The age‐dependent association of Life's Simple 7 with transitions across cognitive states after age 60

Life's Simple 7 (LS7) aims to promote ideal cardiovascular health (CVH). Its association with different cognitive states in the older old is unclear.


Introduction
Dementia is a debilitating clinical syndrome characterized by cognitive decline severe enough to interfere with daily functioning. A body of evidence suggests that poor cardiovascular health (CVH) is closely linked to dementia development and is a potential target for dementia prevention [1].
Life's Simple 7 (LS7) was introduced to define ideal levels of seven modifiable cardiovascular risk factors and promote ideal CVH [2]. Considering the paramount role of CVH in brain health, LS7 was also recommended to promote optimal brain health [3]. LS7-defined ideal CVH in middle and old ages has been associated with a lower risk of dementia [4]. However, the impact of LS7-defined CVH on the development of cognitive impairment, no dementia (CIND), the intermediate stage between normal cognitive aging and dementia, and CIND progression to dementia have yet to be assessed. Furthermore, the clinical relevance of conventional cardiovascular risk factors in cognition in late life has been questioned [5], but whether LS7 is a potentially useful tool to promote cognitive health in the older old (i.e., ≥80 years) has not been investigated.
This population-based cohort study aimed to improve the understanding of the association between LS7 and cognition in old age (i.e., ≥60 years). We hypothesized that LS7-defined better CVH would be associated with a lower risk of the progression from normal cognition to CIND and dementia, with a differential impact in younger versus older old individuals.

Study design and participants
This study used data from the Swedish National Study on Aging and Care in Kungsholmen (SNAC-K), which is an ongoing multidisciplinary project that includes people aged 60 years or older living in the central Stockholm area, Kungsholmen [6]. The SNAC-K study population consists of randomly selected samples from eleven age cohorts: 60, 66, 72, 78, 81, 84, 87, 90, 93, 96, and 99+ years. Between 2001 and 2004, 3363 (73.3%) of the 4590 eligible and invited individuals attended the baseline examination. Then, individuals aged <78 and ≥78 years were followed every 6 and 3 years, respectively. In the baseline and each followup examination, trained physicians, nurses, and psychologists performed comprehensive assessments, including interviews regarding demographics and lifestyle, clinical examinations, and cognitive testing following standard procedures. The whole examination lasted for 5-6 h. Information on the use of medications was collected and recorded according to the Anatomical Therapeutic Chemical (ATC) Classification System. Diseases and symptoms were recorded according to the International Classification of Diseases, 10th revision (ICD-10) [7]. In addition, data from the Swedish National Patient Register and the Swedish Death Register were obtained and linked to the SNAC-K data. In this study, the follow-up data were available until December 2019. The SNAC-K and linkage to register data were approved by the Regional Ethical Review Board in Stockholm, Sweden. Written informed consent was obtained from all participants or a proxy if the participant was cognitively impaired.
For the present study, we excluded people who were diagnosed with dementia (n = 240) or lacked information on dementia diagnosis (n = 10) at baseline from the 3363 eligible participants, leaving 3113 (92.6%) dementia-free participants. Among the 3113 dementia-free participants, 367 withdrew (refused participation, lost contact, or moved) prior to the first follow-up examination. As a result, the analytical sample for examining the association between LS7 and dementia (the first analytical sample) consisted of 2746 (81.7%) dementia-free participants with follow-up information available. Lastly, we excluded participants without any CIND information from baseline to the end of the follow-up period (n = 276), leaving 2470 (73.4%) people in the analytical sample for the associations of LS7 with transitions across cognitive states (the second analytical sample).

Assessment of LS7 metrics
According to American Heart Association's recommendations, we defined and categorized each LS7 component into poor, intermediate, and ideal CVH levels [2], corresponding to scoring 0, 1, and 2. Detailed criteria for the LS7 components and deviations from the criteria are shown in Table 1.
Smoking status was ascertained through a questionnaire administered by nurses (Table S1). Body mass index (BMI) was calculated from weight (kg) and height (m) which were measured in light clothes without shoes. Physical activity was ascertained through a self-administrated questionnaire (Table S1). Diet was assessed with a validated 98item food frequency questionnaire, which collected information on the consumption of a list of 98 dishes and beverages [8]. Total cholesterol was measured from peripheral blood samples. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) in the sitting position were measured twice from the left arm, each after a 5-min rest, in a quiet room with a constant temperature. BP levels were calculated as the mean of the two times' BP measurements. Hemoglobin A1c was measured from peripheral blood samples and used as a proxy for fasting blood glucose. The use of cholesterollowering agents (ATC code: C10), antihypertensive medication (ATC codes: C02, C03, C04, C07, C08, and C09), or antidiabetic medication (ATC code: A10) was ascertained through self-reporting or reviewing drug prescriptions or drug containers. We summed up the scores of the seven components to a total LS7 score. A higher total LS7 score indicates better CVH.

Assessment of outcomes
The operationalization of CIND has been described elsewhere [9]. Briefly, psychologists assessed cognitive function in five domains: episodic memory (free recall), executive function (trail making test, part B), language (category and letter fluency), visuospatial abilities (mental rotations), and perceptual speed (digit cancellation and pattern comparison) [9]. We standardized each test score into a Z score using the baseline mean and standard deviation (SD) and calculated the average Z scores for each domain. CIND was defined as scoring ≥1.5 SDs below the age group-specific mean in at least one cognitive domain but not meeting the criteria for dementia.
All-cause dementia was ascertained via structured interviews (e.g., lifestyle factors and health history), clinical examinations (e.g., chronic health conditions, use of medications, and physical functioning), and cognitive tests (e.g., the mini-mental state examination, the clock drawing test, and the digit span forward and backward tests) according to the Diagnostic and Statistical Manual of Mental Disorders, fourth edition criteria and following a threestep procedure [10]. Briefly, the examining physician made the first diagnosis, and another physician made the second diagnosis, blinded to the first diagnosis, by reviewing all records from interviews and clinical examinations. In the case of discordant first and second diagnoses, a neurologist external to the data collection made a final diagnosis. For the deceased participants without a prior diagnosis of dementia, physicians in the SNAC-K research group made diagnoses by reviewing medical charts. We also identified dementia cases among the deceased participants through linkage to the Swedish Cause of Death Register (ICD-10 codes: F00, F01, F02, F03, F051, and G30).

Assessment of covariates
The highest level of attained education was ascertained by nurses through interviews and categorized into elementary, high school, and university. Social networks were quantified based on network size (e.g., marital status, living arrangement, number of children) and network support (e.g., satisfaction with social connections) measured through nurse interviews [11]. Stroke (ICD-10 codes: I60, I61, I62, I63, and I64) and heart failure (ICD-10 codes: I110, I130, I132, I27, I280, I42,  I43, I50, I515, I517, I528, Z941, and Z943) were diagnosed by physicians in the SNAC-K through interviews, reviewing medical charts, clinical lab parameters, and clinical examination [7]. In addition, extra stroke and heart failure cases were identified through the Swedish National Patient Register [7]. The stage of heart failure was graded following the New York classification of heart failure [12]. Disability status was evaluated through a structured interview investigating activities of daily living, including bathing, dressing, toileting, transferring, feeding, and continence. Functional dependence was defined as dependence in performing one or more activities of daily living. APOE genotyping was done with matrix-assisted laser desorption/ionization time-of-flight analysis on a modified Sequenom MassARRAY platform at Karolinska Institutet.

Statistical analysis
Baseline characteristics of participants in the whole sample and by age groups were described by means and SDs for continuous variables and numbers and percentages for categorical variables. The baseline characteristics of the SNAC-K participants included in and excluded from the study sample were compared using logistic regressions.
A three-state multistate model using the R package "msm" was used to evaluate the relationship between total LS7 scores and dementia risk in the first analytical sample, first in the whole sample and then in younger old adults (<78 years) and older old adults (≥78 years) separately. The age cut-off of 78 years instead of 80 years was chosen following the categorization of the age groups and the follow-up schemes for the age groups in the SNAC-K. Age was used as the time scale, and participants were followed from baseline to death, withdrawal, or end of the study, whichever came first. Models were adjusted for time-varying age, sex, education, and social network scores. We also explored the association between total LS7 scores and dementia by APOE ε4 allele carriership with stratified analysis.
A four-state multistate model was applied to estimate the relationship between total LS7 scores and transitions across cognitive states (i.e., normal cognition, CIND, dementia, and death) in the second analytical sample, first in the whole sample and then in individuals <78 and ≥78 years. To facilitate post hoc comparisons of survival times by CVH levels, we also dichotomized total LS7 scores into worse (total LS7 score: 0-7) and better (total LS7 score: 8-14) CVH categories based on the mean of the total LS7 scores. The median of total LS7 scores generated the same cut-off for dichotomization. We only created these two CVH categories considering the limited number of certain transitions in our study. We first analyzed the association of total LS7 scores and then LS7 categories (better vs. worse CVH) with transitions across cognitive states. We adjusted for the following covariates in the multistate models: time-varying age, sex, ever having CIND before, education, and social network scores.
To better quantify the effects of LS7-defined CVH in younger old age on cognitive states in late life, we predicted cognitive impairment-free survival times (survival times spent in normal cognition) and overall survival times at ages 60, 65, 70, and 75 years by CVH categories using the R package "elect." The survival times for females and males were predicted separately, considering the sex differences in health and survival times [13,14].
We performed the following sensitivity analysis to test the robustness of the associations between LS7 and transitions across cognitive states: (1) repeating the analysis after removing people with stroke or moderate or severe heart failure at baseline as those individuals may have lower BP or lower BMI levels due to the diseases but can be classified as having an intermediate or ideal level of BP and BMI; (2) repeating the analysis after removing people with SBP/DBP <90/60 mm Hg or BMI <18.5 kg/m 2 as those people may be frail rather than having ideal CVH; and (3) removing people with both CIND and dependence in one or more activities of daily living at baseline to explore the potential effects of reverse causation. We used multiple imputations by chained equations to deal with missingness in LS7 and covariates. All statistical analyses were performed with R software version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria). More details of the statistical analysis can be found in the Supporting Information section.

Results
The mean age of participants in the analytical sample for the LS7-cognitive state association was 72.7 (SD, 10.1) years (Table 2). Of those participants, 61.9% were females, more than 85% had an educational level of high school or above, 9.4% had a stroke, and 1.6% had moderate or severe heart failure ( Table 2). The baseline characteristics of participants in the analytical sample for the LS7dementia association were similar to the analytical sample for the LS7-cognitive state association (Table S2). Compared with those excluded from the analysis of the LS7-cognitive state association, those included were generally younger, had higher education levels and social network scores, and were less likely to have a stroke or moderate or severe heart failure (Table S3).

LS7 and dementia
During a mean follow-up period of 9.2 (SD, 4.6) years, 439 (16.0%) out of 2746 people developed dementia. The results from the three-state multistate models showed that the total LS7 score was not associated with dementia risk in the overall sample (Table 3). When stratifying by age groups, a 1-point increment in the total LS7 score was significantly associated with a lower hazard of dementia (adjusted hazard ratio (HR): 0.87 [0.78-0.97]) in younger old adults but not in older old adults ( were comparable in younger old adults (Table 3).

LS7 and transitions across cognitive states
The baseline dementia-free participants with CIND information were followed for an average of 9.7 (SD, 4.4) years. Of the 1833 people who had normal cognition at baseline, 451 (24.6%) developed CIND, and 209 (11.4%) developed dementia. Of the 550 people with CIND at baseline, 138 (25.1%) developed dementia, and 195 (35.5%) experienced reversion from CIND to normal cognition. The exact numbers of transitions across the cognitive states are shown in Tables S4 and S5. LS7 was not associated with any transitions across cognitive states in the overall sample ( transitioning from normal cognition to dementia in younger old adults (Fig. 1A). However, LS7 was not associated with CIND reversion to normal cognition or CIND progression to dementia in younger old adults (Fig. 1A). Neither total LS7 scores nor LS7 categories were associated with any transitions across the cognitive states in older old adults (Fig. 1B).

LS7 and survival times
There was an expansion of cognitive impairmentfree survival times in addition to an expansion of overall survival times in both males and females (Fig. 2).  Fig. 2B).

Sensitivity analysis
The sensitivity analyses did not yield significantly different results from those of our main analyses with regard to the associations of LS7 with transitions across cognitive states (Tables S6-S8).

Discussion
Our study found that better CVH according to LS7's criteria was related to a lower risk of developing CIND and dementia in younger old adults but not in older old adults. LS7 was not associated with CIND reversion to normal cognition or CIND progression to dementia. People with better LS7-defined CVH had an expansion of cognitive impairment-free years of the life of 2-3 years. This is the first study that investigated the associations of LS7 with transitions across all cognitive states along the cognitive continuum and specifically investigated LS7's association with cognitive decline in younger old and older old adults separately.
Several cohort studies have investigated the relationship between LS7 and dementia risk [4], some of which investigated LS7 in middle age [15][16][17][18][19], and others studied LS7 in old age [17,[20][21][22][23][24][25]. All studies supported a protective effect of LS7-defined better CVH against dementia, except a study using data from the UK Biobank, which consisted of a relatively younger population and had a relatively short follow-up period [18], and another study that did not consider the diet component of LS7 [20]. Our study adds to the literature that a higher LS7 score was related to a lower risk of dementia in younger old age but not in older old age. In line with our results, one study showed that LS7 was related to a lower hazard of dementia in both APOE ε4 allele noncarriers and carriers, even though the association reached statistical significance only in APOE ε4 allele noncarriers [25].
The relationship between LS7 and cognitive decline was less studied in previous literature. Few studies have assessed the relation between LS7 and cognitive decline and found that a higher LS7 score was associated with a slower cognitive decline [4]. Consistent with our results, a study using data from SNAC-K showed that such an association only existed in younger old adults but not in older old adults [26]. Only one study investigated LS7 in relation to cognitive impairment and found that ideal CVH was related to a lower incidence of cog-nitive impairment [27], which was consistent with our results. A further notable finding of our study was that LS7 was related to the direct transition from normal cognition to dementia. Direct transitions from normal cognition to dementia may represent dementia cases that had a fast progression course, with intermediate states that might be captured only with a more frequent assessment over time. Our results did not support an effect of LS7 on CIND progression to dementia, possibly due to the limited number of transitions from CIND to dementia, which led to insufficient power of the analysis.
No study has examined the relationship between LS7 and reversion from CIND to normal cognition. However, previous data have shown that people without hypertension and people without a stroke history had a higher probability of reversion from CIND to normal cognition, suggesting that maintaining good CVH might be related to CIND reversion to normal cognition [28]. Our results did not find an association between LS7 and reversion from CIND to normal cognition. Future studies may verify our results and investigate whether individual LS7 factors are more relevant in CIND reversion to normal cognition than the overall LS7.
Poor CVH has previously been related to greater white matter hyperintensity volume, silent brain infarcts, lower gray matter perfusion, and lower brain volume, which may mediate the association between CVH and cognitive function [29,30]. In addition, a recent study showed that LS7 was related to levels of cerebrospinal Alzheimer's disease biomarkers, suggesting a link between better CVH and a lower degree of Alzheimer's disease pathology [31]. Furthermore, poor CVH may increase the risk of dementia by interacting with the β-amyloid burden or promoting β-amyloid and tau deposition, the pathological hallmarks of Alzheimer's disease [32][33][34][35]. These pathways could account for the associations of ideal CVH with a lower risk of CIND and dementia in our study. Notably, the relationship between LS7 and direct transition from normal cognition to dementia was stronger than the relationship between LS7 and incident CIND or CIND progression to dementia. It is possible that people who developed dementia without a prior diagnosis of CIND had experienced CIND but were not detected in our study due to a fast progression from CIND to dementia. Therefore, better CVH may be more related to dementia types with fast progression than dementia types with a gradual progression.
On the other hand, our results did not support the role of LS7-defined ideal CVH in cognition in older old adults. There are several reasons that LS7-defined ideal CVH may not be as meaningful for promoting brain health in older old adults as in younger old adults. First, LS7 is primarily designed for the primordial prevention of cardiovascular diseases; its application in older adults is hampered by the fact that cardiovascular risk factors and cardiovascular diseases in old age are prevalent [36]. Second, not all LS7 components are relevant in terms of cognitive aging across the lifespan. For instance, a recent study showed that, while diabetes consistently predicted dementia risk from middle to late life, higher SBP did not predict dementia risk after 65 years [37]. This means that age may be crucial when defining CVH profiles to promote brain health. Recently, American Heart Association introduced Life's Essential 8, which includes sleep duration as an additional CVH component [38]. Given that sleep duration has been associated with dementia risk [39], future studies comparing old-age Life's Essential 8's and LS7's associations with cognitive decline and dementia are warranted. Third, the ideal levels of conventional cardiovascular risk factors with regard to maintaining cognition in older old adults have been questioned previously. Elevated levels of BP, BMI, and fasting glucose in late life have been associated with better cognition [5]. Therefore, LS7's criteria of ideal CVH may not apply to older old adults in terms of combating cognitive decline and dementia. Fourth, the reasons that the cardiometabolic factors in LS7 meet the criteria for being intermediate or ideal may be heterogeneous in older adults. For instance, lower levels of BP in late life can result from pre-terminal BP decline, which has been associated with heart failure and dementia [40]; lower BMI levels in late life may result from cerebrovascular disease pathology or Alzheimer's disease pathology [41].  [42]. This implies that defining ideal levels of CVH components in late life without taking the history of the factors in earlier life into account may be inappropriate. Lastly, previous data showed that in the older old, although many risk factors, including cardiovascular risk factors, seem less influential for dementia development, age remains a strong risk factor for dementia [43,44]. It is possible that in the older old, the aging process is a major contributor to dementia development, making the effects of those cardiovascular risk factors less relevant [45].

Study strengths and limitations
The strengths of our study include the availability of information from the older old, comprehensive measurements of the LS7 components, the use of a comprehensive neuropsychological battery to define CIND, the use of multistate modeling to estimate transitions across cognitive states, and the prediction of clinically meaningful quantities such as survival time in different cognitive states.
Several limitations need to be acknowledged. First, operationalizations of physical activity deviated from American Heart Association's recommendations. Second, there were sparse cases of some transitions, for instance, transitions from normal cognition to dementia and from CIND to dementia, which can impact the power of our analysis. Third, there may be residual confounding from unmeasured confounders (e.g., neurodegeneration) and imperfectly measured confounders (e.g., educational level). Fourth, selective survival is a common problem in dementia research, which may also lead to underestimating the effects of LS7 on cognition. Lastly, the study population consists of people with relatively higher socioeconomic status than people in other regions in Sweden, limiting the generalizability of our study results to other populations.

Conclusions
In conclusion, our study suggests that maintaining ideal CVH according to LS7's criteria may protect against cognitive decline but only in younger old adults, postponing the onset of both CIND and dementia and increasing years of life lived without cognitive impairment substantially. The relevance of ideal CVH recommended by LS7 in older old adults needs further investigation.

Author contributions
Xin Xia, Chengxuan Qiu, Laura Fratiglioni, and Davide L. Vetrano contributed to the conception of the work. Xin Xia and Davide L. Vetrano contributed to the design of the work. Laura Fratiglioni, Giulia Grande, and Erika J. Laukka contributed to the data acquisition. Xin Xia, Debora Rizzuto, and Jie Guo contributed to the data analysis. All authors contributed to the interpretation of the study results. Xin Xia drafted the manuscript. All authors gave a critical review of the manuscript and final approval of the manuscript.

Acknowledgments
We would like to thank all SNAC-K participants and SNAC-K staff for their tremendous contributions to this work. We also thank Dr.