Although our primary aim was to study the largest possible number of neurological conditions (which we expected to have been analyzed in umbrella reviews), we discovered that the umbrella reviews had studied only 12 neurological conditions. For instance, we could not find umbrella reviews on risk or protective factors for some common or major chronic neurological disorders such as migraine, headache, brain cancer, or epilepsy. Therefore, future umbrella reviews should be considered regarding the non-purely genetic risk factors of these conditions. Interestingly, almost all studies had focused on neurological disorders with high prevalence and in resource-rich countries, which could be indicative of the disproportionally lower number of publications regarding meta-research for global health neurology, namely neurological diseases of resource-poor countries (e.g., meningitis, neurocysticercosis), as well as for rare/orphan neurological disorders.
Principal findings
We studied 115 distinctly named risk/protective factors with a marked association with chronic non-communicable neurological disorders, including biomarkers, habits, dietary factors, medical history or/and comorbid diseases, drugs, and exposure to toxic environmental agents. Fourteen factors exhibited a decreased risk for an extensive number of non-communicable neurological disorders, with these factors ranging in strength from class I to class IV. Below, we provide our insight into some of these findings.
Notably, the following associations appeared with class I evidence: (a) in neurodegenerative diseases, dementia, and AD, Mediterranean diet was a protective factor; (b) in MS, smoking, anti-EBNA IgG seropositivity, as well as infectious mononucleosis were risk factors; (c) in ALS, lead was a risk factor; (d) in PD, physical activity was protective, while constipation was a risk factor (although serious concerns were previously raised [
50]); (e) in AD, late-life depression and type 2 diabetes mellitus were risk factors; (f) in dementia, depression at any age, life depression, frequency of social contacts, and benzodiazepine use were risk factors; (g) in vascular dementia, type 2 diabetes mellitus was a risk factor; (h) in stroke mortality, high uric acid levels were a risk factor.
While several risk and protective factors had class III and IV evidence of being significant in the occurrence of these neurological conditions, three of them, namely tobacco smoking, hypertension, and serum uric acid, exerted a mixed risk and beneficial effect. Based on the I2 metric, heterogeneity was present in published reports, and few studies were consistent with non-heterogeneous evidence when data had a prediction interval excluding the null.
With regard to dietary factors, we found substantial evidence highlighting the potential role of Mediterranean diet in lowering the risk of dementia, AD, cognitive impairment, neurodegenerative diseases, and stroke. Until now, several meta-analyses have reported quite solid evidence of the beneficial effect of Mediterranean diet in AD and other dementias, i.e., major categories of neurodegenerative disorders (for an example, see umbrella reviews [
81] and [
98]). However, discrepancies have been reported regarding the cardiovascular benefits of Mediterranean diet across socioeconomic groups [
99]. Because of different reporting methods across studies in the field, development of standardized tools is imperative for the assessment of the effectiveness of Mediterranean diet in preventing cognitive impairment and neurodegenerative diseases. In a similar context, a recent study, consisting of a series of meta-analyses and including more than 130 million person-years of data from more than 240 original studies, presented quite solid evidence of low glycemic index food intake in stroke reduction [
88].
In parallel, negative associations between coffee consumption and PD and AD [
100] have been reported, with these findings being consistent across study designs and geographical settings. The biological mechanism(s) underlying this protective effect remain(s) unclear. For example, regular coffee intake enhances insulin sensitivity and, hence, reduces the risk of diabetes mellitus type 2, which itself is a strong risk factor for cognitive decline [
101]. Also, recent meta-analyses, having considered the plausible roles of numerous modifiers, suggest that a 3.5-cup/day coffee intake is inversely associated with all-cause mortality, an association that has remained undiluted even after adjusting for major modifiers, such as aging, smoking, and alcohol consumption [
102].
This systematic review of umbrella reviews revealed counterintuitely a significant association of high serum uric acid levels with a decreased risk of several neurological diseases (i.e., AD, PD, dementia, MS, neuromyelitis optica, and ALS), while diabetic peripheral neuropathy and stroke mortality were associated with an increased risk. Our credibility assessment revealed that, with the exception of PD (with class II evidence) and stroke mortality (with class I evidence), these significant associations were within class IV evidence [
82]. Hence, no definitive conclusion could be made in favor or not of intensive lowering of serum uric acid levels in light of a putative higher risk for neurological diseases [
103,
104]. Further mechanistic studies are needed in this field, using appropriate animal models for each distinct disease entity. Also, clinical trials of increasing serum uric acid in neurological disorders have been conducted [
105,
106].
According to class I–IV evidence, physical activity was found to exert a beneficial effect against PD, AD, and all types/vascular dementia. Physical exercise can increase serum uric acid levels, which has been associated with a lower risk of developing PD and dementia [
82,
104]. However, patients with PD may be unable to exercise much owing to neurological dysfunction, which might indicate reverse causation [
107].
Serum vitamins B
12, C, and D levels were associated with a lower risk of different neurological conditions, such as MS (as also reported recently [
108]), AD, dementia/cognitive impairment, and PD. Around 80% of these meta-analyses represented heterogeneous evidence (
I2 > 50%), which cautioned against false interpretations. The observed heterogeneity most likely arose from different comparison groups in prospective, retrospective, and case-control studies, causing some of the meta-analyses to be derived from studies with diverse, contrasted categories of serum vitamin B
12, C, and D levels [
49‐
51]. Furthermore, strong evidence links the presence of anti-EBV antibodies to MS (for further discussion, see [
109]).
Our meta-umbrella review provides some evidence for a positive association of exposure to farming, pesticides, low-frequency electromagnetic fields, organic solvents, and
C. pneumonia infection with the occurrence of several neurological conditions (such as MS, PD, and ALS). However, most of these associations were based on class III and IV evidence, which could have resulted from the substantial heterogeneity among the primary studies. Hence, these associations warrant cautious interpretation. We also suggest that the findings on chronic cerebrospinal venous insufficiency should be interpreted with caution, considering both the wide range of the corresponding confidence intervals and previous reports in the field [
110,
111].
Framing our
meta-umbrella review into the broader context of studies reviewing risk and protective factors for neurological disorders, we noticed that, in another comprehensive review of systematic reviews, for example, exposure to pesticides was identified as the commonest risk factor for AD, ALS, and PD, whereas smoking was associated with AD and MS [
112].
Smoking as an exemplar of studying risk and protective factors for neurological disorders
Below, we discuss the findings on the effects of tobacco smoking in a separate section. This choice was made because we consider that, with all the body of evidence surrounding this field, smoking should represent an exemplar for studying risk and protective factors for neurological disorders, or, as other authors have previously claimed, represents
the poster child of causal relations [
113]. We found that tobacco smoking is linked to an increased risk of MS (class I evidence), dementia (class IV evidence), and vascular dementia (class IV evidence), but also to a decreased risk of PD (class II evidence). A positive association exists between tobacco smoking and MS, with convincing (i.e., class I) evidence of, at least, a modest effect [
51], even though confounding effects cannot be totally denied. More broadly, tobacco smoking has been included in the five principal risk factors that could explain around two out of three initial manifestations of demyelination (further reviewed in [
114,
115]). Mechanistically, adverse immuno-modulatory effects, demyelination, and the disruption of the blood-brain barrier could be accountable for the positive association between smoking and MS, even though this remains to be proven [
116]. Of note, the effects of smoking are now well-established regarding lung inflammation, the latter also linked to a high risk for MS [
115]. Of particular interest is also the role of oral tobacco (
snuff) usage, which was considered to be associated with a lower risk of MS, potentially through nicotine-mediated effects on subunits of immune cells expressing acetylcholine receptor [
115].
Another possibility could be that people suffering from a certain neurological disorder, such as MS, prefer to smoke, whereas those unaffected choose to stop smoking more easily, as previously observed in patients with schizophrenia [
117]. Therefore, there is concern that, since retrospective studies had been included in the initial meta-analyses, these could have introduced a bias in the relevant results of this
meta-umbrella approach. Perhaps, in this specific field, it would have been probably wiser to select only the meta-analyses of prospective studies among the umbrella reviews. Similarly, another possibility could be to consider only umbrella reviews that have examined
credibility ceilings [
118], in order to assess effect estimates in combination with other sensitivity analyses (i.e., to include only prospective studies to assess temporality and reverse causation, or to perform the so-called
credibility ceilings, which take into consideration limitations regarding the methodology of the studies) [
37,
69,
119]. Nonetheless, this option would have been a rather laborious process in the context of this, already extensive,
meta-umbrella approach. Besides, it is commonly known that extensively performing sub-analyses in many subgroups could be linked to artificially increasing events of statistical significance. In every case, the
teaching example of cross-sectional studies on lung cancer and smoking (in which case, patients with lung cancer tend to quit smoking) for causing inverse causation should always be kept in mind [
37].
With regard to PD, the potential underlying genetic and non-genetic roots (or/and bias) of the association between tobacco smoking and PD are reviewed elsewhere [
120]. However, caution is needed in distinguishing epidemiological terminology (e.g.,
suggesting that longer duration of smoking is needed for a risk reduction, as cited in the above study) from core public health messages.
In every case, we feel that the core message of promoting tobacco smoking cessation as an effective public health intervention should remain undiluted because of its several well-established positive health effects [
121,
122], irrespective of whether tobacco cessation might also decrease the incidence and/or severity of MS [
123] and regardless of genetic susceptibility to smoking habits [
121,
124]. Thus, we feel that the example of smoking, acting both as a risk factor for certain diseases and a protective factor for others, should not serve as an opportunity of potentially diluting a key public health message, or even counseling MS-affected patients or their family members who are at higher-than-normal risk in favor of smoking [
125].
Additional features with class I evidence
Below, we wish to highlight some additional features with class I evidence. Chronic occupational exposure to lead presented a higher risk for ALS. Arguably, lead toxicity represents a major underlying mechanism in ALS-related pathogenicity [
126,
127]. In humans, lead toxicity manifests as clinical symptoms similar to those in ALS, such as weakness originating in the finger extensors or the wrist and, ultimately, spreading to additional muscles. Also, the blood of patients with ALS revealed higher levels of lead exposure-biomarkers than the respective levels in healthy controls [
127]. In potential future research, lead toxicity should not be considered
in vacuum but rather in association with other heavy metals and welding, even though the latter two are classified in lower levels of evidence (class II and class III, respectively). Moreover, when lead levels are taken into account, the association of another heavy metal, i.e., copper, with ALS risk becomes attenuated, suggesting a chief role of lead [
128], even though certain isotopic compositions of copper have been detected at higher levels in the CSF of ALS patients than of AD patients or healthy controls in other studies [
129]. Interestingly, occupational exposure to silica has also been implicated in ALS risk [
130]; thus, it would be worth exploring whether silica (which belongs to the same family of the periodic table as lead does) could explain these traits.
Constipation was positively associated with PD (class 1 evidence). A prospective cohort study reported a significant association with a similar effect size in meta-analyses (reviewed in [
50]). Another study reported that constipation could be a symptom of PD but also a premorbid symptom preceding motor dysfunction symptoms of PD by at least 10–20 years [
131]. Nowadays, constipation is regarded as a manifestation of PD via the peripheral nervous system, a condition in which the threshold for the appearance of symptoms may be decreased. This is perhaps because of the larger functional reserve of the midbrain dopamine and integrated basal ganglia motor systems to control movement [
132]. In any case, the connection between PD and gut dysfunction seems quite solid. In this context, laboratory studies have demonstrated an abnormal deposition of α-synuclein within the enteric nervous system, while, recently, the gut-to-brain α-synuclein’s spread (which is related to the Braak hypothesis) through the vagus nerve has been demonstrated in mouse models [
50,
133,
134]. Moreover, a recent study of a huge cohort of 1.6 million subjects reported that the physiological human appendix contains intraneuronal α-synuclein and misfolded aggregates, and that removing the appendix early in life reduces the risk of developing PD [
135]. Lastly, any causal association between beta-2-adrenoreceptor antagonist (beta-blocker) and higher risk for PD appears weak in terms of its evidence [
136].
Our meta-umbrella review assessed specific risk factors related to dementia and AD. While only late-life depression and type 2 diabetes mellitus were positively associated with AD, depression at any stage in life was linked to all types of dementia. In fact, late-life depression was markedly associated with both dementia (vascular/all types) and AD [
137]. It is still obscure whether depression is a risk factor for developing dementia or just a prodrome of dementia manifested by progressive cognitive decline [
138]. The class II evidence of the association of type 2 diabetes mellitus with all types of dementia might reflect type 2 diabetes mellitus-driven susceptibility to different types of dementia, with a modest increase in the risk for AD [
49].
Low levels of social interaction markedly affected the occurrence of dementia. Thus, social networking, along with educational and leisure activities, are modifiable protective factors, which might aid in the maintenance of cognitive function with increasing age (for systematic reviews of modifiable factors in dementia, see [
139,
140]). The above could reflect the notion of
brain reserve, which describes an individual’s ability to not develop the disease phenotype despite brain pathological changes that are either age- or disease-specific [
49,
141].
Lastly, while serum 25-hydroxy-vitamin D has been investigated in umbrella reviews of neurological disorders, the same does not hold true for 1,25-hydroxy-vitamin D, as the latter has only been assessed in cancer [
52].
Overall, despite this extensive body of evidence, we wish to emphasize that the majority of epidemiologically identified risk and protective factors do not lie at the bottom of the
health impact pyramid, in which the main social and economic determinants of health, such as education, race, housing, and income, are included [
142] (for an umbrella review on how these determinants affect health, see [
143]). Interestingly, modification of these factors is expected to have the most pronounced impact at the population level, even though they have received significantly less research attention than socioeconomic determinants—an issue of health equity we have attempted to address elsewhere [
124]. Thus, core public health actions should be undertaken not only top-down but also bottom-up, i.e., tackling not only the disease-specific but also the fundamental determinants of health [
144‐
146].
In addition, there are potentially less appreciated or less easily quantifiable risk and protective factors, such as (a) the family environment (now-studied through
Family-Wide Association Studies [
147]); (b) the accumulation of physical and emotional stress along the human lifespan [
148]; (c) living in urban versus rural environments, and in slum versus non-slum urban environments [
149‐
151]; and (d) specific nutritional habits, such as milk and milk product consumption [
152]. These factors may be worth exploring in the future, regarding their association with specific and integrated neurological conditions, thus combining
epidemiological and environmental neuroscience [
153]).
Implications for target groups
Major implications for several target groups, namely patients and their caregivers, healthy subjects, clinicians, researchers, environmental health specialists, policy makers, and educational institutions, could be anticipated from this meta-umbrella review. In a way similar to umbrella reviews in other fields [
34,
154], this
meta-umbrella study provides the opportunity to (a) stimulate more comprehensive, patient-centered approaches, allowing truly informed decisions during genetic counseling or/and coaching for lifestyle changes [
154‐
156]; (b) enhance the accuracy of predicted onset and natural history of neurological conditions at high-risk populations, especially if coupled with polygenic risk scores [
157], and, in doing so, our study can help advocate disease prognostication based on the identified risk and protective factors; (c) offer guidance on future prevention interventions to mitigate amenable risk factors and promote protective factors in the general population, especially in young and middle-aged individuals, in whom the so-called
window of opportunity still exists [
158]; in that context, our approach could assist in promoting campaigns on brain health aimed towards the general public and could increase the level of awareness of neurological conditions, following the successful examples of campaigns regarding cancer and cardiovascular conditions; (d) assist policy makers at the local, national, regional, international, and global level to draft new guidelines or update existing ones, and to explore how modifying risk and protective factors should be incorporated into national health plans; (e) stimulate additional mechanistic, translational, and clinical research on the etiology of neurological conditions and the many unanswered questions; and equally importantly, (f) assess the associations between several risk and protective factors and specific neurological conditions in terms of their natural history and magnitude, which represents a gap in the literature; (g) generate a broader discussion on the role of umbrella and
meta-umbrella review approaches as the highest level of evidence in the
meta-research field; (h) serve as teaching material for courses on preventive neurology offered by the relevant medical education institutions; (i) contribute to helping physicians understand the contribution of
environmental elements as risk factors for neurological disorders, to assist environmental health specialists in the appreciation of the ties between the nervous system and environmental health; and (k) to address these factors (i.e., mitigate the risk factors and enhance the protective factors) by taking action starting from early and middle adulthood, thus ultimately reducing to some extent the incidence and, hence, the prevalence of neurological disorders.
Although others have argued that deciphering how the mechanistic effects of certain risk and protective factors are different between distinct neurological disorders (e.g., AD and ALS) [
159], we feel that maintaining a public health lens approach is always crucial. In this case, the commonality of some risk and protective factors could present an opportunity for
holistic policy making (e.g., promoting Mediterranean diet to prevent a wide spectrum of neurological disorders), and it could also serve as an impetus in developing
transdiagnostic approaches in neurology, similarly to psychiatry [
34] (for further discussion on the
transdiagnostic theory, see [
160,
161]).
In addition, our approach would advocate the development of appropriate statistical tools to account for the fractions of affected neurological populations vs. risk and protective factors, in alignment with previous approaches (e.g., [
162]). In the latter context, this study may lead to developing criteria and tools that are essential in the investigation of the quality of umbrella reviews, allowing inter-comparisons between such analyses. In the same direction, there is a need for consistent a priori
publication of protocol for umbrella reviews, in alignment with previous calls [
163]. Further adherence to common, standardized methodologies could be improved in accordance with previous suggestions (as commented in [
54]).
Many of the class I evidence results in our study re-affirm previous opinions on implementation science (for a discussion on geopolitical factors affecting implementation of policies on chronic diseases, see [
10]). As previously supported [
164],
scientists should stop advocating the need for yet another clinical trial on the cognitive benefits of healthy lifestyles and lobby decision-makers to implement societal polices to actively promote propitious lifestyles. This approach will substantially produce benefit not only for the brain but for the society at large. Complicated situations in which a factor has both a beneficial and a risk effect (e.g., hypertension in PD and dementia) provide an opportunity to highlight the broader potential discrepancies between public health and precision medicine [
124]. Interestingly, this gap in the research literature also calls for
implementation science research to guide health policy and to be a major component of the so-called
population health science [
32].
This study has several strengths, as it is public health policy-, clinical science-, and
meta-research-oriented; thus, it represents a call-for-action, similar to similar calls in other diseases [
165]. The first strength includes the use of a methodical and systematic approach in gathering and evaluating all published, appropriate-quality umbrella literature regarding protective and risk factors for the chronic non-communicable neurological disorders. This may be quite useful to the busy clinician who may not have adequate time to perform reviews on his/her own [
166], and who, in turn, is offered an
overarching and up-to-date knowledge on a wide array of contributing epidemiological factors. In this context, our approach attempts to address the challenge of evaluating evidence provided by a number of high-quality meta-analyses and, in turn, umbrella reviews [
167].
Our systematic,
overarching approach allows studying the top
evidence. In general, umbrella designs are of special value when applied to provide an overall picture that can inform guidelines. For instance, rather than examining one risk factor or disease, a meta-umbrella review can consider multiple risk factors for multiple disease processes. For this reason, we ensured that four researchers participated in the search and quality assessment of published studies, thus, enhancing the validity of our analyses. Notably, whereas earlier non-umbrella reviews examined 14 neurological disorders (and the associated risk and protective factors) as part of National Health Guidelines [
63], our study attempted to review all neurological disorders in the sum of relevant umbrella reviews. This approach becomes more important if we consider that umbrella reviews themselves are already regarded as the highest level of evidence in the hierarchical pyramid.
In the same context, by performing an umbrella review, a neurology-wide notion is created based on the extent to which systematic reviews produce similar results and conclusions and, in turn, unravel the consistency or contradiction of evidence in this field (as commented in [
2]). In this sense, by co-examining several factors in parallel, umbrella reviews, and, as a corollary, our
meta-umbrella approach, can detect irregular patterns in the associations observed (e.g., as the case for farming and pesticides in [
167]). To enhance the encompassing character of our study, we included biomarkers acting as surrogate measurements of epidemiological or/and clinical features. Although these surrogates might lower the strength of the evidence, they can still be indicative of the above features and, thus, offer potentially valuable observations.
Our meta-umbrella approach should be viewed as a
third-generation systematic review study design, allowing, among others, to discuss research gaps and to determine the potential mechanistic underpinnings of such associations. In other words, because of the advent in the wealth of umbrella reviews conducted, this
meta-umbrella review should be seen as a
logical next step, where available umbrella reviews will serve as the
analytical unit of the review, in which the
meta-umbrella approach will allow selecting and including the
highest level of evidence,
notably other umbrella reviews (i.e., being the
meta equivalent of umbrella reviews, following their description in [
2]).
This study obviates the need to perform further research on many different factors demonstrating no significant association, as shown in previous studies [
125]. In that context, it is important to cautiously interpret
p values and CIs. Indeed, when a
p value is within the non-significant range, this means that either the groups have no difference or, alternatively, that the participants’ number is too low to show if such a difference exists. Similarly, when a confidence interval lies in both sides of the line of no effect, this means that either the groups have no difference, or, alternatively, that the number of participants (sample size) is too small to show if such a difference exists. Conversely, a confidence interval not crossing the line of no effect can provide reassurance on the strength or weakness of the evidence, and importantly, whether the study’s results are definitive, with no further need to perform additional studies on the particular issue. Therefore, these notions should be considered while interpreting the results of this meta-umbrella study.
Some limitations should be acknowledged when considering this meta-umbrella study. First, the population and interventions discussed could be perceived as too broad. Our study, however, was in alignment with its overarching goal. Also, the wide presentation of risk and protective factors could be perceived as attenuated if not followed by investigations on their biological plausibility, thus calling for studies to bridge the gap between epidemiology and pathobiology of neurological diseases.
Second, our analyses of the existing umbrella reviews heavily relied both on the quality of information and on the meta-analytical methods employed by the included umbrella reviews, an issue that may have some major caveats. Indeed, all negative aspects from the original studies, which may harbor distinct study designs, classification criteria, and sample sizes that should preferably not be combined in meta-analyses, can be
transferred to the meta-analyses, umbrella reviews, and herein, our
meta-umbrella review (discussed further in [
78]). For example, it has been suggested that, in the case of only a few studies (i.e., with less than one thousand patients) examining a specific risk or protective factor, which may otherwise have a very strong effect, this factor may still be classified as class IV evidence [
37]. This is because, by the very nature of the study design, the quality of evidence offered by (and the conclusions derived from) this
meta-umbrella approach cannot surpass that of the umbrella reviews [
125]. Likewise, the biases and limitations of the meta-analyses, including those relevant to confounding factors, are
transferred to the umbrella and meta-umbrella reviews [
37,
54]. Collectively, our
meta-umbrella review depends on the choices of the umbrella systematic reviewers, putting confidence on their choices [
154].
A third caveat is that a risk or protective factor, or even systematic reviews not previously analyzed in umbrella reviews, may be neglected in a
meta-umbrella study. For instance, the role of physical and emotional stress is not discussed in (or even captured) umbrella reviews, while the same holds true for pregnancy-related maternal health, epilepsy, myasthenia, Tourette syndrome, and Huntington disease, to name a few examples (in contrast to other kind of reviews examining these factors [
112]). Likewise, certain factors, such as air pollution and other environmental risks, appear under-investigated by the umbrella reviews examined, despite their major contribution to the pathophysiology of chronic diseases [
168]. Similarly,
intermediate risk factors (as defined in [
169]), such as birth weight, or other increasingly recognized factors, such as social networks, the impact of which on health outcomes is being steadily better understood [
170], could have been neglected by umbrella reviews. Another major example is that studies on early life risk factors, including in utero exposure, seem absent in umbrella reviews, even though the early life-related factors (and even those evoked in a transgenerational manner by progenies, e.g., through epigenetic regulations [
171‐
173], or those described in the capacity-load model [
174]) play crucial roles in the development of adult diseases [
56,
57,
175]. Interpreting these studies with caution is essential as, at least in theory, several early life factors may influence the intercept of adult disease manifestations, whereas other factors may exercise an influence later in the disease course. The above limitation could be studied in detail in the future.
That said, with regard to the majority of risk and protective factors, which cannot be examined using standard epidemiological approaches or are not frequently assessed, the level of evidence pertaining to these factors will be most likely low given the scarcity of the existing relevant data (as further discussed in [
34]). Similarly, in light of the massive amount of primary and secondary research studies, the possibility of a medical field being totally unexplored at the systematic review/meta-analysis level is not high [
37].
Umbrella reviews and, as corollary, our
meta-umbrella approach, are inherently biased in favor of commonly assessed factors, towards more common neurological disorders (perhaps with the exception of neuromyelitis optica), or towards those of adult but not pediatric populations, because, so far, all umbrella reviews have been conducted on these disorders. This is because of the lack of umbrella reviews examining factors that are less common in rare or adult neurological disorders, besides the conditions included. Assessing the input of genetic and environmental factors in rare diseases by comparing evidence from common diseases could be hampered by the heterogeneity of studies (further discussed in [
78]). Perhaps, key events may lie in the interactions of genetic and non-purely genetic factors, such as epigenetic modifications (e.g., DNA methylation, chromatin modifications, etc.), the gut microbiome, and others, summing up to the so-called
exposome—a field that is still in its infancy [
176]. Likewise, our study may be biased towards more easily quantifiable factors, neglecting other crucial factors, such as stress—a factor established nowadays for its multiple negative health effects, including neurological disorders [
177‐
181].
Despite our efforts to be as thorough as possible in employing a sensitive literature search strategy, no umbrella reviews except in the English language were found in our systematic literature search. However, we cannot exclude the likelihood that some relevant papers in other languages might have been overlooked, even though Moher et al. have shown that omitting studies in non-English languages may not introduce considerable bias [
182]. Similarly, while we assessed the main results of the reviews, we may have overlooked the unreported protective and risk factors that could have been significant. Likewise, some existing evidence might have been omitted in our meta-umbrella review because some studies might have been omitted or were not included in the prior meta-analyses or, because of the blossoming field of umbrella reviews, some studies might have been published after September 21, 2018. To this end, and following the example of previous studies [
89], we also performed an additional search for all relevant publications (umbrella reviews) that have appeared in PubMed until January 1, 2020, so that the reader remains updated (Additional file
3) [
46,
83,
84,
183‐
202]. Moreover, despite our search strategy to include gray literature and policy documents, following previous urging [
2], no relevant documents were identified.
It is possible that conclusions drawn in some studies may have been confounded by biases other than sample size or by residual confounding effects per se [
203]. For example, reverse causality might operate in the associations assessed, and it might have affected the findings regarding the nature of linkage. This consideration highlights the need for prospective studies to demonstrate the direction of causality, and the nature of linkage. This issue becomes important when a broad range of disorders are considered (e.g., association of smoking with stroke vs. that with lung cancer metastases to the brain). Similar considerations could also include other sources of bias, either unknown or known, e.g., the so-called survivor bias (as in ALS, or stroke). This latter bias refers to factors that are highly prevalent following the disease onset; these factors can turn out to be protective. The phenomenon is worse in studies assessing the prevalence of a disease, because factors contributing to both the disease’s onset and its progression may be underestimated (as reviewed in [
159]). Moreover, several additional types of bias remain unaddressed. These could include sex bias, how controls are selected, how the measures of exposure are defined, and the composition of the population examined in terms of demographic and geographical variation (for further discussion on these issues, see [
125]). More broadly, causal inference vs. investigation of mere
correlates nowadays requires sophisticated methods (e.g., multivariable Mendelian randomization or Bayesian approaches), the description and usage of which exceed the aims of the present study (for a discussion, see [
113,
204]).
Another source of bias could refer to the heterogeneity of the data in combining risk and protective factors for multiple different disease processes. Multiple risk factors, which may be ill-defined or based on studies that are observational in design, may be associated with very distinct disease processes, with differing pathophysiologies. For example, PD is particularly prone to recall bias as it tends to have a longer prodromal phase with symptoms that might not be diagnosed as PD until manifestation of the later characteristic motor symptoms. Likewise, the comparator groups for different neurological disorders included in this meta-umbrella review are expected to be different. Importantly, the so-called
vibration of effects in observational studies, which is linked to how the selection of adjusting variables leads to results’ variability, should be considered while assessing our results [
205].
The overarching character of presenting more-than-one protective and risk factors for the sum of neurological conditions should be regarded through the public health and clinical counseling lenses rather than from a mere epidemiological perspective. One of our chief goals was to reduce reporting bias—this is, indeed, one of the main reasons for which systematic reviews (in our case a systematic review of umbrella reviews) are performed. In doing so, we wish to highlight (a) the challenges preventive medicine specialists can face (and the need for balanced, informed decisions outweighing benefits vs. harms in both societal and personalized approaches) during counseling services, especially with regard to how well they can communicate their message, e.g., when the same factor is protective for one neurological disease and risk factor for the other, and (b) the potential inherent difficulties of forming public health guidelines.
In the same context, other limitations could include (a) the potentially different, or, in some cases, potentially inaccurate definitions regarding healthy control groups among meta-analyses or/and individual studies leading to inaccuracies, thus introducing another source of bias (commented in ([
34]); and (b) the fact that, in some cases, the protective and risk factors examined may be due to the interaction of other, more
primordial factors (e.g., educational level as result of parental education and family wealth status), therefore representing risk markers, or that there is an apparent heterogeneity in the definition of these factors [
54]. That said, and in alignment with previous studies [
34], we pursued a
pragmatic approach by showing confidence to the choice of the primary systematic reviewers—and, this should be recognized [
37].
Of note, the possibility of non-uniform diagnostic criteria for the disorders used for study selection in the primary meta-analyses gathered by the umbrella reviews is also a limitation that should be considered when comparing different factors for a particular disorder, especially in the context of umbrella reviews assessing a single factor vs. several disorders; this is particularly important for disorders such as AD and cognitive impairment. For instance, some umbrella reviews might restrict pooling to studies that have used only particular criteria, even though the original meta-analysis might have included more studies with more permissive criteria (such as “assessment by attending doctor”). Ideally, extracting detailed information from the primary meta-analyses (or the summary table with meta-analysis characteristics available in some umbrella reviews) would be needed, as it is difficult to assess these particularities at the level of the meta-umbrella review; however, such a task could not be easily performed in our study because (a) our initially designed pragmatic approach followed the definition of the umbrella reviews, (b) there were no informaticians at our library facilities to assist in this task, (c) umbrella reviews, in most cases, do not include the inclusion criteria of the original studies, and (d) if we had decided to set out for the above task, then we would have had to list the inclusion criteria of all the primary studies, which were presumably different.
Moreover, the issue of overlapping systematic reviews cannot be excluded, although we have undertaken every effort to retain systematic reviews and meta-analyses that included the highest number of primary studies. Therefore, as discussed elsewhere [
167], similar direction (i.e., positive or negative association) and order of magnitude may be due to similar biases existing in these overlapping studies. It would have been, nevertheless, preferable if we had created a data matrix previously described as
Corrected Covered Area, to assess overlapping (for further description, see [
42]).
Of note, our approach presents additional difficulties in comparing the effect sizes spanning the sum of factors investigated, due to the issue of directly comparing the effect sizes between different meta-analyses (e.g., HR, RR, mean difference, and standardized mean difference), let alone when these effect sizes stemmed from different study designs (e.g., HR presents difficulties if used and interpreted in cross-sectional studies) or when distinct effect size metrics are not converted to a single metric of reference. Thus, our study’s results should be interpreted more through a qualitative rather than a quantitative lens [
37]. Additionally, pooled effects cannot be estimated—the same is pertinent to umbrella reviews of meta-analyses. In this context, data could have been presented in forest plots without depicting pooled effects [
37]. Furthermore, another effect size, the attributable risk (AR), has the potential of assessing public health impact, because it allows capturing how much an outcome could have been prevented from occurring if there was no exposure to a certain factor [
206]. Nevertheless, AR has not been frequently used in systematic reviews [
89]. Thus, future studies should be encouraged to focus on calculating the AR for neurological disorders based on the exposure to all risk and protective factors, to ultimately guide health and public policy interventions. In addition, previous studies have suggested replacing the term
risk factor with
predictor and
explanatory factor for risk stratification and causal studies, respectively [
70]. Similarly, others have proposed the notion of
risk markers to reflect that many risk factors, such as immigration and ethnicity, may stem from the interaction of other risk factors [
34]. Collectively, future umbrella and
meta-umbrella studies should agree on the effect size to be used for comparisons and on the preferred terminology regarding risk and protective factors (for the lack of standardization in medical terminology, see [
207]). In the same context, an additional consideration could be that, as shown in Table
4, most umbrella reviews discussed herein applied their own definition regarding the credibility of epidemiological evidence, consequentially hindering inter-comparisons and, thus, calling for a uniform set of criteria in future umbrella reviews.
Table 4
Level of evidence in various studies
| High: concordance between meta-analyses of RCTs and meta-analyses of observational studies; low: meta-analyses of RCTs with results contrary to those from meta-analyses of observational studies | High: meta-analyses of prospective studies with no heterogeneity, no potential confounding factors identified, and agreement of results over time and among meta-analyses, including studies with different designs; medium: meta-analyses of prospective studies with no heterogeneity and no potential confounding factors identified; low: meta-analyses of prospective and case-control studies with no heterogeneity and no potential confounding factors identified | High: meta-analyses of prospective studies lacking information on heterogeneity and potential confounding factors; medium: meta-analyses of prospective and case-control studies lacking information on heterogeneity and potential confounding factors; low: meta-analyses of case-control studies or meta-analyses of any other study design with significant heterogeneity (I2 > 50%) and potential confounding factors | Limited studies included in meta-analyses (n ≤ 3) or evident contrasting results from meta-analyses with the same level of evidence |
| Statistical significance with p < 10− 6, more than 1000 cases (or > 20,000 participants for continuous outcomes), the largest component study reported statistically significant effect (p < 0.05); 95% PI excluded the null; no large heterogeneity (I2 < 50%), no evidence of small-study effects (p > 0.10) and excess significance bias (p > 0.10) | Statistical significance with p < 10− 6, more than 1000 cases (or > 20,000 participants for continuous outcomes), the largest component study reported statistically significant effect (p < 0.05) | Statistical significance with p < 10− 3, more than 1000 cases (or > 20,000 participants for continuous outcomes) | The remaining statistically significant associations with p < 0.05. |
| Significance threshold reached at p ≤ 0.001 for both random and fixed effects calculation; > 1000 cases (or > 5000 total participants if the metric was continuous); not large heterogeneity between studies (I2 < 50%); 95% PI excluding the null value; no evidence of small-study effects (if it could be tested) | Significance threshold reached at p ≤ 0.001 for both random and fixed effects calculation; > 1000 cases (or > 5000 total participants if the metric was continuous); not considerable heterogeneity between studies (I2 = 50–75%) | Significance threshold reached at p ≤ 0.001 for random effect calculation; 500–1000 cases (or 2500–5000 total participants if the metric was continuous) | Significance threshold reached at p ≤ 0.05 for random effects calculation |
| Evidence existed from both observational studies and RCTs, and association/effect was of the same direction, statistically significant at p ≤ 0.001, and free from bias | Evidence existed from both observational studies and RCTs, and association/effect was of the same direction and statistically significant at p ≤ 0.001, but excess significance could not be tested; or evidence existed from RCTs and effect was statistically significant at p ≤ 0.001 and with no contrary results from observational data (that is, systematic reviews, if any exist, are also definitive or suggestive and meta-analyses of observational studies, if any exist, are in the same direction) | Suggestive: Evidence from RCTs with an effect at 0.001 ≤ p ≤ 0.05 and with no contrary results from observational data (same as above); or evidence from meta-analyses of observational studies showing an association at p ≤ 0.001, with no contrary results from randomized data (that is, meta-analysis of RCTs, if present, have effects in the same direction) and, if it could be tested, no evidence of small-study effects (p ≥ 0.10), not very large heterogeneity (I2 ≤ 75%), no evidence for excess significance, based on cumulative evidence of more than 500 disease events (or more than 5000 total participants if type of metric was continuous) | [Substantial effect unlikely]: Evidence from observational studies or RCTs enough to conclude that a substantial effect is unlikely based on the magnitude and the significance level |
| More than 1000 cases, significant summary associations (p < 0.001) per random effects calculations, no evidence of small-study effects, no evidence for excess significance bias, PI not including the null, and not large heterogeneity (I2 ≤ 50%)effects and excess significance | [No such category exists in these studies] | Nominally significant summary associations (p < 0.05) per random effects calculations, no evidence of small-study effects, no evidence for excess significance bias, and not large heterogeneity (I2 < 50%) | All other risk factors with nominally significant summary associations (p < 0.05); |
| The associations that fulfilled all the following criteria: statistical significance according to random effects model at p < 10− 6; based on more than 1000 cases; without large between-study heterogeneity (I2 < 50%); 95% PI excluding the null value; and no evidence of small-study effects and excess significance | Associations with > 1000 cases, p < 10− 6, and largest study presenting a statistically significant effect (with 95% CI excluding the null value) | The associations supported by > 1000 cases and a significant effect at p < 10− 3) | All other risk factors with nominally significant summary associations (p < 0.05) |
| The classification was based on AMSTAR (A Measurement tool to Assess Systematic Reviews), as following: Q1: A-priori design; Q2: Duplicate study selection and data extraction; Q3: Search comprehensiveness; Q4: Inclusion of gray literature; Q5: Included and excluded studies provided; Q6: Characteristics of the included studies provided; Q7: Scientific quality of the primary studies assessed and documented; Q8: Scientific quality of included studies used appropriately in formulating conclusions; Q9: Appropriateness of methods used to combine studies’ findings; Q10: Likelihood of publication bias was assessed; Q11: Conflict of interest-potential sources of support were clearly acknowledged in both the systematic review and the included studies. |
| Not available levels of evidence (*) |
Another limitation is that our
meta-umbrella review was not registered for its goal or protocol at a database like PROSPERO, although (a) our search criteria were predefined and typical of systematic reviews, (b) our exclusion criteria were not intending to alter the number or nature of umbrella reviews identified and (c) no
re-synthesis or quantitative synthesis with regard to outcomes of interest or of any other data in any sort of meta-analysis (as discussed in [
2]) took place. Besides, there has been no established protocol to assess the quality of umbrella reviews; thus, AMSTAR application should be treated with caution.
Other limitations could be that (a) we merged the systematic reviews referring to observational studies with those of clinical trials, given that observational studies should be treated with high cautiousness when referring to risk and protective factors (commented in [
208]); (b) we did not distinguish between the two major categories of observational studies, i.e., case-control vs. cohort studies; and (c) we did not perform a sensitivity analysis. Recently, interesting methodologies have been proposed to reconcile the distinct study designs of observational studies and randomized trials [
209]. Nonetheless, we feel that the above
merging of distinct study design types, which holds an undeniable bias in principle, has not impacted our results, because of (a) our
pragmatic approach, (b) the non-quantitative approach applied herein, and (c) the lack of evident
internal discrepancy among the findings reviewed. Besides, both study types offer
real-word results (commented in [
167]). Another similar concern could be why retrospective studies were included in this
meta-umbrella review, especially for those influential factors that cannot be randomized (smoking). Nevertheless, selecting for retrospective studies to include can affect, in principle, the meta-analyses (e.g., there are meta-analyses of observational or cohort studies) and not a meta-umbrella review. In this sense, performing a sensitivity analysis may not be feasible for a
meta-umbrella review. In addition, the fact that umbrella reviews only use statistical testing to show the existence of bias and cannot provide evidence of their nature and extent should be taken into account, alongside with the fact that umbrella reviews cannot supply any comparative ranking, as is done in a network meta-analysis.
Lastly, our search strategy for umbrella reviews did not allow identification of studies titled
systematic reviews of systematic reviews,
meta-reviews,
systematic meta-reviews (examples of this
study types include [
210‐
212]),
systematic reviews of meta-analyses, meta-epidemiological studies, overviews, field-wide meta-analyses, series of systematic reviews and meta-analyses, synthesis of systematic reviews, meta-meta-analyses, research-on-research studies, comprehensive reviews, reviews of meta-analyses, or
combinations of the above (examples in [
42,
213‐
216]). However, omitting such studies does not negate our primary goal to consider only the data reported in reviews clearly defined as umbrella reviews. Besides, many of these types of studies, such as
overviews, do not clearly describe systematic reviews, let alone in a homogeneous manner [
97].
Moreover, an a posteriori search strategy using similar search strings revealed only one study with potential applicability (i.e., [
217]). Under all circumstances, a call for uniform terminology in the field of
meta-research could be made, in alignment with previously expressed opinions [
207], in particular those arguing that the definitions of systematic reviews is surrounded by ambiguity and lack of clarity [
218].