To assess LAF among individuals in Sweden, comprehensive register and health care data from multiple nationwide sources were linked [33
]. This linking was based on the unique 10-digit personal ID numbers assigned at birth or immigration to all Swedish residents for life, information on which is nearly 100% complete. These numbers were replaced with serial numbers to preserve anonymity. Our database contains data from four sources:
1. The Swedish Multigeneration Register, which contains information on family relationships (siblings, parent-offspring). The register contains information on index persons registered in Sweden between January 1, 1961 and December 31, 2008 and born between January 1, 1932 and December 31, 2008.
2. The Swedish Hospital Discharge Register, which contains information on all hospital diagnoses for all people in Sweden for the period 1987–2009. Each record includes the main discharge diagnosis.
3. The Swedish Cause of Death Register, which contains data on date of death for the period 1961–2010.
4. The Total Population Register, which includes data on year of birth, gender, country of birth and education.
This study was approved by the Ethics Committee of Lund University, Sweden.
Cases of LAF were identified in the Swedish Hospital Discharge Register by the use of the ICD (International Classification of Diseases) codes 427D (ICD-9) and I48 (ICD-10) and age ≤60 years at the first diagnosis of AF. Individuals with one or more of the following discharge diagnoses within the 5-year period prior to the first diagnosis of AF were excluded: Hypertension (401–405 (ICD-9) and (I10-I15 (ICD-10)); Heart failure (428 (ICD-9) and I50 (ICD-9)); Coronary heart disease (410–414 (ICD-9) and I20-I25 (ICD-9)); Morbi rheumatici cordis and Valvular disease (390–398, 421 and 424 (ICD-9) and I00-I09 and I33-I39 (ICD-10)); Cardiomyopathy (425 (ICD-9) and I42 and I43 (ICD-10)); Myocarditis (422 (ICD-9) and I40 and I41 (ICD-9)); Pericarditis (420 (ICD-9) and I30-I32 (ICD-10)); Other heart disease (429 (ICD-9) and I51 (ICD-10)); Thyrotoxicosis (242 (ICD-9) and E05 (ICD-10)); and Diabetes mellitus (250 (ICD-9) and (E10-E14 (ICD-10)). These diagnoses, together with the same diagnoses from the Cause of Death Register, were also used to define cardiovascular disease (CVD) outcome in model 1.G (see below).
AF in parents and siblings was defined by the ICD codes 427D (ICD-9) and I48 (ICD-10). Parental history of AF was defined as AF in at least one parent during the study period. Sibling history of AF was defined as AF in at least one sibling sometime during the study period.
The validity of the diagnosis of AF has been evaluated, and diagnoses were found to be correct in 97% of cases in the Hospital Discharge Register [37
]. Diagnoses of other cardiovascular disorders such as stroke and myocardial infarction have an approximate 95% validity [37
]. Generally, the validity in the Hospital Discharge Register is approximately 85-95% [37
The analyses were based on a database containing information on all cases of LAF during the period 1987–2009 (n=29,660, mean age=50.1 years (SD=9.6), 74% men, AF recurrence rate=49.6%).
We used Cox proportional hazards models in order to investigate recurrence of AF within 10 years. Cases were followed from date of LAF diagnosis during the study period until AF recurrence, death, emigration or the end of the follow-up period (December 31, 2009 or a maximum of 10 years) (whichever came first). In the first analysis, a parent-offspring analysis, we investigated all LAF proband cases whose parents both lived in Sweden sometime between 1987 and 2009 (n=16,160). In model 1.A, parental history of AF was included as a covariate (yes/no), and in model 1.B we also included sex, age at diagnosis of LAF (centered at the mean value), and terms for the interaction between parental history of AF and age/sex. The interaction terms were only included in the model if the p-values were <0.05. In model 1.C the variable parental history was categorized as no parental history of AF, one parent with AF and two parents with AF. In model 1.D, we investigated the association of parental history of LAF with time to first recurrence of AF in proband cases.
In order to further evaluate the results, we also investigated time from first until second recurrence of AF (model 1.E), as well as time from diagnosis of LAF until second recurrence of AF (model 1.F, n=7,370). We additionally investigated time to first recurrence of AF in individuals who did not experience any other CVD outcome during the 10-year follow-up period (model 1.G, n=9,071).
In the second analysis, a sibling analysis, we examined all LAF proband cases with at least one sibling living in Sweden sometime between 1987 and 2009 (n=20,373). In model 2.A, sibling history of AF was included as a covariate (yes/no), and in model 2.B we also included sex, age at diagnosis of LAF (centered at the mean value), and terms for the interactions between sibling history of AF and age/sex. The interaction terms were only included in the model if the p-values were <0.05. We also adjusted both models for number of siblings to the proband case (not reported in the tables).
In the third analysis, we merged datasets I and II and only analyzed individuals who were included in both datasets (13,525 cases). In model 3.A, sibling history of AF and parental history of AF were included as covariates (yes/no), and in model 3.B we also included sex, age at diagnosis of LAF (age at LAF), terms for the interactions between sibling history of AF and parental history of AF and terms for the interactions between sibling history of AF/parental history of AF and age/sex. The interaction terms were only included in the model if the p-values were <0.05.
In order to take into account the non-independence of observations from the same family, we used a robust sandwich estimator in all models [34
]. We present hazard ratios (HRs) and the corresponding 95% CIs [39
]. The proportional hazards assumption was fulfilled for the variables of interest.
In order to investigate familial transmission of first time LAF hospitalization we used a case-cohort approach to determine familial risks with odds ratios (ORs) [34
]. We conducted two main analyses: proband-sibling and proband-parent. In these analyses, we studied all LAF proband-relative pairs that could be matched to five control pairs in the Swedish population. For example, in the proband-sibling analysis we selected all sibling pairs for which at least one sibling was diagnosed with LAF and matched each of them to five control pairs. The control pairs were chosen randomly from individuals who lived in Sweden at the time of the probands’ diagnosis of LAF and comprised pairs of individuals who were not diagnosed with LAF or AF prior to the time of the proband’s diagnosis of LAF. Furthermore, both individuals in the control pair also had to have lived in Sweden sometime during the period 1987–2009. Control pairs were matched based on year of birth, sex, country of birth and level of education (the year before the date of diagnosis). Analyses were conducted by conditional logistic regression [34
]. As an example, in the proband-sibling analysis, AF in sibling (yes/no) was used as the independent variable. We present odds ratios (ORs) and the corresponding 95% CIs, according to previous studies of familial risks [34
]. As a proband could be included several times, we adjusted for non-independence by using a robust sandwich estimator [34
]. In all analyses, less than 1% of the proband pairs could not be matched to five controls and were excluded from the analysis.
All calculations were performed using SAS version 9.3.