Comparison of risk factors for ischemic stroke and coronary events in a population-based cohort

The Malmö Diet and Cancer Study (MDCS) is a large population-based cohort with a prospective design [13]. In total, 28,449 subjects (11,246 men and 17,203 women) from the eligible population of 68,905 individuals participated in baseline examinations between 1991 and 1996. We excluded those with a history of stroke or CE (n = 841). To rule out acute inflammatory conditions and laboratory errors, those with a total leukocyte count higher than 20 × 10⁹/L were also excluded [14]. Finally, all participants with missing values for co-variates were also excluded from the final study population. Therefore, complete information regarding risk factors, leukocyte count and covariates was available for the total number of 26,519 participants. The traditional risk factors and total and differential leukocyte count were explored with relation to incident CE and ischemic stroke in this study population.

A random sample of participants from the MDCS were invited between October 1991 and February 1994 to study the epidemiology of carotid artery atherosclerosis. This sub-cohort, consisting of 6103 individuals (2572 men and 3531 women), was named the Malmö Diet and Cancer Cardiovascular cohort (MDC-CV) [15]. Seven inflammatory markers namely alpha1-antitrypsin, orosomucoid, haptoglobin, complement C3 (C3), C-reactive protein (CRP) and soluble urokinase plasminogen activator receptor (suPAR), were explored in relation to incident CE and ischemic stroke in this sub cohort. For this analysis, cases of CE and stroke, that had occurred between baseline screening and this re-examination, were excluded (n = 147) as well as those missing values for co-variates (n = 1161) resulting in a study population of 4795. Lastly, the missing number of measurements for each of the inflammatory markers were excluded in separate analyses for each inflammatory marker.

Study populations for both the cohorts are illustrated in a flow chart in Additional file 1: Fig. 1.

The baseline measurements of the MDCS consisted of a self-administered questionnaire, anthropometric measurements and collection of non-fasting blood samples, described in detail previously [13].

Blood pressure (mmHg) was measured once, after ten minutes of rest, while the subject was in supine position, and using a mercury-column sphygmomanometer. Waist circumference (cm) was measured midway between the lowest rib margin and the iliac crest. Information regarding smoking, use of anti-hypertensive medications and education level was obtained from the questionnaire. Smoking status was dichotomized into current or former smokers and non-smokers. Education level was divided into two categories; ≤ 9 years of schooling and > 9 years of schooling (reference category). Apolipoprotein A1 (ApoA1) and apolipoprotein B (ApoB) concentrations were determined by Quest Diagnostics (San Juan Capistraon, CA, USA) in serum samples that had been stored at − 80 °C until the analysis was performed in 2013 using an immunonephelometric assay that was run on a Siemens BNII (Siemens, Newark, DE, USA) [16]. Total and differential leucocyte counts (neutrophils, lymphocytes, mixed cells) were determined in heparinized whole blood using Sysmex K 1000 (Sysmex Europe, Norderstedt, Germany) at the central laboratory of Malmö Hospital, within 2 h from blood sample collection. Diabetes mellitus (DM) at baseline was defined based on a self-reported physician’s diagnosis of DM, or use of anti-diabetic medication, and retrieved from several local and national registers, which have previously been described in detail [17].

In the MDC-CV, additional fasting blood samples were drawn. The low-density lipoprotein (LDL) concentration was calculated according to the Friedewald’s formula. The plasma levels of ceruloplasmin, orosomucoid, haptoglobin, alpha1-antitrypsin and C3 were determined using Cobas c-systems (Roche Diagnostics GmbH, Germany) [18]. High-sensitivity CRP in plasma was analyzed using the Tina-quant®CRP latex assay (Roche Diagnostics, Basel, Switzerland) on an ADVIA®1650 Chemistry System (Bayer Healthcare, NY, USA). SuPAR was determined using the commercial ELISA suPARnostic kit (ViroGates, Copenhagen, Denmark) [19].

The two outcomes of interest in the study were incident CE and incident ischemic stroke. Participants were followed from baseline examination until first CE or ischemic stroke event, emigration from Sweden, death, or end of follow-up (December 31st, 2016), whichever came first.

Information regarding incident CE and ischemic stroke were obtained from local and national registers. Briefly, the Swedish Hospital Discharge Register and the National Cause of Death Register were used for CE. A CE was defined as a fatal or non-fatal myocardial infarction (International Classification of Diseases 9th (ICD-9) codes 410) or death due to ischemic heart disease (codes 411, 412 and 414 (ICD-9) or I21-25 (ICD-10)).

Up until 2010, incident stroke cases were retrieved by linkages with both the Stroke Register of Malmö [20], and the Swedish National Hospital Discharge Register [21], the latter for those who suffered from a stroke in hospitals outside of Malmö. In the Malmö stroke register, which has monitored stroke incidence in Malmö 1989–2010 [20], all stroke diagnoses were validated by review of medical records (63% of all stroke cases in this study), according to the WHO stroke definition. Computed tomography or magnetic resonance imaging of the brain, or autopsy had been done in all stroke cases classified as ischemic, intracerebral hemorrhage or subarachnoid hemorrhage. If no imaging had been performed the case was classified as unspecified stroke. From 2011 to 2016, all stroke cases were retrieved from the Swedish National Hospital Discharge register. International Classification of Diseases ICD-9 code 434 and the ICD-10 code I63 was used to identify incident cases of ischemic stroke in this register.

Variables with a skewed distribution (C3, CRP and suPAR) were log-transformed. Baseline characteristics are reported separately for subjects without incident CE or ischemic stroke, subjects with only incident CE, subjects with only incident ischemic stroke, and subjects with both incident CE and ischemic stroke during follow up. Continuous data were described as means ± standard deviations (SD) or median (25–75%) for skewed distribution, and as proportions for categorical variables. Differences in the baseline characteristics of the study population according to incident disease status were tested using chi-square for categorical variables and ANOVA for continuous variables.

Analysis was conducted using Cox proportional hazards regression models to investigate associations between various risk factors and CE and ischemic stroke as endpoints. Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated using standardized values (Z scores) for the continuous variables and specified reference categories for the dichotomous variables. Follow-up time until the end, death, emigration or incident CE/ischemic stroke was used as time-scale. For the analysis conducted in MDCS, Model 1 was adjusted for age and sex. Model 2 was additionally adjusted for waist circumference, DM, smoking, systolic blood pressure, use of anti-hypertensive medication, ApoA1, and ApoB.

For the analysis in the MDC-CV, Model 1 was adjusted for age and sex. Model 2 was additionally adjusted for waist circumference, smoking, DM, systolic blood pressure, use of anti-hypertensive medications and LDL cholesterol.

The heterogeneity in the strength of association between CE and ischemic stroke was assessed using a modified method of Lunn-McNeil competing risks models using a data duplication method [22]. Briefly, this consists of duplicating the dataset, so that each individual appears in two strata. The failures (CE or ischemic stroke) are then sorted by strata, and a stratified Cox regression is performed, which thus allows the estimation of separate hazard ratios (HRs) for the two outcomes. The HRs (95% CI) obtained from this analysis is identical to the values obtained from the separate Cox regression in the original data set. Another analysis is then done with just one variable for the exposure of interest while rest of the model is kept the same. The likelihood ratio test is then conducted to compare the second model to the first one (with differential effects). P-values for the difference in association of a given exposure to the separate outcomes are derived from this likelihood ratio test, and the null hypothesis is that both outcomes are equally associated with the risk factors. Compared to the original Lunn–McNeil model, the modified version has events in both strata if the participants had developed both CE and ischemic stroke. The HRs obtained from this approach are identical to results from separate Cox models run for each outcome, which was verified in standard Cox regression models.

The proportional hazard assumptions were tested by incorporating the time-dependent effects of covariates, and also by calculating Schoenfeld residuals to inspect it visually. It was found that there was some deviation from the assumption for certain exposures. The analyses were, therefore, rerun for follow up time intervals before and after the median. However, visual inspection of the Schoenfeld residuals of the deviations and dividing up the follow up time indicated that the effect of this deviation on HRs was minor.

To further examine the robustness of the analysis, a sensitivity analysis was conducted after excluding those with both ischemic stroke and CE. Furthermore, as part of our sensitivity analysis, we stratified the ischemic stroke cases into those with and without atrial fibrillation (AF) before the ischemic stroke event. The analysis was repeated after excluding those with AF before ischemic stroke event.

The baseline characteristics of the study population are summarized in Tables 1 and 2 for the MDCS and MDC-CV, respectively. In relation to the disease status, those who developed incident CE, ischemic stroke or both had higher cardiovascular risk factors as compared to those with no disease status in the MDCS (Table 1).

Similarly, in the MDC-CV (Table 2) those with both incident CE and ischemic stroke showed relatively higher levels of inflammatory markers.

In the MDCS, there were 2270 incident ischemic stroke events (mean follow-up time 19.43 ± 5.82 years) and 3087 incident CE events (mean follow-up time 19.33 ± 5.87 years) in the MDCS (Table 3). Both CE and ischemic stroke developed in 431 individuals, which, after exclusion resulted in 1838 incident ischemic stroke and 2656 incident CE events (Additional file 2: Table 1). In the MDC-CV, there were in total 320 and 447 incident cases of ischemic stroke (mean follow up 20.78 ± 5.69 years) and CE (mean follow up 20.68 ± 5.78 years), respectively, whereas 93 individuals had both incident CE and ischemic stroke (Table 2).

The age and sex adjusted and multivariable adjusted HRs for the multiple risk factors are shown in Table 3 and Fig. 1 for MDCS, and in Table 4 and Fig. 2 for MDC-CV. In Model 2, waist circumference, smoking, systolic blood pressure, DM, ApoB, age, sex, total leukocyte count, and neutrophil count were positively associated with risk of both CE and ischemic stroke. Schooling of ≤ 9 years was associated with an increased risk of incident CE but was not associated with incident ischemic stroke. ApoA1 levels were associated with reduced risk of both outcomes. In Model 2, lymphocyte count did not show significant association with either CE or ischemic stroke. Similar results were observed in the sensitivity analysis when participants with both CE and ischemic stroke were excluded (Additional file 2: Table 1).

In the MDC-CV, the inflammatory markers were associated with higher risk of CE and ischemic stroke in Model 1. However, after adjusting for potential confounders, only orosomucoid, and suPAR were significantly associated with increased risk of ischemic stroke. On the other hand, orosomucoid, alpha1-antitrypsin, ceruloplasmin, CRP and suPAR were associated with incident CE after adjustment for confounders.

Some important differences between the outcomes were found in relation to the baseline risk factors. Table 3 reports these data for the MDCS. In the final model, current smoking, ApoB, male sex and ≤ 9 years of schooling were preferentially associated with incident CE as compared to ischemic stroke (p for equal association = 0.038, < 0.001, < 0.001, and 0.040, respectively). ApoA1 was more protective for CE (HR_CE 0.81, CI 0.78–0.84) as compared to ischemic stroke (HR_st 0.94, CI 0.90–0.98) (p for equal association < 0.001). Age showed stronger association with ischemic stroke (HR_st 1.84, CI 1.75–1.93) as compared to CE (HR_CE 1.69, CI 1.62–1.76) (p for equal association = 0.011). Conversely, although some leukocytes were positively associated with both increased risk for CE and ischemic stroke, no difference in strength of association was observed for the two outcomes. Similar results were observed after excluding those both incident CE and ischemic stroke (Additional file 2: Table 1).

In Table 4, the associations of inflammatory markers in relation to incident CE and ischemic stroke are shown. Although most inflammatory markers were associated with both CE and ischemic stroke, no significant heterogeneity was found. Interestingly, in the sensitivity analysis (Additional file 2: Table 2) alpha1-antitrypsin showed a stronger association with CE after excluding participants with both CE and ischemic stroke. However, it was not significantly associated with risk of ischemic stroke.

After excluding ischemic strokes cases with AF (n = 503), in the MDCS, the difference in association was no longer significant for current smoking, age and education level. However, ApoA1, ApoB and male sex showed similar stronger association for CE (Additional file 2: Table 3). In the MDC-CV, after excluding those with AF, the results were essentially the same (Additional file 2: Table 4).