A systematic literature search was performed in the PubMed, Cochrane Library, and EMBASE databases for studies published up to August 3, 2022, by the pre-established search strategy, using the keywords “cerebral hemorrhage,” “ICH,” “intracerebral hemorrhage,” “low-density lipoprotein cholesterol,” and “LDL-C.” The searches were conducted without language restrictions and adapted for each electronic database. The specific terms used for searching in each database, along with the number of records retrieved, are detailed in Supplementary Tables S1–S3. Besides, reference lists of retrieved articles were further screened to identify any eligible studies that didn't come up on the initial search.

Eligible studies satisfied the following selection criteria: (1) either retrospective or prospective cohort studies in nature; (2) enrolled sICH patients who were diagnostically confirmed by computed tomography (CT) or magnetic resonance imaging (MRI) examinations; (3) assessed the correlation between baseline serum LDL-C levels and the mortality risks in sICH patients, and (4) provided with univariate or multivariate-adjusted effect estimates [odds ratio (OR) with corresponding 95% confidence interval (CI)] for the association between LDL-C levels and sICH risks of mortality. The studies were excluded if: (1) these were either reviews, letters to the editor, comments, or meeting abstracts; (2) they included ICH participants with primary traumatic injuries; and (3) these were either duplicate publications or multiple articles based on the same cohort studies with overlapping data. In this case, the one with the most comprehensive results, or the largest sample size, was included.

Two investigators (JL and GL) independently searched and identified eligible literature for extracting relevant data for the meta-analysis, per the pre-determined selection criteria and data extraction strategies concerning the PRISMA recommendations (#PROSPERO CRD42022318318) as illustrated in Supplementary Table S5 (20). The extracted features included the first (co)author name(s), year of publication, study area and design, demographics of participants (gender variation, sample sizes, and mean/range of ages), clinical characteristics (scores on the NIHSS and GCS scales, and ICH volumes), outcome measurements, adjusted ORs with 95%CIs, and adjusted parameters in the multiple factor analysis (MFA). Once completed, investigators exchanged their data audit forms, and if there was any discrepancy, a group discussion was conducted to arrive at a consensus.

The Newcastle-Ottawa Scale (NOS) rating was applied to assess the methodological qualities of eligible articles (21). The scale includes three subscales of subject selection, comparability across groups, and ascertainment of exposure (Supplementary Table S4). Nine NOS stars referred to the maximum score for each article, of which 7–9 NOS stars corresponded to high, 4–6 stars to moderate, and ≤3 stars to low quality.

The ORs and 95%CIs were computed to measure the effects of altered LDL-C levels on mortality risks in sICH patients. The random-effect model (REM) was executed for meta-analyses. The Cochran's Q-test and I² index were used to determine the heterogeneity across studies (22). Results with I² values of > 50% or P-values < 0.05 indicated substantial heterogeneity. Subgroup analyses were performed by grouping studies based on the study area (e.g., China vs. Western) and design (prospective clinical study, PCS, vs. retrospective clinical study, RCS). The sensitivity analysis was conducted by sequential dropping of individual studies to measure the contribution of the respective study to the overall risk assessment. Egger's test was employed to determine any potential publication bias of enrolled studies (23). Stata v12.0 (Stata Corp., USA) was used for all statistical analyses.

The results of the literature search and the literature screening process are shown in Figure 1. Full-text documents were retrieved for 629 articles (226 in PubMed, 372 in EMBASE, and 31 in Cochrane) from electronic databases for meta-analyses. No additional eligible study could be identified in the reference lists of included studies. After eliminating 127 duplicate articles, 502 studies were obtained in the initial screening. Of these, 492 irrelevant articles were excluded by reviewing their titles and abstracts. Then, 10 articles were subjected to full-text evaluation. Finally, 8 cohort studies (16–19, 24–27) were selected for the analysis.

Table 1 summarizes the baseline characteristics of eligible studies. Eight selected articles published between 2009 and 2022 were analyzed. Four studies were from China (17, 25–27), and the rests were from the USA (16), Finland (18), and Spain (19, 24). There were two retrospective and six prospective cohort studies, having sample sizes ranging from 88 to 75,433 and a total of 83,013 participants. The study subjects were all acute sICH patients diagnosed by imaging methods such as CT or MRI. For each included study, the subjects' ages were in the range of 59 to 74 years. Additionally, the in-hospital mortality risk was reported in 4 studies (16–18, 26) and 5 studies (18, 19, 24, 25, 27) documented the 3-month mortality risk in sICH patients. The lower or higher LDL-C level cutoff values varied obviously across the original studies. Multivariate analyses revealed the association between LDL-C level changes and ICH mortality risks. The NIHSS and GCS scores, ICH volumes, OR (95%CI), and adjusted factors are listed in Table 1.

Most studies were rated with NOS scores ranging between 7 and 9 stars (Supplementary Table S4), with six studies scoring ≥8 stars, indicating all finally included studies were of high quality. The main sources of bias in these studies were recall bias and confounding bias.

Figure 2 shows the forest plot for in-hospital and 3-month mortality risks. Four studies including 77, 855 patients, estimated the risk of in-hospital mortality and the patients' LDL-C levels, indicating no significant association between the elevated LDL-C levels and risks of in-hospital mortality (OR: 0.92; 95%CI: 0.63-1.33; P = 0.647) in sICH patients, with substantial heterogeneity across studies (I² = 91.4%; P < 0.001).

Moreover, five studies including 6,122 patients showed that alteration in LDL-C levels and the risk of 3-month mortality had a strong association, with a pooled risk estimate of 0.51 (0.33, 0.78; P = 0.002) and no statistically significant heterogeneity (I² = 47.8%; P = 0.104) across studies. In summary, a higher serum LDL-C level may have a beneficial effect on decreasing the risk of death in sICH patients.

For the in-hospital mortality, the pooled results of the three subgroups (China, PCS, and RCS) were not statistically significant and highly heterogeneous (P > 0.05; I² > 50%). However, the pooled results for the Western subgroup were significant (OR: 0.67; 95%CI: 0.57–0.78; P < 0.001), with almost no heterogeneity among the studies (I² = 0%, P = 0.466) (Table 2).

For the 3-month mortality, the pooled estimates and size of heterogeneity for both subgroups of China and PCS were consistent with the original pooled results, whereas the pooled results for the Western and RCS subgroups were not significant (P > 0.05) (Table 2).

Overall, the study design variation did not have any effects on the LDL-C level's association with the in-hospital mortality risks of sICH patients, unlike the association with the 3-month mortality risk. Interestingly, the region of the study conducted seemed to influence the association of LDL-C level modulation with either in-hospital or 3-month mortality risk in sICH individuals. In other words, a higher LDL-C level may be associated with decreased in-hospital mortality risks in the West and 3-month mortality risks in China.

As shown in Table 3, the sensitivity analysis revealed that the pooled risks were significant for both in-hospital and 3-month mortality risks in sICH, and no major change was observed in the pooled estimation even if any individual study was eliminated at each turn. Moreover, the change in the pooled effect size for in-hospital mortality risks, from 0.78 (0.51, 1.22) to 1.05 (0.82, 1.52), was not significant (P > 0.05). Likewise, the leave-out one study sensitivity analysis of 3-month mortality also indicated a significant (P < 0.05) stable pooled OR (95%CI), ranging from 0.47 (0.34, 0.65) to 0.62 (0.44, 0.89).

The Egger's test could not identify any publication bias for both in-hospital (P = 0.724) and 3-month mortality (P = 0.100) risks for analyzed studies (Table 3). The funnel plot asymmetry analysis could not be performed for each subgroup due to the inclusion of fewer than 10 articles (28).

Several different explanations regarding underlying mechanisms may be considered in this context.

First, LDL-C plays key functions in many physiological processes, such as maintaining vessel wall integrity. Reportedly, most healthy individuals are born with LDL-C levels ranging from 40 to 60 mg/dL (33). While lower LDL-C levels can be independently related to brain microhemorrhages under acute ICH conditions (34), which can further contribute to poor treatment response, stroke recurrence (35), and perihematomal edema expansion (36). The possible mechanisms are as follows: (1) disintegration of the endothelium (37); (2) necrotic death of medial smooth muscle cells (MSMCs) (38); (3) inhibited platelet aggregation (39); (4) accelerated osmotic membrane rupture in erythrocytes (40), and (5) impaired synaptic reconstruction (41). ICH patients with genetically reduced levels of LDL-C are more likely to carry APOE ε2/ε4 allele, indicating recurrent ICH (42, 43). Theoretically, high LDL-C levels could rescue ICH patients from hematoma enlargement and mortality risks. Further prospective studies are essential to assess an optimum LDL-C level to prevent adverse circumstances in ICH.

Second, the inclusion of observational studies prevented establishing a causal relationship in this meta-analysis. An excessively low LDL-C level can reversibly increase the mortality risk (44), but it may not necessitate a causal relationship in all cases (25). Whether LDL-C levels change over time due to physiological and clinical reasons is still unclear. A longitudinal study shows a subacute reduction in LDL-C levels preceding sICH (45), and both LDL-C and triglyceride levels can decline at the disease onset and restore to normal levels during the recovery phase (46–48). Debilitation and/or disease also can lower LDL-C levels (49, 50), which might be a surrogate for malnutritional or a sign of severe disease (51, 52), thus predisposing the individual to increased stroke mortality (53), indicating lower LDL-C level could induce chronic health problems and higher mortality, especially in the elderly (54). Hypocholesterolemia can lead to sepsis, adrenal failure, and increased mortality risks in critically ill patients (46). It is hypothesized that the pathophysiological abnormality of ICH patients causes higher mortality and reduction in LDL-C levels, in parallel, which may not be rescued by treatments directed to normalizing a single dysfunction (55). Thus, adjusting the LDL-C level to an optimum may not likely modify the death risk in sICH subjects. Furthermore, elderly pre-ICH statin users are at higher risk of comorbidities due to their lower LDL-C levels, just like individuals under antithrombic medications with an elevated risk of bleeding and stroke (56). Due to insufficient data on pre-ICH statin use, we could not evaluate the relationships between statins, sICH outcomes, and low LDL-C levels. However, we investigated the relationship between sICH outcomes and serum LDL-C levels. Besides, the pre-statin exposure rate among sICH is relatively low, without any independent correlation with worse treatment responses in sICH patients (18, 57), which is consistent with other meta-analyses suggesting that pre-ICH statin application might not have any association with post-ICH mortality risks (56, 58–60). Above all, the reciprocal correlation between the LDL-C level at admission and mortality risks in sICH patients should be interpreted with caution. Further large-scale studies are needed to further elucidate the impact of low LDL-C levels and ICH death risk.

Although conventionally lowering lipid levels is considered the best to control atherosclerotic cardiovascular diseases (61), the effect of statins on optimally regulating LDL-C levels still needs more validation to gauge the risk of sICH in atherosclerotic patients. Since chronic use of statins increases bleeding complications, while its sudden discontinuation can predispose to ischemia, the application of statins in acute sICH patients is highly debated (62). Similarly, a recent study (10) advised against abruptly discontinuing statin medication in cases of acute ICH without first conducting a thorough health assessment. In contrast, a prior Markov decision analysis endorsed that avoiding statins is expedient in all ICH survivors after weighing these competing risks and benefits (63). However, individualized treatment decisions based on expert consultation have been advocated by recent guidelines (64). Furthermore, due to the limited data availability, our meta-analysis was unable to identify the threshold value, and we could not comprehend the linear or non-linear association of serum LDL-C with mortality risk in sICH patients. The previous report showed low LDL-C levels (<100 mg/dL) containing subjects had enhanced 3-month mortality and were more prone to hematoma enlargement than their counterparts with higher LDL-C levels (110–129 mg/dL) (25). Particularly, <70 mg/dL can increase the risk of hematoma expansion and in-hospital mortality by several folds compared to subjects with higher LDL-C levels (17, 19, 65). Statin treatment in ICH patients with relatively low LDL-C should be approached with caution due to the increased lifetime risk. Hence, randomized controlled trials are urgently necessary to resolve the complicacy of statin use in patients with higher risks of ICH.

We found that the elevated level of LDL-C exhibits an inverse relationship with the in-hospital and 3-month mortality risks, respectively, among post-ICH patients in western countries and China. To the best of our knowledge, this is the first meta-analysis reporting an assessment of LDL-C level variation in relation to sICH mortality risk. The strengths of the study include its extensive literature search, stringent criteria, and thorough scrutinization of included studies. Despite the limited number of articles, the sample size was considerably large, and the accuracy of the meta-analysis results was relatively satisfactory. Moreover, the methodological quality control for included reports matched the international standard. Despite the influence of confounding factors, like pre-treatment medications, ICH volumes, admission NIHSS scores, and hypertension, all studies produced highly objective and reasonable observations after multiple factor adjustments. Furthermore, the sensitivity analyses and Egger's test suggested the good reproducibility and high credibility of the pooled results across the included studies. Finally, we strictly adhered to the PRISMA recommendations for study screening, data extraction, quality control checking, and analyses to reduce recall bias, especially for observational studies.

Yet, we note that this review does have some limitations. First, being a study-level meta-analysis, this study could not consider methodological variations in analyzed articles. It wasn't possible to adopt a uniform adjustment of variables for all studies. Also, a standardized analysis of the age and sex of the participants could not be performed to identify potential confounding factors for heterogeneity. An individual-level meta-analysis will be appropriate to evaluate the relationship between ICH mortality risks and low LDL-C levels in terms of demographics, existing medications, and comorbidities.

Second, in all cases, only one-time measured post-ICH serum LDL-C level values were included for analysis, which might introduce classification bias. There were no pre-ICH onset data for these patients, and possible variations in LDL-C levels during the hospital stay and follow-up periods were not considered. It remains unclear whether an acute sICH condition induces LDL-C level alteration and/or whether the reverse phenomenon is the actual culprit in elevating the mortality risk. Time-average serum LDL-C level analysis can be useful to verify the association of pathological LDL-C levels with higher mortality risks.

Third, different cutoffs for LDL-C levels were used in the prediction of mortality risks across studies, and the methodological heterogeneity might lead to the discrepancy in results, which prevented us from determining the optimal value of LDL-C level in better predicting ICH survival. And the enrolled studies were ineligible for the dose-response analysis because they lacked the necessary data (e.g., the number of cases in each LDL-C level). Additionally, abnormally low LDL-C levels might have underpowered our results. We expect future studies to analyze clinically uncommon subsets of patients critically for better understanding.

Fourth, all eligible studies involved cohort analyses, which had a higher susceptibility to bias from confounding factors than randomized controlled trials. Residual confounding biases are sometimes unavoidable due to the omission of or insufficiently measured variables. Some important confounding factors related to mortality risk should be adjusted as comprehensively and consistently as possible in the statistical model, while the severity of ICH (GCS and NIHSS scores), hypertension, pre-medications, divergent inciting causes of ICH (blood pressure, cerebral amyloid angiopathy, anticoagulation, and vascular anomalies), complete neuroimaging data (hematoma shape, volume, and site), and time from ICH occurrence to cranial CT scan were not fully captured in several of these studies. Inadequate adjustment may have resulted in an overestimation of the risk estimate.

Fifth, most participants were Han Chinese and Caucasians, which limited the generalizability of our observations to the general population, but it might have some reference values. Caution should be paid to extrapolating these results to other ethnic groups. The association and optimal range of LDL-C level change may differ across ethnic groups due to their varying baseline LDL-C levels, environments, and individual risk factors.