Preventing cardiovascular disease in hypertensive patients with type 2 diabetes

Introduction Type 2 diabetes affects 3–6% of the adult populations in most countries and its prevalence increases with age up to 25–30% [1]. In addition, another 10–15% of adults have impaired glucose tolerance. Diabetes usually coexists with a cluster of other major cardiovascular disease (CVD) risk factors including dyslipidaemia and hypertension. Diabetes doubles the risk of CVD events and mortality [2], also in hypertensive patients who already have a high risk due to their high blood pressure [3]. Since many hypertensive patients have diabetes, this combination will account for a large proportion of the premature CVD morbidity and mortality in hypertensive patients. Diabetes itself seems to increase the risk of acute myocardial infarction (AMI) and stroke as much as a previous history of AMI or stroke [4]. In Type 2 diabetic patients, the risk of coronary death is significantly associated with a cluster of risk factors related to endogenous hyperinsulinaemia [3,5]. The main aim of the management of hypertensive diabetic patients must be the reduction of the overall CVD risk, rather than treatment of glycaemia or hypertension alone. Nevertheless, prevention of premature CVD events in such patients also demands a particular attention regarding the treatment of hypertension in diabetic patients [6]. Targets for blood pressure lowering in diabetic hypertensive patients The vital importance of effective blood pressure control for the prevention of CVD events in diabetic hypertensive patients has been clearly demonstrated in a number of controlled outcome trials [7–9]. In this respect, diabetic patients do not differ from non-diabetic ones: lowering high blood pressure values is essential. Effective blood pressure control may be even more important than glucose control in hypertensive patients with diabetes [10]. Also, in diabetic hypertensive patients, the treatment targets probably should be even lower that in non-diabetic hypertensive patients. Various guidelines differ to some extent regarding the target blood pressure in diabetic patients; 140–130/80–85 mmHg is commonly used. Most of the hard evidence on benefits of antihypertensive drug therapy in diabetic patients available today comes from subgroup analyses of a rather limited number of trials. In the HOT trial, there was a 51% reduction in CVD event rates in diabetic patients in the < 80 mmHg target group compared with the < 90 mmHg target group [8]. There is a lack of outcome trials specifically designed for diabetic hypertensive patients. Only a few thousand diabetic hypertensive patients have been included in the previous trials that were not designed to test the efficacy of treatment in diabetic patients a priori. Nevertheless, these trials have provided important information on possible treatment benefits in this high-risk group. On the other hand, the clinical and epidemiological knowledge about the very high absolute risk of CVD in diabetic hypertensive subjects has, however, existed for more than 20 years [3]. Therefore, it is difficult to understand why comparative evidence-based data on the potential for preventing CVD with different anti-hypertensive regimens in such patients is still lacking. There has been a debate about the most suitable treatment to be prescribed to diabetic hypertensive patients to prevent premature CVD morbidity and mortality [11]. Since firm data are still lacking on this question, the debate has sometimes been stretched far beyond the facts. Also, the lack of hard outcome data has stimulated the use of various intermediate surrogate parameters. Claims for a superior metabolic profile (glucose, lipids, insulin or uric acid) are the most common [12]. Yet, not a single study thus far has shown that these possible metabolic effects of antihypertensive drugs have to do with the prognosis of treated hypertensive patients. The finding that treatment based on diuretics or β-blockers may trigger ‘latent’ diabetes is generally accepted [13]. It must be emphasized that diabetic patients, however, are no longer at risk of developing diabetes. From this point of view, it is a completely different matter to prescribe antihypertensive treatment to patients with diabetes than to those at risk of developing diabetes. Trial results comparing different antihypertensive drug regimens The Swedish Trial in Old Patients with hypertension-2 (STOP-2) originally reported that elderly hypertensive patients were equally protected by a regimen based on the dihydropydidine calcium-channel antagonists (CaA) as by an angiotensin converting enzyme (ACE)-inhibitor or on a conventional regimen based on diuretics and β-blockers [14]. Several other recent trials comparing the effects of various antihypertensive agents have also come to the conclusion that the outcome is little influenced by the type of antihypertensive drugs [15–18]. The new subgroup analysis from the STOP-2 trial published in this issue of Journal of Hypertension now reports that ACE inhibitors may be useful in the prevention of non-fatal AMI in diabetic hypertensive patients [19]. There were 17 AMI events in the ACE inhibitor group and 32 in the CaA group (P = 0.025). This result resembles that originally reported from the ABCD trial [20]. On the other hand, other unspecified cardiovascular deaths were somewhat increased in STOP-2 on ACE-based than CaA-based therapy (19 versus seven deaths, respectively). There has been a discussion about the validity of the ABCD trial data and the original conclusion regarding the superiority of ACE inhibitors over CaA [20] was probably incorrect. Although the authors of the STOP-2 trial subgroup analysis [19] correctly underline some of the limitations of their post-hoc analysis, these limitations should alert the careful reader about the ultimate significance of the subgroup analyses. Furthermore, performing many subgroup analyses on many outcomes increases the probability of false-positive findings. Adjustment for multiple comparisons is not often seen in reports based on subgroup analyses from trials that have been targeted and powered for testing the outcome in the total trial population. Sometimes it may be difficult to do such an adjustment since the data are analysed separately from the overall results. Therefore, it would be important to lay down the rules as to how P-values derived from subgroup analyses of the trials should be interpreted, i.e. what can be called statistically significant. Subgroup analyses In controlled clinical trials, subgroup analyses are done in two different ways: (i) stratifying by a patient characteristic (exposure phenotype), or (ii) evaluating specific end-points only (stratifying by outcome pheno-type). The interpretation of the findings from a subgroup analysis is never without problems. The main problem related to the stratification by the exposure phenotype in the data analysis is that the randomization groups within a stratum (e.g. diabetes) may not be fully comparable regarding some factor that can influence the outcome. Some of these factors may have been measured and data can be compared and appropriate statistical adjustments can be made, but there are always factors that have not been measured and these can result in spurious results. This exposure phenotype problem can be taken into account, in principle, by a priori stratified randomization for the phenotype of interest, e.g. diabetes. There are, however, too many phenotypes to permit them all be used in the stratified randomization. There are additional problems related to the subgroup analyses using stratification for the outcome phenotype: (i) the completeness of the event ascertainment and the validity of diagnostic classification and (ii) statistical problems with multiple comparisons. For example, in the previously published ABCD trial the case-fatality in AMI was extremely low, around 10% [20]. In comparison, the community-based AMI registries have shown that 35–50% of the AMI events in diabetic patients are fatal [21]. Such a large discrepancy requires an explanation. In the STOP-2 trial, the reduction in AMI incidence in the ACE-inhibitor group compared with the CaA group was compensated by a similar increase in deaths due to ‘other CVD’ in the ACE-inhibitor group [19]. Multiple comparisons The subgroup analysis problem of multiple statistical comparisons when using several end-points can be handled by statistical means. In practice, it means that a correction for multiple comparisons (Bonferroni correction) is made by multiplying P-values by the number of comparisons made in the subgroup analyses. The question is: how many comparisons were actually done? In a final publication, data may be presented for a few outcome phenotypes only, although a large number of outcome phenotypes may have been measured in the study. Often non-significant findings from subgroup analyses are left out from the final publication. In theory, and also in practice, the number of subgroup analyses should be equal to the number of different outcome parameters used in the trial. This argument can be justified by the fact that the investigators often carry out all possible subgroup analyses. In outcome trials related to hypertension this number is around 10. Therefore, I propose that P-values in subgroup analyses for outcome phenotype are systematically multiplied by 10. Thus the ‘borderline’ statistical significance obtained from such subgroup analyses is P < 0.005. Anything higher than this should be considered as a false-positive or a chance finding. According to this principle. the P-value for AMI incidence between ACE inhibitor and CaA groups in the subgroup analysis of the STOP-2 data would be 0.25, which is clearly non-significant. The problem of multiple testing is especially problematic in the contemporary era of genome-wide search for susceptibility genes for complex diseases such as hypertension [22]. Hundreds or thousands of comparisons are made in these studies. Therefore, it has been proposed to use very conservative P-values for the identification of a potential disease susceptibility locus or marker [23]. This is important because the next phase of the search for actual genes is very expensive and time consuming. In comparison, we should be similarly conservative in making inferences from subgroup analyses of controlled clinical trials. The consequences after wrong conclusions from the trial data are similarly time consuming and costly. Treatment policies cannot be based on doubtful findings. Genetic studies use replication of the findings from one study in another study as a means to overcome this problem. In clinical trials, the replication is also important, especially when comparing the efficacy between different treatment modes. Meta-analyses One of the reasons for many false-positive findings or chance findings is that the sample size has been too small. It is somewhat surprising how many clearly underpowered trial results have been recently published, some of them dealing with diabetes and hypertension. A justification for small study sizes could be an innovation and to generate new hypotheses, but it seems that often such studies are just derived from post-hoc findings from secondary data analyses. Meta-analyses can be helpful in pooling the data from small studies and obtaining more stable estimates for the relative risk and benefits from treatments. It should be remembered, however, that the quality of the data in a meta-analysis can only be as good as the data in the original studies. If studies have problems in case ascertainment of end-points or the diagnostic validity is poor, a meta-analysis will not help. Pooling such data will only create more problems with the interpretation, even though the number of events will increase and thus results may look more solid. It is unfortunate that even nowadays some authors are very flexible in their inclusion criteria when carrying out meta-analyses on trial data [24]. It is much better to wait until proper data are available than to carry out meta-analyses on data that are not unequivocally acceptable. The future We must find out the most efficient ways to prevent CVD events in diabetic hypertensive patients due to their very high risk. Debates based over questionable data do not contribute much to solve this important problem. Also, the trial data clearly show that in most diabetic hypertensive patients it is not possible to achieve a satisfactory blood pressure control with one anihypertensive agent alone [8–10,15]. Sometimes even three or more agents are needed, at least in patients with long-term sustained hypertension. Thus, the argument about a possible superior antihypertensive agent may not be the central issue in treating hypertension in diabetic patients. We should instead search for the best possible combination therapy for these high risk patients. In addition, it would be probably necessary to initiate antihypertensive drug treatment in these patients at lower blood pressure levels than the current practice. Diabetic hypertensive patients who have a high risk of premature death and severe CVD events may not be interested in treatments that might marginally reduce some milder non-fatal events, if there is no reduction in mortality. The ongoing large ALLHAT trial [25] will provide very important data on this issue in the future since it has approximately 14 000 diabetic patients randomized to different treatment regimens. The expectations from the ALLHAT trial may, however, be too high, as illustrated by a recent report on the doxazosin arm of the trial [26]. Although it is the first and a sufficiently large outcome trial on this issue, it also has certain limitations in its design and in the choice of the treatment regimens used in this trial. Fortunately, other ongoing outcome trials comparing various treatment regimens in hypertension together include at least another 14 000 diabetic patients, and there are several additional trials under development. Thus there is a good hope that future results will throw better light on this pending question. Until then, it may not be very helpful to try to derive far-reaching conclusions from the existing and insufficient data.