Psoriasis and cardiovascular risk: Strength in Numbers

In this issue of the JID, Wakkee and colleagues report a self-described exploratory cohort study and conclude that psoriasis may not be an independent risk factor for ischemic heart disease hospitalizations and that there is only a slight and borderline increased risk of ischemic heart disease among psoriasis patients (Wakkee, Herings et al. 2009). This negative result needs to be interpreted in light of the study’s limitations, the complexity of the relationship between varying levels of psoriasis severity, patient age, and CV risk, and in the context of the rapidly growing literature on this topic.

When interpreting a negative study two basic principals need to be evaluated. First, is the data on exposure, outcome, and co-variables (such as confounders) valid? By valid, we mean does the study measure what it truly intends to measure? Database studies, especially those using administrative data in which the primary purpose is for payment are prone to error in that the electronic code entered by a health care provider or administrator may not reflect the true clinical state of the patient(Strom 2005). Such errors (called misclassification bias) may mask a true association and yield falsely negative results. Multiple approaches are necessary for validating a database code including medical record review to determine the predictive value of an electronic code to validly reflect the true clinical state(Rawson and D'Arcy 1998). For example, in the General Practice Research Database (GPRD), we have demonstrated that the positive predictive value of a psoriasis code is about 90% based on medical record review conducted by GPs 3–4 years after the entry of a psoriasis diagnostic code(Neimann, Shin et al. 2006). Similarly, other investigators have shown in GPRD that the positive predictive value of an acute MI code is 90% based on review of medical record data that included factors such as diagnostic EKG changes, typical chest pain presentation, increased cardiac enzymes, etc(Meier, Jick et al. 1998; Hammad, McAdams et al. 2008). Wakkee and colleagues do not present data on the positive predictive value of their coding algorithm to measure the exposure of interest (psoriasis) or the outcome of interest (acute IHD hospitalizations) using the gold standard of medical record review. Therefore, we cannot exclude misclassification bias as a source of error that explains their negative results. Furthermore, Wakkee did not report the results of their multi-variable model, so we are unable to determine if their approach was able to confirm the expected relationships between cardiovascular risk factors and hospitalization for acute IHD (von Elm, Altman et al. 2007).

The second basic principal is that of statistical error. The key question being, did the study have statistical power to detect a clinically meaningful association if one truly exists? Wakkee shows in Table 2 that the HR for acute IHD hospitalization (primary endpoint) and acute myocardial infarction (secondary endpoint) adjusted for prior use of antihypertensive, antidiabetic, and lipid lowering drugs, and a measure of health utilization is 1.05 (95% CI 0.95 to 1.17) and 0.94 (0.80, 1.11), respectively. However, in the general population, the overwhelming majority of patients (about 80%) have mild psoriasis (≤ 2% BSA) and a small minority (about 5%) have severe disease (> 10% BSA) and therefore, the summary analysis is driven by patients with mild disease(Stern, Nijsten et al. 2004; Kurd and Gelfand 2009). Although their primary finding was not statistically significant their point estimate of the association of psoriasis overall (e.g. not restricting to severe disease) with acute IHD is very similar to our results and those of other investigators(Gelfand, Neimann et al. 2006; Brauchli, Jick et al. 2009). This small association requires additional studies to confirm, but could be important from a public health point of view given that an estimated 125 million people are affected by psoriasis world wide(Foundation).

Of special interest however, is the impact of more severe psoriasis on cardiovascular risk. Wakkee and colleagues do not report the hazard ratio with 95% CI in patients who have more severe psoriasis (such as patients requiring treatment with systemic or phototherapy). As a result, we are unable to determine if Wakkee’s study had appropriate statistical power to detect a meaningful association between severe psoriasis and acute IHD nor can we determine if their point estimate of association was similar to other studies(von Elm, Altman et al. 2007). Finally, as we, and others, have demonstrated, the relative risk of cardiovascular events is greatest in younger psoriasis patients with severe disease and therefore, this finding should be accounted for in multi-variable models(Mallbris, Akre et al. 2004; Gelfand, Neimann et al. 2006; Brauchli, Jick et al. 2009). This finding is referred to as effect modification or statistical interaction and is generally hard to detect due to limitations of statistical power. Wakkee and colleagues attempted to evaluate effect modification by stratifying on an arbitrarily selected age (i.e., 65); however, this simplistic approach is problematic in that age is the most important risk factor for CV disease. Importantly, atherosclerosis and ischemic heart disease are diseases of aging. Cardiovascular risk follows a quadratic, hyperbolic curve in which the competing risk of age outweighs accelerating risk factors for atherosclerosis such metabolic syndrome and familial hypercholesterolemia(Austin, Hutter et al. 2004; Gelber, Gaziano et al. 2008). Therefore, we can not determine if the lack of effect modification by age reported by Wakkee is valid or if it was due to lack of statistical power, or selecting the incorrect age upon which to dichotomize.

All good science depends on other investigators confirming the findings using robust approaches. Older approaches, which are more sensitive to bias, have yielded conflicting findings(McDonald 1989). For example, Stern compared the rates of CV mortality in a cohort of patients with generally severe psoriasis derived from a clinical trial of PUVA at tertiary care medical centers compared to the general US population and found that the point estimate for psoriasis was protective, but not statistically significant (SMR for CV mortality 0.83 (90% CI 0.7–1.0)(Stern and Lange 1988). The PUVA trial patients however, are not representative of the US population (e.g. selection bias) and the problem is compounded by the fact that CV mortality rates vary widely (by 2 fold or greater) across the US (Labarthe 1998; Rothman K.J., Greenland S. et al. 2008). As a result, non-population based studies that do not have an appropriate internal control can be very difficult to interpret.

In contrast, modern epidemiological approaches using population-based methods which minimize bias and enhance generalizability of the results have found very similar point estimates for the adjusted relative risk of coronary heart events by Wakkee (hospitalization for IHD), Brauchli (MI), and Gelfand (MI) in psoriasis patients overall (Rothman 1988; Gelfand, Neimann et al. 2006; Brauchli, Jick et al. 2009). Moreover, as shown by Brauchli and our group, the point estimates of the adjusted relative risk of MI in severe psoriasis (based on treatment patterns) is clinically important and exceed or is similar to the relative risk of MI conferred by major cardiovascular risk factors such hypertension, diabetes, and hyperlipidemia (Gelfand, Neimann et al. 2006; Brauchli, Jick et al. 2009). To put this in perspective, the annual risk of a patient with severe psoriasis (based on treatment history) who is in their forties having an MI that is attributable to psoriasis and not traditional CV risk factors is estimated to be about 8 times greater than their risk of developing a melanoma (SEER; Gelfand, Neimann et al. 2006).

More recently, other investigators, using varying approaches and populations have shown that severe psoriasis is an independent risk factor for atherosclerotic cardiovascular disease as defined by outcomes such as database derived codes, imaging of coronary and carotid arteries, or measurements of endothelial function and arterial stiffness(Ludwig, Herzog et al. 2007; Balci, Balci et al. 2008; El-Mongy, Fathy et al. 2009; Gisondi, Fantin et al. 2009; Gladman, Ang et al. 2009; Prodanovich, Kirsner et al. 2009). Wakkee notes that information bias (e.g. psoriasis patients may be screened more carefully for CV disease) may explain the positive associations however, this is very unlikely. First, in our studies, analyses limited to patients seen regularly by the GP had minimal impact on the point estimates(Gelfand, Neimann et al. 2006; Gelfand, Dommasch et al. 2009). Wakkee’s approach was to adjust for this potential bias by including the total number of hospitalizations (except for CV diseases) in the 6 months prior to cohort entry in addition to adjusting for prescriptions for hypertension, hyperlipidemia, and diabetes. Although adjusting for prior health care utilization did not appear to meaningfully confound the association (e.g. the point estimates changed by < 10%), it is important to note that this approach may not be appropriate for assessing information bias as prior use of health care services could be in the causal pathway of the association. For example, diabetes is associated with higher use of health care utilization compared to those without diabetes, however, diabetes is still a risk factor for MI and adjusting for prior health care utilization is only likely to falsely attenuate this association. Second, MI is a relatively hard endpoint, with our results extending to CV mortality as well, and the magnitude of association of severe psoriasis with MI and CV mortality is similar to or exceeds the association of traditional major CV risk factors(Mehta, Azfar et al. In press). It is very unlikely that GPs are looking more carefully for MI in psoriasis patients than in their patients with diabetes and hypertension, and CV mortality would be robust to observation bias. Furthermore, other studies which have used diagnostic testing in psoriasis and control patients (e.g. coronary artery CT scans) would not be subject to detection bias of the outcome and have similarly found psoriasis to be an independent risk factor for CVD(Ludwig, Herzog et al. 2007; Balci, Balci et al. 2008; El-Mongy, Fathy et al. 2009; Gisondi, Fantin et al. 2009).

Wakkee and colleagues also raise the important point that incompletely measured or unknown confounders can explain the associations we and others have observed. This possibility is difficult to exclude, but appears less likely as several studies that have looked at CV disease in small cross sectional studies using well defined cases and controls have confirmed the independent association even when tightly controlling for confounding through direct measurement of risk factors(Ludwig, Herzog et al. 2007; Balci, Balci et al. 2008; El-Mongy, Fathy et al. 2009; Gisondi, Fantin et al. 2009). Moreover, we have shown in our studies of stroke and cardiovascular mortality that an unknown confounder would have to be very common in the general population (20% prevalence), strongly associated with psoriasis (OR of 2.7) and have a stronger association with CV mortality (OR of 6.5) or stroke (OR of 4.3) than traditional CV risk factors in order to render our results null(Gelfand, Dommasch et al. 2009; Mehta, Azfar et al. In press).

Despite the numerous recent publications investigating psoriasis and cardiovascular risk, there are still a lot of fundamental questions that remain unanswered. In particular, we do not know what degree of psoriasis severity (e.g. measured by BSA) and duration is necessary to meaningfully increase CV risk. We also do we know if successfully treating psoriasis will lower cardiovascular risk and mortality in these patients(Gelfand 2007; Gelfand, Troxel et al. 2007). Until more research is done, the basic public health message remains the same. Patients with psoriasis, especially if disease is moderate to severe should be educated about their potentially increased risk for cardiovascular disease and undergo appropriate medical evaluations and treatment of modifiable risk factors(Friedewald, Cather et al. 2008; Kimball, Gladman et al. 2008).