In the current investigation of three cohort studies, we extend findings in previous publications exploring the associations of obesity and metabolic status with incident dementia. Our findings implicated that metabolic status had a stronger association with dementia than obesity. When compared with MNNO, MANO and MAO presented higher risks for incident dementia. However, we didn’t observe evidence suggesting that the cumulative incidence of dementia in MNO group was different from MNNO. Moreover, participants with MANO who recovered to normal metabolic status showed a decreased risk of incident dementia when compared with their stable counterparts.

Previous studies showed that there were close associations of obesity and metabolic abnormality with incident dementia in middle-aged adults [23, 36,37,38,39,40]. A meta-analysis including 14 cohort studies found that compared to non-obesity, obesity was associated with approximately 31% higher risk of incident dementia in populations below the age of 65 years [40]. In terms of metabolic status, Qureshi et al. found a positive association between MetS and dementia in mid-life individuals [23]. In our study with three prospective cohorts, we observed consistent results for elevated risk of incident dementia in participants with obesity or abnormal metabolic status.

To date and to our knowledge, only a few cohort studies have examined the associations of metabolic heterogeneity of obesity with incident dementia in middle-aged adults, with inconclusive findings [10, 11, 41]. In the Whitehall II (median follow-up = 20.8 years), MNO was associated with a substantially higher risk of incident dementia in individuals aged < 60 years (HR: 1.69, 95% CI: 1.16–2.45 compared with MNNO) [10]; while in another cohort study based on the National Health Insurance System of Korea (median follow-up = 5.4 years), an inverse association was observed between MNO and incident dementia (HR: 0.90, 95% CI: 0.87–0.93) [11]. Our observations in three well-characterized cohort studies suggested no significant difference in dementia risks between MNNO and MNO groups. Differences in population characteristics and methodologies for assessing obesity and metabolic status may contribute to the inconsistency across studies. For instance, the Whitehall II study primarily involved a male population [10], which was inconsistent with the three cohorts in the current study. Additionally, Wang and colleagues used blood pressure and blood-based biomarkers, including C-reactive protein (CRP), TG, low-density lipoprotein cholesterol (LDL-C), HDL-C, and HbA1c, to define metabolic status [41], which was different from the other studies [10,11,12,13]. In addition, variations in adjustments for confounding may contribute to the different associations observed across investigations. Furthermore, most previous studies were limited by small sample sizes or short-term follow-up, which may result in reverse causation induced by early physiological and biochemical changes resulting from preclinical dementia. Our study provided valuable information with relatively long-term follow-up, but future studies are still warranted to clarify the causal association between MNO and dementia.

The comparisons of the associations of obesity and abnormal metabolic status with dementia findings are novel. Participants with abnormal metabolic status had a higher HR for incident dementia than participants with higher BMI. In addition, individuals with MANO had a 33% higher risk of incident dementia, while MNO was not associated with dementia. Thus, our study provided additional evidence that metabolic abnormality had a greater impact on elevated dementia risk than obesity.

In addition to the baseline BMI-metabolic phenotypes, our study also investigated the associations of changes in metabolic status with incident dementia in populations with and without obesity, which were few examined previously [10, 13]. Machado-Fragua and colleagues constructed five trajectories to identify different transition patterns, but they failed to explore the associations of metabolic status transitions with dementia in individuals with obesity due to the absence of corresponding trajectories [10]. Cho et al. investigated the associations between BMI-metabolic phenotypes transitions and AD among participants aged > 60 years [13], which were different from current study populations. Therefore, our findings of the associations between changes in metabolic status and risk of incident dementia in middle-aged adults are unique. The findings suggested potential risks of accelerated dementia progression when metabolic status transitioned from normal to abnormal in both populations with and without obesity, although they did not reach statistical significance in the main analyses. In contrast, attenuated dementia development was observed in participants who transitioned from MANO to MNNO as compared with stable MANO.

Although the underlying biological mechanism remains unclear, insulin resistance, oxidation, and inflammation pathways could potentially explain the observed associations [42,43,44,45]. For example, both obesity and metabolic abnormality can increase the expression of proinflammatory cytokines [43, 44], which contribute to neurodegeneration and neurotoxicity, leading to the onset of dementia [46,47,48]. Compared with MAO group, individuals with MNO had lower oxidative stress, greater plasma adiponectin concentration, and lower skeletal muscle ceramide content. These factors may confer protection against the higher degree of insulin resistance observed in MAO group [45]. On the other hand, obesity is positively associated with the secretion of nerve growth factors, which might potentially protect against dementia by affecting the cholinergic system [49]. Consequently, the direction of the pooled effect of obesity remains uncertain, which might clarify the observations that a combination of abnormal metabolic conditions may have a greater influence on dementia development than obesity alone, and the risk of dementia for participants with MAO was higher than participants with MNO.

Our findings have several critical clinical and public health implications. First of all, our results suggested that metabolic abnormality plays a more significant role in dementia than obesity, indicating that individuals with abnormal metabolic status are crucial targets for dementia prevention. However, since obesity is a major contributor to abnormal metabolic status, a combination of effective weight management and maintaining a healthy metabolic status is essential for the prevention of dementia. Secondly, it is imperative to incorporate the assessment of metabolic status into standard clinical procedures. Individuals with MANO are frequently overlooked in health management due to their seemingly normal weight, while regular screening for metabolic status can facilitate the early identification of these at-risk individuals, enabling timely interventions that may decelerate the progression of dementia. In addition, given that abnormal metabolic status is reversible and individuals without obesity can decrease the risk of dementia by restoring normal metabolic status, public health initiatives should focus on enhancing metabolic health education among middle-aged adults, promoting consistent metabolic screenings, and ensuring prompt interventions.

The strengths of the current study include long-term follow-up in population-based cohort studies, which enabled us to robustly investigate the associations of obesity, metabolic status, and metabolic heterogeneity of obesity with incident dementia. Since the continuum of dementia includes a long latent phase, our study offered valuable insights into the exploration of early risk factors for incident dementia. Notably, the observed associations remained consistent after excluding individuals who developed incident dementia within five years of follow-up, suggesting that our findings are not merely cross-sectional. Furthermore, this study included three prospective cohorts from different ethnicities, all with rigorous study designs and large sample sizes, which strengthens our findings. The consistency of results across these cohorts underscores the generalizability of our conclusions, and diverse sensitivity analyses further ensured the robustness of the results.

However, our findings should be interpreted with caution due to some limitations. First, we utilized HbA1c as a substitute for fasting glucose due to the limited number of fasting samples in the UKB, which differs from the other two cohorts. Nevertheless, the American Diabetes Association recommendations support using this metric as a reliable stand-in for glucose values [25]. Second, mild dementia may go undetected, potentially resulting in misclassification bias. Notably, previous studies showed acceptable positive predictive values for the defined dementia outcome in the UKB, ARIC, and FOS [26,27,28]. Third, as the cohort studies mainly consisted of a Western population, the generalizability of our findings to diverse cultural and ethnic backgrounds may be limited. Fourth, we were unable to investigate the associations of weight status transitions with the risk of dementia due to the insufficient number of participants who experienced weight changes across the two waves of data collection. Longitudinal studies of a larger scale are still warranted to construct a more complex transition model and clarify the causal associations between transitions in BMI-metabolic phenotypes and dementia. Finally, given the observational design of this study, residual confounding and other non-causal explanations should be considered.