Introduction

Heart failure (HF), defined as the mismatch of supply and demand of cardiac perfusion, is recognized as a major health burden in the world [1]. The relationship linking HF and dementia has been established [2,3,4,5].

In addition to an aging society, it is also important to anticipate the onset of dementia [2,3,4,5,6]. Nonetheless, a score validated for dementia risk discrimination in patients with HF is currently unavailable. The existing CHA2DS2-VASc and AHEAD scores have been validated for risk stratification of thromboembolism and mortality in the HF population [7,8,9,10]. It would be novel to compare the CHA2DS2-VASc and the AHEAD scores in this HF patient population for stratifying dementia risk to see if the more recent scoring strategy is beneficial. Hence, we used such a large sample size from this huge Taiwanese national data base with adjusting for the covariates that are the familiar risk factors for dementia to examine the predictive performance of the existing two scores.

Methods

Data source

This retrospective cohort study was conducted using Taiwan’s National Health Insurance Research Database (NHIRD). The National Health Insurance (NHI) program was implemented in Taiwan since 1995, and it covers nearly 99% of all residents [11]. For this study, we used a subset of the NHIRD containing health care data including files of inpatients claims, and Registry of Beneficiaries. These files record the disease based on International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). The Research Ethics Committee of China Medical University and Hospital in Taiwan approved the study (CMUH-104-REC2-115).

Sampled participants

After examination of the medical claims, 387,595 patients with a diagnosis of HF (ICD-9-CM code 428) were identified from inpatients claims during 2000–2011. The dates of their first hospitalization diagnosis of HF were defined as their index dates. Patients with a history of dementia (ICD-9-CM codes 290, 294.1 and 331.0) prior to the index date, missing information for age or gender were excluded. The CHA2DS2-VASc score [12, 13] and AHEAD score (9, 10) were calculated for each patient to measure dementia risk. Preexisting comorbidities with hyperlipidemia, hyperthyroidism, sleep disorder, gout, chronic obstructive pulmonary disease (COPD), head injury, depression, and alcoholism-related disease were included for analyses. Follow-up began from the index date and continued until a diagnosis of dementia, withdrawal from the NHI program, or at the end of 2011.

Statistical analysis

We calculated the number and percentage for categorical variables and the mean and corresponding standard deviation (SD) and median (Q1–Q3) for continuous variables (including age, CHA2DS2-VASc score, AHEAD score, and follow-up period). The Kaplan–Meier method was applied to estimate the cumulative incidence of dementia stratified by CHA2DS2-VASc score or AHEAD score, and the Log-rank test was used to examine the statistical significance of the differences of CHA2DS2-VASc score or AHEAD score. The incidence density rate (per 1000 person-years) of dementia was estimated according to the CHA2DS2-VASc score and the AHEAD score. Univariable and multivariable Cox proportional hazard models were performed to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) of dementia associated with HF stratified by CHA2DS2-VASc score or AHEAD score. The multivariable model for CHA2DS2-VASc score was simultaneously adjusted for atrial fibrillation, hyperlipidemia, COPD, hyperthyroidism, sleep disorder, gout, chronic kidney disease, anemia, head injury, depression, and alcoholism-related disease. The multivariable model for AHEAD score was simultaneously adjusted for gender, CVA or TIA, vascular disease, hypertension, COPD, hyperlipidemia, hyperthyroidism, sleep disorder, gout, head injury, depression, and alcoholism-related disease. The proportional hazard model assumption was also examined using a test of scaled Schoenfeld residuals. In the model evaluating the dementia risk throughout overall follow-up period, results of the test revealed a significant relationship between Schoenfeld residuals for CHA2DS2-VASc score and follow-up time, and for AHEAD score and follow-up time, suggesting that the proportionality assumption was violated (p value = 0.047 and p value < 0.001). In the subsequent analyses, we stratified the follow-up duration by median (< 2, ≥ 2 years) to deal with the violation of proportional hazard assumption. The area under the curve of receiver operating characteristics (AUROC) was used to assess the predictive accuracy of CHA2DS2-VASc score and AHEAD score in predicting dementia. The DeLong test was used to examine the difference between the two scores. All statistical analyses were performed using the SAS package (Version 9.4 for Windows; SAS Institute, Inc., Cary, NC, USA). A two-tailed p value < .05 was considered significant.

Results

Table 1 displays the demographic characteristics and comorbidities of the HF patients. The mean age among HF patients is 72.0 (SD=13.7) years. More patients are male (n = 201,548; 52.0%), and hypertension is the most prevalent comorbidity (60.8%). The mean CHA2DS2-VASc score and AHEAD score are 4.33 and 1.63, respectively. The mean follow-up of dementia is 2.91 years.

Table 1 Baseline characteristics of heart failure patients
Full size table

The incidence of dementia increases from 0.26% to 3.84% when a CHA2DS2-VASc score increases from 1 to ≥ 6 (Fig. 1a) or increases from 0.80% to 2.96%, while an AHEAD score increases from 0 to ≥ 3 (Fig. 1b). The incidence density rate of dementia increases from 0.64 per 1000 person-years for HF patients with a CHA2DS2-VASc score of 1 to 18.4 per 1000 person-years for those with a CHA2DS2-VASc score of ≥ 6 (Table 2). The risk of dementia increases from 3.22 (95% CI = 2.37–4.37) in HF patients with a CHA2DS2-VASc score of 2 to 25.3 (95% CI = 18.9–33.8) in HF patients with a CHA2DS2-VASc score of ≥ 6 as compared with those with a CHA2DS2-VASc score of 1 (trend test, p < .001). The risk of dementia is still higher in HF patients with a CHA2DS2-VASc score of 2 to >= 6 compared to those with a CHA2DS2-VASc score of 1 stratified by the follow-up duration (median follow-up duration < 2 years and median follow-up duration ≥ 2 years). The incidence density rate of dementia increases from 1.81 per 1000 person-years in HF patients with an AHEAD score of 0 to 15.3 per 1000 person-years in HF patients with an AHEAD score of 3 or above. Compared to patients with HF with an AHEAD score of 0, the corresponding adjusted hazard ratios (aHRs) of dementia are 4.54 (95% CI = 4.10–5.03), 5.58 (95% CI = 5.04–6.19), and 6.12 (95% CI = 5.50–6.82) for those with an AHEAD of 1, 2, and ≥ 3, respectively (trend test, p < 0.001). The risk of dementia is still higher in HF patients with an AHEAD score of 1 to >= 3 compared with those with an AHEAD score of 0 stratified by the follow-up duration (median follow-up duration < 2 years and median follow-up duration ≥ 2 years).

Fig. 1
figure 1

The incidence of new-onset dementia continuously correlates with CHA2DS2-VASc score (a) and AHEAD score (b) in patients with heart failure who need hospitalization

Full size image
Table 2 Incidence and hazard ratios (HRs) of dementia in heart failure patients according to CHA2DS2-VASc and AHEAD scores
Full size table

In the subgroup of HF patients with AF, compared to patients with a CHA2DS2-VASc score of 1, the corresponding aHRs of dementia are 4.37 (95% CI = 2.00–9.53), 13.5 (95% CI = 6.38–28.6), 21.7 (95% CI = 10.3–45.7), 25.7 (95% CI = 12.2–54.1), and 31.0 (95% CI = 14.7–65.2) for those with a CHA2DS2-VASc score of 2, 3, 4, 5 and ≥ 6, respectively (trend test, p < 0.001) (Table 3). Compared to HF patients with AF with an AHEAD score of 1, the corresponding aHRs of AF are 5.47 (95% CI = 4.44–6.73), and 6.14 (95% CI = 4.96–7.60) for those with an AHEAD score of 2, and ≥ 3, respectively (trend test, p < 0.001).

Table 3 Incidence and hazard ratios (HRs) of dementia in heart failure patients with a history of AF according to CHA2DS2-VASc and AHEAD scores
Full size table

The cumulative incidences of dementia associated with increasing CHA2DS2-VASc score and AHEAD score over time are clearly demonstrated by Kaplan–Meier curves in Fig. 2a, b (Log-rank test p < 0.001). The AUROC for CHA2DS2-VASc score in predicting dementia (0.61, 95% CI = 0.60–0.61) is significantly higher than the AHEAD score (0.55, 95% CI = 0.54–0.55) (DeLong test p < 0.001) (Fig. 3).

Fig. 2
figure 2

Cumulative incidence curves of new-onset dementia stratified by CHA2DS2-VASc score (a) and AHEAD score (b) in patients with heart failure who need hospitalization

Full size image
Fig. 3
figure 3

Receiver operating characteristic (ROC) curve for CHA2DS2-VASc score and AHEAD score in predicting new-onset dementia in patients with heart failure who need hospitalization

Full size image

Discussion

We performed a large-scale nationwide cohort investigation based on the inpatients claims and Registry of Beneficiaries provided by the National Health Research Institutes of Taiwan to identify the association between comorbidities (estimated by two scores), and incident dementia in a total of 387,595 patients hospitalized for HF.

The strength of this article appears to be that the predictive values of the CHA2DS2-VASc and AHEAD scores have not been tested in patients with HF using a large-scale data base providing a large study power in exchange for loss of granularity. This study is part of the growing literature on the association between HF and dementia.

Compared with serious acute medical problems such as stroke and heart attack, dementia can be underestimated or unnoticed by patients or physicians, since such illness is relatively not life-threatening. However, HF patients who have dementia are more likely to have untoward physiologically and psychologically sequelae [2,3,4,5,6]. Indeed, there is an increasing body of evidence that HF is associated with cognitive decline and aging brain diseases [2,3,4,5,6, 14,15,16,17]. In the light of the data obtained, the implication of this study is that patients with HF have much higher rates of dementia, and those who have dementia tend to have more aggregate comorbidities. The CHA2DS2-VASc score, originally validated for thromboembolic stroke prediction risk in patients with AF [18,19,20], seems to outperform the AHEAD score for dementia risk discrimination in this population.

Interestingly, the moderate predictive ability of the CHA2DS2-VASc and AHEAD scores seems to be similar in patients with HF, although the former has a better stratification power based on the DeLong test. The potential mechanism for the observed difference of the predictive ability of the exist two scores might be related to the components of the scoring systems since more atherosclerotic factors are captured in the CHA2DS2-VASc as opposed to the AHEAD score [7,8,9,10]. Considering that the CHA2DS2-VASC score considers many cardiovascular risk factors, the results might simply reflect the impact of such variables on the onset of dementia, an aspect which has been widely stated in the literature. Further studies to explore other new scoring schemes including the possible risk factors not captured in CHA2DS2-VASc or AHEAD scores might be helpful in dementia risk stratification, and make a clinically important distinction in this group in terms of the modest predictive value of the existing scores.

Limitations

First, no information about drug therapy or clinical conditions of the patients are reported, which may affect the results of the study. Thus, several potential confounders are unmeasured, and the possibility of incomplete statistical analysis cannot be excluded. Second, even if a total of 387,595 HF patients were recruited, this is a retrospective nationwide cohort study using the NHIRD of the Taiwan NHI Program, and the retrospective nature of this registry study is subject to certain limitations. Finally, the completeness of ICD coding for the diseases is another major concern. That is, missing or incomplete coding might affect the interpretation of the results. Nevertheless, this particular national database has been previously validated and reported to be of high accuracy.

Conclusions

The CHA2DS2-VASc score appears to be more predictive of dementia than the AHEAD score in patients with HF who need hospitalization. The modest predictive value of both scores implies that other sophisticated models might be needed for dementia risk stratification in this population.