Location specific risk factors for intracerebral hemorrhage: Systematic review and meta-analysis

Objective To conduct a systematic review and meta-analysis of studies reporting on risk factors according to the location of the intracerebral hemorrhage. Methods We searched PubMed and Embase for cohort and case-control studies reporting on ≥100 patients with spontaneous intracerebral hemorrhage, that specified the location of the hematoma and reported associations with risk factors published until June 27th 2019. Two authors independently extracted data on risk factors. Estimates were pooled with the generic variance-based random effects method. Results After screening 10 013 articles, we included 42 studies totaling 26 174 patients with intracerebral hemorrhage (9 141 lobar and 17 033 non-lobar). Risk factors for non-lobar intracerebral hemorrhage were hypertension (risk ratio 4.25, 95% confidence interval 3.05-5.91, I2=92%), diabetes (RR 1.35, 1.11-1.64, I2=37% ), male sex (RR 1.63, 1.25-2.14, I2=61%), alcohol overuse (RR 1.48, 1.21-1.81, I2=19%), underweight (RR 2.12, 1.12-4.01, I2=31%), and being black (RR 2.19, 1.21-3.96, I2=96%) or Hispanic (RR 2.13,0.94-4.81, I2=71%) in comparison with being white. Hypertension, but not any of the other risk factors, was also a risk factor for lobar intracerebral hemorrhage (RR 1.83, 1.39-2.42, I2=76%). Smoking, hypercholesterolemia and obesity were associated with neither non-lobar nor lobar intracerebral hemorrhage. Conclusions Hypertension is a risk factor for both non-lobar and lobar intracerebral hemorrhage, although with double the effect for non-lobar intracerebral hemorrhage. Diabetes, male sex, alcohol overuse, underweight, and being black or Hispanic are risk factors for non-lobar intracerebral hemorrhage only. Hence, the term "hypertensive intracerebral hemorrhage" for non-lobar intracerebral hemorrhage is not appropriate.The uploaded data is supplemental data belonging to the manuscript "Location specific risk factors for intracerebral hemorrhage: Systematic review and meta-analysis" Funding provided by: HartstichtingCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100002996Award Number: 2012 T077Funding provided by: HartstichtingCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100002996Award Number: 2014 T060Funding provided by: ZonMwCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100001826Award Number: 015.008.048Funding provided by: ZonMwCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100001826Award Number: 016.126.351The uploaded data is supplemental data belonging to the manuscript "Location specific risk factors for intracerebral hemorrhage: Systematic review and meta-analysis" Methods Search strategy and selection criteria We registered our protocol in PROSPERO (CRD42019117543). We searched PubMed and Embase for cohort, case-crossover, and case-control studies on risk factors for ICH published until June 27th 2019 according to the PRISMA statement methodology.18 We used different combinations of the keywords intracerebral hemorrhage and synonyms; cohort, case-control, case-crossover or longitudinal study; and potential risk factors and synonyms (data available from Dryad; see e-1 for detailed search strategy). For this review we did not assess use of (antithrombotic) medication as a risk factor nor did we study genetic risk factors. We used the studies selected in our previous systematic review and meta-analysis for studies published before 2001. We checked reference lists of all included publications and the citation list of our previous systematic review and meta-analysis for additional articles,7 and repeated this until no further studies were found. We applied no language restrictions. Titles and abstracts and subsequently full-text versions were screened independently by two investigators (WMTJ and KW) using the following inclusion criteria: 1) Included patients were 18 years or older; 2) ICH had to be confirmed by CT, MRI, or autopsy in 100% of cases, not only based on International Classification of Diseases (ICD) codes; 3) ICH location had to be specified; 4) A cohort, case-crossover or case-control design; 5) ICH had to be analysed as a separate entity, not in combination with subarachnoid hemorrhage; 6) Reporting on at least 100 patients with ICH. If studies included patients with ICH caused by a vascular malformation, tumour, coagulation disorder (use of antithrombotic medication was allowed), or hemorrhagic transformation of infarction, data extraction needed to allow exclusion of these patients; if not the study was excluded. Conflicts regarding inclusion were resolved by consensus with a third reviewer (CJMK). We used Covidence (www.covidence.org) for standardized screening of articles. Data extraction Data were extracted independently by two reviewers (WMTJ and KW) using a pre-specified and piloted extraction form (data available from Dryad; e-2). Discrepancies in extracted data were resolved by discussion, and if necessary a third reviewer (CJMK) was consulted. In case of multiple publications on overlapping cohorts, we included the study that best matched our inclusion criteria and with the largest amount of data relevant to the review. We extracted data on study period, study design, country of study, in- and exclusion criteria, number of cases and controls, mean or median age, proportion of males, and risk factors. Risk factors were assessed according to lobar and non-lobar (deep and infratentorial) ICH location and if possible, for deep (basal ganglia, thalamus and intraventricular) and for infratentorial hemorrhages (brainstem and cerebellum) separately. We assessed methodological quality, including risk of bias, of the included studies according to the Newcastle-Ottawa Scale (NOS) for cohort and case-control studies.19 Statistical analysis Estimates of cohort and case-control studies were first analysed separately and then combined if 95% confidence intervals (CIs) of the pooled estimates from cohort studies overlapped with those of case-control studies. We also combined maximally adjusted estimates, when available, with unadjusted estimates, if 95% CIs of the pooled unadjusted estimates overlapped with pooled adjusted estimates. If studies used different definitions for risk factors, we standardised risk factors across studies whenever possible or otherwise we accepted the criteria used in the studies. Risk factors reported in at least three studies were combined in meta-analyses; for the different subgroups of a risk factor we accepted two studies for meta-analyses. For the included studies odds ratios (ORs), relative risks (RRs) and hazard ratios (HRs) with corresponding 95% CIs, whichever were available, were obtained for the various risk factors and pooled with the generic variance-based method, weighing individual study results by the inverse of their variance. Heterogeneity was assessed using I2 statistics.20 We used a random-effects model because of the heterogeneous study characteristics. Because ORs accurately estimate RRs when risks of disease are small, we combined ORs with RRs and HRs from the longitudinal studies.21, 22 We performed a sensitivity analysis for studies with a high-quality, defined as studies with >5 points (arbitrarily chosen) on the NOS. Meta-analyses were performed in R (R programming, version 1.1.456), using the meta package (version 4.9-4).23 Data availability Data used in this study are available to qualified investigators on request to the corresponding and senior authors