The number of scientific articles in print increases each year, and it is impossible to read all of these, even when limited to one’s own field of interest. For example, almost 30’000 papers on childhood asthma are published each year (PubMed search).
47
Vol. 16 No. 4 2005 Fortbildung / Formation continue
The number of scientific articles in print in-
creases each year, and it is impossible to
read all of these, even when limited to one’s
own field of interest. For example, almost
30’000 papers on childhood asthma are pu-
blished each year (PubMed search). From
time to time, however, even the busiest cli-
nician will want to read a particular article ca-
refully and completely. It would be extremely
helpful if such designated article reading
could be done quickly and efficiently, without
jeopardizing the quality of information gle-
aned from the article. In this review, I pro-
pose a simple 5-step method to read an ar-
ticle on therapy critically and carefully, which,
after a bit of exercise, can be done in 5 mi-
nutes. The five steps are presented in ta-
ble 1.
Step 1. Be active!
We all know that physical activity is healthy,
and promoting exercise in childhood is con-
sidered to be one of the key measures to pre-
vent obesity. Yet, when reading a scientific
paper, we usually take the «couch potato» ap-
proach: we sit on our seats, we watch the
screen (or the printed paper), and we eat it
all up. As a result, we usually take the results
(and the authors’ conclusion) at face value.
Needless to say, this is not a very useful (nor
a very exciting) way of collecting knowledge.
Conversely, much more information can be
gleaned from a paper if the reader takes an
active approach. Just by asking two ques-
tions before one starts reading a paper, the
reader can quickly decide whether the paper
is worth reading at all (table 2).Sometimes, the answer to question 1 can be
easily obtained by just looking at the title. For
example, in the Impact study, the title tells
you that the study is about the effects of pa-
livizumab on hospitalizations for respiratory
syncytial virus (RSV) infections. (IMpact Stu-
dy Group, 1998) Most commonly, however,
this information must be obtained from the
Methods section (see step 2). Perhaps even
more importantly, the size of the treatment
effect that one considers useful can be de-
cided upon before one starts reading the pa-
per. Quite commonly, effects of treatment are
described to be «significant», which is useful
from a statistical point of view, but utterly use-
less from a clinical point of view. For exam-
ple, in a clinical trial comparing budesonide
and fluticasone in childhood asthma, the ef-
fect on morning peak flow of fluticasone was
said to be «significantly higher» than that of
budesonide, but in fact, the difference in peak
flow between the two groups was only 7
L/min, (Hoekx et al., 1996) which, of course,
is completely irrelevant (figure 1).
Step 2. Read the methods section
Although this sounds boring and uninviting,
reading the methods section is the key towards
useful interpretation of data from any article.
In particular, the validity of study design needs
to be determined. In a study on treatment ef-
fects, the points mentioned in table 3are of
particular importance. (Guyatt et al., 1993)
The importance of randomization cannot be
overemphasized. Any study that is not ran-
domized is an observational study, which are
useful for generating hypotheses, not for tes-
ting them. (Guyatt et al., 1993) For example,
in a quite well-known observational study,
children using inhaled steroids had far bet-
ter lung function than children whose pa-
tients refused inhaled steroids because of
fear of side effects, and who used cromolyn
instead. (Agertoft and Pedersen, 1994)
Quite correctly, the authors presented this as
an observational study, and designed a pro-
per prospective randomized controlled stu-
dy to test the hypothesis that their study ge-
nerated. (Pauwels et al., 2003).
Exceptionally, a study is presented as a ran-
domized trial when, in fact, it is not. This was
the case with a Finnish study comparing the
effects of hypoallergenic versus cow’s milk
formula as supplementary feeding during the
first days of life on the development of ato-
pic disease later in life. This study was pres-
ented as a randomized study, both in the
title and in the abstract. After reading the
Methods section, however, it was obvious
that allocation to treatment was actually not
random, but «depended on the month of
birth and the hospital». (Saarinen et al., 1999)
How to read an article about therapy
in 5 minutes
Paul L P Brand, Zwolle (NL)
Table 1: 5-step method to read an article
about therapy in 5 minutes
1. be active!
2. read the methods section
3. analyze results for yourself
4. no surrogate end points,
no subgroup analyses
5. draw your own conclusion
Table 3: Step 2: Read the methods
section, and decide whether the results
of this study on treatment are valid.
1. was the assignment of treatment to
patients randomized? And was the
randomization list concealed?
2. was follow-up of patients sufficiently
long and complete?
3. were all patients analyzed in the groups
to which they were randomized?
Some less important points:
4. were patients and clinicians kept blind
to treatment?
5. were groups treated equally, apart
from the experimental therapy?
6. were the groups similar at the start of
the trial?
Table 2: Step 1: be active! Questions
to consider before you start reading
1. is the outcome measured an important
one in my view?
2. what effect size do I think is relevant?
Figure 1: Results of a study in which the effects on morning
peak expiratory flow (PEF) of 8-wk treatment with fluticasone
proprionate (FP) and budesonide (BUD) by dry powder in-
haler were compared in children with asthma. According to
the abstract, morning PEF was «significantly higher» in the
FP group than in the BUD group. The mean difference in mor-
ning PEF was 7 L/min (95% confidence interval 1–14 L/min)
which is an irrelevant difference clinically.
Fortbildung / Formation continue
48
Vol. 16 No. 4 2005
Such systematic randomization is known
to introduce considerable bias. (Altman and
Dore, 1990) Consequently, the study should
be regarded as observational, and the results
should be interpreted with caution.
Similarly, lack of concealment of allocation
can undermine a study’s validity. If an allo-
cation list is open for investigators to view,
purposeful (biased) allocation is a real pos-
sibility. As a rule of thought, if the methods
section does not mention allocation to be
concealed, it should be regarded as unblin-
ded, and, hence, biased. For example, in a
study on the preventive effect of probiotics
on the development of atopic disease, con-
cealment of allocation was not mentioned,
and this issue has been raised in evidence-ba-
sed journals discussing this study. (Kalliomä-
ki et al., 2001) The prevalence of atopic ec-
zema in the placebo group in this study is un-
usually high (almost 50%) when compared to
similar groups of patients in other European
countries, and it is possible that this was cau-
sed by purposeful open allocation.
When looking at table 3, it may come as a
surprise that blinding of treatment is listed
under «some less important points». This is
because the degree of bias introduced by
lack of randomization or unblinding of tre-
atment allocation is larger than the degree
of bias introduced by unblinding. This is par-
ticularly true when «hard» end points are
used, such as death or results of a laboratory
test. Conversely, if «soft» subjective end po-
ints such as symptom scores or diagnoses
are used, unblinding may become a real pro-
blem. For example, in an unblinded study on
the effects of chiropractic therapy on crying
behaviour in infants with colic, the therapy
was found to be highly effective as judged by
the hours of crying parents recorded in a dia-
ry; (Wiberg et al., 1999) in a similar, but blin-
ded study, both the active and the sham chir-
opractic treatment were associated with sig-
nificant improvements in crying, indicating a
strong placebo effect. (Olafsdottir et al., 2001)
Step 3. Analyze results
for yourself
Reading scientific articles reallybecomes
interesting if one begins to analyze results for
oneself. In keeping with step 1 (Be active!), the
reader is encouraged to determine beforestar-
ting to read the paper which effect size one
considers to be relevant (for a continuous endpoint), or which number needed to treat (NNT)
one thinks to be acceptable (table 4).
Studies with a continuous end point
Consider the study in figure 1once again. The
mean difference in morning PEF after 8 wks
of treatment was 7 L/min, with a 95% con-
fidence interval (CI) from 1 to 14 L/min. This
means with 95% confidence, the true diffe-
rence in morning PEF between asthmatic
children treated with FP and BUD would lie
between 1 and 14 L/min. Is this clinically re-
levant? I don’t think so.
A useful way of looking at this graphically is
represented in figure 2.
Studies with a dichotomous end point
The usual outcome parameter in such stu-
dies is a percentage. In the Impact studymentioned previously, the percentage of hos-
pital admissions for RSV infections was
4.8% in the palivizumab group and 10.6% in
the placebo group. (IMpact Study Group,
1998) Such a difference is usually present-
ed as a relative risk reduction,probably be-
cause the figure (in this case 55%!) looks im-
pressive. It is the absolute risk reduction
(ARR), however, that matters, because the
number needed to treatcan be easily com-
puted as the inverse of this ARR. In this
example, the ARR is 10.6% – 4.8% = 5.8%,
and the NNT is 1/5.8% = 17.2. This means
that 17 patients should be treated with pali-
vizumab in order to prevent one RSV hospi-
talization. Whether or not this is relevant or
acceptable, is up to the reader to decide. In
this particular case, the discussion has fo-
cused around the cost effectiveness of this
approach.
Step 4: no surrogate end points,
no subgroup analyses
In step 1, we have decided which outcome
measure we consider important. Facing dif-
ficulties in obtaining the most relevant out-
comes easily and reliably, clinical trial desi-
gners often resort to measuring something
that can be easily measured, but has a more
or less distant relationship to the truly rele-
vant outcome. This is called a surrogate end
point. Surrogate end pointsare fraught with
difficulties, which may be best explained with
an example. (Greenhalgh, 1997) Obviously,
in HIV disease, the relevant outcome para-
meter is death, but in clinical trials, surrogate
end points such as CD4 counts are com-
monly used because they are easily measu-
rable. It is assumed that CD4 counts predict
death, but in a large and quite famous study,
early treatment with zidovudine in HIV pa-
tients slowed the decline of CD4 counts…
but had no effect on survival (paper quo-
ted in Greenhalgh 1997). Thus, the reader
should decide if the outcome measure in a
particular therapeutic study is a relevant one,
and if it is a surrogate end point, whether the
evidence linking the surrogate end point to
the actual outcome measure of interest is
sound. It usually isn’t.
The issue of subgroup analyses is even more
contentious. It is common practice for a cli-
nical trial to report results both from the stu-
dy as a whole and in different subgroups. This
is onlyOK if two conditions are met (table 5).
(Yusuf et al., 1991)
Table 4: step 3: Analyze results for yourself
Studies with a continuous end point:which
treatment effect do I think is relevant?
Calculate mean difference and 95% CI of
difference
Studies with a dichotomous end point:which
number needed to treat (NNT) do I consider
acceptable?
Calculate NNT as 1 divided by the absolute
risk reduction
Figure 2:A graphical method to distinguish between sta-
tistical significance and clinical relevance. The squares re-
present the difference in treatment effect between two stu-
dy groups; the vertical lines represent the 95% CIs around
this mean difference. The thick horizontal line represents
the zero difference line; the thin horizontal line represents
a clinically relevant difference (the reader decides where
this line should lie!). A CI crossing the zero (thick) line re-
presents a study with evidence of no effect. Studies with
a CI above the zero line represent studies with evidence
of effect, but if the CI is below the thin line, the effect is
clinically irrelevant (which, in practice, is of equal impor-
tance as a study with evidence of no effect). The Hoekx
study is an example of such a study. (Hoekx et al., 1996)
This figure is adapted from a textbook in Dutch on evi-
dence based medicine. (Offringa et al., 2003)
49
Vol. 16 No. 4 2005 Fortbildung / Formation continue
As a result of biological variation, results in
different subgroups can be expected to dif-
fer from the overall study result, but this is
a quantitative rather than a qualitative phe-
nomenon. A subgroup result which is in the
opposite direction or strikingly different from
the overall result should, therefore, raise an
eyebrow. The most common example of this
is a study with an overall negative result («evi-
dence of no effect», see step 3), but a signi-
ficant treatment effect in one or more sub-
groups. For example, this was the case in a
study where cetirizine was given to infants
with atopic eczema in the hope of preventing
the development of asthma. (Warner, 2001)
The overall study result was zero, but infants
with house dust mite or grass pollen allergy
were found to benefit from cetirizine thera-
py. The paper concluded that cetirizine «tru-
ly delays or, in some cases, prevents the de-
velopment of asthma in a subgroup of infants
with atopic dermatitis…». From a methodo-
logical point of view, this conclusion is inva-
lid. The study was not designed to answer the
research question in a subgroup, and the
subgroup analysis was not planned before
the study started (in technical terms, it was
a «post hoc analysis»). If the overall result was
zero and one subgroup benefited from active
therapy, then another subgroup must have
benefited from placebo therapy. Clearly, we
wouldn’t draw the conclusion that these pa-
tients should be treated with placebo. The in-
verse conclusion, however, is just as invalid.
What canbe done is consider subgroup ana-
lyses as hypothesis generating exercises
(just like the observational studies mentio-
ned in step 2). It is quite valid to conclude
from the cetirizine study that infants with ato-
pic eczema and house dust mite or grass pol-
len sensitization mightbenefit from cetirizine,
but this hypothesis must be tested in a stu-
dy that was designed to do so before this
conclusion can be accepted. Meanwhile, pa-
tients in clinical practice should not be tre-
ated along this hypothesis (at least not if one
wants to practice evidence based medicine).
Step 5: draw your own conclusion
The example of the cetirizine study brings us
to step 5: draw your own conclusion. Clear-
ly, my conclusion differs from the one
drawn by the authors of the study, and I am
convinced that myconclusion is the right
one. Even though the paper was published in
a major high-quality journal, I take the liberty
of disagreeing – respectfully – with its au-
thors’ conclusion. This, perhaps, is the
most important of my five step method. I
would encourage each and every reader to
draw their own conclusions after reading an
article. This can be done quite easily – just
follow the steps in table 1, and with a little
practice, you will be able to do so yourself.
I guarantee you: it is a rewarding exercise,
and makes scientific reading much more
worthwhile (and fun).
One final note: if you need a memory aid to
remember the five steps – it’s easy: the first
letters of each of the five steps form the last
name of the author of this paper.
Reference List
– IMpact Study Group. Palivizumab, a humanized respi-
ratory syncytial virus monoclonal antibody, reduces
hospitalization from respiratory syncytial virus infec-
tion in high-risk infants. Pediatrics 1998; (102): 531–
537.
– Hoekx J C M, Hedlin G, Pedersen W, Sorva R, Holling-
worth K, Efthimiou J. Fluticasone propionate compared
with budesonide: a double-blind trial in asthmatic chil-
dren using powder devices at a dosage of 400 mg.day-
1. Eur Respir J 1996; (9): 2263–2272.
– Guyatt G H, Sackett D L, Cook D J, Evidence-Based Me-
dicine Working Group. Users’ guides to the medical li-
terature. II. How to use an article about therapy or pre-
vention. A. Are the results of the study valid? JAMA
1993; (270): 2598–2601.
– Agertoft L, Pedersen S. Effects of long-term treatment
with an inhaled corticosteroid on growth and pulmo-
nary function in asthmatic children. Respir Med 1994;
(88): 373–381.
– Pauwels R A, Pedersen S, Busse W W, Tan W C, Chen
Y–Z, Ohlsson S V, Ullman A, Lamm C J, O’Byrne P M.
Early intervention with budesonide in mild persistent
asthma: a randomised, double-blind trial. START inves-
tigators group. Lancet 2003; (361): 1071–1076.
– Saarinen K M, Juntunen-Backman K, Järvenpää A–L,
Kuitunen P, Lope L, Renlund M, Siivola M, Savilahti E.
Supplementary feeding in maternity hospitals and the
risk of cow’s milk allergy: a prospective study of 6209
infants. J Allergy Clin Immunol 1999; (104): 457–461.
– Altman D G, Dore C J. Randomisation and baseline com-
parisons in clinical trials. Lancet 1990; (335): 149–153.
– Kalliomäki M, Salminen S, Arvilommi H, Kero P, Kos-
kinen P, Isolauri E. Probiotics in primary prevention of
atopic disease: a randomised placebo-controlled
trial. Lancet 2001; (351): 1076–1079.
– Wiberg J M M, Nordsteen J, Nilsson N. The short-term
effect of spinal manipulation in the treatment of in-
fantile colic: a randomized controlled clinical trial with
a blinded observer. J Manipulative Physiol Ther 1999;
(22): 517–522.
– Olafsdottir E, Forshei S, Fluge G, Markestad T. Ran-
domised controlled trial of infantile colic treated withchiropractic spinal manipulation. Arch Dis Child 2001;
(84): 138–141.
– Offringa M, Assendelft W J J, Scholten R J P M. Inleiding
in evidence-based medicine. Klinisch handelen geba-
seerd op bewijsmateriaal. Bohn Stafleu van Loghum,
Houten 2003.
– Greenhalgh T. How to read a paper. Papers that report
drug trials. BMJ 1997; (315): 480–483.
– Yusuf S, Wittes J, Probstfield J, Tyroler H A. Analysis and
interpretation of treatment effects in subgroups of pa-
tients in randomized clinical trials. JAMA 1991; (266):
93–98.
– Warner J O. A double-blind, randomized, placebo-con-
trolled trial of cetirizine in preventing the onset of asth-
ma in children with atopic dermatitis: 18 months’ tre-
atment and 18 months’ posttreatment follow-up. ETAC
Study Group. J Allergy Clin Immunol 2001; (108):
929–937.
Correspondence:
Paul L P Brand
Princess Amalia Children’s Clinic
Isala klinieken
Zwolle, the Netherlands
Tel. +31 38 424 5050
Fax +31 38 424 2734
p.l.p.brand@isala.nl
Table 5: conditions that must be met
before results from subgroup analyses
can be accepted
1. the subgroup analysis was planned
before the study (and not reported
because the data looked so exciting)
2. the result in the subgroup is in the same
direction as in the whole group
Weitere Informationen
Autoren/Autorinnen
Paul L P Brand , Princess Amalia Children’s Clinic Isala klinieken Zwolle, the Netherlands