This study was approved by the Ethics Committee of Lund University, Sweden. Data used in this study contained information on all individuals registered as residents of Sweden [21]. It included individual-level information on age, sex, occupation, geographic region of residence, hospital diagnoses, and dates of hospital admissions in Sweden (1964–2008), as well as date of emigration, and date and cause of death [21]. The dataset was constructed using several national Swedish data registers (reviewed by Rosen and Hakulinen) [22], including, but not limited to, the Swedish National Population and Housing Census (1960–1990), the Total Population Register, the Multi-Generation Register, and the Swedish Hospital Discharge Register [23]. The data were released to us from the National Board of Health and Welfare and Statistics Sweden.
Information retrieved from the various registers was linked, at the individual level, via the national 10-digit personal identification number assigned to each resident of Sweden for his or her lifetime. Registration numbers were replaced by serial numbers to preserve anonymity. As well as being used to track all records in the database at the individual level, these serial numbers were used to check that individuals with hospital diagnoses of ischemic or hemorrhagic stroke appeared only once during the follow-up (for the first hospital diagnosis of ischemic or hemorrhagic stroke during the study period).
The follow-up period for analysis of data in the present study started on January 1, 1987 and continued until hospitalization for ischemic or hemorrhagic stroke, death, emigration, or the end of the study period (December 31, 2008). Data for first hospitalization for ischemic or hemorrhagic stroke during the study period were retrieved from the Hospital Discharge Register (1987–2008). This study did not include data for hospital outpatients or patients treated at primary health care centers.
Predictor variable
The predictor variable was hospitalization for an IMD, diagnosed according to ICD-7, ICD-8, ICD-9, and ICD-10 (Additional file 1 Table S1).
Outcome variable
Diagnosis of ischemic stroke was based on the 9th, and 10th revisions of the International Classification of Diseases (ICD-9, and ICD-10). Cases of ischemic stroke were identified using the following ICD codes: 433, 434, 435, 437.0, and 437.1 (ICD-9); and I63 (not I636), I65, I66, I67.2, and I67.8 (ICD-10).
Diagnosis of hemorrhagic stroke was also based on ICD-9, and ICD-10. Cases of hemorrhagic stroke were identified using the following ICD codes: 431 and 432 (ICD-9); and I61 and I62 (ICD-10).
Individual-level variables adjusted for in the model
The individual-level variables were sex, age, time period, geographic region of residence, socioeconomic status (SES), and comorbidity.
Sex: male or female.
Age was divided into 5-year categories. Subjects of all ages were included in the study.
Time period was divided into five time periods in order to allow for adjustment for any change in hospitalization rates over time: 1987–1991, 1992–1996, 1997–2001, 2002–2008.
Geographic region of residence was included as an individual-level variable to adjust for possible differences in hospital admissions for ischemic or hemorrhagic stroke between different geographic regions in Sweden. It was categorized as: 1) large city (city with a population of >200,000 (i.e., Stockholm, Gothenburg, or Malmo); 2) Southern Sweden (both rural and urban); and 3) Northern Sweden (both rural and urban).
Occupation was used as a proxy for SES. We classified each individual’s occupation into one of six categories: 1) blue-collar worker, 2) white-collar worker, 3) professional, 4) self-employed, 5) farmer, and 6) non-employed (Individuals without paid employment). Homemakers and students without an occupation were categorized on the basis of their husband’s, father’s or mother’s occupation. If that was not possible, they were included in the “non-employed” category. For individuals aged <20 years, parental occupation was used.
Comorbidity was defined as the first hospital diagnosis at follow up (1987–2008) of the following: 1) chronic lower respiratory diseases (490–496 (ICD-9), and J40–J49 (ICD-10)); 2) obesity (278A (ICD-9), and E65–E68 (ICD-10)); 3) alcoholism and alcohol-related liver disease (291 and 303 (ICD-9), and F10 and K70 (ICD-10)); 4) type 2 diabetes mellitus (250 (age >29 years) (ICD-9), and E11-E14 (ICD-10)); 5) hypertension (401–405 (ICD-9), and I10–I15 (ICD-10)); 6) atrial fibrillation (427D (ICD-9), and I48 (ICD-10)); 7) heart failure (428 (ICD-9), and I50 (ICD-10)); 8) renal disease (580–591 and 753B (ICD-9), and N00-N19, Q61 (ICD-10)); 9) sepsis (036,038 (ICD-9), and A39-A41 (ICD-10)); and 10) coronary heart disease (410–414 (ICD-9), and I20-I25 (ICD-10)).
Statistical analysis
Person-years at risk (i.e., number of persons at risk multiplied by time at risk) were calculated from the time at which subjects were included in the study (in 1987 or later) until first hospitalization for ischemic or hemorrhagic stroke, death, emigration, or the end of the study period. Person years for IMD patients were calculated from discharge of first hospitalization for IMD (IMD patients with previous stroke before the first IMD hospitalization or at the same hospitalization as the first IMD hospitalization, were excluded). The expected number of cases was based on the number of cases in the reference group. SIRs were calculated as the ratio of observed (O) and expected (E) number of ischemic or hemorrhagic stroke cases using the indirect standardization method [24]:
(1)
Where denotes the total observed number of cases in the study group; (expected number of cases) is calculated by applying stratum-specific standard incidence rates obtained from the reference group to the stratum-specific person-years (n) of risk for the study group; represents the observed number of cases that the cohort subjects contribute to the jth stratum; and J represents the strata defined by cross-classification of the following adjustment variables: age, sex, time period, SES, geographic region of residence, and comorbidity [24]. Ninety-five percent confidence intervals (95% CIs) were calculated assuming a Poisson distribution [24]. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC, USA).