions, and the exclusion of patients with psychiatric diagnoses or type 1 diabetes limits generalizability across broader clinical populations. Furthermore, the single total score cannot provide isolated domain profiles for specialized tracking. The investigators concluded that the D-23 represents an efficient, psychometrically sound candidate instrument that can streamline patient-reported outcome assessments during public primary care consultations, but emphasized that cross-validation in independent, larger cohorts must occur before deploying the tool in routine clinical practice.

Subject of Research: Psychometric evaluation and short-form derivation of the Diabetes-39 instrument in Mexican adults with type 2 diabetes

Article Title: Confirmatory factor analysis of the diabetes-39 and derivation of a 23-item short form in Mexican adults with type 2 diabetes

Article References: Pérez-Ortiz, A., Becerra-Gálvez, A. L., Ortega-Luyando, M., Mancilla-Díaz, J. M., Alvarez-Rayón, G. L., Anguiano-Serrano, S. A., & Amaya-Hernández, A. (2026). Confirmatory factor analysis of the diabetes-39 and derivation of a 23-item short form in Mexican adults with type 2 diabetes. Frontiers in Clinical Diabetes and Healthcare, 7, Article 1979834. https://doi.org/10.3389/fcdhc.2026.1979834

Image Credits: AI Generated

DOI: 10.3389/fcdhc.2026.1979834

Keywords: type 2 diabetes, health-related quality of life, confirmatory factor analysis, Diabetes-39, psychometrics, patient-reported outcomes, primary care, clinical assessment, Confirmatory, factor, analysis, derivation