Data source and study design

We conducted a nationwide retrospective observational study using the Spanish National Hospital Discharge Database (RAE-CMBD, Registro de Actividad de Atención Especializada–Conjunto Mínimo Básico de Datos). This registry, maintained by the Spanish Ministry of Health, is mandatory and captures over 95% of all hospital admissions nationwide [18,19]. The unit of analysis was the hospital discharge episode rather than the unique patient. Consequently, all estimates reported in this study refer to hospitalizations or discharge episodes. Because the anonymized RAE-CMBD does not provide a longitudinal patient identifier, multiple hospitalizations involving the same individual during the study period may have been counted as separate observations.

Our analysis focused on the post-acute phase of the pandemic and included all hospital discharge records from January 1, 2021, to December 31, 2024.

Selection criteria and population stratification

The study population consisted of hospitalizations for patients aged ≥40 years with a diagnosis of T2DM in any diagnostic position (ICD-10 code E11.x). We excluded episodes with missing data for sex, age, or discharge status, as well as cases of type 1 diabetes. To evaluate the clinical burden, the cohort was stratified according to the presence of all-cause dementia and its primary subtypes: AD and VaD. These conditions were identified using a specific cluster of ICD-10 codes, as detailed in S1 Table.

Clinical variables and comorbidity assessment

To identify disparities in patient profiles and hospital outcomes, we evaluated a refined core set of clinical variables, categorized into: (i) cardiometabolic comorbidities (including hypertension, dyslipidemia, obesity, heart disease, stroke/TIA, and CKD); (ii) neurological and neuropsychiatric conditions (including Parkinson’s disease, depression, and substance use); and (iii) the acute infectious context (specifically COVID-19 diagnosis). A comprehensive list of all conditions and their corresponding ICD-10 codes is provided in S1 Table. Additionally, overall comorbidity burden was assessed using the Charlson Comorbidity Index (CCI), modified by excluding T2DM and dementia from the calculation to avoid redundancy [20,21].

Outcome measures

The primary clinical endpoint was all-cause IHM, defined as death from any cause during the hospitalization period, regardless of the primary diagnosis at admission. Secondary outcomes included the length of hospital stay (LOHS) in days, admission to the intensive care unit (ICU), and total estimated hospitalization costs.

Statistical analysis

Annual prevalence rates were calculated for all-cause dementia, AD, and VaD among T2DM hospitalizations, with comparisons between sexes and age groups. Descriptive data were presented as means with standard deviation (SD) for continuous variables and as percentages for categorical data.

Temporal trends in categorical outcomes were evaluated using the Cochran–Armitage test for trend, with calendar year treated as an ordered variable from 2021 to 2024. Comparisons of categorical variables between groups were performed using two-sided chi-square tests, whereas continuous variables were compared using Student’s t-test or ANOVA, as appropriate. Because the large sample size could result in statistically significant p-values for small absolute differences, the magnitude of the observed changes was considered alongside statistical significance.

To determine independent predictors of both the presence of dementia and IHM, we developed multivariable logistic regression models. These models were adjusted for demographics, CCI, and the specific comorbidities in the “core set.” Results were expressed as adjusted odds ratios (aOR) with 95% confidence intervals (CI). All analyses were performed using SPSS version 29.0 or Stata version 14, with significance set at a two-sided p < 0.05.

Ethical considerations

The data were obtained from an anonymized administrative database provided by the Spanish Ministry of Health. Under Spanish legislation [22], the use of de-identified public registries for epidemiological research does not require specific ethics committee approval or individual patient consent. Data were accessed for research purposes on September 1, 2025. The authors had no access to information that could identify individual participants during or after data collection, as all records in the RAE-CMBD are fully anonymized by the Ministry before provision.

Overall trends in prevalence

From 2021 to 2024, the RAE-CMBD recorded 2,786,184 T2DM-related hospitalizations among individuals aged ≥40 years. Within this population, all-cause dementia was identified in 180,623 admissions, corresponding to 6.48% of all T2DM-related hospitalizations. Across the study period, the annual prevalence of all-cause dementia ranged from 8.75% to 9.45% in women and from 4.35% to 4.84% in men, with women exhibiting a consistently higher prevalence than men in all years. Regarding specific subtypes, the annual prevalence of AD ranged between 4.25% and 4.51% in women and 1.75% and 1.90% in men, while VaD prevalence was 1.73% and 1.24%, respectively. Overall, total T2DM-related hospitalizations increased from 653,517 in 2021 to 728,951 in 2024. Although the temporal trend test was statistically significant for all-cause dementia, the absolute variation over the study period was modest, ranging from 6.16% in 2021 to 6.55% in 2024. The statistical significance of this trend should therefore be interpreted in the context of the large number of hospital episodes analyzed. The annual frequency of all-cause dementia and its primary subtypes among T2DM-related hospitalizations, stratified by sex and study year, is presented in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Annual prevalence of dementia among hospitalizations with type 2 diabetes mellitus, by sex and dementia subtype. Spain (2021–2024).

https://doi.org/10.1371/journal.pone.0358874.t001

Demographic and clinical profiles: The frailty and comorbidity paradox

Hospitalizations involving T2DM and comorbid dementia presented a significantly different profile than those without dementia with baseline characteristics summarized in Table 2. The dementia group was characterized by an older mean age (84.06 vs. 74.46 years) and a markedly higher proportion of women (57.83% vs. 39.89%). Detailed demographic and clinical profiles of the dementia cohort, further stratified by sex, are provided in S2 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of T2DM-related hospitalizations according to all-cause dementia status, Spain, 2021–2024.

https://doi.org/10.1371/journal.pone.0358874.t002

Hospitalizations involving all-cause dementia had a lower mean modified CCI score than those without dementia (1.14 vs. 1.31), as shown in Table 2. Heart failure (21.67% vs. 20.69%), atrial fibrillation (26.43% vs. 23.27%), and chronic kidney disease (29.35% vs. 24.58%) were more frequent among hospitalizations involving dementia. Prior stroke/TIA was also more frequent in hospitalizations involving dementia than in those without dementia (17.09% vs. 9.76%).

In-hospital outcomes according to dementia status

Hospitalizations involving all-cause dementia had lower ICU admission rates (1.33% vs. 7.58%; p < 0.001), higher frequencies of palliative-care coding (3.76% vs. 1.65%; p < 0.001), shorter hospital stays, and lower median estimated hospitalization costs (€6,129.3 vs. €8,602.94; p < 0.001) than hospitalizations without dementia (Table 3). A comprehensive breakdown of in-hospital mortality according to specific concomitant conditions and dementia status is available in S3 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Annual in-hospital outcomes among T2DM hospitalizations by dementia status and sex, Spain (2021–2024).

https://doi.org/10.1371/journal.pone.0358874.t003

Multivariable analysis of factors associated with dementia and in-hospital mortality

The results of the multivariable logistic regression models identifying factors independently associated with the presence of dementia and IHM are summarized in

Table 4 and Table 5. Subtype-specific predictors for the presence of and mortality associated with AD and VaD are detailed in S4 and S5 Tables.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Multivariable logistic regression for factors associated with the presence of all-cause dementia among hospitalizations with T2DM. Spain (2021–2024).

https://doi.org/10.1371/journal.pone.0358874.t004

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Multivariable logistic regression analysis of factors associated with in-hospital mortality among T2DM-related hospitalizations with all-cause dementia, Spain, 2021–2024.

https://doi.org/10.1371/journal.pone.0358874.t005

Regarding the probability of having a dementia diagnosis, age ≥ 85 years emerged as the most powerful independent predictor (aOR 32.75; 95% CI 31.43–34.14), with women in this age group showing an even higher likelihood (aOR 41.95; 95% CI 39.02–45.1) compared to men. A prior history of stroke/TIA was strongly associated with all-cause dementia (aOR 2.35; 95% CI 2.32–2.39), and this association reached its peak for VaD with a massive aOR of 5.89 (95% CI 5.69–6.09). Other significant neurological comorbidities included PD (aOR 2.91) and depression (aOR 1.43). Notably, a higher modified CCI (CCI ≥ 2) was associated with a lower probability of being diagnosed with dementia (aOR 0.42; 95% CI 0.41–0.43), suggesting a potential “comorbidity paradox” in the clinical coding of these complex patients.

In the multivariable model of IHM among T2DM-related hospitalizations with all-cause dementia, age ≥ 85 years showed the strongest positive association with IHM (aOR 2.14; 95% CI 1.86–2.46). COVID-19 was also associated with higher odds of IHM (aOR 1.27; 95% CI 1.22–1.32). Female sex was independently associated with lower odds of IHM (aOR 0.90; 95% CI 0.87–0.92). Obesity (aOR 0.81; 95% CI 0.77–0.85) and depression (aOR 0.78; 95% CI 0.74–0.82) also showed inverse associations with IHM. In contrast to the model evaluating dementia coding, a higher comorbidity burden (CCI ≥ 2) was associated with higher odds of IHM (aOR 1.24; 95% CI 1.18–1.31).

Trends in prevalence and clinical profiles

Our findings show that dementia diagnoses remained frequent among T2DM-related hospitalizations in Spain (6.48% overall; 9.13% in women), consolidating the upward trend observed during the 2011–2020 period [6]. This high prevalence reinforces the well-documented bidirectional relationship between metabolic disorders and neurodegeneration, where T2DM patients face a 1.5 to 2.5 times higher risk of developing cognitive impairment [4,5].

Regarding dementia subtypes, AD remains the most frequent form in our sample (2.88%), though VaD (1.44%) represents a critical complication driven by cumulative glycemic toxicity and neurovascular damage [6,20,21]. The modest annual changes observed between 2021 and 2024 suggest a phase of stabilization following the disruptive impact of the COVID-19 pandemic on hospital admission patterns [6,17].

Sex-related differences in dementia and in-hospital mortality

A central finding of this study was the marked sex difference in dementia coding. All-cause dementia was recorded in 9.13% of T2DM-related hospitalizations involving women and 4.64% of those involving men. This difference may partly reflect the older age distribution among women and sex-related differences in dementia risk described in previous studies [13,14,22,23].

Female sex was independently associated with lower odds of in-hospital mortality among T2DM-related hospitalizations with all-cause dementia (aOR 0.90; 95% CI 0.87–0.92). Although this pattern may be described as a female survival advantage, it should not be interpreted as evidence of an intrinsic biological protective effect. Residual confounding, differences in admission patterns, unmeasured disease severity, functional status, treatment eligibility, and other clinical or social factors may have contributed to the observed association.

Clinical profiles: The vascular and neuropsychiatric burden

Our findings revealed a lower CCI in the dementia group compared to patients without cognitive impairment (1.14 vs. 1.31), despite the former being significantly older. This ‘comorbidity paradox’ likely reflects a coding bias inherent in administrative databases, where the identification of advanced dementia or acute life-threatening complications may lead to the under-reporting of secondary chronic conditions in the discharge summary.

The strong association between prior stroke/TIA and vascular dementia coding (aOR 5.89; 95% CI 5.69–6.09) highlights the relevance of cerebrovascular disease in this population [6,20]. This pattern is consistent with the contribution of cumulative neurovascular injury to cognitive decline among people with T2DM, alongside neurodegenerative mechanisms [22,23]. However, the effectiveness of any specific cerebrovascular preventive intervention cannot be inferred from this observational analysis. Furthermore, the coexistence of PD (6.82%) and the positive association with depression (aOR 1.43) indicate a substantial neuropsychiatric comorbidity burden [17]. Recent literature suggests that the multimorbidity of diabetes and depression post-stroke doubles the risk of loss of independence and progression to dementia, aligning with the high vulnerability detected in our cohort [23–25].

Differences in care intensity according to dementia status

Hospitalizations involving all-cause dementia had lower ICU admission rates (1.33% vs. 7.58%) and higher frequencies of palliative-care coding (3.76% vs. 1.65%) than hospitalizations without dementia. They also had shorter hospital stays and lower median estimated hospitalization costs. These findings indicate differences in the intensity and orientation of hospital care. However, the RAE-CMBD does not include information on dementia severity, baseline functional status, frailty, advance directives, patient or family preferences, ICU eligibility, treatment-limitation decisions, or the clinical rationale for palliative-care referral. Therefore, the observed coding patterns should not be interpreted as direct evidence of a systematic transition to end-of-life care [26]. The observed IHM of 17.06% represents all-cause mortality occurring during hospitalizations involving dementia and should not be interpreted as mortality directly attributable to dementia. Higher mortality frequencies were observed among hospitalizations with concomitant COVID-19 or heart failure [27,28]. However, the RAE-CMBD does not include direct measures of dementia severity, frailty, baseline functional status, acute physiological severity, or the specific cause of death. Therefore, the mechanisms underlying these differences cannot be determined from the present analysis.

Strengths and limitations

The primary strength of this study is the use of the RAE-CMBD database, ensuring nationwide representativeness that exceeds 95% of Spanish admissions [17]. The stratification by dementia subtypes and sex provides significant clinical value beyond studies that treat dementia as a single, homogeneous entity. This granular approach is essential because neurological risks in T2DM are highly heterogeneous; treating these conditions as uniform can mask distinct pathophysiological patterns and risk clusters [29,30]. Furthermore, our clinical differentiation is supported by evidence suggesting that T2DM may contribute to neurodegeneration through non-amyloidogenic pathways, such as chronic neuroinflammation and microvascular damage, rather than solely accelerating classic Alzheimer’s pathology, making the distinction between VaD and AD clinically meaningful for targeted prevention [31]. Finally, the inclusion of a sex-sensitive lens addresses a robust determinant of cognitive decline, as the synergy between metabolic status and comorbidities like depression has been shown to manifest differently across sexes [32].

However, several limitations must be acknowledged. Clinical conditions were identified from discharge coding, which may have resulted in undercoding, particularly for milder dementia [33–35]. Residual confounding is likely because the RAE-CMBD does not capture dementia severity, cognitive or functional assessments, frailty, glycemic control, diabetes duration, socioeconomic status, medication use, advance directives, patient or family preferences, ICU eligibility, treatment-limitation decisions, or the clinical rationale for palliative-care referral. The unit of analysis was the hospital discharge episode rather than the unique individual; therefore, the same person may have contributed more than one hospitalization, and within-person correlation arising from recurrent hospitalizations could not be accounted for. Consequently, the reported percentages should be interpreted as episode-level frequencies among T2DM-related hospitalizations and not as estimates of population prevalence among unique persons.

Finally, the observational and episode-based design precludes causal inference [36,37].