Predicting dementia in primary care patients with a cardiovascular health metric: a prospective population-based study

The number of people with dementia is expected to increase due to the aging of the global population [1]. Along with this increase come high and rising costs for care, rendering dementia a worldwide public health problem [2]. In the absence of pharmacological treatment that goes beyond symptom delay or reduction, the study of modifiable risk factors for dementia has high priority [3]. Converging evidence points to a central role of cardiovascular risk factors such as physical inactivity, smoking, diabetes mellitus, midlife hypertension, midlife obesity, and hyperlipidemia in late-life cognitive decline [4].

Several studies suggested that substantial fractions of dementia cases could be prevented if modifiable cardiovascular risk factors were targeted by intervention [5‐7]. Evidence-based prevention programs, however, have only recently been developed [8] and are far from implementation on the population level [3]. Primary care practices might offer a suitable setting for delivering preventive measures, as they provide the facilities to assess and control cardiovascular risk factors for dementia in the older general population. In this context it would be desirable to have a tool for the assessment of cardiovascular health (CVH) that can be easily employed by general practitioners (GPs) and that facilitates the communication between patient and physician.

A range of CVH-metrics have been employed to predict cognitive decline and dementia [9]. Among the more recent is the Life’s Simple 7 metric that was developed by the American Heart Association (AHA) [10]. The Life’s Simple 7 assess CVH based on seven parameters: Smoking, body mass index (BMI), physical activity, dietary habits, total cholesterol, blood pressure, and fasting glucose. Cut-offs are applied to each parameter to categorize the patients CVH-status as poor, moderate, or ideal. Based on these classifications two sub-score for health behaviors and blood parameters, as well as a total score can be calculated. The simple three-step approach has illustrative value and has been combined with a traffic-light system in primary care to improve patient-physician communication and track changes in CVH as an effort to prevent stroke in primary care [11].

So far, however, only few studies investigated the relationship between the Life’s Simple 7 and cognition. Low scores indicating poor CVH were associated with decreased performance on measures of cognitive functioning in a cross-sectional investigation [12] and predicted incident cognitive impairment in previously unimpaired and stroke-free persons [13]. Furthermore, high scores indicating good CVH in young adulthood were related to better cognitive functioning in mid-life [14]. To the best of our knowledge, the Life’s Simple 7 have not been used to predict dementia diagnosed according to clinical criteria. Also, no study so far examined whether a CVH-metric can be employed to assess dementia risk in primary care, where most older people receive their health care and where dementia prevention programs are likely to be implemented.

As age is the most important predictor for dementia, CVH-metrics need to provide additional information to be useful. Yet, this characteristic cannot always be assumed. A commonly employed CVH-metric, for example, was found to lose all predictive validity for dementia death when age was included in the model [15].

The aim of our study was to determine the potential of a CVH-metric to identify individuals with increased cardiovascular risk of dementia at routine primary care visits. Since the data of the present study were collected before the Life’s Simple 7 was introduced, it was not possible to fully adhere to the criteria proposed by the AHA. Instead, the association between scores on a CVH-metric whose construction was based on the Life’s Simple 7 and incident dementia was examined in a large sample of older primary care patients. In particular, it was investigated whether behavioral variables, blood parameters, or a combination best captured the cardiovascular risk for dementia and whether possible associations are independent of age, sex, and educational level.

At baseline, 3908 participants were examined. Three-hundred sixty-one participants were excluded from the analysis (Fig. 1). The baseline characteristics of the included 3547 persons are displayed in Table 2. The median observation time was 6.7 years. In total, 296 (8.3 %) new cases of dementia occurred over the course of the study. Most diagnoses (N = 170, 57.4 %) pertained to “dementia not otherwise specified” (ICD-10 F03) so that no sub-analysis for different types of dementia was performed.

Over the course of the study, 388 patients were lost due to death and 50 due to a change of the insurance company. For all other cases information about incident dementia was available until the study end. That is, for merely 50 participants there is only limited information about incident dementia (from start of the observation until change of insurance).

The proportional hazards assumption was checked by visual inspection of Kaplan-Meier estimator plots and found to be valid. Table 3 displays the results of both the unadjusted and adjusted Cox proportional hazards regression models for the individual components. In the unadjusted model, moderate and poor levels of physical activity, as well as poor levels of fasting glucose were statistically significantly associated with an increased risk of dementia. Poor levels of BMI, however, were associated with a decreased risk. Adjusting for age, sex, and education revealed significant associations of physical activity and fasting glucose with dementia that were similar to the unadjusted analyses, yet slightly decreased in strength. Poor smoking behavior (i.e., current smoking) was now associated with a significantly increased risk of dementia, possibly due to the fact that most smokers in the sample were of younger age. Poor BMI status was not anymore a significant predictor, indicating a stronger relationship of BMI with demographic variables than with dementia. All other models remained unchanged by the adjustment.

Table 4 displays the unadjusted and adjusted Cox proportional hazards regression models for the behavioral index, the blood parameters index, and the total score. None of the scores was able to predict dementia, neither with nor without adjustment. Sensitivity analyses with varying definitions of the clinical endpoint generally confirmed the previously established associations and, thereby, support the validity of the original endpoint. Using the more liberal endpoint (i.e., diagnosis only in at least one billing quarter or setting) most associations between CVH-variables and dementia remained unchanged in strength and direction. Counting the 157 cases with only one diagnosis as non-dementia cases most associations decreased in strength but retained their direction.

The present study investigated whether a CVH-metric combining health behaviors and blood parameters could be used to identify primary care patients at cardiovascular risk of dementia and, hence, in need of medical intervention. In sum, the results do not support the suitability of the employed CVH-metric for a valid assessment of the cardiovascular risk of dementia in older primary care patients.

Employing a CVH-metric that combines several factors to assess dementia risk is only justified if the sum scores carry predictive value that goes beyond the information gained from the individual components. Also, the metric should add information about dementia risk independently of age and other demographic variables. Both assumptions were not found to hold true for the CVH-metric employed in the study at hand. Only the individual components smoking, physical activity, and fasting glucose were significant predictors. Sub-scores for health behaviors and blood parameters, as well as a total score were not able to capture the cardiovascular risk of dementia. With regard to the research question it can be concluded that the employed metric is not suited to assess the risk of dementia attributable to cardiovascular disease in a demographically heterogeneous sample of primary care patients.

Inspecting the demographically adjusted associations between the individual CVH-components and dementia reveals the problems underlying the sum-scores. Current smoking, sedentary behavior, and impaired fasting glucose were scored as reflecting poor CVH and showed the expected relationship with an increased dementia risk. Contrary, poor levels of BMI, blood pressure, and total cholesterol were actually related to a (statistically non-significant) decreased risk of dementia. By adding the individual components into sum scores, they likely cancelled each other out and rendered the derived scores useless. That is, the sum scores actually masked the valuable predictive information that is included in the single components.

The problems encountered with the CVH-metric as tool to assess the cardiovascular risk of dementia reflect emerging evidence that suggests non-linear relationships of blood pressure and BMI with dementia where age acts as a moderating factor. While high blood pressure at mid-life seems to increase the risk of dementia, the association appears to reverse at late-life, when higher levels are associated with a decreased risk [23]. The results of the study at hand support this notion. Similarly, overweight and obesity at mid-life are assumed to increase the risk of dementia [24], while a higher BMI in late-life seems to be associated with a decreased risk [25, 26]. In addition, the relationship of hypercholesterolemia with dementia is somewhat ambiguous and requires further investigation [27, 28], yet, the condition would be expected to relate to an increased risk [29]. In the study at hand, no relationship was found. The ordinal three-step conceptualization of the CVH-metric (poor, moderate, ideal) appears unsuited to capture these complex associations of blood pressure, BMI, and cholesterol with dementia. As a consequence, a CVH-metric that might be well suited to predict, for example, mortality [30] is likely not applicable for the assessment of the cardiovascular risk of dementia.

Reverse causality also needs to be considered when interpreting the results. It is possible that preclinical dementia actually causes weight loss and a decrease in blood pressure, which can be observed years before a clinical diagnosis of dementia is made. In this case, hypotension and underweight would rather constitute early consequences of incipient dementia than risk factors. These potential associations might account for the above described interaction of BMI and blood pressure with age in the pathogenesis of dementia. When administered at only one point in time at higher age, CVH-metrics would not be able to capture these decade-long processes and red-flag potential underlying cognitive decline. Instead, CVH-metrics needed to be administered repeatedly starting at mid-life in order to track, for example, progressive weight loss that might indicate incipient dementia.

The CVH-metric employed in the present study slightly deviates from the Life’s Simple 7 with regard to the employed variables and cut-offs. Consequently, the findings may not fully pertain to the Life’s Simple 7. The most important difference was that diet was not included. Even though a healthy diet seems to be associated with a decreased risk of dementia [31, 32], previous studies found no or only weak associations between diet and cognition in the context of the Life’s Simple 7 [12‐14]. Therefore, it seems unlikely that the omission of diet significantly decreased the predictive validity of the CVH-metric. Also, physical activity was defined as vigorous activity, hence, not taking into account light or moderate activity.

Previous studies [12‐14] reported stronger associations between the Life’s Simple 7 total sum score and cognitive impairment than the association between the CVH-metric and dementia. Next to differences in the components of the metrics, these studies employed measures of specific cognitive functions as outcome variables. Contrary, in the present study a clinical diagnosis of dementia was used. It is possible that others studies found associations between poor general CVH and more subtle cognitive changes that would not suffice to justify a diagnosis of dementia.

If the goal is to develop a CVH-metric with high predictive validity, it should only include components that show unambiguous associations with dementia. Smoking [33, 34], sedentary behavior [35, 36], and hyperglycemia [37], are well-known to substantially contribute to late-life cognitive decline. Focusing on these variables might offer a starting point for the assessment and reduction of the cardiovascular risk of dementia. Metrics could also be used to monitor improvements in CVH. In the Life’s Simple 7, improved CVH would be reflected in increasing scores. It remains to be investigated, however, whether improving scores actually translate into a decreasing dementia risk.

The strengths of the present study include that CVH was assessed at routine primary care examinations by the participants’ usual GPs. The results, therefore, reflect current possibilities to monitor and control cardiovascular risk factors in the health care system. Only for a very small number of participants there was incomplete information about incident dementia available. This fact reduces the possibility of bias due to non-random drop-out. These diagnoses were made in in- and outpatient settings according to usual clinical practice and, thereby, increase the study’s relevance and validity in the health care system. Some limitations had to be accepted. The use of ICD-codes as clinical endpoints may bear the risk of falsely classified cases. In order to reduce the possibility of overstated results due to false positives and increase the validity of the endpoint, more conservative criteria were applied to the outcome variable and sensitivity analyses were conducted. False negative cases also might have been an issue, assuming that dementia is underdetected in community-dwelling persons. However, dementia case definitions based on health insurance claims data have been shown to be sufficiently valid [21, 22]. Two-thirds of the eligible patients did not enroll in INVADE. Given that the study was aimed at the identification and treatment of cardiovascular disease, it is possible that more health-conscious people decided to participate. Potentially, these persons showed better CVH than the general older population, which might somewhat reduce the generalizability of the study’s findings.

The study at hand was the first to examine the suitability of a CVH-metric to assess the cardiovascular risk of dementia for older persons at routine primary care examinations. The metric’s sum-scores were not associated with dementia risk and, therefore, do not seem to be suited for that purpose. Due to complex relationships of BMI and blood pressure with dementia, individual components cancel each other out and render the sub- and sum-scores useless. CVH-metrics that are suitable for other purposes cannot be readily applied for assessing the risk of dementia due to cardiovascular disease on the population level.

AHA, American Heart Association; AOK, Allgemeine Ortskrankenkasse; BMI, body mass index; CI, confidence interval; CVH, cardiovascular health; DBP, diastolic blood pressure; GP, general practitioner; HR, hazard ratio; IBM, International Business Machines Corporation; ICD-10, International Classification of Diseases and Related Health Problems 10th revision; INVADE, Intervention Project on Cerebrovascular Disease and Dementia in the District of Ebersberg; SBP, systolic blood pressure; SPSS 22, Statistical Package for the Social Sciences version 22