Traditional Framingham risk factors fail to fully account for accelerated atherosclerosis in systemic lupus erythematosus
The frequency of coronary heart disease (CHD) and stroke are increased in systemic lupus erythematosus (SLE), but the extent of the increase is uncertain. We sought to determine to what extent the increase could not be explained by common risk factors.
The participants at two SLE registries were assessed retrospectively for the baseline level of the Framingham study risk factors and for the presence of vascular outcomes: nonfatal myocardial infarction (MI), death due to CHD, overall CHD (nonfatal MI, death due to CHD, angina pectoris, and congestive heart failure due to CHD), and stroke. For each patient, the probability of the given outcome was estimated based on the individual's risk profile and the Framingham multiple logistic regression model, corrected for observed followup. Ninety-five percent confidence intervals (95% CIs) were estimated by bootstrap techniques.
Of 296 SLE patients, 33 with a vascular event prior to baseline were excluded. Of the 263 remaining patients, 34 had CHD events (17 nonfatal MIs, 12 CHD deaths) and 16 had strokes over a mean followup period of 8.6 years. After controlling for common risk factors at baseline, the increase in relative risk for these outcomes was 10.1 for nonfatal MI (95% CI 5.8–15.6), 17.0 for death due to CHD (95% CI 8.1–29.7), 7.5 for overall CHD (95% CI 5.1–10.4), and 7.9 for stroke (95% CI 4.0–13.6).
There is a substantial and statistically significant increase in CHD and stroke in SLE that cannot be fully explained by traditional Framingham risk factors alone.
As the overall prognosis for persons with systemic lupus erythematosus (SLE) has improved, arterial vascular disease, including coronary heart disease (CHD) (1–14), stroke (8, 9, 15–20), and peripheral vascular disease (21), has become an increasingly important cause of morbidity and fatality. Strokes are reported in 2.6–20% of SLE patients (9, 15–20) and have a high recurrence rate (20, 22) and result in greater mortality than in persons of similar age and sex but without SLE (22). The risk of myocardial infarction (MI) may be increased as much as 9-fold in SLE (9); a second study shows a 50-fold increase in MI in women with SLE ages 35–44 years (1). Cardiac involvement is also common in unselected patients with asymptomatic SLE (23–25).
SLE has features of accelerated atherosclerosis as seen in diabetes mellitus, in that vascular complications appear early in the course of the disease. The pathogenesis of this vascular disease is not understood. It remains unclear to what extent the excess risk of cardiovascular events observed reflects differences in the values of common risk factors. None of the reported studies has fully controlled for the effects of these risk factors. Manzi et al (1) adjusted only for age and sex, while Ward (26) did not take into account the effect of smoking and cholesterol. It is important to dissect the degree to which the disease, its treatment (i.e., corticosteroids), or coexistent “traditional” risk factors are involved, since the optimal prevention could differ correspondingly.
We have quantified the relative risks of developing stroke, nonfatal MI, CHD overall, and CHD death in SLE patients compared with the general population, while also taking into account the differences in vascular risk factors at the beginning of followup. Specifically, we tested the hypothesis that the rates of these outcomes among SLE patients, adjusted for their baseline risk profiles, are higher than the rates expected based on the risk factor effects estimated in the Framingham heart study (27). In addition, we have estimated 95% confidence intervals (95% CIs) for the identified relative risks, since the precision of the estimates is important given the relatively small sample sizes typical of SLE studies.
PATIENTS AND METHODS
Patients and sources of data
This study was based on the retrospective review of cardiovascular outcomes in two prospectively followed cohorts. The study subjects consisted of all patients with SLE meeting the ACR criteria (28) in the Montreal General Hospital Lupus Registry and in the Notre-Dame Hospital Lupus Registry who were followed up since 1977 and 1983, respectively. For both registries, standardized clinical and laboratory data have been collected prospectively using a protocol at the time of outpatient visits. In addition, the same data abstractor reviewed retrospectively all available medical records at both registries to determine the presence or absence of cardiovascular outcomes prior to and after the first registry visit. The cardiovascular outcomes were reviewed until December 1996 unless the patient died or was lost to followup at an earlier date.
Outcomes and risk factors
Only patients who had no evidence of previous cardiovascular disease at the baseline visit were studied. Specifically, patients with angina, MI, congestive heart failure (CHF) due to CHD, transient ischemic attack, or stroke prior to baseline were excluded. This ensured that our study patient sample was comparable to the participants of the Framingham study who were asymptomatic at the baseline visit (27).
Four outcomes considered in the original Framingham study (27) were assessed through retrospective chart review: nonfatal MI, the development of CHD, CHD death, and stroke. CHD was defined as at least 1 of the following 4 cardiac events: nonfatal MI, angina pectoris, CHF due to CHD, or CHD death. These 4 outcomes were assessed in separate analyses, and in each analysis a patient rather than an event was the unit of analysis, so that a given patient could contribute 1 event at most to a particular analysis. For example, in the analyses of CHD development, a hypothetical patient who had angina followed by 2 nonfatal MIs and, finally, a CHD death would be counted as a single outcome. However, the same patient could contribute to separate analyses of different outcomes. Thus, the hypothetical patient discussed above would be counted in 3 separate analyses, focusing, respectively, on CHD development, nonfatal MI, and CHD death, each time as a single event.
The baseline risk factors (age, sex, total serum cholesterol, diastolic blood pressure, systolic blood pressure, left ventricular hypertrophy [LVH], diabetes mellitus, and current cigarette smoking) were assessed at entry into the respective registry. They were defined as in the Framingham models (27).
A conservative approach was used to handle missing risk factor data. To ensure that we did not overestimate the observed:expected relative risks, missing values were replaced by the higher risk category (smoking, diabetes) for binary variables and by the 90th percentile of the sex-specific distribution observed in the Framingham study for continuous variables (total serum cholesterol, systolic and diastolic blood pressure). We chose the 90th percentile because our recent study of the impact of continuous risk factors on CHD mortality indicated that risks level off in the upper end of the distribution (29). The exception was LVH: given its rarity (the prevalence of LVH was ∼4% in the Framingham population), LVH was considered present only if it was explicitly recorded in the patient's chart. Data were missing for ≤1% of baseline risk factors except for information on smoking (data were missing in 9% of the cohort) and LVH (electrocardiogram was not performed in 26%).
The independent groups t-test and Fisher's exact test were used to compare subgroups on continuous and binary risk factors, respectively. Yearly event rates for particular outcomes were estimated assuming the exponential distribution of time-to-event.
To estimate adjusted relative risks associated with SLE, we calculated the ratio of observed to expected number of patients with a given vascular event. Theoretically, an alternative approach would require collecting data, during a comparable period, on subjects without SLE in the same population and then using a conventional multivariable regression model in which the effect of SLE would be adjusted for common vascular risk factors. This was not done for several reasons, but the chief reason was that the observed number of outcomes made it difficult to model simultaneously the effects of several variables and still obtain adequate precision, since a minimum of 5 outcomes for each independent variable would be needed in the binary regression model. Even if estimates were obtained, their precision would be poor.
Our approach relied instead on the Framingham estimates of the effects of common vascular risk factors (27), which are derived from a large, prospectively followed up, population-based cohort. These are generally accepted as an accurate quantification of the impact of particular risk factors and are the basis for assessing cardiovascular and cerebrovascular risks in individual patients (30–33). All guidelines for CHD prevention are based on risk predictions derived from the Framingham equations (34). Moreover, the predictions based on Framingham models have been validated in our previous study against the results of several prospective CHD studies (31), and there is some recent evidence that these models approximate well the risk factor effects observed in various Western populations (35–37). The approach described below allows us to draw on the strengths of the Framingham study to estimate the relative risks associated with SLE.
Estimating the expected rate of events
First, for each SLE patient, the Framingham logistic regression equations were used to estimate the probability of a given outcome (stroke, nonfatal MI, CHD death, or CHD overall) during the 8 years after the initial visit, based on the patient's risk factors at that visit: patient's age, current smoking status, blood pressure, total serum cholesterol, diabetes mellitus, and presence/absence of LVH. In addition, the equations used to estimate the expected probability of a particular event included an interaction between cholesterol and age, since it was found in the Framingham study that the impact of cholesterol decreased significantly with increasing age, especially for males (27). Separate models have been published for each sex and for systolic and diastolic blood pressures, and they were used to calculate two separate probabilities, based on diastolic and systolic blood pressures, respectively (27). These values were multiplied by regression coefficients from the corresponding sex-specific Framingham model and converted into probabilities using logistic transformation.
Next, we used an exponential model, which assumes a constant hazard rate over time to adjust the 8-year probability for the actual duration of the time interval over which the cardiovascular outcomes were assessed for a given patient. This was calculated as the time to death from any cause, the time the patient was lost to followup, or the time to the end of the study, whichever came first.
Estimating the ratio of the observed to the expected events
Since outcomes for different patients are clearly independent of each other, the expected total number of patients with a given type of event can be estimated by simply summing up the patient-specific probabilities. The point estimate of SLE-related relative risks was then obtained as the ratio of the number of patients observed with a given outcome to the expected number of patients with this outcome. The resulting risk ratio has a non-normal distribution. Therefore, to quantify the variation around this estimate, a bootstrap analysis of the data was performed (38, 39) based on repeated resampling with replacement. For each of 10,000 bootstrap samples, the algorithms for computing the expected and the observed numbers of outcomes were replicated and the observed:expected ratios were calculated. This allowed us to obtain an empirical distribution of the 10,000 observed:expected ratios that approximates quite accurately the underlying distribution of our estimate. The boundaries of the 95% CI for the observed:expected ratio were estimated, respectively, as the 2.5th and the 97.5th percentiles of this empirical distribution.
For each outcome, the above procedure was repeated twice, using either the systolic- or diastolic-based Framingham model to estimate probabilities of a given event for individual patients. The expected Framingham-based probabilities of outcomes for individual patients were, on average, slightly higher for the model employing diastolic rather than systolic blood pressure, thereby resulting in lower estimates of the relative risks in the former model. To simplify reporting and to err on the conservative side, only results from the diastolic blood pressure–based models are reported.
The Framingham cohort consisted only of individuals ≥30 years of age (27), and 92 of our 263 SLE patients (35%) were under age 30. To estimate the expected probability of an outcome for these patients, in the main analysis, the Framingham estimate of age effect (as well as age–cholesterol interaction) was extrapolated below the age of 30. Additionally, we repeated the analyses with the conservative assumption that the risk does not decrease further when baseline age decreases below 30 years. Accordingly, those young patients under age 30 were assigned the probability of an outcome corresponding to the age of 30.
Empiric validation of the observed:expected approach
To further validate our approach, the same method was applied to a different population for which there was no a priori reason to expect a systematic increase in the risk of cardiovascular events. Specifically, we used data from an observational study of asymptomatic patients with carotid bruits (40, 41) from a source population (Quebec) and a time period (1989–1994) similar to those of our SLE subjects (42) and examined a subgroup with a low degree of stenosis (i.e., those with <50% reduction in arterial diameter). Individuals with a mild degree of carotid stenosis have vascular event rates similar to those in the general population of elderly persons (43). Moreover, both adjusted and unadjusted risk ratios for stroke and transient ischemic attack among subjects with carotid stenosis of <50% are similar to those among subjects without stenosis (41, 44). Therefore, we expected that the number of vascular events observed among asymptomatic subjects with mild carotid stenosis should agree well with the number predicted based on Framingham models, yielding relative risk estimates close to 1.0 using our approach.
While asymptomatic patients with mild stenosis were substantially older (mean age 62.0 years) and were followed up for shorter time periods (mean followup 3.9 years) than our SLE patients, both age and followup time were adjusted for using the approach described above. Among 207 asymptomatic patients with mild stenosis, 6 had strokes and 8 had CHD events, including 3 nonfatal MIs and 3 CHD deaths. For each of the 4 outcomes, the ratio of the observed to the expected number of events was close to 1.0, as follows: 1.45 (95% CI 0.45–2.79) for stroke, 0.82 (95% CI 0.31–1.41) for CHD overall, 0.67 (95% CI 0.20–1.52) for nonfatal MI, and 1.16 (95% CI 0.02–2.67) for CHD death. These results are consistent with the a priori hypothesis that risks among patients with mild stenosis are similar to those found in the Framingham study. This further validates our approach by showing that the observed:expected ratio for SLE patients will reflect relative risks specific to this clinical population, and that the Framingham reference population can be used.
We identified 204 SLE patients from the Montreal General Hospital and 92 SLE patients from the Notre-Dame Hospital. Of these patients, 21 at the Montreal General Hospital and 12 at the Notre-Dame Hospital had had a previous vascular event (nonfatal MI [n = 9], congestive heart failure [n = 3], angina [n = 12], and stroke [n = 12]) and were excluded. Fourteen of these 33 patients had had the vascular event after SLE diagnosis but prior to baseline assessment for the present study. Not unexpectedly, the 33 excluded subjects tended to have more abnormal levels for the Framingham risk factors at baseline (after the event) than did the remaining 263 SLE patients (183 from Montreal General Hospital, 80 from Notre-Dame Hospital) who had no clinical evidence of atherosclerotic vascular disease at the baseline visit. The 33 excluded subjects were significantly older and had significantly higher mean systolic and diastolic blood pressures as well as higher mean total serum cholesterol (Table 1). The risk factor distributions for the Montreal General and Notre-Dame patients were comparable, but the latter had significantly lower mean cholesterol (Table 1).
Table 1. Distributions of risk factors for patients with systemic lupus erythematosus, by center and for excluded patients*
|Male, no. (%)||23 (13)||6 (8)||7 (21)|
|Age in years, mean ± SD||39 ± 14||35 ± 13||46 ± 16‡|
|Systolic BP in mm Hg, mean ± SD||122 ± 17||123 ± 21||129 ± 17‡|
|Diastolic BP in mm Hg, mean ± SD||76 ± 11||77 ± 11||81 ± 9‡|
|Current smoker, no. (%)||79 (59)§||43 (55)§||19 (61)§|
|Cholesterol in mmoles/liter, mean ± SD||5.2 ± 1.5||4.4 ± 1.0||5.5 ± 1.4‡|
|Diabetes mellitus, no. (%)||11 (6)||1 (1)||1 (3)|
|Left ventricular hypertrophy, no. (%)||3 (6)§||1 (2)§||3 (15)§|
The frequencies of outcome events were similar at the Montreal General and Notre-Dame Hospitals (Table 2). For 71 of the 263 SLE patients (27%), followup ended before the date of closure of the study, resulting in a total followup of 2,271.8 patient-years (mean ± SD 8.6 ± 4.9 years/patient). Among the 263 SLE patients, 34 had CHD events (12.9%), including 17 with angina (6.5%), 17 with nonfatal MIs (6.5%), 7 with CHF (2.7%), and 12 with a CHD death (4.6%). Sixteen of the 263 patients had a stroke (6.1%). Overall, 44 of the 263 patients (16.7%) had either a CHD event or a stroke (6 patients had both a CHD event and a stroke).
Table 2. Disease duration, followup, and numbers of patients with cardiovascular events, by center*
|Duration of SLE from diagnosis to baseline in years, mean ± SD||4.2 ± 5.8||4.6 ± 5.4||4.3 ± 5.7||NA|
|Followup duration from baseline in years, mean ± SD||8.5 ± 5.0||9.0 ± 4.7||8.6 ± 4.9||NA|
|Angina, no. (%)||12 (6.6)||5 (6.3)||17 (6.5)||0.75|
|Nonfatal myocardial infarction, no. (%)||15 (8.2)||2 (2.5)||17 (6.5)||0.75|
|Congestive heart failure due to coronary heart disease, no. (%)||7 (3.8)||0 (0.0)||7 (2.7)||0.31|
|Death due to coronary heart disease, no. (%)||9 (5)||3 (3.8)||12 (4.6)||0.53|
|Overall coronary heart disease, no. (%)†||25 (13.7)||9 (11.3)||34 (12.9)||1.50|
|Stroke, no. (%)||11 (6.0)||5 (6.3)||16 (6.1)||0.70|
The 44 patients with a CHD or stroke event had higher baseline levels for most of the risk factors (Table 3). The levels were significantly higher for age, systolic and diastolic blood pressure, and cholesterol. The mean duration of SLE was on average 1 year longer in patients who had events, although the difference was not statistically significant and was reduced to 0.5 years after adjusting for age.
Table 3. Baseline risk factors in systemic lupus erythematosus (SLE) patients with and without subsequent outcome events*
|Male, no. (%)||7 (16)||22 (10)||0.300|
|Age in years, mean ± SD||44 ± 17||36 ± 13||0.003|
|Systolic BP in mm Hg, mean ± SD||129 ± 19||121 ± 18||0.003|
|Diastolic BP in mm Hg, mean ± SD||81 ± 10||76 ± 11||0.008|
|Current smoker, no. (%)||28 (68)‡||94 (55)‡||0.159|
|Cholesterol in mmoles/liter, mean ± SD||5.5 ± 1.2||4.8 ± 1.4||0.006|
|Diabetes mellitus, no. (%)||0 (0.0)||12 (5)||0.230|
|Left ventricular hypertrophy, no. (%)||1 (4.0)‡||3 (3)‡||1|
|Previous SLE duration in years, mean ± SD||5.8 ± 7.5||4.8 ± 5.6||0.326§|
The analysis, which took into account the patients' baseline risk factors, revealed a striking increase in the incidence rates for all 4 outcomes in SLE patients contrasted with those expected (Table 4). For individual patients, the probabilities of specific events ranged from a minimum below 0.01 for all 4 outcomes to a maximum of 0.09 for nonfatal MI, 0.17 for CHD overall, 0.07 for CHD death, and 0.27 for stroke.
Table 4. Relative risk of vascular outcomes in systemic lupus erythematosus compared with those expected based on Framingham models*
|Nonfatal myocardial infarction||17||1.7||10.1||5.8–15.6|
|Death due to coronary heart disease||12||0.7||17.0||8.1–29.7|
|Overall coronary heart disease||34||4.5||7.5||5.1–10.4|
One of the most striking findings was that among initially asymptomatic SLE patients, the risks of CHD or stroke were both >7-fold higher than what would be expected based on their individual vascular risk factors alone. The increased risk is most dramatic for particularly serious CHD events: 10-fold for nonfatal MI and 17-fold for CHD death. The relative risk estimates are systematically, but only marginally, higher using systolic blood pressure–based models (data not shown). The corresponding 95% CIs show that the presence of SLE inflates vascular risks at least 4-fold compared with the predictions based on the Framingham models (Table 4). For each outcome, even the lowest of the 10,000 estimates of the ratio was >1.0 (data not shown), further indicating very high statistical significance of the impact of SLE on vascular outcomes. Sensitivity analyses demonstrated the robustness of our relative risk estimates, since changes in imputing risk corresponding to age 30 years for patients who were under age 30 at the baseline visit did not change the observed:expected ratio estimates beyond the second decimal place (data not shown).
The results of the present study demonstrate a significantly elevated risk for CHD and stroke in SLE. Most important, they show that the elevated risks cannot be fully explained by altered levels of the traditional Framingham risk factors considered in our analyses. The results are consistent with the increased risks noted in small case series and in uncontrolled studies as well as in several controlled studies. In a small population-based study of SLE in two health districts in southern Sweden, Jonsson et al (9) observed 8 MIs when 0.87 were expected. Although not tested statistically, this represented a 9-fold increase in risk. Using the California Discharge Database to evaluate hospitalization for acute MI, CHF, and cerebrovascular accident in persons with SLE, Ward (26) reported an ∼2-fold increase in these outcomes in those ages 18–44 years. Manzi et al (1) assessed angina and MI in 498 women with SLE who were seen over a 14-year period at the University of Pittsburgh Medical Center. These investigators reported a 52.4-fold increased risk of MI in women ages 35–44 years (95% CI 21.6–98.5) and a 4.2-fold increase in women ages 55–64 years (95% CI 1.7–7.9) compared with women in the Framingham Offspring Study. Although not statistically significant, angina was increased 2.3-fold in these same age ranges of 35–44 and 55–64 years, and for women ages 45–54 years, MIs took place with >2-fold the expected frequency.
The present results demonstrate that the risks of all relevant cardiovascular outcomes are increased even after accounting for all traditional Framingham risk factors (27) that were considered in our analyses. Thus, even when the baseline vascular risk factors are accounted for, nonfatal MI, death due to CHD, overall CHD, and stroke are increased 10.1-fold, 17.0-fold, 7.5-fold, and 7.9-fold, respectively, in patients with SLE.
The present study is not without limitations. While the traditional Framingham risk factors considered in our analyses include almost all those commonly taken into account in epidemiologic studies of CHD (34, 45), it is possible that some of the excess risk in SLE may be accounted for by factors such as family history, low levels of high-density lipoprotein cholesterol, obesity, or lack of exercise, for which data were not available in our study. However, even if these additional factors have been found in some studies to have statistically significant independent effects on cardiovascular risks, their reported effects seem far too weak to fully account for the dramatic risk increases revealed by our analyses.
The study was retrospective and therefore more open to error than if these same patients had been followed up prospectively with the specific hypothesis concerning CHD and stroke determined in advance. Nonetheless, the present study may have been more likely to miss mild outcome events than the Framingham study. This is particularly likely for mild angina, which may have been mistaken for pleuritis or pericarditis due to SLE (1). If this occurred, it would have biased the results conservatively.
Use of the Framingham models based on asymptomatic subjects required exclusion from the analysis of 33 patients with CHD or stroke events before their baseline evaluation. Among the 33 subjects excluded for this reason, there were 14 who had their first event before the baseline visit but after their first diagnosis of SLE. These subjects were eliminated because pre-event risk factor values were not available. Eliminating these patients probably reduced the risk estimates of the impact of SLE on the risks of cardiovascular events.
The explanation for the marked increase in risk for vascular outcomes in SLE is probably multifactorial and may differ for cardiovascular and cerebrovascular events. Immune complex–induced endothelial damage, vasculitis, antiphospholipid antibody–induced thrombosis, the effects of Libman-Sacks endocarditis, hypertension from renal involvement or corticosteroid therapy, and corticosteroid-induced central obesity, hyperglycemia, or hypercholesterolemia have all been suggested as causal factors and probably play a role.
The present study shows that the excess of cardiovascular events in SLE cannot be explained by the baseline values of the traditional Framingham risk factors that we considered in our analyses, and probably arises from the underlying disease and/or its treatment. Indeed, the fact that patients who had cardiovascular events during the followup period showed a slightly longer prebaseline duration of SLE than those without events, even after adjusting for the differences in age, seems to corroborate the hypothesis that at least some of the excess risk is due to SLE. The excess is sufficiently dramatic that clinicians should consider aggressive intervention to control known risk factors such as smoking, hypertension, and hypercholesterolemia pending the identification of SLE-specific risk factors and the results of SLE-specific intervention studies. However, it is not clear that even more aggressive application of what is currently recommended for traditional risk factor modification will affect the accelerated atherosclerosis observed.