We found that diabetes mellitus is the cardiovascular risk factor most convincingly associated with SMI. Meta-analysis of the highest quality studies revealed almost a two-fold risk of diabetes in schizophrenia-like illnesses but not bipolar affective disorder. Conversely, meta-analysis revealed no association between SMI and hypertension. Similar findings were observed for total cholesterol levels, but these studies were limited by their design and so conclusions must be guarded. There were inadequate numbers of comparative studies of other lipids, such as HDL cholesterol, or of the metabolic syndrome to conduct a meta-analyses. Lower HDL cholesterol levels in people with SMI found in two studies [
13,
47] were not confirmed by another [
20]. Five studies on the metabolic syndrome revealed an excess risk in people with SMI, and two, involving 718 people with SMI, confirmed such an excess of metabolic syndrome statistically. Raised "Framingham" or ten year cardiovascular risk scores may only be demonstrable in SMI when differences in effects are examined separately in different age groups and sexes. For instance excess risk scores may only be detectable in those over 40.
Quality and variability of published studies
There were very few high quality comparative studies on cardiovascular risk factors and the metabolic syndrome in people with and without SMI. Many studies in this review were limited by small, convenience samples of people with SMI such as those in specific clinics or inpatient units, compromising the generalisability of their findings. Furthermore, several studies reporting higher levels of cardiovascular risk in SMI did not obtain raw comparison data within their study to allow statistical assessment of the importance of their findings. Cardiovascular risk factors are increasing rapidly in the general population, hence the need for relevant contemporary comparison figures. There were no studies designed to compare the longitudinal development of cardiovascular risk factors between people with and without SMI.
The meta-analysis for diabetes did not detect any heterogeneity between inpatient and community samples with schizophreniform illnesses, suggesting consistency between settings despite the sampling bias inherent to inpatient samples. However the result for people with bipolar disorder was significantly different from that for schizophrenia (figure
2). The marked heterogeneity score from the hypertension meta-analysis suggested that there was considerable variation between studies which may have arisen from the differing sampling methods and/or different definitions of hypertension employed in different papers.
Based on our second level of evidence, (namely studies utilizing general population figures for comparison data) the estimated excess risk of diabetes varied between a zero and a fivefold risk (figure
3). Only three of these studies permitted calculation of confidence intervals for diabetes risk ratios and two of these were not significant at the 5% level (figure
3). This wide variation in the magnitude of diabetes risk between studies may reflect differences in 1) the sampling and definition of SMI, 2) the source of the comparison group (and their inherent risk for diabetes), and 3) the definition of diabetes or hyperglycemic outcomes. Furthermore, studies that rely on identification of cardiovascular risk factors in routine clinical practice may be flawed due to differential screening rates in people with and without SMI. In the past, people with SMI may have been less likely to be screened for diabetes. More recently there is evidence that people prescribed certain second generation antipsychotics are more likely to receive screening for diabetes. The direction of bias due to differential screening rates may therefore extend in either direction, leading to underestimation or overestimation of diabetes prevalence in SMI, compared to people without.
This review included several small studies that may have lacked statistical power to detect real differences in risk factors or the metabolic syndrome. Few studies have investigated effect modification by age, but there is some support of this phenomenon when comparing people with and without SMI [
3,
13].
Strengths and limitations
This is the first review to systematically appraise quality and synthesise data from comparative studies of diabetes, hypertension and lipid levels in people with and without SMI.
We paid critical attention to the quality of studies, in terms of the representativeness of samples, the outcomes measured and we present a large volume of comparative results regarding the prevalence of cardiovascular risk in SMI.
We have grouped these studies according to levels of quality within additional file
2; tables 1–5, especially regarding their selection of comparison data and where possible explored the role of different diagnoses and sampling methods in the meta-analysis. There were insufficient papers to allow us to further subdivide the results.
The review was labour intensive, and like other systematic reviews there was inevitably a delay between the search and publication of the review. The search strategy retrieved over 14,000 papers. However narrowing the search terms was not acceptable because the restricted search missed several important papers of which we were aware.
Therefore, during the production of this review further evidence may have emerged subsequent to our original search. From our knowledge of the field we are aware of one quality paper meeting our criteria involving both SMI and controls which was published in 2007. Mackin et al [
49] reported data consistent with the main findings of this review. They compared metabolic parameters in 90 people with severe mental illnesses and 92 without. They report increased rates of cardiovascular risk factors in SMI including impaired glucose metabolism, lower HDL cholesterol and raised LDL cholesterol, raised triglycerides and increased metabolic syndrome. In common with our findings, blood pressure was not raised in this study [
49].
These recent results have not been included in our analysis because this would bias the systematic nature of our review. Legitimate inclusion of this paper would require re-running of the search for other papers from 2007 and reviewing potentially thousands of new titles. This would be beyond the scope of our current funding.
We acknowledge the difficulty of synthesizing data from multiple studies. In this field many existing studies of cardiovascular risk factors have been opportunistic and have not ensured their SMI samples are representative nor that comparison data are comparable in terms of ethnicity and socio-economic deprivation. This may further explain the observed variation in results and heterogeneity. Furthermore several papers which are commonly cited as evidence for increased cardiovascular risk in SMI could not be included as they contained insufficient data or no comparison data.
In particular, studies that rely on clinical diagnoses for outcome definition are problematic, since many people may have cardiovascular risk which is undetected (such as abnormal lipids). Screening for lipids, blood pressure and glucose probably occurs in less than a third of people with SMI during routine practice [
50]. These points may explain why other narrative reviews conclude the risk of diabetes may be even higher in people with SMI [
7,
8].
A further challenge is the employment of different definitions of outcomes such as diabetes, hypertension and metabolic syndrome in different studies. We minimised this problem by only including papers which compare risks in people with and without SMI using the same definition, thus focusing on the relative rather than absolute risk. The relative risk is less sensitive to the employment of differing definitions.
The poor epidemiological quality of studies in this field has been highlighted by a complementary systematic review [
51] examining the relative diabetogenic risk of first and second generation antipsychotics. The authors found methodological weaknesses in most studies and were only able to make tentative conclusions about the possible role of second generation antipsychotics in the aetiology of diabetes.
Explanation of excess risk
No studies longitudinally assessed predictors of diabetes (or other cardiovascular risk factors) in SMI. The relative contribution of second generation antipsychotics [
2,
4,
6], lifestyle[
10], family history [
10], social deprivation and SMI itself (perhaps through chronic stress models) are still debated [
7]. This possible role of SMI itself is supported by two small studies suggesting metabolic disturbances may be observable in newly diagnosed or drug naïve people with SMI [
52,
53]. However our findings reveal that we do not have an accurate estimate of the contribution of either second generation antipsychotics or SMI itself, to support or refute these theories. We know that the relative risk of metabolic harm differs between the second generation antipsychotics [
2,
4,
6,
10] but the absolute attributable risks are unclear.
To understand excess cardiovascular disease in SMI we must improve our knowledge of cardiovascular risk factors in two areas. First, we must accurately determine the extent of the excess of diabetes, dyslipidaemia, hypertension and the metabolic syndrome in representative samples of people with and without SMI, taking into account the effects of age, gender and socio-economic status. This can be achieved in studies with adequate comparison data that explore the level of risk apportionable to SMI and to relevant confounders, especially socio-economic status [
13]. Secondly, to determine the risk attributable to antipsychotic medication, lifestyle, stress and addictions to cardiovascular risk we require data from carefully designed prospective studies. Retrospective estimates of such exposures in SMI are limited by inaccurate reporting and recall bias. Finally, these studies should be adequately powered and should test specific mechanistic hypotheses, rather than measuring multiple risk factors.