Age and Ageing 2017; 46: 722–728 © The Author 2017.

r 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.
doi: 10.1093/ageing/afx105 This is an Open Access article distributed under the terms of the Creative Commons Attribution
Published electronically 24 June 2017 License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution,
and reproduction in any medium, provided the original work is properly cited.

REVIEW OF RESEARCH METHODS

Systematic reviews: guidance relevant for studies

of older people
SUSAN D. SHENKIN1,2, JENNIFER K. HARRISON1,2,3, TIM WILKINSON4, RICHARD M. DODDS5,6,

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

JOHN P. A. IOANNIDIS7
1
Geriatric Medicine Unit, University of Edinburgh, Edinburgh, UK
2
Centre for Cognitive Ageing and Cognitive Epidemiology, University of Edinburgh, UK
3
Alzheimer Scotland Dementia Research Centre, University of Edinburgh, UK
4
Centre for Clinical Brain Sciences, University of Edinburgh, Edinburgh, UK
5
Academic Geriatric Medicine, University of Southampton, Southampton, UK
6
AGE Research Group, Institute of Neuroscience, Newcastle University, Newcastle, UK
7
Department of Medicine, Health Research and Policy, Biomedical Data Science, and Statistics, and Meta-Research Innovation
Center at Stanford (METRICS), Stanford University, California, USA
Address correspondence to: S. D. Shenkin. Tel: 0131 242 6481. Email: [email protected]

Abstract
Systematic reviews and meta-analyses are increasingly common. This article aims to provide guidance for people conducting
systematic reviews relevant to the healthcare of older people. An awareness of these issues will also help people reading
systematic reviews to determine whether the results will influence their clinical practice. It is essential that systematic reviews
are performed by a team which includes the required technical and clinical expertise. Those performing reviews for the first
time should ensure they have appropriate training and support. They must be planned and performed in a transparent and
methodologically robust way: guidelines are available. The protocol should be written—and if possible published—before
starting the review. Geriatricians will be interested in a table of baseline characteristics, which will help to determine if the
studied samples or populations are similar to their patients. Reviews of studies of older people should consider how they
will manage issues such as different age cut-offs; non-specific presentations; multiple predictors and outcomes; potential
biases and confounders. Systematic reviews and meta-analyses may provide evidence to improve older people’s care, or
determine where new evidence is required. Newer methodologies, such as meta-analyses of individual level data, network
meta-analyses and umbrella reviews, and realist synthesis, may improve the reliability and clinical utility of systematic
reviews.

Keywords: systematic review, meta-analysis, methodology, older people

Introduction at around 8% (593 [7.8%] in 2006; 1846 [7.5%] in 2015),

but this is likely to be a gross underestimate since many sys-
Systematic reviews and meta-analyses are increasingly com- tematic reviews do not get indexed with specific age ranges,
mon, with probably more systematic reviews than new ran- and most reviews include individuals regardless of age.
domised trials now published every year [1]. Medline has Clinical trials in the same period increased less steeply—
indexed at least 227,334 articles as ‘systematic reviews’, from 27,125 in 2006 to 34,270 in 2015. The attraction of
increasing more than 3-fold in the last decade (from 7,579 systematic reviews and meta-analyses is clear: the increasing
in 2006 to 24,517 in 2015). The increase in the number of number and complexity of primary research studies makes
meta-analyses has been even steeper [1]. The proportion of it challenging for researchers—or clinicians—to be aware
these systematic reviews specifically indexed in Medline as of, assimilate and critically evaluate all the available
including participants over age 65 has remained fairly static evidence.

722
 Systematic reviews

A good systematic review will not only identify the pub- training.cochrane.org). This advice is most relevant for ran-
lished and unpublished literature on a specific question, but domised controlled trials (RCTs) of interventions in health-
summarise the findings, critically appraise the included stud- care, e.g. pharmacological treatments such as thrombolysis in
ies, and make recommendations on how this should influ- acute ischaemic stroke [6] or complex interventions such as
ence clinical practice and future research. It may or may not multi-component interventions to prevent delirium [7].
include a meta-analysis (a statistical summary of the evidence). However, much research relevant to older people is not
This is a key step in reducing research waste (http://blogs.bmj. in the form of randomised controlled trials: this may be
com/bmj/2015/10/29/how-systematic-reviews-can-reduce- due to the difficultly of performing RCTs for the questions
waste-in-research). However, many published reviews and of interest to specialists in the care of older people, encom-
meta-analyses are redundant, misleading or serve conflicting passing complex multidisciplinary, multi-component inter-
interests [1], and may compound the limitations of primary ventions (e.g. stroke units) where it is not easy to tease out
studies, rather than critically exposing them [2]. This article the individual components. The population of interest typ-

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

aims to provide guidance for people conducting systematic ically includes frail older people with multiple medical con-
reviews relevant to the healthcare of older people, while ditions and ranges of functional impairment, thus there is
acknowledging that there is no universal definition of ‘older very large heterogeneity, and the restricted eligibility criteria
person’. An awareness of these issues will also help people of RCTs may leave out many types of patients and settings.
reading systematic reviews to determine whether the results Where RCTs are not possible, feasible or ethical, observa-
will influence their clinical practice. tional study designs (cohort or case–control studies) may
provide some evidence for healthcare interventions. For a
discussion, with case examples, of how observational studies
Guidance on conducting a systematic may provide moderate to (rarely) high strength evidence in
review systematic reviews of healthcare interventions see [8].
Observational methods are also a main method for exploring
Protocols associations between harmful risk-factors and outcomes, e.g.
Before performing a systematic review, the reviewer the association between obesity and dementia [9], and there
should first ensure that the question has not already been are specific aspects that should be considered [10].
answered, and check whether a review is already ongoing else- Whether a systematic review is of trials, or observational
where. To check for published reviews, one can search data- studies, the assessment of risk of bias of each included
bases which collate systematic reviews and meta-analyses (e.g. study, and of the review overall is key. Authors should
https://discover.dc.nihr.ac.uk/portal/search/signals), the ensure that they do not ‘sanctify results from poor studies’
Cochrane Library (http://www.cochranelibrary.com/) or bib- [2] and include clear critical appraisal of included data.
liographic databases such as Medline, PsycInfo or CINAHL. For specific guidance on conducting systematic reviews of
Databases e.g. PROSPERO (https://www.crd.york.ac.uk/ observational studies see online step by step guidance
PROSPERO/) are available to register a planned review. (http://www.ccace.ed.ac.uk/research/software-resources/
PROSPERO is an open access resource that does not pro- systematic-reviews-and-meta-analyses) and guidance for
vide critical review of protocols, but provides a searchable early career researchers in epidemiology [11]. The MOOSE
repository to check if this review is already complete, ongoing guidelines [12] provide a standard for systematic reviews of
or planned. Once the new protocol is finalised, it would be observational studies, but have not been updated since
useful to register it if possible. PROSPERO only allows regis- 2000. The STROBE statement (https://www.strobe-
tration of topics related to health outcomes. Other databases statement.org), outlines the issues that should be considered
e.g. http://www.researchregistry.com/ allow registration of in the reporting of epidemiology studies. Software is available
systematic reviews on any topic, at any stage in the review (e.g. Covidence https://www.covidence.org/, EPPI-Reviewer,
process. The protocol can be updated later to make clear any http://community.cochrane.org/tools/review-production-
subsequent changes. tools/eppi-reviewer/about or Rayyan—https://rayyan.qcri.
org/) which keeps all data secure remotely, and allows col-
laboration between authors in different locations: this can
Practical advice facilitate independent assessment, allow accountability and
Detailed guidance on performing systematic reviews is is a method to keep track of the process. It is not a substitute
available elsewhere (https://www.york.ac.uk/crd/guidance/). for adequate training in review methodology and statistical
We have suggested some good practice points (Figure 1). literacy.
The PRISMA-P [3] and PRISMA [4] statements provide
guidance for designing and reporting systematic reviews,
respectively. A checklist is also available of what should be Issues to consider in studies of older people
included. The Cochrane Handbook of systematic reviews of
interventions [5] provides comprehensive guidance on all Geriatricians are well aware of the complexity of dealing
aspects of systematic reviews and meta-analysis, and online with older patients, and the paucity of evidence to guide
modules are freely available from Cochrane Training (http:// their practice. When performing a systematic review

723
S. D. Shenkin et al.

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

Figure 1. Good practice points for performing systematic reviews and meta-analyses in ageing research.

relevant to the healthcare of older people, there are some which include older people. Reviewers should predetermine
specific issues that should be considered. how they will deal with studies, particularly randomised
controlled trials, which have different recruitment strategies
relating to age. Many studies exclude patients over a certain
Age age (e.g. 75 years), in which case geriatricians will be wary
There is no generally agreed criterion for definition of ‘old- about extrapolating the results to their clinical practice. If
er people’. A cut-off of over 60 or 65 years is often used there is no age restriction, can results be examined specific-
(http://www.who.int/healthinfo/survey/ageingdefnolder/en/). ally for older people (e.g. >65; 75; 85 years)? Even if older
However, in the developed world, the syndromes most rele- people are recruited, is there evidence of selection bias, i.e.
vant to geriatricians (e.g. dementia, frailty) are most common are included patients different from those in routine clinical
after age 80. In other areas, such as sub-Saharan Africa, ‘older practice? Research studies generally recruit people who are
people’ may be defined as over age 50 (http://www.who. healthier, better educated and of higher (less deprived)
int/healthinfo/survey/ageingdefnolder/en/index.html). It socioeconomic status. It is important to include a baseline
is important to be clear which studies will be eligible for table of participant characteristics so that readers can deter-
inclusion: those with participants at all ages, or only those mine if the results are directly relevant to their own patients

724
 Systematic reviews

or extrapolations need to be made. Many studies will not interventions. Can the findings be extrapolated to a clinical
present the overall age range, with mean and SD, but population? If data on populations of interest (e.g. age over
instead split the sample into age decades, or other categor- 90; care home residents; people lacking capacity) are lack-
ies. This can make it difficult to compare findings between ing, identify and discuss this gap in the literature.
studies—and authors of systematic reviews should attempt
to obtain original information from study authors. We
would encourage authors of individual studies to include Study design
overall age range, mean and SD as well as reporting in cat- Be clear if the studies included are intervention studies (e.g.
egories, and to make the individual-participant data avail- randomised controlled trials) or observational studies (e.g.
able for reuse as soon as possible. For reviews, as a prospective or retrospective cohort studies, or case–control
minimum, mean or median age (in years) and a description studies), and what methodological considerations are
of the distribution (standard deviation, range or inter- important for each. Note that observational designs identify

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

quartile range) is required. If there is a wide age range, then associations, but cannot prove causation: e.g. the association
it is helpful to report the numbers and proportion over 65 between diabetes and risk of falls may be due to complica-
years. Also, reviewers should consider providing informa- tions of diabetes, or hypoglycaemia, or unmeasured con-
tion on the numbers of the oldest participants, e.g. numbers founders such as hypotension or other unknown or
and proportion of those aged over 80 or 85 years. convoluted reasons [16].

Non-specific presentation of disease Confounding

Systematic reviews can be challenging when there is lack of Particularly for observational studies, it is important to con-
agreement on what to include as predictor or outcome vari- sider if the results could be explained by another variable
ables, or the definition of participants. For example, there is that may be unobserved. Confounding variables may make
variability in what is meant by frailty [13], or ‘care home it appear—incorrectly—that an observed exposure or pre-
resident’ [14]. Many studies will focus on a single predictor dictor is (or is not) causally associated with an outcome. In
or outcome. It is important to consider if this outcome is studies of older people, potentially important confounders,
relevant, or a ‘surrogate’ which may, or may not relate to particularly from earlier life, should be considered, such as
clinical outcomes, e.g. body sway, rather than number of socioeconomic status. For example, socioeconomic status
falls. In addition, although one outcome may be reported, it probably explains the observed association between caffeine
is important to consider other potentially relevant out- intake and IQ [17].
comes: e.g. treatments for incontinence will report the
impact on continence, but may not report whether there
were also impacts on rate of falls or cognition. It is also Effect modification
important to consider how a review will deal with outcomes This occurs in studies where an exposure has a different
which are reported inconsistently across trials (e.g. different effect in different subgroups: effect modifiers are associated
cognitive tests or functional assessments). How will missing with the outcome, not the exposure. For example, older (or
outcome data be handled? Is it missing at random, or will frailer, or cognitively impaired) people may respond differ-
missing data influence the result? Can additional informa- ently to an intervention than younger (or fitter, or cogni-
tion be obtained from the primary investigators on out- tively normal) people.
comes that are not reported? Consider contacting the
corresponding author from the original paper, or any
author whose current contact details can be found. What Bias
impact does death of participants have on the results? It is Bias is an incorrect estimation of the association, e.g.
also important, particularly in clinical trials, to determine between exposure and outcome, due to a systematic error.
whether the outcomes are those that were planned in the The assessment of bias is a critical component of systematic
original protocol [2]. reviews of RCTs and observational studies, and is the main
component of what has often been termed ‘quality assess-
ment’. Careful consideration should be given in advance as
Inclusion/exclusion criteria to how studies will be assessed (see [14] for some guidance
Are the people included similar to routine clinical practice? relevant to studies of older people: including the Cochrane
For example, are people excluded on the basis of age, or Risk of Bias tool [19] for RCTs, ROBANS [18] for obser-
co-morbidities that are common in older people? Be aware vational studies). Summary scores are generally discouraged
of international variation in healthcare practices: the infra- as they can mask complexity and give equal weight to
structure in a residential home, or support by families, var- aspects of quality assessment that could have minimal, or
ies widely internationally. There is also international large, impact on the applicability of results. Note, however,
variation in attitudes to ageing by older people themselves, that there can be considerable inter-individual variation in
and others [15] which may affect the impact of risk of bias assessments (e.g. [19]).

725
S. D. Shenkin et al.

For all studies including older people, reviewers should reviews have been helpful to prompt changes in clinical
consider selection bias—that results will be distorted by practice, when change has not followed publication of indi-
including people in the study that are not representative of vidual trials (e.g. thrombolysis for acute ischaemic stroke).
patients that are looked after in clinical practice (e.g. the Reviews can also be used to support business cases for ser-
hypertension study SPRINT which excluded people aged vice redesign (e.g. comprehensive geriatric assessment) or to
≥75 with co-morbidities such as diabetes, heart failure, identify gaps in evidence where research and clinical change
stroke, dementia [20])—and survival bias—that only those is required (e.g. music therapy in the acute hospital). When
who survive are included in the results, which may give assessing the relevance of the review to the care of older
erroneous conclusions about efficacy of treatments or risk- people, readers should consider the quality of the systematic
factors for outcomes (e.g. better function appeared to pre- review itself as well as the studies that are included in it.
dict care home admission in this study as those with poorer Are included subjects relevant to clinical practice? Is the
function died before discharge [21]). intervention feasible in routine care? Tools such as

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

AMSTAR [23] suggest items which should be considered in
Meta-analysis determining quality of a published systematic review.
A statistical synthesis of the results [22] is not required for
a systematic review (either of RCTs or observational data), The future of systematic reviews and meta-
but this should be performed where possible, while being analysis
aware of how to interpret the results in the presence of het-
erogeneity. It is good practice to contact authors of the ori- Given the limitations of published reviews and meta-
ginal studies to obtain additional data (e.g. if data have been analyses, it is essential that clinicians treat the results with
presented in different age bands in different studies): this appropriate scepticism, and make informed choices for any
can be variably successful. It is important to have sufficient proposed change in practice. Although systematic reviews
statistical expertise in the systematic review team, and to and meta-analyses may be seen as the ‘top’ of the evidence
combine this with clinical knowledge to assess if statistical pyramid, this only applies if the reviews are performed to
synthesis is appropriate and how it should be interpreted. A the highest standard. The quality of reporting of published
summary effect can be tantalisingly succinct, but this should reviews has improved since the publication of the PRISMA
not dissuade authors from critically discussing the limita- statement [4], but the large majority of reviews and meta-
tions of the included studies and, therefore, the summary analyses continue to be flawed, redundant, misleading or
evidence. Different meta-analyses of the same question can clinically not useful [1]. The requirement that authors dis-
come to different conclusions due to inclusion of studies close any conflict of interest is important: results and con-
with diverse protocols and methods. clusions are often biased when authored by company
If presenting data in a forest plot, check that the axes employees, or if a study is industry sponsored [24].
are clearly labelled, so that the size of the effect, and its Crucially, there is an increasing call from funders as well as
potential clinical significance, can be gauged. Include confi- researchers that all clinical trials (and observational studies)
dence intervals of the summary estimate. Is a fixed-effect are reported (e.g. www.AllTrials.net) and that individual
or random-effect model planned, or will a prediction inter- level data is made available as soon as possible. This
val be presented: this provides a predicted range for the requires additional infrastructure and funding, especially for
true treatment effect in an individual study [19]? Reviews studies that require significant storage space such as
often quote formal measures of heterogeneity such as the I2 imaging data (e.g. Alzheimer’s Disease Neuroimaging
statistic, but it is also important to carefully scrutinise the Initiative http://adni.loni.usc.edu/).
table of baseline characteristics to see if there is relevant There are several methodologies which may improve the
heterogeneity between the individual studies included. reliability and clinical utility of systematic reviews [1], and
Whether meta-analysis is possible or not, systematic reviews these are discussed here.
should narratively summarise and critically appraise the
results across all included studies.
Individual patient meta-analysis
Using systematic reviews to inform clinical Where the data from individual studies are combined (while
practice and service redesign retaining the clustering of patients within their individual
study)—has become increasingly common [25]. It is often
It is typically discouraged for systematic reviewers them- limited by the provision of original data, and may result in
selves to make ex officio guideline-level recommendations. analysis of only a subset of otherwise eligible studies, and
Usually, systematic reviews end with the critical appraisal thus can be prone to reporting bias. The meta-analysis of
and synthesis of the evidence. However, these systematic data across multiple observational studies can allow a sum-
reviews are used eventually by multiple users, ranging from mary estimate of risk-factors, e.g. the effect of smoking and
clinical practitioners and readers to guideline committees to smoking cessation on cardiovascular mortality in older peo-
inform clinical practice and for service redesign. Systematic ple [26], but is still at risk of residual confounding. Selection

726
 Systematic reviews

for inclusion of studies based on already known results can single interventions. However, the lack of predefined ques-
be a problem in meta-analyses of individual level data [27]. tions/outcomes risks focussing on positive studies, and it
may be most useful as an exploratory method for complex
Network meta-analysis and multidisciplinary interventions.
Network meta-analysis allows multiple treatments to be com-
pared using direct comparisons within randomised controlled Conclusions
trials and indirect comparisons using a common comparator
Despite their limitations, systematic reviews are useful for
[28]. This allows both direct and indirect evidence to be used
practising geriatricians to inform clinical care, and to iden-
to synthesise the published evidence, such as the effective-
tify where further research is needed. It is essential that any
ness and adverse effects of antidepressants [29]. Treatment
review, either using traditional or more novel methodolo-
networks can quantitatively analyse data for all treatment
gies, is done to a high standard, considering both general

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

comparisons for the same disease. Multiple treatments meta-
issues for good quality reporting, but also paying particular
analysis can rank the effectiveness of many treatments in a
attention to specific issues relevant to the health of older
network. However, for this to add useful information to sim-
people.
pler methods of data synthesis, large amounts of data are
required, and care must be taken in interpretation of the dir-
ect and indirect comparisons, which can sometimes disagree. Key points
• Systematic reviews must be planned and performed in a
Prospective meta-analysis transparent and methodologically robust way: guidelines
Trials can be designed with the explicit predefined purpose are available, and a carefully designed protocol will help.
of future combination in a meta-analysis [2]: this approach • A table of baseline characteristics will help practising geria-
can combine studies with different designs, interventions, tricians in relating the studied populations to their patients.
comparisons, setting and populations. It is ideal for diverse • Geriatricians should consider how reviews deal with: differ-
interventions such as falls prevention programmes [30], but ent age cut-offs; non-specific presentations; multiple predic-
is uncommon as they require significant collaboration, co- tors and outcomes; potential biases and confounders.
ordination and advance planning. • Reviews may determine where new evidence is required
Umbrella reviews [2] are systematic reviews of systematic to improve the care of older people.
reviews: they may summarise all systematic reviews or • Meta-analyses of individual level data, network meta-
meta-analyses performed on a given topic, e.g. all treat- analyses and umbrella reviews offer interesting possibil-
ments for a condition or set of conditions; or all risk- ities for more in-depth or broader views of the evidence.
factors assessed for some disease (e.g. Alzheimer’s disease,
[31]); or all associations that a specific risk factor has been
evaluated for in relationship to a variety of outcomes/dis-
eases (e.g. [32]). These reviews may permit understanding
of the amount and credibility of the evidence in a large Supplementary data
field, knowledge gaps and the main sources of heterogen- Supplementary data are available at Age and Ageing online.
eity and bias.
Rapid reviews [33] are gaining popularity as methods to
make reviews and subsequent treatment decisions more Conflicts of Interest
rapidly than allowed by full systematic reviews. If appropri-
None declared.
ately supervised they may be useful for training, and may
markedly speed up the review process, but as yet there is
no consensus on definition or methodology. Similarly, par- Funding
ticipation in adjudication of studies for systematic reviews
as in Cochrane Crowd (http://crowd.cochrane.org) can J.K.H. is supported by a PhD studentship from Alzheimer
provide some experience and training. Cochrane crowd are Scotland and The University of Edinburgh Centre for
not a substitute for a systematic review by a team with Cognitive Ageing and Cognitive Epidemiology, part of the
appropriate expertise. cross council Lifelong Health and Wellbeing Initiative (MR/
L501530/1). Funding from the Biotechnology and Biological
Sciences Research Council (BBSRC) and Medical Research
Realist synthesis Council (MRC) is gratefully acknowledged. SDS is a member
Realist synthesis has been suggested as a pragmatic of The University of Edinburgh Centre for Cognitive Ageing
approach for evaluating complex interventions [34] e.g. and Cognitive Epidemiology (MR/L501530/1). T.W. is funded
models of care for older people living in care homes [35]. It by an MRC Clinical Research Training Fellowship (MR/
tries to establish, ‘what works for whom in what circum- P001823/1). R.M.D. is funded by the NIHR Integrated
stances and in what respects?’ [36], rather than looking at Academic Training Programme.

727
S. D. Shenkin et al.

References 23. Shea BJ, Grimshaw JM, Wells GA, Boers M, Andersson N,
Hamel C et al. Development of AMSTAR: a measurement
The very long list of references supporting this review has meant that tool to assess the methodological quality of systematic
only the first 36 are listed here. The full list of references is avail- reviews. BMC Med Res Methodol 2007; 7: 10.
able on the journal website http://www.ageing.oxfordjournals.org/ 24. Ebrahim S, Bance S, Athale A, Malachowski C, Ioannidis JP.
as Appendix. Meta-analyses with industry involvement are massively pub-
1. Ioannidis JP. The mass production of redundant, misleading, lished and report no caveats for antidepressants. J Clin
and conflicted systematic reviews and meta-analyses. Milbank Epidemiol 2016; 70: 155–63.
Q 2016; 94: 485–514. 25. Riley RD, Lambert PC, Abo-Zaid G. Meta-analysis of indi-
2. Ioannidis J. Next-generation systematic reviews: prospective vidual participant data: rationale, conduct, and reporting.
meta-analysis, individual-level data, networks and umbrella BMJ 2010; 340: c221.
reviews. Br J Sports Med 2017. http://dx.doi.org/10.1136/ 26. Mons U, Muezzinler A, Gellert C, Schottker B, Abnet CC,
bjsports-2017-097621. Bobak M et al. Impact of smoking and smoking cessation on

Downloaded from https://academic.oup.com/ageing/article-abstract/46/5/722/3887244 by guest on 06 December 2019

3. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, cardiovascular events and mortality among older adults:
Petticrew M et al. Preferred reporting items for systematic meta-analysis of individual participant data from prospective
review and meta-analysis protocols (PRISMA-P) 2015 state- cohort studies of the CHANCES consortium. BMJ 2015;
ment. Syst Rev 2015; 4: 1. 350: h1551.
4. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred report- 27. Flegal KM, Ioannidis JPA. A meta-analysis but not a system-
ing items for systematic reviews and meta-analyses: the atic review: an evaluation of the global BMI mortality collab-
PRISMA statement. Ann Intern Med 2009; 151: 264–9. W64. oration. J Clin Epidemiol 2017.
5. Collaboration TC. Cochrane handbook for systematic reviews 28. Mills EJ, Thorlund K, Ioannidis JP. Demystifying trial net-
of interventions Version 5.1.0. JPT H, S G, editors 2011. works and network meta-analysis. BMJ 2013; 346: f2914.
8. O’Neil M, Berkman N, Hartling L, Chang S, Anderson J, 29. Thorlund K, Druyts E, Wu P, Balijepalli C, Keohane D,
Motu’apuaka M et al. Observational evidence and strength of Mills E. Comparative efficacy and safety of selective sero-
evidence domains: case examples. Syst Rev 2014; 3: 35. tonin reuptake inhibitors and serotonin-norepinephrine
10. Egger M, Davey Smith G, Schneider M. Systematic reviews reuptake inhibitors in older adults: a network meta-analysis.
of observational studies. In: Egger M, Davey Smith G, J Am Geriatr Soc 2015; 63: 1002–9.
Altman D eds. Systematic Reviews in Healthcare: Meta- 30. Province MA, Hadley EC, Hornbrook MC, Lipsitz LA,
Analysis in Context. London: WIley, 2008: pp. 211–27. Miller JP, Mulrow CD et al. The effects of exercise on falls in
11. Denison HJ, Dodds RM, Ntani G, Cooper R, Cooper C, elderly patients. A preplanned meta-analysis of the FICSIT
Sayer AA et al. How to get started with a systematic review in Trials. Frailty and injuries: cooperative studies of intervention
epidemiology: an introductory guide for early career research- techniques. JAMA 1995; 273: 1341–7.
ers. Arch Public Health 2013; 71: 21. 33. Haby MM, Chapman E, Clark R, Barreto J, Reveiz L, Lavis JN.
12. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, What are the best methodologies for rapid reviews of the
Rennie D et al. Meta-analysis of observational studies in epi- research evidence for evidence-informed decision making in
demiology: a proposal for reporting. Meta-analysis of obser- health policy and practice: a rapid review. Health Res Policy Syst
vational studies in epidemiology (MOOSE) group. JAMA 2016; 14: 83.
2000; 283: 2008–12. 34. Rycroft-Malone J, McCormack B, Hutchinson AM, DeCorby
15. Laidlaw K, Power MJ, Schmidt S. The Attitudes to Ageing K, Bucknall TK, Kent B et al. Realist synthesis: illustrating
Questionnaire (AAQ): development and psychometric prop- the method for implementation research. Implement Sci
erties. Int J Geriatr Psychiatry 2007; 22: 367–79. 2012; 7: 33.
18. Kim SY, Park JE, Lee YJ, Seo HJ, Sheen SS, Hahn S et al. 35. Goodman C, Dening T, Gordon AL, Davies SL, Meyer J,
Testing a tool for assessing the risk of bias for nonrando- Martin FC et al. Effective health care for older people living
mized studies showed moderate reliability and promising val- and dying in care homes: a realist review. BMC Health Serv
idity. J Clin Epidemiol 2013; 66: 408–14. Res 2016; 16: 269.
19. Higgins JP, Altman DG, Gotzsche PC, Juni P, Moher D, 36. Pawson R, Greenhalgh T, Harvey G, Walshe K. Realist
Oxman AD et al. The Cochrane Collaboration’s tool for asses- review—a new method of systematic review designed for
sing risk of bias in randomised trials. BMJ 2011; 343: d5928. complex policy interventions. J Health Serv Res Policy 2005;
22. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. 10(Suppl 1): 21–34.
Introduction to Meta-Analysis. Chichester: John Wiley &
Sons, 2009. Received 29 May 2017; editorial decision 8 June 2017

728