The impact of education level on all-cause mortality in patients with atrial fibrillation

Data source

The study population has been previously described¹². Briefly, data from the Swedish National Patient Registry, the General Population Registry, and the Swedish Cause of Death Registry were cross-linked through the Epidemiological Centre of the Swedish National Board of Health and Welfare and Statistics Sweden.

These nationwide registries facilitate the collection of patient data, including age and sex, using personal identity numbers unique to each resident of Sweden. Sweden has a comprehensive public health insurance system, ensuring high-quality data in the registries¹³. All diagnoses cited in this study were based on the International Classification of Diseases (ICD), utilizing the ICD-9 system from 1987 to 1996 and the ICD-10 system from 1997 to the present.

We calculated cumulative Charlson Comorbidity Index (CCI) scores for AF patients to analyse the association of education level with all-cause mortality in the presence of other diseases. The CCI was developed to provide an estimate of mortality risk associated with 20 comorbid conditions, including cardiovascular disease^14,15. Identification and coding of comorbidities using ICD-9 and ICD-10 was conducted according to the recommendation of Quan et al¹⁶. In the absence of a universally adopted CCI categorization for cardiac disease¹⁵, we classified the CCI scores as 0–2 (low mortality risk), 3–4 (moderate mortality risk), and ≥ 5 (high mortality risk). Because of the low number of individuals and deaths, the first two categories (0–4) were merged into a low mortality risk classification, predominantly comprising the original moderate risk category. If multiple diagnostic codes were present within a comorbid condition group, they were counted as one.

We used the CHA₂DS₂-VASc score¹⁷ to describe the thromboembolic risk of the patients.

Study design and population

This nationwide, register-based, retrospective cohort study encompassed all individuals hospitalized in Sweden with a primary or secondary diagnosis of AF registered for the first time during an index admission from 1 January 1995 through 31 December 2008. Exclusion criteria were unknown education status and age under 30 years to ensure high-quality data on the highest education level. Patients over 85 years of age at diagnosis were not included. (10) Mortality endpoint data were available through 31 December 2009.

Atrial fibrillation was defined as 427 D (DA, DB, DC, DD, and DW) in ICD-9, and I48 and I48.9 (A, B, C, D, E, F, P, and X) in ICD-10. To minimize the risk of information bias, we did not distinguish among paroxysmal, persistent, and permanent AF or atrial flutter (AFl). Our intention was to avoid missing any AF diagnoses, and our choice of study population allowed patients with an AFl diagnosis without concomitant or alternating AF to be included. Validation of AF diagnoses in Danish registries showed that the ICD codes had a high positive predictive value for AF. Specific codes for AFl were rarely used, and the registries were unreliable in distinguishing between the two diagnoses¹⁸. Allowing patients with both primary and secondary diagnoses of AF meant that our population included both patients in whom AF was the cause of admission and patients in whom AF was detected when they were admitted for other reasons.

The study population was divided into three cohorts based on the highest level of education according to the categories of the International Standard Classification of Education (ISCED) as follows: primary (ISCED 0–2), secondary (ISCED 3–4), and academic (ISCED 5–8) education. After a 30-day blanking period, each subject was followed for up to five years, until death, emigration, or the end of the study on 31 December 2009.

Risk of all-cause mortality was evaluated in subpopulations comprising CCI risk score categories, heart failure, coronary artery disease (CAD), acute myocardial infarction (AMI), and cerebrovascular event (CVE) [stroke, ischaemic stroke, and transient ischaemic attack (TIA)] as well as cancer present at index hospitalization or diagnosed in the preceding five years. The ICD code definitions for stratification factors were based on Andersson et al¹². To reduce misclassification and accurately assess the impact of CVE, we combined stroke, ischaemic stroke, and TIA into a single variable.

Statistical analysis

Categorical variables are presented as percentages and continuous variables as mean ± SD. Examined comorbidities were compared using the Chi-squared test and age using one-way analysis of variance (ANOVA) across education groups.

Unadjusted Kaplan-Meier failure plots were used to illustrate the cumulative all-cause mortality risk across education groups, and Cox regression models were used to compare all-cause mortality risk of education groups separately by sex. Cox regression provides hazard ratio (HR) with 95% confidence interval (CI) as association measure. The analysis was conducted unadjusted and adjusted for age, year category of AF diagnosis (1995–1999; 2000–2004; 2005–2008), AF as the primary diagnosis at hospitalization, and CCI score with eight categories (1, 2, 3, 4, 5, 6, 7, ≥ 8). Age was modeled using a restricted cubic spline with four knots according to Harrell^19,20. Because of non-proportional hazards, stratified Cox regression was used. Non-proportional hazards were evaluated with Schoenfeld residuals and tested with the STATA PHTEST command, with a p value < 0.01 regarded as significant for non-proportionality.

To evaluate whether adjusted mortality risks across education groups differed in patients with low (0–4 score) and high (> 5 score) CCI scores, interaction modelling was performed. A similar interaction analysis was performed to assess mortality risk across education groups for patients with predefined comorbidity diagnoses (heart failure, CAD, AMI, CVE, cancer). To avoid over-adjustment by using the CCI score, the models were adjusted for hypertension and the comorbidities represented by the CCI score (diabetes, peripheral vascular disease, dementia, chronic kidney disease, chronic obstructive pulmonary disease, chronic pulmonary disease, rheumatic disease, mild liver disease, moderate or severe liver disease, hemiplegia or paraplegia, renal disease, HIV, and peptic ulcer), except for CAD and AMI, respectively.

When evaluating the potential interaction of education level with CAD, AMI was not included. In the interaction analysis, non-proportional associations were evaluated by interactions with follow-up time (30 days–2.5 years and > 2.5–5 years follow-up, or 30 days–1 year, > 1–2.5 years, and > 2.5–5 years).

All statistical analyses were performed using STATA versions 16 and 17 (StataCorp, College Station, TX, USA).

Ethics

The study procedures complied with the principles of the Declaration of Helsinki²¹ and were approved by the Regional Ethical Review Board in Uppsala, Sweden (Dnr 2009/273). All data were anonymised by the Swedish National Board of Health and Welfare and Statistics Sweden prior to being provided to the authors, and requirement for informed consent was waived by the ethical board.