Question 1

Looking at the same figure (remember that Studies 1, 2 and 5 are cohort studies, 3 and 4 are case‐control studies, and they are arranged in order of year of publication,
Earliest first), Which of the following statements is/are reasonable? The order in
Which the studies are presented:

A)Is fine as it is
B)Is the best possible because it takes the time dimension into account
C)Allows us to see clearly the substantial heterogeneity of the results
D)Would be improved if the studies were grouped by type

Is fine as it is
Allows us to see clearly the substantial heterogeneity of the results
Would be improved if the studies were grouped by type

Accepted Answer

Given the small number of studies a case could be made for each of
these.When we look at where the RRs are sited there is overall modest heterogeneity
only (which in practice we would confirm by a statistical test), largely driven by the very
unstable results (underpowered and uncertain, as shown by the very wide CIs) of
studies 3 and 4.And we can also see that all the point estimates (RRs) are positive,
which in itself is helpful.But if there were many more studies and some showed an
inverse relation (RR <1.0), it could become harder to see if a pattern existed.It looks like
from these data as is the RR from case‐control studies are systematically higher than
those from cohort studies - and this is not uncommon - so grouping separately by
study type (D) is often a good first level proxy for ranking results by their quality.We
then see that there is very little heterogeneity within the two groups, with prospective
studies (which are probably of higher quality) showing more modest effects.(Where
there are sufficient prospective data, it is not uncommon for reviewers to exclude all
case‐control studies from consideration.)
Simply ordering studies by year of publication is generally not helpful if this mixes
research using different designs of varying quality.However there are always
exceptions, and we give an example in the text where this did work very well because
this order correlated closely with increasing precision of measurement of the exposure
(HPV infection).

Question 2

Having decided on your research question, you establish your criteria for which studies to include.Which (if any) of the criteria listed below should not be a basis for
Including a study in your review? (Select all that apply)

A)It must be an analytical epidemiological study in humans
B)It must include individual data on body size and on the presence or absence of kidney cancer
C)It must either include an estimate of the effect of body size on kidney cancer risk (e.g. OR, RR), or include data allowing such an effect to be estimated
D)It must report results for at least 50 cases of renal cancer
E)It must be a prospective study (cohort or RCT)
F)Noneoftheabove

Noneoftheabove

Accepted Answer

All of these are valid criteria.The one exception might be excluding small
studies (D); this would only be appropriate if there were many large studies in your
literature search in which case you might exclude smaller studies that would only have
limited influence on the overall summary result (they might also be of lower quality).A,
B and C are important criteria for this review and, if sufficient data are available, many
reviewers will now restrict their review to prospective studies (E).An intermediate step
could be to include population‐based case‐control studies and exclude those with
hospital controls and cross‐sectional studies.This is because of the potential for
'reverse causality' in a cross‐sectional study, and for major selection bias in a hospital‐
based case‐control design, given the relation of body size to so many diseases.Some
reviewers would include them, but should then consider their results separately from
the better designs.
You will also find there are many measures of body size.You might wish to include all
(with separate analyses for each), only those relating to obesity (if that is the basis for
the biological mechanism you believe to be operating), only those most commonly used
(e.g.body mass index or BMI), or perhaps only newer more precise measures (e.g.
waist‐hip ratio, or some direct measure of fat mass).Similarly, "kidney cancer" is not
one disease: some forms occur in children only (Wilm's tumours), and among adults
only about 85% fall into the common histological category known as renal cell
carcinomas (RCC), so you might exclude all tumours except RCCs of adults.

Question 3

After applying your inclusion/exclusion criteria you have results for over 100 studies. Which of the following strategies (select all that apply) will help you organise the
Information about the studies and present the results most clearly?

A)Include a simple table giving brief descriptions of each study in one table row (probably as an online appendix given the very large number of studies)
B)Include key elements for each study including the study design, sample size, exposure measure, effect size and confounder control
C)Order the studies by study type (cohort or RCT) and sample size, largest first
D)Provide a forest plot in which you use the highest RR you can find in each paper to maximise the visual impact
E)Conduct a meta‐analysis to give a simple summary effect estimate no matter what the pattern of the data is

Include a simple table giving brief descriptions of each study in one table row (probably as an online appendix given the very large number of studies)
Include key elements for each study including the study design, sample size, exposure measure, effect size and confounder control
Order the studies by study type (cohort or RCT) and sample size, largest first

Accepted Answer

This is the biggest challenge of all for a review team, and various
standardised approaches have been used (see text references).A useful rule of thumb
is to provide at least (i) a simple table (as an appendix if very large) giving brief
descriptions of each study in one table row, which might include the reference, year,
type, number of participants (and response rate), confounder control, and key results
(A, B).It is also important to think carefully about how to order the studies.A good
starting point is to separate them by study type, which as you have seen can of itself
bring order to an apparently highly varied set of results, and further ranking within
types might then be by sample size (C) or perhaps by publication date (this is especially
useful if patterns may have changed over time).While a figure plotting the pattern of
study results (a forest plot, D) is also very valuable (and this should follow the same
order as the table), this should not simply show the highest RR from a paper but the
most appropriate estimate for the relationship of interest (often that with the most
complete control for confounding).Similarly, while a meta‐analysis can also be very
helpful, in the presence of substantial heterogeneity between studies you should be
very cautious about calculating a summary estimate (E).If you do then no matter what
caveats you apply to it, it will be remembered and used.

Question 4

You are commissioned to write a quantitative systematic review.Which of the following research questions is framed in the most epidemiologically appropriate
Way?

A)What is the relation between body size and genito‐urinary cancers?
B)What is the relation between obesity and genito‐urinary cancers?
C)By how much does obesity (measured as body mass index, BMI) increase a woman's risk of developing genito‐urinary cancer?
D)By how much does obesity (measured as BMI) increase a woman's risk of developing renal (kidney) cancer?
E)By how much does body size increase risk of renal cancer?

Accepted Answer

One of the key elements of a research question is its specificity.Genito‐
urinary cancers include cancers at all different sites in the genitor‐urinary (GU) tract and
of course these differ between men and women as well as having different sets of risk
factors.So while we might ultimately want to know about GU cancer more broadly, it
will require a series of reviews of individual cancers first.Thus A, B and C are not
appropriate, despite the more precise nature of C - even among women there are
major differences in the causes of different GU cancers; e.g.cancer of the uterine body
is related to obesity, but cancer of the uterine cervix is not.Clearly substantial
background knowledge is needed to make these judgements, and you should never
wade into a review without such insights.Both D and E focus on one cancer (although
there are subtypes of renal cancer too), and both include a quantitative element (how
big is the effect?).However D is the more precise of the two, with respect to both
exposure measurement (body size includes height and weight, as well as obesity) and
the sex‐specificity of the outcome.However both men and women have kidneys and
many studies of the association will include both sexes, so why exclude men? And while
obesity is the exposure of public health concern given its modifiability and influence on
many other diseases, there is also intrinsic biological interest in the relations between
height and disease.So perhaps if E included further specification of the subset of body
size measures to be included, and an intention to conduct sex‐specific as well as overall
analyses, that would be optimal.

Question 5

Before starting out to do a systematic review, the most important requirement is:&#10;A)A lot of time&#10;B)A clear research question&#10;C)To have read a narrative review on the same topic, preferably by a Professor&#10;D)To have read the appropriate guidelines on how to conduct a systematic review&#10;E)Ready access to relevant online databases and search engines

Accepted Answer

Unless the question to be addressed is thoughtfully and carefully framed the
large amount of effort required (and A is also true) can be largely wasted.It is also
important to have good access to the relevant search engines and databases (E) and to
be familiar with guidelines for conducting reviews (D).Note, while C is not the ideal, you
will want to read widely on a topic before starting a formal review, and there are
informative personal views, they just cannot be taken as definitive if they lack a formal
structure to ensure they are systematic.

Question 6

Which of the following statements about the conduct and reporting of systematic reviews and meta?analyses are true? (Select all that apply)&#10;A)The review should be considered as a research paper in its own right, and structured accordingly&#10;B)Literature search strategies must be detailed enough to permit their repetition by the reader&#10;C)All funding sources must be specified&#10;D)Weaknesses of the primary data sources (included studies) and of the review itself should be discussed&#10;E)A formal causal argument should be presented in the Discussion

Accepted Answer

All of these elements are important.The MOOSE (Stroup et
al., JAMA, 2000 http://www.ncbi.nlm.nih.gov/pubmed/10789670) and PRISMA
(Liberati et al., BMJ, 2009 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2714672/)
guidelines provide valuable information about how to structure and report the results
of a systematic review.

Question 7

The following figure displays results from a systematic review and meta‐analysis of the effects of sun exposure on mental health.The RRs represent the association between
Higher sun exposure and good mental health.Studies 1, 2 and 5 are cohort studies, 3 and 4
Are case‐control studies.They are arranged in order of year of publication, earliest first. Which of the following statements are true? (Select all that apply) The summary
Result of the meta‐analysis:

A)Tells us for certain that sunlight is good for mental health
B)Is meaningless because of the extreme heterogeneity of the RRs
C)Is of practical public health importance
D)Is helpful in assessing whether the relation might be causal
E)Is statistically significant

Accepted Answer

E is clearly correct, as the lower bound of the CI is greater than 1;
combining results across the studies decreases the likelihood of chance as an
explanation for the association.D is also correct as the summary RR gives the best
overall expectation of the true strength of the relation IF we do not think the
heterogeneity is so great that it is inappropriate to combine the data across studies, i.e.
that B is not correct.In this situation all of the estimates are >1.0 so this assumption is
probably not too unreasonable, although it is notable that the retrospective case‐
control studies (3 & 4), which would be more susceptible to selection and recall bias,
have given much higher estimates than the cohort studies.Excluding the case‐control
studies would have given a much lower pooled RR.[NB In assessing heterogeneity we
are also assessing the consistency of the evidence across studies and study types, again
contributing to an assessment of causation.] Before we could conclude that C was true
we would need more information on how important a 35% increase in mental health, as
measured here, could be.To establish A we would need more information (e.g.
regarding the likely validity of the study results) and to do a formal causal analysis,
taking account of a number of factors in addition to the strength of association
(relatively small in this instance - especially if we put more emphasis on the results
from the cohort studies).

Question 8

In a systematic review of aspirin as a possible preventive of lung cancer (Feskanich et al., British Journal of Cancer, 2007
Http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2360462/pdf/6603996a.pdf), the
Authors identified all papers with original data that assessed the relation between
Aspirin use and lung cancer incidence or mortality (N=18) and then excluded:
- Retrospective analyses (N=4)
- Prospective studies that did not account for smoking (N=3)
- Studies with no quantitative measure of aspirin use (N=2).
Which of the following statements are true? (select all that apply)

A)The results of the review will be biased because 9 of the 18 studies were excluded
B)It is appropriate to exclude retrospective studies because of potential biases in aspirin recall and selection of controls
C)It is appropriate to exclude retrospective studies because their results are likely to be confounded
D)It is inappropriate to exclude the retrospective studies because there is only a small number of studies overall
E)It is appropriate to exclude studies with no control for confounding by smoking because this is likely to influence the association between aspirin and lung cancer

Unlock Deck

Unlock for access to all 10 flashcards in this deck.

Unlock Deck

k this deck

Accepted Answer

It is highly likely that the results of retrospective studies could be
biased due to selection bias and/or recall bias (B) and it is also likely that smoking and
aspirin use could be associated making smoking a likely confounder (E).Confounding is
no more a problem in retrospective studies than prospective studies (C) and study
designs that are very likely to give biased results should be excluded regardless of how
many potentially eligible studies were identified (D).After excluding the studies with
potentially biased results, the results of the review should be more reliable (A).

Question 9

You may have noted that the x&#8208;axis of the plot uses a logarithmic scale (the distance from 1-2 is the same as that from 2-4 and from 4-8).This is:  &#10;A)To ensure the correct visual balance between RRs that are positive (>1.0) and those that reflect an inverse association (<1.0)&#10;B)A mistake - linear (arithmetic) scales are easier to work with and interpret&#10;C)Irrelevant, since the range of RRs is quite small&#10;D)To allow a very wide range of effect sizes to be included on the one plot

Accepted Answer

Looking closely at the scale, the distance from the intersection with the y‐
axis (the 'no effect' point, RR= 1.0) to an RR of 2.0 (double the risk) is identical to the
distance from this point in the other direction to an RR of 0.5 (half the risk).This
ensures that a point on the plot representing a study with RR=2.0 is equally as far to the
right of the y‐axis as the point for a study with RR= 0.5 is to the left; i.e.on a logarithmic
(log) scale a doubling of risk has the same visual impact as halving it.With an arithmetic
scale, RRs below 1.0 are compressed between 0 and 1.0 y‐axis giving a false impression
of the balance of effects.The log scale does also allow a wider range of RRs to be
included (D) but this is not relevant here and in fact the choice of scale is not critical.

Question 10

You are of course concerned about the quality of studies included in a systematic review.In general RCTs are the 'gold standard' design, especially for interventions,
But there are many circumstances when they are infeasible, unethical or otherwise
Inappropriate.Identify for which of the following research questions an RCT would
Not be the optimal design.

A)Does beta‐carotene prevent skin cancers in Scandinavians?
B)Does smoking cigarettes cause heart attacks in women?
C)How effective are large‐scale population‐based screening programmes in lowering mortality from breast cancer?
D)Does long‐term aspirin use lower mortality after a heart attack among male doctors?
E)None of the above

Unlock Deck

Unlock for access to all 10 flashcards in this deck.

Unlock Deck

k this deck

Accepted Answer

This question is not directly amenable to an experimental approach, as it is
clearly unethical to randomise anyone to being a smoker.An inverted approach -
randomising women who smoke to a quit smoking programme - could test some
aspects of the hypothesis if the anti‐smoking programme worked, numbers were large
and follow‐up was long.(The latter two factors compromised an experiment to test D in
an RCT of over 20,000 male doctors followed up for 12 years.This had to be abandoned
without the question being answered because the death rates were very low in both
arms of the trial).And while RCTs are crucial in evaluating whether under research
conditions a screening programme can save lives (C), once the programme has been
implemented as a wide‐scale service, its evaluation rests largely on simple descriptive
measures of population mortality (see Chapter 15).For A an experimental design is
critical to getting the right answer.

Deck 11: Assembling the Building Blocks: Reviews and Their Uses