Risk of ischemic stroke in patients with hypertrophic cardiomyopathy in the absence of atrial fibrillation – a nationwide cohort study

Patient characteristics

There were 17,371 patients who met the study inclusion criteria and 17,371 matched AF subjects without HCM treated as the reference group. The median follow-up time was 7.3 years. Clinical and demographic characteristics were listed in Table 1 and there was no significant difference between two groups. The mean age was 61 years and around 45% subjects were women. Nearly 70% subjects had hypertension, 25% had diabetes and 40% had hyperlipidemia. However, there were significantly more prescriptions of antiplatelet and ACEI/ARB in HCM patients than those of reference group.

Outcome of ischemic stroke

The overall incidence of ischemic stroke was 4.9% of HCM patients without AF and 4.5% of reference group during a median follow-up period of 7.3 years (25–75%, 5.2–9.3 years). Predictors of IS in patients with HCM without documented atrial fibrillation were age, CHA₂DS₂-VASc and sudden cardiac death in multivariate analysis model (Supplementary Table 1). In both groups, we classified them into 4 different subgroups of age between 20-39, 40-64, 65-74 and above 75 years. We observed progressive increased annual ischemic stroke rate (100-person year) in HCM subjects (20-40 years, 0.167, 95% CI 0.079-0.208; 40-65 years, 0.372, 95% CI 0.271-0.398; 65-75 years, 0.743, 95% CI 0.687-0.792 and >75 years, 1.048, 95% CI 0.942-1.216) (Table 2). There was a similar trend of annual risk of ischemic stroke in matched reference group (20-40 years, 0.114, 95% CI 0.075-0.213; 40-65 years, 0.389, 95% CI 0.285-0.456; 65-75 years, 0.575, 95% CI 0.432-0.684 and >75 years, 0.528, 95% CI 0.473-0.625). Overall, treated the general population with AF as reference group, the relative risk of ischemic stroke for HCM patients was comparable (HR 0.965, 95% CI 0.854 – 1.091, p=0.215). However, the relative risk of ischemic stroke in different age groups showed continuously increased from younger to older age group strata. There was no difference of risk of ischemic stroke among subjects with age 20-40 years (HR 0.595, 95% CI 0.151 – 2.346, p = 0.712), and 40 – 65 years (HR 0.642, 95% CI 0.151 – 2.346, p = 0.712). By contrast, there was an significantly increased risk of ischemic stroke for HCM subjects with age 65 – 75 years (HR 1.278, 95% CI 1.070 – 1.335, p = 0.025) and > 75 years (HR 1.757, 95% CI 1.435 – 2.152, p = 0.012) when compared with matched general population with AF. The results of subdistribution Cox proportional model also demonstrated that HCM patients without AF in the older age group were associated with a higher risk of ischemic stroke during the follow-up period than the matched controls (Supplementary Table 2). In the sensitivity analyses, we stratified our cohort into 6 subgroups on the basis of age with the same interval of 10 years (age 25–34, 35–44, 45–54, 55–64, 65–74 and > 75 years). The major finding remained unchanged in the sensitivity analysis (Supplementary Table 3). The Kaplan-Meier survival curves were illustrated in Figure 1. The overall cumulative incidence curves showed no significant different between HCM patients without AF and matched controls (log-rank test, p = 0.157) (Figure 1A). However, the cumulative incidence curve with the log-rank test demonstrated that patients in the older age subgroup were associated with a higher risk of ischemic stroke during the follow-up period (age 20–44 years, p=0.445; age 45-64 years, p=0.993; age 65–74 years, p=0.005 and age ≥ 75 years, p<0.001) (Figure 1B–1E).

We also stratified our cohort according to CHA₂DS₂-VASc score into 5 groups (score 0, 1, 2, 3, 4 and ≥ 5). Compared with matched general population, HCM subjects had significantly increased risk of ischemic stroke in those with CHA2DS2-VASc score ≤ 2 (score 0, HR 2.062, 95% CI 1.816 – 4.209, p value 0.026; score 1, HR 2.261, 95% CI 1.940 – 3.300, p value 0.018; score 2, HR 1.383, 95% CI 1.023 – 2.201, p value 0.034) (Figure 2). However, those with score ≥ 3 had similar risk of ischemic stroke compared with control.

Limitations

There are several limitations to this retrospective cohort database study. First, baseline comorbidities and estimation of incidence of ischemic stroke were on diagnostic codes registered by the physician. Although administrative databases are increasingly used for clinical research and high concordance between claims records in the NHIRD and patient self-reports at surveys, our study is still potentially susceptible to errors arising from not completely coding all disease. However, the large size of the database in this study should be sufficient to reach an accurate statistical conclusion [20]. Second, the conventional risk of AF and ischemic stroke in HCM patients was not recorded, such as left ventricle wall thickness, left atrial diameter and continuous electrocardiogram monitoring [3]. Therefore, we cannot investigate the impact of these parameters on the incidence of stroke. Third, given the conceived better prognosis of elder HCM patients, the physician would less aggressively investigate HCM-related diseases. As a result, AF might be under-diagnosed in clinical practice. Fourth, our study enrolled only Taiwanese participants, and we do not know whether our result could be extrapolated to non-Asian populations; therefore, further study should be conducted.

Source of data

This large-scale, longitudinal cohort study used integrated medical and pharmacy claims data from National Health Insurance Research Database (NHIRD) in Taiwan. The National Health Insurance program has provided compulsory universal health insurance in Taiwan since 1995. More than 98% of the total Taiwanese population of 23 million is covered by the program [21]. The NHIRD contains nearly complete claims history of diagnosis and procedures, provided as the International Classification of Diseases Ninth Revision Clinical Modification (ICD-9-CM) codes, and drug dispensing for every beneficiary. The Bureau of National Health Insurance performs routine validations of the diagnoses by reviewing the original medical charts of all of the patients who applied for catastrophic illness registration. To comply with data privacy regulations, personal identities were encrypted and all data were analyzed in a de-identified manner. The protocol for this study was approved by the Institutional Review Board of National Taiwan University Hospital.

Study population and outcomes

We investigated the database of NHIRD during year of 1996 to 2013. The index date for the study cohort was identified as the date of the first-time that had a diagnosis of HCM (ICD9-CM: 425.1). We identified all patients who were above 20 years old. The exclusion criteria included the following: (1) prior diagnosis of ischemic stroke (ICD9-CM code: 434, 434.1, 435, 436, 437, 434.91) before index date, (2) ambulatory visit for AF (ICD9-CM code, 427.3) during follow-up (Figure 3). All subjects were then followed from the index date to the date of first ischemic stroke event, or Dec. 31, 2013 if no event occurred with a median follow-up duration of 7.3 years (25–75%, 5.2–9.3 years). Comorbidity was defined by diagnoses at hospital discharge or in clinic records. For our study population, we searched the database to see if they had hypertension (ICD-9-CM codes: 401.X-405.X), diabetes mellitus (250.X, 249.X), hyperlipidemia (272.X), coronary artery disease (ICD9-CM code, 411.X-414.X, V17.3, V81.0), heart failure hospitalization (ICD9-CM code, 428.0-428.3, 429.9), chronic kidney disease (ICD9-CM code, 585.X-588.X). Medications that were dispensed at time of index date, including antiplatelet, angiotensin converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), beta-blockers, calcium channel blockers (CCB) and statins were identified.

The primary endpoint was diagnosis of ischemic stroke, based on ICD-9-CM coding (ICD9-CM code, 427.3, 435.X, A299) in any ambulatory visit and discharge diagnoses. The main purpose of the study was to investigate the risk of long-term development of ischemic stroke of HCM patients in the absence of AF. Therefore, we used latent class models to identify matched subjects without HCM but with AF from database of NHIRD sharing a similar underlying age, sex, and comorbidities including hypertension, diabetes, hyperlipidemia and chronic kidney disease. Of note, we excluded subjects not receiving oral anti-coagulants during follow-up. The index date of matched subjects was defined as the diagnosis of AF. Each HCM patient and his/her corresponding matched control had the same date of the diagnosis of HCM/AF. Our study design with the same index date of both groups was similar as new user (incident user) designs, preventing the problem of depletion of susceptibles. Subjects in our study were classified into two groups of patients with HCM in the absence of AF and matched subjects with AF.

Statistical analyses

For comparison of the baseline characteristics between two groups, paired t-test (continuous variables) and the chi-squared (categorical covariates) test was performed. To reduce the potential bias and to make the two groups more comparable, the propensity score was applied. We derived a propensity score, which is the logit (probability) for receiving beta-blockers treatment from a logistic regression model by using matching variables listed in Table 1 (age, sex, CHA2DS2-VASc score, hypertension, diabetes, coronary artery disease, hyperlipidemia, hospitalization for heart failure and chronic kidney disease). The predicted accuracy of the logistic model was assessed with an area under the receiver operating characteristic curve (C statistic), which was 0.824 (95% CI 0.792 – 0.846). According to the propensity score, patients were selected by 1:1 matching without replacement using the nearest neighbor method. A caliper width of 0.15 standard deviations (SDs) was used for matching as our previous studies [21, 22]. The annual risk of ischemic stroke was calculated for patients who were stratified into 4 groups on the basis of age (20 to 39 years, 40 to 64 years, 65 to 75 years, and above 75 years). For survival analyses, multivariate Cox’s proportional hazard regression analyses were used to derive the adjusted HR for developing ischemic stroke by using matched population with AF as controls [23]. The model was adjusted for age, gender, risk factors (hypertension, diabetes mellitus and hyperlipidemia), comorbidities (coronary artery disease, chronic kidney disease, hospitalization for heart failure), and medication usage. On the other hand, considering that all-cause death is a competing risk for ischemic stroke in HCM population, we also estimated the association between each HCM patient and incidence of ischemic stroke during follow-up by using subdistribution Cox proportional hazard. To assess the effect of aging, we also did subgroup analyses for different age interval. In the primary analysis, patients who were stratified into 4 groups on the basis of age. Since there was not same window for all the groups, we performed a sensitivity analysis with fixed interval of 10 years between 6 subgroups (age 25-34, 35-44, 45-54, 55-64, 65-74 and > 75 years) to test the robustness of our study design and results. For survival analysis, we adopted cumulative incidence function curves for incidence of ischemic stroke in HCM patients and matched controls and comparison was conducted using the Fine and Gray test [24]. Data are displayed as mean ± SD for continuous variables and as proportions for categorical variables. All of the analyses were conducted using the Statistical Package for the Social Sciences (SPSS) for Windows, Version 22.0 (SPSS, Inc., Chicago, Illinois). Values of P<0.05 were considered significant; all are reported as 2 sided.