Study design made easy

Diogo Bruno
Arteriae, Lisbon, Portugal

Correspondence:
Diogo Bruno, Arteriae Unip. Lda.
R. Cid. Cádiz 14 4.º Dto-A, 1500-157
Lisboa, Portugal
[email protected]
+351 966 625 510

Abstract
Analysis of statistical data is an important
part of any medical writer’s skill set,
especially those professionals working in
publication and regulatory areas. Under-
standing the various study designs is
key to a thorough understanding of
study methodology. Nevertheless, many
medical writers come from a non-clinical
background and have a knowledge gap
when it comes to study design options.
This article describes the main types of
study design. Case report, cross-
sectional, case control, cohort, quasi-
experimental, randomised controlled
trials, and systematic reviews and meta-
analyses studies are explained and their
uses, advantages, and limitations discussed.

The naked truth is that mankind still lacks

time machines. If we had them, there would
be no need for epidemiology as one could,
for example, easily observe a group of
individuals exposed to smoking over the
course of their lives and, then, travel back in
time and reobserve them after persuading
them to stop smoking. Epidemiologists try
to determine whether an exposure (i.e. risk
factor) is associated with an outcome (i.e.
disease), such as smoking and lung cancer in
this example.1
The first step in a study is to define the
hypothesis to be tested. After this, one must
determine which study design is the most
appropriate and/or feasible to test this
hypothesis.

Overview of study designs

Broadly-speaking there are two approaches
to study the association between an

26 | September 2016 Medical Writing | Volume 25 Number 3

 Bruno – Study Design Made Easy

Meta-analysis
Systematic
Review
Randomised
Controlled Trial associated with the outcome. Another use
Quasi-experimental Study for this design is in the case of descriptive
survey studies when the main aim is to
Prospective Cohort Study describe a population in a given time
period.5
Retrospective Cohort Study Cross-sectional design lends itself well to
descriptive statistics, where no association
Case Control Study between exposure and outcome or causal
relationship is sought, and the intention is
Cross Sectional Study
solely to describe the properties of the
Case Report observed data. On the other hand, infer-
ential statistics aims to drive an association
between exposures and outcomes through
hypotheses and estimates. The designs
Figure 1: Main types of study design. described in the next sections are better
Study designs organised in order of statistical validity from the highest validity on approaches to describe these associations as
the top of the pyramid to the lowest validity in the bottom of the pyramid. cross-sectional studies give no indication of
the sequence of events and are prone to
prevalence-incidence bias (e.g. high mortal-
exposure and an outcome: 1. interventional study design for medical writers, the ity conditions will be under-represented as
or experimental and 2. observational or randomised controlled trial (RCT). they will have low prevalence even in the
non-experimental studies.2 When analysing case of high incidence).
their scientific validity, experimental studies Case report
are of higher quality when compared to Case report articles are considered the Case control study
observational studies. They usually involve lowest level of evidence and findings usually Contrary to the cross-sectional studies, the
the study of a factor that can be controlled require formal verification through robust case control design aims to establish an
by the investigator and enrolled individuals epidemiological studies. However, they can association between risk factors and
are randomly assigned to being exposed or represent the emergence of new issues and disease (i.e. uses inferential statistics). A
not to that factor. Observational studies, on key ideas. Namely, they can provide group of patients with the study disease
the other hand, lack randomisation and, as important information for patient care that (cases) is selected and compared to a
such, various other factors might be is not detected in clinical trials or other group of healthy individuals similar to the
unevenly distributed between the studied studies seen as more robust in design. They group of cases in every other aspect
groups; as a consequence of these usually describe in detail an individual (controls). Information about risk factors
confounding factors, a true association is clinical case that shows: 1. a rare variation of is then collected retrospectively and is used
more difficult to ascertain.2 a condition, 2. an unexpected drug adverse to compare both groups and to find
In addition, studies can be characterised event, 3. clues on the pathogenesis of a measures of association.6
as retrospective or prospective based on disease, 4. an unexpected association Consequently, this design is often used
when the subjects are enrolled into the between factors, 5. a unique therapy, or 6. a to study infrequent or rare diseases in which
study.3 These differences will be further unique anatomical variation. prospective studies would be difficult to
explained when we explore cohort studies. perform. To study rare diseases in prosp-
As a consequence of these differences, Cross-sectional study ective designs, a great number of patients
study designs are often organised as a Cross-sectional studies analyse data taken would have to be enrolled, rendering them
pyramid in order of validity (Figure from a sample at a specific point in unfeasible.7 Additionally, case control
1).4 Unfortunately, the most time. They are usually applied studies may have more power than cohort
valid studies are often more for public health purposes as studies as it is easier to have larger samples.
expensive, more time-con- Unfortunately, the they give a snapshot of the rate Finally, instead of measuring the risk of
suming, and more difficult to most valid studies of an outcome of interest (i.e. disease based on exposure, we measure the
manage. are often more prevalence of a condition) in a odds of exposure based on disease. There-
In the next sections, I will expensive, more population. Moreover, res- fore, relative risk is not applicable as a
describe each study design time-consuming, earchers also describe patient measure of association. It is the disease that
further with a special emph- and more difficult characterstics and important is selected at the study onset, so the odds
asis on the most important to manage. risk factors thought to be ratio is used.

www.emwa.org Volume 25 Number 3 | Medical Writing September 2016 | 27

Study Design Made Easy – Bruno

Past Present Future

Cross Sectional

Randomised controlled trial

Cross Control Study (RCT)
RCTs are the only studies truly experimental
Retrospective Cohort Study in nature as the effect of an investigator-
chosen intervention is studied in randomly
assigned subjects from a study sample
Prospective Cohort Study deemed to be representative of a population.
Frequently, a study group is exposed to an
Randomised Control Study intervention (e.g. a novel treatment) and its
effects on one or more outcomes of interest
are studied by comparing the exposed group
Figure 2: Retrospective and prospective study designs. to a control group not exposed to the
Main types of study design and their relation to the studied time-points. intervention (e.g. placebo) or exposed to a
standard intervention (e.g. already establish-
ed therapy or standard of care).10 Figure 3
Cohort study Less often, ambispective or ambi- gives a representation of this study design.
By definition, in cohort studies a group of directional cohort studies are performed The RCT’s inherent characteristics make
subjects comprising a sample deemed to that, as the name implies, combine retro- it a robust study design. Randomisation of
represent the population of interested is spective and prospective information intervention allocation is used to decrease
followed over time whilst collecting data on including past, present, and future confounding effects and allocation bias. The
risk factors and outcomes. It differs from timepoints.8 characteristics that might affect the relation-
cross-sectional studies in the sense that, Regarding the data analysis, unlike case ship between intervention and outcome
although risk factors and outcomes are control studies, the most usual measure is measures will be roughly equal between
studied, subjects are studied over time. the risk ratio of the outcome of interest, study and control groups. Blinding or
Moreover, unlike case control studies, calculated by the risk of the outcome in masking is also frequently used to decrease
individuals are disease-free at the outset of exposed subjects relative to those not study bias. In single-blinded studies, sub-
the study and the risk of development of the exposed to the risk factor. jects are unaware of their group assignment,
disease (outcome) is the measure of decreasing the performance bias that could
interest. Quasi-experimental study occur as this knowledge can affect the
As discussed previously, cohort studies Quasi-experimental studies are sometimes subjects’ response to the intervention. In
can either be retrospective or prospective also called nonrandomised or pre-post double-blinded studies, group allocation is
when relating to time of subject enrolment intervention studies and are used to not known to both study subjects and
(Figure 2). Retrospective studies are also evaluate the effects of specific interventions investigators. This further decreases bias by
called historical cohort studies and study or policy changes. It is a design chosen when avoiding differences in treatment administ-
events from the past up until the present it is not logistically feasible or ethical to ration between treatment arms (perform-
time. The obvious advantage is that the conduct a RCT. By definition, these studies ance bias) or the over- or under-estimation
information is readily available, however lack randomisation and are conducted after of the effects of an intervention (assessment
tracing subjects might prove difficult and a policy change comes into effect. As policy bias).11 Studies that have no blinding are
investigators have to rely on the quality of changes can inadvertently be non-beneficial, characterised as open-label.
the recorded information (e.g. electronic investigators compare specific outcomes The RCT allows investigators to control
health records, patient recollection) which before and after a policy change to the intervention and establish causality with
is often low.3 determine if it was of benefit.9 a good degree of certainty providing
On the other hand, prospective cohort Unlike previous designs, the strongest evidence of an
studies are those studying events from the these studies include an RCTs frequently give association (efficacy or safety
present time until a time in the future. The intervention chosen by the rise to higher quality data). However, to calculate
study design allows investigators to incorp- researchers (policy change at a evidence, however the sample size, researchers
orate any exposure or baseline charact- given time-point). However, researchers and must have prior knowledge
eristics to be studied so the study is more they lack randomisation and, medical writers about the expected effect size
complete; however, the follow-up time can consequently, are sensitive to have to be aware of and sometimes ethical issues
be long, especially for infrequent outcomes, confounding effects and causal the limitations of prevent the comparison of an
and such studies can have a high loss to relationships, and are, there- RCTs when intervention with a placebo
follow-up (dropout) rate. fore, less valid than RCTs.9 analysing the results (an inert treatment in blinded

28 | September 2016 Medical Writing | Volume 25 Number 3

 Bruno – Study Design Made Easy

Outcome +

Population Test Group

Outcome –
Study Sample Randomised
Outcome +
Control Group

Outcome –

Figure 3: Schematic representation of a randomised controlled trial.

Subjects in a sample are randomly assigned to different interventions (test and control groups)
and then followed to compare the risk of the outcome between groups.

studies) or no intervention (in open-label quantitative appraisal.14 While meta- taking hormone replacement therapy. More
studies).12 analyses virtually always include systematic recent studies, however, attribute the
RCTs of new drugs are often classified in reviews of the literature, this is not always differences to a lack of external validity of
phases. Phase I trials involve testing in true for systematic reviews. the WHI study that had an older study
healthy volunteers (except for novel population (average age of 63 vs. 57-59
oncology drugs) with dose escalation to Discussion and conclusion years) and a higher percentage of users who
assess safety (i.e. side effects and toxicity) It is true that observational studies are more had gone through menopause more than 10
and to determine if it is appropriate to check prone to bias and confounding effects than years previously.18
for efficacy. Phase II trials involve a small RCTs and that RCTs frequently give rise to In the end, the pyramid shown in Figure
group of patients to assess safety and higher quality evidence, however research- 1 stands true for most study examples and
efficacy. Phase III trials involve a large group ers and medical writers have to be aware of studies higher in the pyramid have more
of patients to further assess and establish the limitations of RCTs when analysing the valid and robust findings. However, medical
safety, efficacy, and effectiveness. Phase IV results. They have more internal validity writers have an increasingly active role as
trials are those performed during post- when compared to prospective cohort consultants and in literature review to
marketing surveillance.13 studies; that is, the causal inference or properly counsel clients in strategies and
relation is properly demonstrated in the evidence-based medicine. A sound
Systematic review study sample, as bias and confounding knowledge of statistical methods is therefore
and meta-analysis factors are often adequately controlled. essential for any contemporary medical
Systematic reviews are studies that try to However, the strategies aimed at increasing writer, and understanding the key
collect all available evidence about a subject internal validity, such as controlling the advantages and limitations of the several
of interest and critically appraise it. intervention and the strict inclusion and study designs presently at our disposal is
Systematic reviews of RCTs are often used exclusion criteria, may undermine the another small step to achieve that goal.
to guide guidelines and other aspects of external validity or generalisability of the
evidence-based medicine. They usually study findings. One pivotal example is the
involve a thorough search on a research finding of increased rates of coronary heart References
question in multiple article databases and disease (CHD) in postmenopausal women 1. Adami HO, Hunter D, Trichopoulos D,
indexes, such as Web of Science, Embase, taking hormone replacement therapy in the editors. Textbook of Cancer
and PubMed.14 Women’s Health Initiative (WHI) RCT.15 Epidemiology. 2nd ed. Oxford: Oxford
Meta-analyses not only try to collect all Contrary to the WHI, two previous University Press; 2009.
available evidence but also combine the prospective cohort studies based on the 2. dos Santos Silva I. Cancer
results of similar papers to give an app- Nurses’ Health Study Cohort suggested a Epidemiology: Principles and
roximate pooled measure of association (e.g. reduced CHD risk.16,17 At first glance, Methods. 1st ed. Lyon: IARC Press;
odds ratio or risk ratio) using specific differences were attributed to a lack of 1999.
statistical methodology. In summary, randomisation in the observational studies. 3. Levin KA. Study design IV. Cohort
systematic reviews give a qualitative As a consequence of the WHI study, studies. Evid Based Dent.
evaluation, whilst meta-analyses aim at a millions of women worldwide stopped 2003;7(2):51–2.

www.emwa.org Volume 25 Number 3 | Medical Writing September 2016 | 29

Study Design Made Easy – Bruno

4. Haynes B. Of studies, syntheses, 12. Wechsler ME, Kelley JM, Boyd IO, 18. Hernán MA, Alonso A, Logan R,
synopses, summaries, and systems: Dutile S, Marigowda G, Kirsch I, et al. Grodstein F, Michels KB, Willett WC,
the “5S” evolution of information Active albuterol or placebo, sham et al. Observational studies analyzed
services for evidence-based healthcare acupuncture, or no intervention in like randomized experiments: an
decisions. Evid Based Nurs. asthma. N Engl J Med. application to postmenopausal
2007;10(1):6–7. 2011;365(2):119–26. hormone therapy and coronary heart
5. Levin KA. Study design III: Cross- 13. Friedman LM, Furberg CD, DeMets disease. Epidemiology. 2008;
sectional studies. Evid Based Dent. DL. Fundamentals of clinical trials. 19(6):766–79.
2006;7(1):24–5. New York: Springer; 2010.
6. Levin KA. Study design V. Case- 14. Higgins J, Churchill R, Cumpston M, Acknowledgements
control studies. Evid Based Dent. Chandler J, editors. Cochrane The author would like to express gratitude
2006;7(3):83–4. Handbook for Systematic Reviews of to Joana Carlos for the proofreading of this
7. Schulz KF, Grimes DA. Case-control Interventions. Ver. 5.0.2. New Jersey: article.
studies: research in reverse. Lancet. Wiley-Blackwell; 2008.
2002;359(9304):431–4. 15. Group TWHIS. Design of the Conflicts of Interest and
8. Mortensen PB. Ambidirectional Women’s Health Initiative clinical trial Disclaimers
studies – an extension of longitudinal and observational study. Control Clin The author declares no conflicts of interest.
studies in psychiatry. Psychiatr Dev. Trials. 1998;19(1):61–109.
1988;6(2):173–81. 16. Grodstein F, Manson JE, Colditz GA,
9. Harris AD, Bradham DD, Baumgarten Willett WC, Speizer FE, Stampfer MJ.
M, Zuckerman IH, Fink JC, A prospective, observational study of
Perencevich EN. The use and postmenopausal hormone therapy and
interpretation of quasi-experimental primary prevention of cardiovascular
studies in infectious diseases. Clin disease. Ann Intern Med. Author information
Infect Dis. 2004;38(11):1586–91. 2000;133(12):933–41. Diogo Bruno, MD has been a medical
10. Levin KA. Study design VII. 17. Grodstein F, Stampfer MJ, Manson JE, writer, translator and proofreader for
Randomised controlled trials. Evid Colditz GA, Willett WC, Rosner B, et three years. He is a senior OB/GYN
Based Dent. 2007;8(1):22–3. al. Postmenopausal estrogen and Specialty Registrar at Hospital Prof.
11. Chan AW, Tetzlaff JM, Altman DG, progestin use and the risk of Dr. Fernando Fonseca, Portugal, has a
Laupacis A, Gøtzsche PC, Krleža-Jerić cardiovascular disease. N Engl J Med. Harvard Medical School certificate in
K, et al. SPIRIT 2013 statement: 1996;335(7):453–61. clinical research, and has presented
defining standard protocol items for works in obstetrics and gynaecology at
clinical trials. Ann Intern Med. several meetings.
2013;158(3):200–7.

Brussels 2016 – save the date

See page 54 for more details

30 | September 2016 Medical Writing | Volume 25 Number 3