Obesity paradox in subarachnoid hemorrhage: a systematic review

Details of the literature search strategy are described in Online Resource 2. Briefly, the search was based on both keywords and index terms and utilized three different databases: PubMed, Scopus, and Cochrane library. A study question and eligibility criteria were formulated using the four-step PICO (Patients, Intervention, Controls, Outcome) principle [17]. Case reports, case series, letters, commentaries, book chapters, animal studies, and descriptive studies without calculated risk estimates (hazard ratio (HR), odds ratio (OR) or relative risk (RR)) were excluded.

To evaluate the heterogeneity and research methods of previous studies, we chose to use the Cochrane Collaboration Handbook and Critical Appraisal Skills Program (CASP) [15]. Based on the appropriate checklists, we focused on selection, detection, and measurement bias; limited and residual confounding; as well as on statistical power and causality between exposures and outcomes. To specify the sources of bias, we created six typical domains to describe these limitations wherein each domain had specific requirements classifying the studies into low, unclear or high-risk-of-bias categories (Online Resource 2). Finally, reviewed studies were grouped into either low-quality or high-quality categories. To reach the high-quality category, all six domains had to be categorized as low-risk-of-bias.

A directional power analysis was performed to estimate sufficient sample size for significant results. The protective effects of being overweight, obese, and morbidly obese were optimistically estimated using an OR of 0.8 for BMI 25–29.9, OR = 0.7 for BMI 30–39.9, and OR = 0.6 for BMI ≥ 40, with a reference 30-day mortality figure for the normal weight category (BMI = 18.5–24.9) approximated to 25% for hospital-based studies and 40% for studies that also included sudden-death SAHs before hospitalization [25]. Standard values were used for significance (p < 0.05) and statistical power (p = 0.8). Additionally, depending on the amount of high-quality studies, I² statistics for heterogeneity and random effects models for calculation of pooled estimates were prepared for possible meta-analysis.

Four studies [8, 11, 18, 46] excluded sudden-death SAHs. Studies missing roughly 25% of all SAHs and about 50% of fatal cases have a theoretical risk for selection bias. Even though one previous study did not find BMI differences between sudden-death and hospitalized SAH patients [27], no comparison was made between fatal cases only. Since the same study found that sudden-death SAH patients have significantly worse overall risk factor profiles than the hospitalized patients [27], even a minimal BMI difference between sudden SAH deaths and hospitalized fatal SAHs may distort hospital-based registers to the extent of losing study reliability. In the Norwegian study with both sudden and hospitalized SAH deaths in the analyses [36], no comparisons between the groups were made.

With regard to obesity assessments, all six studies had several shortcomings. First, three studies [27, 36, 46] measured BMI only at enrollment. Since most adults gain weight until serious illnesses late in life or until the onset of sarcopenia [28], those who were overweight at enrollment are likely to also be overweight or obese at the onset of SAH. Therefore, the use of premorbid BMI is likely underestimating the effect of obesity. To avoid false negative results, future studies with long preictal monitoring periods could perform BMI remeasurements at admission. Second, two studies analyzed obesity as a continuous variable [18, 27], while three used only combined BMI categories, such as non-obese (BMI < 30) [8], non-morbidly obese (BMI < 40) [11], or high BMI (BMI > 25) [46] categories. Therefore, statistical analyses of weight extremes could not be conducted. Since the effect of obesity is unlikely linear (i.e., the fatter the better) and differences by obese severity have also been reported in SAH mortality, the comparison between overweight (BMI 25–29.9), obese (BMI 30–39.9), and morbidly obese (BMI ≥ 40) SAH patients is difficult to conduct if, for example, all patients with BMI values over 25 are combined into one category. Third, two studies estimated BMIs by relying on physician reports [8, 11], and one study used self-administered questionnaires [46] without controlling for measurement bias. All studies used only BMI as an obesity variable. Since BMI sensitivity has been reported to be only 50% for high mass of body fat [30], future studies could also consider other obesity variables with better sensitivity such as waist and hip circumference measurements, magnetic resonance imaging (MRI), dual-energy X-ray absorptiometry (DEXA), or bio-impedance of body.

Only two studies, NIS register-based studies [8, 11] from the USA, fulfilled the criteria of sufficient sample size providing high statistical power for analyses. Any protective effect of high BMI values is likely smaller than our optimistic estimation, so the other four studies [18, 27, 36, 46] likely had insufficient power to achieve significant results. In addition, our directional analysis did not consider effects of potential confounding, thus further underestimating the required cohort sizes.

In a recent large prospective follow-up study, Lindbohm et al. [27] found that the risk for sudden SAH death seems to be highest among people with the worst risk factor profile (i.e., among heavy smokers and people with high blood pressure values). Both smoking [6] and high blood pressure values [14] have been associated with obesity. In terms of smoking, a large study of nearly 300,000 participants [19] reported that the observed obesity paradox in patients with cardiovascular diseases vanished after analyzing only non-smokers. For these reasons, we presumed that both smoking and blood pressure should be treated as confounders in low-risk-of-bias studies of the obesity paradox in SAH. Overall, two studies [27, 46] fulfilled this criterion (Tables 2 and 3).

As potential other confounders, obese patients often suffer from metabolic syndrome, diabetes, coronary artery disease, and peripheral vascular disease. Although these comorbidities might also impact the survival of critically ill patients, no high-quality evidence exists, suggesting an association between these comorbidities and short-term SAH mortality. Moreover, as the prevalence of these comorbidities is low in SAH patients [2, 23], we did not consider these factors as likely confounders, and therefore, they were not regarded mandatory in statistical analyses to reach the classification of low-risk-of-bias. In contrast, other factors, such as SAH severity and rebleeding, have been associated with higher mortality rates; no evidence relates these findings to obesity groups. Furthermore, Rinaldo et al. [34] found an effect of BMI on SAH outcome, and this effect varied depending on the treatment modality (clipping versus coiling). However, the study found no differences after using the validated WHO categories for BMI. In addition, the characteristics between treatment groups varied significantly. Since convincing evidence of treatment confounding in SAH does not exist, we did not include this factor in bias classifications. In brief, future studies about obesity paradox in SAH should try to consider at least smoking and hypertension as potential confounders and perhaps evaluate if composite variables of cardiovascular risk modify the relation of BMI and SAH mortality.

Although the literature search was done systematically utilizing three different databases, we may have missed some relevant publications. For example, large studies with comprehensive and wide-ranging risk factor assessments may have neglected minor and especially negative findings in their indexing. This type of reporting and indexing bias may cause the exclusion of studies like the one by Lindbohm et al. [27]. Furthermore, we used only English search terms, assuming that clinically relevant studies have been conducted in large hospitals and subsequently reported in English. Lastly, we were not able to conduct a meta-analysis due to methodological differences and the absence of high-quality studies.