Multimorbidity patterns and 18-year transitions from normal cognition to dementia and death: A population-based study

Abstract


Introduction
Several chronic diseases, such as diabetes and atrial fibrillation, are associated with accelerated cognitive decline and a higher risk of dementia [1][2][3]. However, the role of chronic diseases in cognitive aging has frequently been studied in isolation. Older adults rarely suffer from a single chronic condition; instead, multimorbidity-the coexistence of at least two diseases in the same person-is rather the rule than the exception after age 60 [4]. In addition, diseases tend to co-occur following specific patterns, within which common pathophysiological mechanisms and risk factors can be recognized. Previous studies showed that specific disease patterns (i.e., cardiovascular, cardiometabolic, and neuropsychiatric multimorbidity) have different prognoses, and some of them play a major role in the development of dementia [5][6][7][8].
Dementia is the end stage of a progressive cognitive and functional decline that begins years before and evolves at varying rates across individuals [9]. Understanding the determinants of such heterogeneous trajectories across the whole cognitive continuum would open the way to more effective preventive strategies. The progression from normal cognition to dementia usually implies an intermediate state of cognitive impairment that can be operationalized as cognitive impairment, no dementia (CIND) [10,11]. This stage represents an important window of opportunity to implement preventive and therapeutic strategies, but it also poses challenges as it is a dynamic and heterogeneous condition with up to 30% of individuals remaining cognitively stable over time or even reverting to normal cognition [12]. It is still unknown how different patterns of multimorbidity influence individuals' transitions through different cognitive stages, namely, the development of CIND, the progression from CIND to dementia, or the reversion to normal cognition.
The aim of this study was to investigate the impact of multimorbidity and of different multimorbidity patterns on transitions across cognitive stages and death over an 18-year period.

Study population
For the present study, we used data from the Swedish National study on Aging and Care in Kung-sholmen (SNAC-K) [13]. SNAC-K is an ongoing population-based longitudinal study that started in 2001; it included randomly selected individuals aged 60 years and older either living at home or institutionalized, from the Kungsholmen district of Stockholm. At baseline (2001)(2002)(2003)(2004), 3363 individuals were enrolled (73% participation rate) and followed up regularly over time at 6 years intervals for the younger cohorts (<78-year old) or 3 years intervals for the older ones (≥78-year old).
Written informed consent was obtained from all participants or a proxy in the case of cognitively impaired individuals. The protocol for all waves of the SNAC-K study was approved by the Regional Ethical Review Board in Stockholm.
For the present study, we excluded individuals with dementia at baseline (n = 240) and one individual with intellectual disability, leaving a final sample of 3122 individuals.
The results of this study are reported following the STROBE recommendations.

Data collection
At all the visits in SNAC-K, clinical, laboratory, functional, and cognitive data were collected by trained nurses, physicians, and psychologists. Home visits were carried out for participants unable to travel to the research center.
Covariates. Sociodemographic characteristics of the participants (i.e., age, birth year, sex, and education) were collected through nurses' interviews. Three levels were defined for educational attainment: elementary, high school, and university or higher. DNA was extracted from peripheral blood samples for apolipoprotein E (APOE) allele genotyping. Participants were defined as APOE-ε4 carriers or noncarriers if they had at least one ε4 allele or none, respectively. Mini-mental state examination was used to assess global cognition.
Cognitive impairment, no dementia (CIND) definition. CIND operationalization was based on a neuropsychological test battery [14]. At each wave, five cognitive domains were assessed: executive function (Trail Making Test, Part B), episodic memory (free recall), visuospatial abilities (mental rotations), language (category and letter fluency), and perceptual speed (digit cancellation and pattern comparison). To define cognitive impairment, the raw scores were standardized into z scores, using the age-specific baseline mean and standard deviation (SD). When more than one cognitive test per domain was available (e.g., language), we created a domain-specific score by averaging the z scores across tests. An individual was deemed to have CIND if, in the absence of dementia, he or she scored more than 1.5 SD below the mean of sameaged participants in at least one cognitive domain [15]. This procedure was also used to identify CIND at follow-up visits, using the means and SD defined at baseline.

Dementia diagnosis.
A clinical diagnosis of dementia was made across all waves according to the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, (DSM-IV) criteria, through a three-step procedure [16]. Two preliminary diagnoses were made: the first by the examining physician and the second by a reviewing physician. In case of disagreement between the two evaluations, the final diagnosis was established by a neurologist external to the data collection process (GG, LF). In the case of participants who died between two visits without a dementia diagnosis, further information was obtained via (1) the clinical charts and medical records of the participant who died and (2) the Swedish National Cause of Death Register.
Chronic disease assessment. Diagnoses were established by SNAC-K physicians based on physical evaluation, medical history revision, and participant and/or proxy interview. Additional information was obtained from laboratory tests, clinical parameters, and medication lists. Diseases were coded following the International Classification of Diseases 10th revision (ICD-10). In the present study, we only considered chronic diseases identified at baseline. A more detailed description of the methodology adopted for the definition and operationalization of chronic diseases in SNAC-K has been provided elsewhere [4].
Vital status. Information on the vital status of the participants was obtained from the Swedish Cause of Death Register.

Statistical analysis
Participants were first classified according to the presence/absence of multimorbidity at baseline; out of the 3122 dementia-free individuals, we identified 429 participants with no multimorbidity. Multimorbid participants (n = 2693) were further grouped into multimorbidity patterns according to their disease combinations at baseline using a fuzzy c-means cluster analysis algorithm, following a methodology developed and described in previous studies [5,17]. This algorithm allows the identification of underlying multimorbidity patterns and assigns a membership probability to each participant; participants are then assigned to the cluster with the highest probability of membership. To buffer statistical noise, for the cluster analysis, we included only the 37 diseases that had a prevalence of at least 2% in our study population (Table S1). The choice of the number of clusters was based on statistical and clinical reasoning. Each multimorbidity pattern was subsequently named according to its characterizing diseases, defined as those conditions showing both an observed/expected ratio >2 and an exclusivity >25%. The observed/expected ratio is defined as the ratio between the prevalence of the disease within a cluster and its prevalence in the whole population, whereas exclusivity is computed by dividing the number of individuals with the disease within the cluster over the number of affected participants in the overall population.
To study cognitive transitions, we used multistate Markov models [18]. These models are used to describe how individuals move across different states over time and have been previously applied to the study of the cognitive continuum [19]. A four-state model was used considering three cognitive states (1: normal cognition, 2: CIND, and 3: dementia) and death as an absorbing state (state 4) (Fig. 1). The only reversible transition allowed was the one between CIND and normal cognition. We used age as the time scale, given the increased risk of developing cognitive impairment with aging. First, we calculated transition intensities and hazard ratios (HRs) with 95% confidence intervals (CI) to investigate the association between cognitive transitions and multimorbidity (i.e., yes/no). Then, multistate Markov models were used again to explore, among multimorbid participants, the association between multimorbidity patterns and cognitive transitions and to estimate the total life expectancy and life expectancy in different cognitive stages for each multimorbidity pattern.
Life expectancies were calculated for cognitively intact individuals at age 75 (mean age of the study population), with an average educational level (high school), and were computed separately for men and women. In case of participants alive and free from dementia with missing information about CIND status, the state was censored. All the analyses were adjusted for age, sex, and education.
We conducted sensitivity analyses adjusting for birth year, to account for birth cohort effect, for previous history of CIND, to consider that a past diagnosis of CIND could influence the future development of negative cognitive outcomes and for APOE genotype.
To fit multistate models and estimate life expectancies, the msm [20] and ELECT [21] packages in R were used. Stata version 17 (StataCorp, TX, USA) and R version 4.2.0 (R Foundation for Sta-tistical Computing) were used for the rest of the analyses.

Baseline characteristics
The mean age of study participants at baseline was 73.6 (SD 10.7) years, 63.4% were women, and 83.7% had a high school or university education. Compared with participants without multimorbidity, multimorbid participants were older (mean difference 9.6 years, 95% CI 10.6-8.6), more likely to be females (65.0% vs. 53.4%), and with lower levels of education (17.1% vs. 7.7% with elementary education).
Among participants with multimorbidity at baseline (n = 2693), five multimorbidity patterns were identified: neuropsychiatric (7.3%), cardiovascular (10.2%), sensory impairment/cancer (14.3%), respiratory/metabolic/musculoskeletal (21.5%), and unspecific multimorbidity (46.7%). None of the diseases included in the unspecific pattern was overrepresented compared to the total population. Moreover, as observed in previous studies [5,17,22], individuals in the unspecific pattern had the lowest number of chronic diseases. Table 1 shows the baseline characteristics of multimorbid participants by multimorbidity pattern. Baseline multimorbidity patterns and their characterizing diseases are shown in Table S2.

Multimorbidity and cognitive transitions
In the multi-adjusted model, compared to nonmultimorbid participants, multimorbid individuals showed an increased hazard of progression from normal cognition to CIND (HR 1.45, 95% CI 1.10-1.92), from normal cognition to death (HR 1.46, 95% CI 1.09-1.94) and from CIND to dementia (HR 2.59, 95% CI 1.14-5.88), and a reduced hazard of reversion from CIND to normal cognition (HR 0.75, 95% CI 0.57-0.99) (Table S3). After further adjustments for birth year and previous CIND history, the results remained consistent, whereas adjusting for APOE genotype, the reduced hazard for CIND reversion was no longer significant (HR 0.79, 95% CI 0.60-1.05). Table 2 shows the HRs for transitions among cognitive stages and death for the different multimorbidity patterns. Among multimorbid participants, in the multi-adjusted model, subjects in the neuropsychiatric pattern showed a 47% (95% CI 0.33-0.85) reduced hazard of reversion from CIND to normal cognition compared to those in the unspecific one. In comparison to subjects in the unspecific pattern, those in the cardiovascular one had an increased hazard of progression from CIND to dementia and of death, regardless of the starting state. A 40% (95% CI 0.39-0.91) reduced hazard of returning from CIND to normal cognition was also seen in participants within the sensory impairment/cancer pattern, as compared to those in the unspecific pattern. No significant differences were found in individuals within the respiratory/metabolic/musculoskeletal pattern compared to those in the unspecific one.

Multimorbidity patterns and cognitive transitions
Further adjustments for birth year and history of CIND did not change the results. After adjusting for the APOE genotype, most of the results remained consistent, except for a no longer significant association between the cardiovascular pattern and the transition from dementia to death (HR 1.32, 95% CI 0.95-1.85) (Table S4).  to the unspecific pattern had a life expectancy of 10.1 years (95% CI 9.5-10.7). Life expectancy for a man with the same characteristics in the respiratory/metabolic/musculoskeletal and sensory impairment/cancer pattern was slightly, but not significantly, reduced, whereas the reduction was more relevant for those in the neuropsychiatric (8.4 years, 95% CI 7.3-9.6) and cardiovascular (6.9 years, 95% CI 6.1-7.6) patterns. Compared to members of the unspecific pattern, who lived on average 7.2 years (95% CI 6.7-7.7) with normal cognition, individuals with neuropsychiatric and even more so cardiovascular multimorbidity, showed a reduction of the time spent with normal cognition (5.9 years, 95% CI 5.0-6.8, and 5.2 years, 95% CI 4.5-5.9, respectively) and anticipation of CIND onset. Moreover, in participants belonging to the cardiovascular pattern, compared to the unspecific, we also observed a reduction of the average time spent in CIND with an anticipation of dementia onset (1.3 years, 95% CI 1.0-1.6, vs. 2.3 years, 95% CI 1.9-2.6,) and a reduced survival with dementia (0.4 years, 95% CI 0.3-0.5, vs. 0.6 years, 95% CI 0.5-0.8). Compared to men, cognitively intact women at age 75 with an average educational level had a longer life expectancy in all multimorbidity patterns. As in men, the largest reduction in life expectancy and in average time lived with normal cognition, with an anticipation of CIND onset, was seen among members of the neuropsychiatric and cardiovascular patterns. Women with cardiovascular multimorbidity also experienced earlier dementia onset and reduced survival with dementia.

Discussion
The main findings of our study can be summarized as follows. First, multimorbidity accelerates cognitive decline in older adults, by increasing the risk of both CIND and dementia development and reducing the likelihood of reverting from CIND to normal cognition. Second, specific disease combinations have a different impact on cognitive trajectories, with the neuropsychiatric, sensory impairment/cancer, and cardiovascular patterns showing the worst prognosis and the lowest hazard of reverting from CIND to normal cognition and the highest risk of dementia development. Finally, older adults suffering from neuropsychiatric and cardiovascular multimorbidity present with significantly shorter total life expectancy, along with a shorter period lived with normal cognitive function and anticipation of CIND and dementia onset.
Multimorbidity and several individual chronic diseases have been shown to negatively impact cognition, which reinforces the notion that a somatic component is involved in the development of dementia [15,23]. We here confirmed that multimorbidity is associated with increased cognitive deterioration in older adults as multimorbid participants showed an increased risk of CIND, a reduced likelihood of reverting to normal cognition, and an accelerated progression from CIND to dementia. Looking at all the possible transitions throughout the cognitive continuum, this finding adds weight to the existing knowledge on the detrimental effects of somatic conditions on cognitive health [24][25][26].
Interestingly, we also found that, within multimorbid individuals, different combinations of diseases have a differential impact on cognition. Chronic diseases tend to cluster in the same person following specific patterns, as they may share common pathophysiological mechanisms and/or risk factors [27,28]. Some studies have explored the impact on the cognition of the combination of two or more chronic conditions affecting similar systems, like cardiovascular or cardiometabolic diseases [8]. We here further characterize the clinical profile of persons with multimorbidity by considering a comprehensive number of chronic conditions and multiple disease combinations simultaneously.
Not surprisingly, heart disease had a strong impact not only on cognition but also on survival. Individuals with cardiovascular multimorbidity had indeed a worse clinical prognosis, as they had an anticipation of CIND and dementia onset, an acceleration of disease progression, and a large anticipation of death. This is in line with previous data showing that heart diseases contribute to cognitive impairment and represent one of the main causes of death worldwide [29,30].
Regarding life expectancies, interestingly, we found a quite short time spent with dementia. However, this result should be interpreted with caution and read in the context of how time in each cognitive stage was estimated. Indeed, we estimated time spent with dementia at a population level, on average, for a 75-year-old individual cognitively intact at baseline. Notably, our findings are in line with a previous study using the same methodology and similar study design [31].
The association between neuropsychiatric comorbidities and dementia is well documented; however, the relationship between these two entities is complex and arguably bidirectional. Indeed, although neuropsychiatric diseases increase the risk of cognitive deterioration, behavioral symptoms may also reflect prodromal manifestations of dementia [32]. Interestingly, we observed a reduced survival in older individuals with neuropsychiatric multimorbidity; this finding is supported by previous evidence of an association between mental disorders and mortality [33].
Consistently with a previous study [5], we found a combination of cancer and sensory impairment in the same cluster, which showed a reduced like-lihood of reversion from CIND to normal cognitive function. This finding apparently conflicts with previous studies that observed an inverse correlation between cancer and dementia [34]. However, sensory impairment is associated with an increased risk of cognitive decline, and hearing loss is listed among the modifiable risk factors for dementia [35,36]. Moreover, by adopting multistate models, we were able to properly consider the competing risk of death, which may have influenced the previously observed inverse association between cancer and dementia.
Only a few studies have explored the biological mechanisms that underlie the association between multimorbidity and dementia so far [37]. One of them showed that multimorbidity is associated with greater brain pathological load that does not involve amyloid deposition [38]. Other possible explanations for this association could be that multimorbidity and dementia share some common risk factors (i.e., sedentary lifestyles, smoking, diet, and low socioeconomic status) [39], or that medications used to treat some chronic diseases, such as neuropsychiatric comorbidities, are known to increase dementia risk. Multimorbidity and related disability often determine social isolation, which has a negative impact on cognitive performance [40]. Finally, several chronic diseases are accompanied by low-grade chronic inflammation, and in a previous study, we have shown that dementia risk for specific disease combinations was further elevated by concurrent chronic inflammation and genetic predisposition (APOE) [5]. However, the exact biological pathways through which specific multimorbidity patterns affect cognitive deterioration still need to be elucidated.

Strengths and limitations
This is, to the best of our knowledge, the first attempt to evaluate the impact of multimorbidity and of specific multimorbidity patterns on transitions between cognitive stages across the whole cognitive continuum, additionally accounting for the reversion from CIND to normal cognition and for transitions from each cognitive stage to death.
One of the strengths of our study is that the use of multistate Markov models allowed us to simultaneously model transitions among different states and examine the role of multimorbidity and multimorbidity patterns on all transitions simultaneously. Moreover, the SNAC-K study is a large population-based study with a long follow-up. At each assessment, participants undergo an extensive evaluation that allowed us to obtain comprehensive information on different chronic conditions that were used to operationalize multimorbidity and multimorbidity patterns. Moreover, the diagnosis of dementia is clinical and follows standardized procedures. Of note, we were able to identify dementia cases even among those individuals who died during follow-up by performing a comprehensive review of clinical records and death registers.
Some limitations should also be discussed. First, participants of our study were observed at regular intervals of 3 or 6 years; therefore, it was not possible to know the exact time of CIND and dementia onset. Second, the Markov model relies on assumptions that can have influenced the results of our analyses; indeed, this model assumes that transition intensities remain constant over time and that the future state of an individual only depends on her/his current state, being independent of time and past states. In the attempt to address these issues, we adopted age as the time scale and included age in the model as a covariate to consider the impact of time on cognitive transitions; furthermore, we performed additional analyses adjusting for previous CIND history to account for the fact that a previous diagnosis of CIND may influence the risk of developing negative cognitive outcomes.
Being SNAC-K a population-based study, the diagnoses of cognitive impairment as well as overt dementia are purely clinical, follow standard procedures, and are made without the use of biomarkers. The lack of a biological characterization of dementia and its subtypes may thus represent a limitation of the current study. Future studies are needed to better untangle the impact of disease combinations on different dementia subtypes and to explore the impact of multimorbidity patterns on specific neuropathological processes. Another limitation of the study is that in the patterns' definition, we did not consider disease severity. Finally, because of the low number of individuals experiencing some transitions, we were unable to stratify the analyses by age groups and to explore whether the effect of multimorbidity patterns varies among different age groups.

Conclusions
Our findings confirm that dementia in older adults is a complex disorder with multiple contributors and emphasize the connection between somatic conditions and cognitive aging. Multimorbidity and multimorbidity patterns have an impact not only on dementia development but also on its earlier prodromal stages and on how individuals transition along the whole cognitive continuum. The identification of specific multimorbidity patterns may provide additional information (compared to a simple yes/no operationalization of multimorbidity) in terms of older adults' risk of adverse cognitive outcomes and could be helpful to develop targeted strategies for dementia prevention. Further studies are needed to clarify the biological mechanisms that underlie the relationship between somatic diseases and cognitive decline.

Author contributions
Martina Valletta, Davide Liborio Vetrano, Laura Fratiglioni, and Giulia Grande contributed to the conception and design of the study. Martina Valletta, Xin Xia, and Albert Roso-Llorach conducted the statistical analyses. Martina Valletta conducted the literature search. All the authors contributed to the interpretation of the results. Martina Valletta drafted the first version of the manuscript. All the authors critically revised the manuscript for important intellectual content. All the authors made a significant contribution to the research and the development of the manuscript and approved the final version for publication. SNAC-K personnel collected the data for the study.

Role of the funding source
The funder had no role in study design, data collection, data analyses, data interpretation, or writing of the report.

Data availability statement
The corresponding and the last author had full access to all the data in the study and have final responsibility for the decision to submit it for publication.