the art and science of writing a paper |
|
Indian Pediatr 2016;53: 811-814 |
|
Reporting Statistics in Biomedical Research
Literature: The Numbers Say it All
|
Amir Maroof Khan and *Siddarth
Ramji
From the Department of Community Medicine, University
College of Medical Sciences and GTB Hospital, and *Department of
Neonatology, Maulana Azad Medical College and associated Lok Nayak
Hospital; New Delhi, India.
Correspondence to: Dr. Amir Maroof Khan, Associate
Professor, Department of Community Medicine, University College of
Medical Sciences and GTB Hospital, Dilshad Garden, Delhi, India.
[email protected]
Editor’s Note: Writing a scholarly article (and
getting it accepted too) is both ‘art’ and ‘science’. Most reputed
journals have a high rejection rate, and extensive editing is required
in most of the manuscripts that are accepted. There is no formal
training in paper writing during medical schooling, but faculty members
of medical colleges are expected to write papers in high impact medical
journals for career promotions. Consequently, there is increasing
incidence of plagiarism, duplicate publication and fraud in
paper-writing. The huge trap of predatory journals is also a challenge
for the scientific community. The articles in this series will aim to
help and guide the readers in writing articles for medical journals.
Simplicity will be the key ‘mantra’ for this series. I hope that readers
will find the series useful; any comments and feedback are welcome.
These may be directly communicated to the authors or to the journal
office at jiap.nic.in. Comments can also be posted on the relevant
thread at www.facebook.com/indianpediatrics.
|
Statistics is the cornerstone of evidence-based
quantitative research. Statistical analysis and outputs pave the way for
clinical and policy decision-making, thereby impacting the health status
of countless individuals across the world. Various studies have shown
that statistical reporting is inappropriate and incorrect in biomedical
journals [1-3]. Uniformity and transparency in statistical reporting
strengthens the validity and reliability of the scientific literature.
Most biomedical researchers are not comfortable with statistics and
hence the reporting of the statistical output is quite varied, confusing
and meaningless to the reader. This article attempts to provide an
overview of good practices of reporting statistics in biomedical
research literature.
The statistical reporting guidelines and styles being
presented in this article draw from instructions for authors of
Indian Pediatrics, International Committee of Medical Journal
Editors (ICMJE) guidelines, Enhancing Quality and Transparency of Health
Reporting (EQUATOR) guidelines, and current practices observed in the
published research literature [4-6].
As different journals have their specific
requirements regarding reporting of statistics, it is necessary that the
authors go through the instructions for authors and some already
published articles of the journals to which they intend to submit their
manuscript for publication. However, the guiding principle of reporting
statistical analyses is to
"Describe statistical methods with enough detail
to enable a knowledgeable reader with access to the original data to
verify the reported results [4]."
Report Statistics When Relevant
"Can we do some statistical analysis and report
from this data here?"
This is a common non-specific question which
biomedical researchers ask their colleagues who help them in
data-analysis. The requirement in this case is to insert some
statistical test result in the manuscript, without giving a thought to
the fact as to how it would fit in with the research objectives. The
question reflects the investigator’s lack of clarity and understanding
of their research/study objectives. Had they been clear about their
research objectives, the question would have been something like this:
"How to compare these means?" Or, "What is the effect of this variable
over another one?" Or, "Is there an association between these
variables?" The point of serious concern is that even the peer-review
process, at times, fail to pointedly enquire about the relevance and the
resultant interpretation of the statistical tests applied.
At times, biomedical researchers are not even aware
of the purpose of statistics being applied in their research. They just
want to have it because it’s a ‘cool’ thing to do or their belief that
it increases the chances of publication. Some are of the notion that
without reporting a ‘p-value’, a study is considered as irrelevant. An
understanding of the conceptual framework of the study is important
before embarking on the statistical analysis. The appropriate
application of statistical tests would then aid in the meaningful
deconstruction and interpretation of the study results. This would also
make the discussion and the conclusion sections more meaningful.
Statistics is a vital part of biomedical literature, but only if it’s
relevant. Else it loses its importance. Statistical tests should be
applied and reported where it is relevant and not just for the sake of
reporting it.
Statistical tests are used in biomedical research
broadly for two reasons:
1. Estimation studies. In this type of
studies, there is no hypothesis statement. The research question is to
find out a ‘population estimate’ from a given ‘sample data’ [7].
Examples of such estimates include determining the
mean weight for age Z-score (WAZ) of under-five children or proportion
of underweight births in the newborn population. In these two examples,
the mean WAZ score and the proportion are the statistics to be estimated
using the sample dataset. In both these cases there is no hypothesis
testing involved. Statistical applications for such objectives will
primarily be restricted to reporting of Standard Errors of Means and
Proportions and their related confidence intervals (CI), and will be
devoid of any p-value. Statistics is employed here to extrapolate the
result from the sample data to the population data in quantitative
terms. Hence, statistical tests and the accompanying p-values are
irrelevant while reporting population estimates.
2. Inferential studies (Hypothesis testing).
These studies intend to determine an association between two or more
variables. These studies usually are designed to test a hypothesis [7].
Case control studies, experimental studies
(randomized or non-randomized, with or without control group), typical
cohort studies have a null hypothesis at the start and fall in this
category. Statistical tests of significance and the accompanying
p-values and effect sizes become relevant and should be reported in
studies having hypothesis testing as an objective.
In a research manuscript, statistical methods and
statistical outputs are reported in the Methods section and the Results
section, respectively. What and how you report the statistical methods
and outputs in your manuscript will also depend on the journal you are
submitting your manuscript to.
Reporting of Statistical Methods in the Methods
Section
The statistical methods reported in the methods
section will depend on the study design and objective of the study.
Box 1 presents the key points to be noted while
reporting statistical methods in the methodology section of a biomedical
research manuscript.
Box 1
Statistical Information to be Included in the Methods
Section |
• All the statistical methods employed in the
study.
• Any data transformation done for the purpose of
statistical analysis.
• Identify any uncommon statistical method
applied, with a reference.
• The order of the statistical methods described
should follow that of the objectives mentioned.
• Report the statistical tests in the context of
the research objectives, and not as generic statements.
• Use appropriate statistical tests for analyzing
paired data.
• Consider the type of distribution while
selecting and reporting statistical tests.
• Mention the statistical analysis software packages used for
data analysis only for complex analysis.
|
In the case of observational studies, this section
should report all methods used, including those for confounder control.
The section must address how missing data was addressed (including loss
to follow-up in cohort studies), subgroup analysis if any that were
planned, and how matching was done for cases and controls (in the case
of case-control studies). When reporting randomized trials, the methods
used to compare the primary and secondary outcomes and any additional
analysis that were planned must also be reported in this section. The
details of what to report in the method sections are available in the
reporting guidelines for each study design [6]. The statistical tests
employed should be mentioned with respect to the variables being
analysed, rather than as standalone general statements. Some examples of
stating the test used could be "Categorical data to test for presence
of association between failure to regain birth weight and the likely
risk factors was analysed using Fishers exact test", or "The
strength of association between the factors and failure to regain birth
weight among the infants studied was determined using odds ratios and
confidence intervals" [8].
This section must also report methods used to
transform raw data prior to analysis such as converting non-normal to
normal distribution, collapsing categories in categorical data, etc.
While reporting common tests used such as Chi-square, Fisher exact,
student t-test, linear regressions, no citations are needed. However,
when reporting more complex analysis, the authors must cite the source,
which should preferably be a standard textbook [9]. There is no need to
report the analytical software used for the basic descriptive analysis.
In the case of statistical analysis involving hypothesis testing,
reporting the analytical software is useful and very much required. The
alpha level used to define statistical significance must also be
mentioned in this section. Box 1 presents the key points
to be noted while reporting statistical methods in the methodology
section of a biomedical research manuscript.
Reporting of Statistical Results
The numerical results must be presented keeping the
study objectives in mind. Box 2 presents the important
points to be kept in mind while reporting statistics in the results
section of a research manuscript.
Box 2
Reporting Statistical Outputs in the Results Section of the
Manuscript |
• Avoid nontechnical uses of technical terms in statistics.
• Explicitly state the groups being compared.
• Report exact P-values, and not just as
significant or non-significant.
• Do not report P-values as 0.000. In such
cases, report as P<0.001.
• Report the effect sizes with their confidence
intervals.
• Usually P-values with effect sizes and the
associated CI are sufficient while reporting the statistical test
results, unless the journal asks for additional details.
• Do not use the term ‘correlation’, a
statistical method to assess the relationship between two continuous
variables, to describe ‘association’.
• Refer to relevant reporting guidelines for the study design
e.g., STROBE, CONSORT.
|
What must be included? For descriptive
statistics, the point estimate and 95% confidence interval should be
reported. For comparative studies, rates, risk, ratios or the mean
difference along with their precision such as 95% confidence interval or
standard deviation must be reported.
If P-values are included, the actual value up
to 1 or 2 decimal spaces may be reported (e.g. P=0.2 or
P=0.41); values less than 0.001 should be reported as P<0.001.
P-values should not be reported as "not significant" or "NS".
When reporting outcome results, first the results of
primary outcomes must be reported and later that of the secondary and
any other sub-group analysis. Post-hoc analysis that had not been
pre-specified must not be reported.
Box 2 presents the key messages regarding
reporting of statistics in the results section of a research manuscript.
What can be omitted? Most statistical softwares
will churn out a plethora of outputs during analysis. Typically the
outputs would include test statistic (e.g., chi-square statistic,
t-statistic, F-statistic), p-value, and degrees of freedom; all of these
can be omitted including p-values (but some reviewers would insist on
the reporting of p-values). In manuscripts reporting randomized
controlled trials, p-values should not be reported when comparing
baselines variables/characteristics. Similarly, regression analysis
outputs would include multiple data outputs which may include
coefficients, R 2, standard
errors and p-value. It is best to avoid reporting these in the results
unless they serve a useful interpretive function. It is also best not to
include complex statistical formulas [10].
Study Design-specific Statistical Results
Observational studies: For these, provide
unadjusted estimates and, if applicable, confounder-adjusted estimates
and their precision (e.g., 95% confidence interval). Make clear
which confounders were adjusted for and why they were included. Report
category boundaries when continuous variables were categorized [11]. If
relevant, consider translating estimates of relative risk into absolute
risk for a meaningful time period.
Randomized controlled trials: For these, provide
results for each primary and secondary outcome, and the estimated effect
size and its precision (such as 95% confidence interval). For binary
outcomes, presentation of both absolute and relative effect sizes is
recommended [12].
When reporting correlation, identify the correlation
being reported – Pearson or Spearman. Report the 95% CI and the Pvalue.
While reporting regression analysis, it is best to depict in a tabular
format.
Conclusions
Reporting statistics in biomedical research
literature has become more transparent, uniform and reliable. The
medical researcher should report relevant statistics and provide
meaningful interpretations in their research manuscripts. Pvalues should
not be reported alone. If at all, they should be reported along with
effect sizes and their confidence intervals. The test statistic and
degrees of freedom can usually be omitted. Both, the summary statistics
and the inferential statistics should be given careful consideration
while reporting. Specific statistical tests require specific components
to be reported. Various guidelines are now available to aid the medical
researcher for reporting methods. It is important that the specific
journal guidelines regarding reporting of statistics should be strictly
followed when preparing and sending a manuscript for publication.
References
1. Hassan S, Yellur R, Subramani P, Adiga P, Gokhale
M, Iyer MS, et al. Research design and statistical methods in
Indian medical journals: a retrospective survey. PLoS One. 2015;10:
e0121268.
2. Jaykaran, Preeti Y. Quality of reporting
statistics in two Indian pharmacology journals. J Pharmacol Pharmacother.
2011;2:85-9.
3. Horton NJ, Switzer SS. Statistical methods in the
journal. N Engl J Med. 2005;353:1977-9.
4. International Committee of Medical Journal
Editors. Recommendations for the Conduct, Reporting, Editing and
Publication of Scholarly Work in Medical Journals Available from:
http://www.icmje.org/news-and-editorials/icmje-recommendations_annotated_
dec15.pdf. Accessed August 18, 2016.
5. Indian Pediatrics. Statistics, Instruction to
Authors. Available from: http://indianpediatrics.net/author1.htm#
Statistics. Accessed July 16, 2016.
6. Enhancing the Quality and Transparency of Health
Research (EQUATOR) Guidelines. Available from:
http://www.equator-network.org/ Accessed July 16, 2016.
7. Hypothesis testing and estimation. In:
Jennifer P, Barton B, Elliot E, editors. Statistics Workbook for
Evidence-based Health Care. 1st ed. West Sussex, UK: John Wiley & Sons;
2009.
8. Namiiro FB, Mugalu J, McAdams RM, Ndeezi G. Poor
birth weight recovery among low birth weight/preterm infants following
hospital discharge in Kampala, Uganda. BMC PregChildbirth. 2012;12:1.
9. Haruhiko F, Yasuo O. A Guideline for Reporting
Results of Statistical Analysis in Japanese Journal of Clinical
Oncology. Jpn J Clin Oncol 1997;27:21-7.
10. Lang TA, Altman DG. Basic statistical reporting
for articles published in clinical medical journals: the SAMPL
Guidelines. In: Smart P, Maisonneuve H, Polderman A (eds).
Science Editors’ Handbook, European Association of Science Editors,
2013.
11. von Elm E, Altman DG, Egger M, Pocock SJ,
Gotzsche PC, Vandenbroucke JP. The Strengthening the Reporting of
Observational Studies in Epidemiology (STROBE) Statement: guidelines for
reporting observational studies. Ann Intern Med. 2007;147:573-7.
12. Schulz KF, Altman DG, Moher D, for the CONSORT
Group. CONSORT 2010 Statement: updated guidelines for reporting parallel
group randomised trials. Ann Int Med. 2010;152:726-32.
|
|
|
|