Influence of Lipid Profiles on the Risk of Hemorrhagic Transformation after Ischemic Stroke: Systematic Review Open Access

Hemorrhagic transformation (HT) of cerebral ischemia may occur either spontaneously or after treatment with recombinant tissue plasminogen activator (rt-PA) [1,2,3]. Although HT may sometimes be related to early recanalization and associated with better clinical outcomes [4], it is more often a predictor of worst clinical outcomes [2]. HT is favored by increasing age [2], stroke severity [5], blood pressure [6] and blood glucose level [7]. Identifying other factors predisposing to HT may lead to new preventive strategies. The risk of spontaneous intracerebral hemorrhage (ICH) increases with lower total cholesterol levels in humans [8] and animals [9], but not in patients treated with intensive lipid-lowering therapy after ischemic stroke [10].

We systematically reviewed published studies to determine the influence of lipid profiles on the risk of HT.

On March 2, 2011, we searched MEDLINE (http://www.ncbi.nlm.nih.gov/pubmed) from 1966 and EMBASE (http://ovidsp.ovid.com) from 1980 for studies that investigated the association between lipid profiles and HT in patients with acute cerebral ischemia. Because there is no medical subject heading (Mesh) term for HT, the search strategy in MEDLINE was based on the association of the following terms: [‘Stroke’ (Mesh) OR ‘Cerebral Infarction’ (Mesh) OR ‘Infarction, Middle Cerebral Artery’ (Mesh) OR ‘Brain Ischemia’ (Mesh) OR ‘Intracranial Hemorrhages’ (Mesh)] AND [‘Lipids’ (Mesh) OR ‘Cholesterol’ (Mesh) OR ‘Cholesterol, HDL’ (Mesh) OR ‘Cholesterol, LDL’ (Mesh) OR ‘Triglycerides’ (Mesh)]. In EMBASE, the search strategy was: stroke (exp) AND (lipid or cholesterol or HDL or high density lipoprotein or LDL or low density lipoprotein or triglycerides).

Studies missed by the electronic search have been identified from the authors’ personal files and by hand searching the references cited in retrieved articles. We excluded abstracts of meetings and book chapters. We did not exclude any language. One reviewer (K.N.) performed the literature search and screened the titles and abstracts of retrieved citations to identify potentially suitable studies. The abstracts were read by 2 reviewers (K.N. and D.L.). Uncertainties and disagreements were resolved by consensus. The 2 reviewers extracted data from included studies.

Studies meeting the following criteria were included in the analysis: (i) adult patients with acute cerebral ischemia; (ii) a systematic search for HT in ischemic stroke patients; (iii) data on total cholesterol, and/or LDL and/or HDL cholesterol, and/or triglycerides, and (iv) a computed tomography (CT) or magnetic resonance imaging (MRI) scan of the brain for the assessment of HT. We excluded studies (i) published as reviews; (ii) including patients with intracranial hemorrhages of other origin than HT, and (iii) based only on autopsies.

The external validity of the studies was defined as follows: level A for population-based studies recruiting consecutive patients; level B for hospital-based studies recruiting consecutive patients; level C for hospital-based studies recruiting a subgroup of consecutive patients encountered with a high frequency (e.g. treated by intravenous rt-PA and age >45 years), and level D for either nonconsecutive patients or patients from a rare subgroup (e.g. patients treated by intra-arterial recanalization).

The first step consisted of reporting for each eligible study, the mean values (SD) of total, LDL and HDL cholesterol, and triglyceride levels separately for patients with and without all types of HT (_allHT), then separately for those with symptomatic HT (_sHT).

The second step consisted of a meta-analysis of eligible studies for the comparison between _allHT versus no HT, for total, LDL and HDL cholesterol, and triglyceride levels, then between _sHT and absence of _sHT for the same variables. We used the weighted mean difference (WMD) method. The research question for the meta-analysis was the influence of baseline levels of lipids on the risk of HT. The comparison was performed between HT and non-HT patients (then between _sHT and asymptomatic HT patients). The outcome was the baseline level of lipids (total, LDL and HDL cholesterol, and triglycerides evaluated separately). Statistical significance of the overall results was expressed with the pvalue in the test for overall effect, with p < 0.05 being considered as statistically significant. According to ranges for interpretationof I² following the Cochrane Handbook for Systematic Reviews of Interventions [11], I² values ≤40% were considered as indicating homogeneity between studies. When I² was ≤40% (p values are reported and p > 0.05 show evidence of no heterogeneity), we used a fixed model for meta-analysis. Statistics were performed with the STATA 11 software. The analysis was conducted according to the MOOSE consensus [12].

The flow chart of the selection of studies is reported in figure 1. Eight studies gathering 1,763 patients were eligible [13,14,15,16,17,18,19,20].

Most patients included in these studies were recruited between 2000 and 2007 [13,14,15,16,17,18,19,20]. All studies were published between 2007 and 2009 [13,14,15,16,17,18,19,20]. Of the 8 studies selected [13,14,15,16,17,18,19,20], 3 were from the same research group [13,18,19]. Two of them (including 168 patients) were not included in the meta-analysis because they did not provide comparisons between patients with and without HT. There might be a possible overlap between these 3 studies including 104, 142 and 26 patients, precluding all influence on the meta-analysis [13,18,19]. Only 1 study had a multicenter design [20], and none was international. None of these 8 studies was conducted in the setting of a clinical trial. Patients were consecutively recruited in 5 studies [13,15,16,17,20], not in 2 [18,19], and the information was not reported in 1 [14]. Patients were, however, prospectively recruited in all studies [13,14,15,16,17,18,19,20]. The lipid profile was not determined specifically to address the question of the relationship with HT in 7 of the 8 studies [14,15,16,17,18,19,20]. Whether it was a prespecified question remains uncertain in 1 study [13]. Five were sponsored by public or institutional grants [13,15,16,18,19], and no detail on sponsoring was provided in 3 [14,17,20].

The characteristics of patients are detailed in table 1.

In 4 studies, patients were in the fasting state [13,18,19,20]. Previous hypercholesterolemia was reported in 6 studies [14,15,16,18,19,20], and ongoing statin therapy in all but 1 study [17]. No information about ongoing fibrate therapy or diet before stroke onset was available in any study. The method used to analyze lipids was detailed in 1 study [15]. All studies reported detailed results for LDL cholesterol levels [13,14,15,16,17,18,19,20], 7 for total cholesterol [13,14,15,16,17,18,19], 6 for HDL cholesterol [13,14,15,16,17,19] and 5 for triglycerides [13,14,15,16,17].

The methods used to assess HT are detailed in table 2, and associated risk factors for HT are detailed in table 3. The methods used for analysis in each individual study are detailed in table 4.

Individual Studies. The comparison of lipid profiles was possible in 2 studies [16,17] for _allHT versus no HT, and in 4 for _sHT versus no _sHT [13,14,15,17] (table 5).

Meta-Analysis. The meta-analysis of comparisons of lipid profiles in patients with and without _allHT included 2 studies [16,17] gathering 688 patients. Detailed results are presented in figure 2 and table 5. LDL cholesterol levels were significantly lower in patients with _allHT (p = 0.008), but not total cholesterol (p = 0.129), HDL cholesterol (p = 0.066) and triglycerides (p = 0.900).

The meta-analysis of lipid levels in patients with and without _sHT included 4 studies [13,14,15,17]. These 4 studies included 727 patients [13,14,15,17]. One study [14] provided only mean values and no standard deviation. We used for this meta-analysis the standard deviation found in the study with the closest number of patients [13]. The detailed results are presented in figure 3 and table 5. Total cholesterol levels were significantly lower in patients with _sHT (p = 0.035), while LDL (p = 0.056) and HDL cholesterol (p = 0.138), and triglyceride levels did not differ (p = 0.851).

This meta-analysis suggested that lower LDL cholesterol levels are associated with _allHT, while lower total cholesterol levels are associated with _sHT, and HDL cholesterol and triglycerides are not associated with the HT risk.

None of the 8 studies meeting the selection criteria [13,14,15,17] was primarily designed to evaluate lipid profiles in HT explaining missing data, and there was heterogeneity for the assessment of HT. Most predictors of HT were taken into account except for silent vascular brain lesions, but a multivariate analysis including these predictors was available in only half of the studies. The external validity of the study population is therefore poor.

The method used to analyze serum lipids was usually not detailed, and the delay between stroke onset and sample collection ranged from admission to the next morning or 7 days.A previous study suggested changes in cholesterol levels at the acute stage of stroke [21], but as this change is minimal [22], this has probably not had any influence on the results. Information on ongoing statin and fibrate therapy, and diet was not available in any study. This may have influenced the results, as statins and potentially fibrates may be associated with a better outcome in ischemic stroke [23,24].

It was not possible to analyze separately patients with and without reperfusion therapy, and according to the type of reperfusion therapy, because individual data were not available, and this would have led to subgroup analyses with a poor statistical power.

The meta-analysis showed a significant association between a low LDL cholesterol level and the presence of _allHT, and between a low total cholesterol level and the presence of _sHT. These results are important, because 1 of the 2 individual studies included in the meta-analysis for _allHT showed a nonsignificant tendency [17] and the other [16] was marginally significant. The influence of HDL cholesterol seems more disputable because none of the studies has shown any significant result for both _allHT and _sHT, and the meta-analyses provided nonsignificant results in opposite directions. There is no statistical association between baseline triglycerides and risk of both _allHT and _sHT, although 1 individual study [15] suggested higher triglyceride levels in _sHT.

Therefore, the association between lipid profile and risk of HT is likely to be mediated by LDL cholesterol. This result is important because LDL cholesterol may be influenced by statins. The question of whether rt-PA is safe in patients with low LDL cholesterol levels or on statin treatment is therefore important: although patients treated with statins have less severe cerebral infarcts [25], a large European multicenter study showed that this effect is lost in patients treated concomitantly with rt-PA [26]. The underlying mechanism of increased HT with low levels of LDL cholesterol is not established. Cholesterol may be important for the integrity of small vessels [8] and of the neurovascular unit.

Although the mechanisms of HT and spontaneous ICH are different, it should be emphasized that the results of this meta-analysis mirror the findings of the INTERSTROKE [27] study showing a link between low LDL cholesterol and the risk of spontaneous ICH but not with those of SPARCL showing no relation between a low LDL cholesterol level and the risk of ICH [10].

This systematic review suggests that patients with low LDL cholesterol levels have an increased risk of HT, but is not conclusive for the influence of HDL, and suggests no influence of triglyceride levels. The results of the meta-analysis are limited by methodological issues: (i) lipid profiles were not determined specifically to address the question of the relationship with HT in 7 studies and were heterogeneously assessed; (ii) no study was conducted to test the hypothesis that is the subject of this meta-analysis, and (iii) heterogeneity for the method of assessment of HT and unclear definition of HT or of its symptomatic presentation. Therefore, there is a risk that unmeasured biases could have affected the results.

The main limitations of this study are that: (i) none of them was designed to address the study question; (ii) the HT definitions were not consistent among studies, and (iii) there were significant issues related to the measurement of lipid parameters. Therefore, these limitations of the data make conclusion quite weak unless confirmed by another large and properly designed study. The questions that should be addressed now are: (i) what is the mechanism of this increased risk of HT with low levels of LDL? (ii) Do statins contribute to the increased risk of HT or do anti-inflammatory properties of statins counterbalance this effect? (iii) Are these results valid in patients who were treated with rt-PA? Prospective studies recruiting non-selected ischemic stroke patients (treated or not by rt-PA) and designed specifically for that purpose as well as studies in animal models are required to solve these issues.

None.

K.N. was supported by grants from the European Neurological Society and from the Groupe Lillois de Recherche en Pathologie Vasculaire.