Ординатура / Офтальмология / Английские материалы / Study Design and Statistical Analysis a practical guide for clinicians_Katz _2006
.pdf
179 Rejection and resubmission
Internet access is especially important if your study has relevance to researchers and clinicians in underdeveloped countries, where access to journals through medical libraries is limited.
11.5 What if my paper is rejected but I am asked to revise and resubmit it?
Be happy! It is rare that any of us have our work accepted on the first submission. Although a revise and resubmit letter is no guarantee of publication, it is the first step to having your paper accepted.
The key to getting your paper over this last hurdle is to adequately address the comments of the editors and the reviewers. In doing this, try not to be defensive. Pretend you are a salesperson: assume the customer (in this case the editor or the reviewer) is right. In other words, do not focus on proving that you were right. For example, if a reviewer complains that no information about recruitment was included in the manuscript, but you know it was included, do not rail against the lazy reviewer. Assume that you did not do a very good job of explaining it. Review that section and see if there is a way you can explain the issue more clearly.
Of course, if the suggestion of the reviewer or the editor is wrong, then you need to explain why you have not chosen to make the change.
In your response to the editor, begin by thanking him or her for inviting you to resubmit your paper and for the suggestions of how to improve the paper. Address each point raised by the editor and the reviewers using numbers so as to make it easy for the editor to follow the changes. With each comment, first explain what the reviewer/editor has asked you to do, then explain how you addressed the issue, and finally include the actual section of the manuscript that you changed (with the page number) so that the editor can review the changes you have made without having to search through the revised manuscript. For example:
Dear Editor:
Thank you for inviting us to resubmit our manuscript: “The impact of designated bicycle lanes on frequency of bicycling to work”. We appreciate your detailed review of the manuscript.
Below we explain each of the issues raised by the reviewers and how we have incorporated their suggestions into their manuscript.
1. Reviewer “A” is concerned that the increases in bicycling to work that we found may not be due to the creation of the designated bicycle lanes but may instead be due to a greater societal interest in fitness.
To address this concern we compared the number of times people bicycled in the park preand post-creation of the bicycle lanes and found no difference.
180 |
Publishing research |
|
|
We have added the following to the Results section:
There were no differences in the frequency that persons stated that they bicycled in the park preand post-creation of the designated bicycle paths (median 1.3 times per month versus 1.2 times per month, respectively; P 0.20) (p. 13).
We have added to the Discussion section:
Although there was a substantial difference in the frequency of bicycling to work with the creation of the designated bicycle paths, we found no difference in frequency of bicycling in the park. If the increases we saw in bicycling to work were due to a general societal interest in fitness, we would have expected to also find an increase in bicycling in the park (p. 15).
11.6 What if my paper is rejected?
Any rejection, professional or personal, is painful. Rejection of a manuscript can be particularly difficult because of the years of effort that go into planning, conducting, and writing up the results of a study. Besides sadness, you may feel anger towards the editors or the reviewers for failing to appreciate the value of your work.
Although these feelings are natural, they are not particularly helpful in deciding your next step. For this reason, it is often best to put the rejection letter aside after reading it and return to it when your feelings have subsided a bit. (I would hasten to add that this is a good strategy in dealing with many of life’s difficulties.)
Once some time has elapsed, reread the reviews and the manuscript. Try to determine as specifically as you can why your paper got rejected. Generally, the reason will fall into one of the following groups:
1.insufficient interest on the part of the journal,
2.an unfair review,
3.flaws in the paper that you can address,
4.flaws in the paper that you cannot address.
You will know that your paper got rejected due to insufficient interest on the part of the journal if the Editors did not send it out for review or if the reviews were positive but your article was still rejected. In this case, it is best to quickly resubmit your article to another journal, ideally one that is more focused on the topic of your manuscript.
Although researchers often feel that the reviews of their paper were unfair, in my experience, this is rarely the case. Harsh, critical, unforgiving: yes. Lazy, biased: sometimes. Unfair: rarely. Most top-notch journals send papers out to multiple reviewers; if all the reviewers have the same negative feeling about your article, there probably is a problem, if not with your findings, then with your ability to communicate your findings.
181 |
Dealing with the media |
|
|
Tip
If your paper gets rejected multiple times because of the same flaws file the manuscript and get to work on a new project.
However, it is possible to have several positive reviews and one very negative one that you believe is unfair. If this is the case, you may try to appeal the decision of the Editor.
Top-notch journals must reject many more papers than they can accept. A paper that is at the borderline may fall to one side or the other depending on the mood of the editorial board at the time your article is being considered by them. If you decide to appeal a decision, it should be based on the importance of the work appearing in that journal. Sometimes it helps if the senior author writes the appeal letter explaining why he or she feels the work is important to publish. Remember, appealing a rejection should only be done if you received mostly positive reviews. Otherwise you are wasting your time and that of the editor. It should also be done in a very respectful way or you run the risk of developing a reputation for being difficult.
If your article was rejected due to flaws that you can address, fix them and resubmit to another journal. If it was rejected due to flaws that you cannot address, you need to decide your next step. Sometimes, a combination of admitting the flaws clearly in the Discussion section and submitting it to a less prominent journal will result in it being published. Sometimes markedly shortening the manuscript into a letter will result in publication. However, if your paper gets rejected multiple times (including as a letter) because of the same flaws, file the manuscript and get to work on a new project.
11.7 How should I deal with the media?
A press release should contain an explanation of the findings in lay terms and why they are important, a quote from the principal authors and/or others in the field, and information on who should be called for further questions (along with telephone and fax numbers, and e-mail address).
Many researchers are unnecessarily afraid of the press. They worry that their results will be misquoted or sensationalized. To avoid this possibility they hide from the media. This is a big mistake as media attention can amplify the impact of your work (not to mention providing unimaginable pleasure to your family and friends!). Also journalists, especially those who write on medicine and science topics, are genuinely interested in correctly capturing the findings of your study – after all, translation of science into lay terms is their job.
Some journals routinely provide advance copies of manuscripts along with press releases to the major media outlets. This makes your job somewhat easier. However, if your journal is not planning to publicize your article and you feel that your article has significance to lay persons, prepare a press release.
Generally, you will want to send out your press release before the date of publication of your article or the presentation of your data at a conference so that media outlets will have sufficient time to prepare their presentations in time for the release date. On the other hand, you may not want the media coverage to begin until the
182
Your press release should include the embargo date (if any).
Tip
Before your interview determine the three most important points of the paper.
Publishing research
date of publication or presentation. This is especially true of important clinical findings. It is very disconcerting as a clinician when your patients ask you about the results of a trial they read about in the newspaper and you have no data source to use in evaluating the claims in the newspaper. To avoid this situation, the press release should state if the material is embargoed (cannot be released) until a particular date. Although, embargoes cannot be enforced, almost all professional media people will respect an embargoed report because they understand that without embargoes it would be impossible to brief the media ahead of the release of data.
Send your press packet to media outlets (television, radio, print, internet) in your area and nationally/internationally (if the findings warrant this). Followup the press release with calls to media people you think might be interested (e.g., the science writer at your local newspaper).
Most journalists will want to interview you on the results. Do not be frightened. They are not trying to catch you off guard. An interview gives you a chance to answer any questions, correct misinterpretations, and help the journalist shape the story.
You may find that a journalist will try to push you to generalize your findings beyond the scope of your work. Do not fall for it. If you are asked a question that goes beyond the data simple state: “the study did not address that question” or “I have no data on that question, but our data do show . . .”
Prior to doing an interview, determine the three most important points of the paper. Then make sure you state these three points during the interview. If you are asked a question that you are not comfortable answering, simply state one of the points.
After the press coverage has passed, call or write to those media people who covered your story well and thank them for doing so. Building a positive relationship with help you the next time you want to get coverage for a story.
Also, do not blame the print writers for the headlines in their newspaper. Someone else does these and they often do not fit the article.170
170For more detailed advice on working with the media see: Stamm, K., Williams, J.W., Noel, P.H., Rubin R. Helping journalists get it right: a physician’s guide to improving health care reporting. J. Gen. Intern. Med. 2003; 18: 138–45.
12
Conclusion
12.1 Would you review the steps for designing and analyzing data from a clinical study?
Step 1 Choose a question that you are genuinely interested in knowing the answer to.
Step 2 Perform a literature search, review the published work, and speak to the experts in the field to learn of unpublished work.
Step 3 State your question in terms of a null and an alternative hypothesis. Step 4 Choose a study design by considering the advantages and disadvan-
tages of the different methods (Chapter 2).
Step 5 Determine the type of univariate, bivariate, and multivariable analyses you will need to perform (Chapters 4–6).
Step 6 Perform a sample size calculation (Chapter 7). Step 7 Develop a study manual (Section 3.2).
Step 8 Submit your research protocol to an institutional review board for approval (Section 2.12).
Step 9 Develop data entry screens (Section 3.3). Step 10 Collect your data (Section 3.2).
Step 11 Enter your data (Section 3.4).
Step 12 Clean, recode, and transform your data, and derive any variables you will need (Sections 3.5–3.8).
Step 13 Review the distribution of all of your variables (Section 4.1).
Step 14 Conduct univariate, then bivariate, and finally multivariate analyses (Chapters 4–6).
Step 15 Write up your results (Section 11.1). Step 16 Send out for publication (Section 11.4).
Step 17 Revise and resubmit (Sections 11.5–11.6).
Step 18 Develop a media strategy to coincide with the publication of your paper (Section 11.7).
Step 19 Bask in your glory!
183
Index
absolute risk, 165
absolute risk difference, 165, 166, 169 Access, 43
accuracy, 145–146, 154 alpha, 132, 134, 135, 136
analysis of variance (ANOVA), 67, 81, 88–89, 105, 108
repeated-measures analysis of variance 108, 111–113
antilogarithm 124 assumptions
censoring, 62, 63, 64
linearity, 92–96, 96–99, 101–102 normality, 54, 55, 56, 57, 79, 80, 81, 82, 83 proportionality, 126
attributable fraction, 166, 167, 168, 169 attributable risk, see absolute risk difference attributable risk percentage, see attributable
fraction authorship, 174–175
bar graphs, 59–60 Bartlett’s test 85
Bayes’ theorem, 142, 148–153, 155 beta, see coefficient
bias, 19, 142, 154, 155, 159–161, 173 biologic plausibility, 156, 160 blinding, 15, 156
blocked randomization, 20–22 BMDP, 171
Bonferroni correction, 83, 90, 105 bootstrap validation, 154
box plots, 56
carryover effects, 18, 19 case–control, 16, 23, 24, 25, 26–31
categorical variable, 35, 36, 41, 42, 100
see also nominal and dichotomous variables causality 6, 155, 158–159, 174
censoring, 62, 63, 64 assumptions of, 64
central limit theorem, 81
chi-squared, 66, 68–72, 77–78, 90, 91, 101–102, 105, 127, 170
clustered observations, 107 Cochran’s Q, 108, 110
coefficient, 93, 96, 97, 98, 99, 124–125 cohort, see prospective cohort study confidence intervals, 58, 59, 74, 75, 130,
131, 144
confounder, 28, 29, 120, 121, 122, 157, 158, 160
consistency checks, 40, 44
continuous variable, see interval variable correlation coefficient,
Pearson’s correlation coefficient 67, 83, 93, 96–97, 99, 132, 135
Spearman rank, 100, 67, 83, 93, 99
Cox regression, see proportional hazards analysis cross-sectional study, 23, 24–25
crossover study, 18, 19 curvilinear, 93, 92, 95
data
cleaning, 45, 34, 38 collection, 32, 38–40, 41, 45 entry, 39, 40, 41, 43, 44, 45, 51 export, 38, 43, 45, 50–51 recoding, 40, 42, 38
sparse data 45–48 transforming, 50, 83
DBASE Plus, 43 deriving variables, 38, 50
diagnostic studies, see predictive studies dichotomous variable, 35, 47, 57–59, 66–77,
77–79, 83, 84–88, 100, 101–102, 108–109, 110, 116–117, 130–131, 133–134, 138
185
186 |
Index |
|
|
discriminant function analysis, 124 distribution
bimodal, 56
Gaussian, see normal distribution Nonnormal, 81, 96, 100
normal distribution, 53, 54, 57, 79, 83 skewed, 54, 55, 62, 80, 83, 113
dose-response, 156, 157, 161 Dunn’s test 83, 91, 92 Dunnett’s test
ecologic study, 24 ecological fallacy, 32
effect size, 128, 132, 133, 162, 163, 164 EpiData, 38, 43
Epi Info, 171, 38, 43–44, 51
equal variance, 79–80, 81, 83, 85, 89 equal allocation randomization, 20, 21, 23 equivalence trials, 137
etiologic studies, see explanatory studies exact tests, 79, see also Fisher’s exact experimental studies, 16
explanatory studies, 141–143
F, 88, 89, 91
F test for the equality of variances, 85 factorial study, 19–20
FileMaker Pro, 43
Fisher’s exact, 66, 72, 79, 109, 171 FoxPro, 43
frequencies, 57
Friedman’s test, 108, 115, 116, 127
Geham’s test, see Wilcoxon test gold standard, 153, 154
Hawthorne effect 16 hazard ratio, 121, 125 histograms, 52, 53, 55
human subjects committees, see institutional review boards
incidence, 61, 64–65
institutional review boards (IRB), 37, 184 intercept, 97, 124
interquartile range, 56
interval variable, 35, 36, 41, 42, 46, 47, 55, 58, 79, 108, 110, 131, 135–136, 138, 146, 151
J-shape, 94
jackknife validation, 154
Kaplan-Meier, 61, 62, 63, 64, 65, 90, 102, 104 Kruskal-Wallis, 67, 83, 89, 91, 108
kurtosis, 57
Levene’s test 85, 89
likelihood ratio, 142, 148, 150, 151, 152, 153 linear regression, 67, 93, 97, 99, 123, 125, 170 logarithmic transformation, 83
log-rank, 104, 105, 106, 117, 135–137
Mann-Whitney test, 67, 83, 86, 87, 88, 91, 99 Mann-Whitney U test, see Mann-Whitney test Mann-Whitney rank sum test, see
Mann-Whitney test matching, 116, 119, 156, 31, 28, 121 masking, see blinding
matched odds ratios, 117
McNemar’s test, 108–109. 116, 117, 171 mean, 53, 88–90, 127, 131, 134
media, 181–182, 183 median, 54, 55, 56
survival 61–64 missing data, 129 mode, 56
multiple linear regression, 123, 137 multiple logistic regression, 123, 124, 137
negative predictive value, 144, 145, 146, 147 nested case-control, 29–31
nominal variable, 36, 77–79, 47, 88–92, 124 noninferiority trials, 137
nonparametric statistics, 83–84, 100 nonrandomized studies, see observational
studies
normal distribution, 54, 57 normal probability plot, 57, 58
number needed to treat, 166, 169–170
observational studies, 16–17, 23–32 cross-sectional, 23, 24–25, 74–75, 158–159 nested case-control, 29–31
prospective cohort, 25–26, 27, 121 case-control, 16, 23, 24, 25, 26–29, 29–31,
68, 75
odds ratio, 67, 73, 75, 76, 117, 125, 142, 156, 157, 167
one-tailed test, 33 one-sided hypothesis, 33
ordinal variable, 35, 46, 52–58, 100–102, 113–116, 118–119, 124
187 |
Index |
|
|
pairwise comparisons, 78, 83, 90, 92, 105–106 Pearson’s correlation coefficient, see correlation
coefficient
poisson regression, 123, 171 polytomous logistic regression, 124
population attributable fraction, 168–169 positive predictive value, 144–145, 147 power, 46, 81, 132, 134, 135, 136, 137, 138 posttest odds, 151, 152
posttest probability, 149, 150, 151 predictive studies, 141, 153 press, see media
pretest odds, 151, 152
pretest probability, 148, 149, 150, 151, 155 prevalence, 8, 10, 58, 168
prevalence ratio, 73, 75
prognostic studies, see predictive studies proportionality assumption, 126 proportional odds logistic regression, 124
proportional hazards analysis, 51, 123, 125, 137 prospective cohort study, 25–26, 29, 30, 31 proximal marker, 139
r, see Pearson’s correlation coefficient R (statistical program), 171 randomization, 18, 22, 122, 156, 158
blocked, 20–22
equal allocation, 20, 22, 23 stratified, 21, 22–23 unequal allocation, 22
randomized controlled trial, 17–20, 161 rate ratio, 67
receiver operating characteristic curve, 147 relative risk, 30, 73, 124, 158, 165–170
hazard ratio, 73, 121, 156 rate ratio, 73, 107, 156 risk ratio, 73, 142, 156, 168
relative hazard, see hazard ratio repeated-measures analysis of variance, 111–112 reverse causality, 6, 24, 25, 158–159, 160, 174 risk ratio, 67, 73, 107
ROC curve, see receiver operating characteristic curve
S-Plus, 171
sample size, 2, 3, 4, 36, 81, 84, 127, 163, 183 SAS, 43, 50, 171
scatterplot, 92
sensitivity, 142, 143–144, 145, 146, 147, 148, 150, 151, 154
skewed distribution, 55, 83, 113
skewness, 57 skip logic, 40, 44 slope, 97, 98
social-desirability, 173, 160 sparse data, 45–48
Spearman’s rank correlation coefficient, 67, 83 specificity, 142, 143–144, 145, 146, 147, 148,
150, 154 spectrum bias, 154 spline functions, 125
split-group validation, 154 SPSS, 43, 171
standard deviation, 54, 55, 130, 131, 132, 135 STATA, 50, 171
stratified randomization, 21, 22–23 stratification, 122, 156
statistical software packages, 38, 43, 49 Student’s t test, see t test Student-Newman-Keul’s test, 91 study manual, 38, 39, 183
SUDAAN, 171
survival analyses, 63, 119
t test, 67, 81, 83, 84, 85, 88, 90, 108, 111, 127, 170
Paired, 108–111, 112, 113–115, 116–119, 127 threshold, 94, 95
time, 61
treatment efficacy, 16 treatment effectiveness, 16
transformation of variables, 50, 83, 126 two-sided hypothesis, 32
type I error, 33, 132 type II error, 133
U-shaped, 93, 94
value labels, 42 validity, 154 variable
categorical, see nominal and dichotomous variables, 35, 36, 41, 100
continuous, see interval variable dichotomous, 35, 47, 58–59, 66–76, 77–79, 83,
84–88, 100, 101–102, 108–109, 110, 116–117, 130–131, 133–134, 138
interval, 35, 36, 41, 42, 46, 47, 58, 79, 108, 110, 131, 135–136, 139, 146–151
nominal, 35, 36, 77–79, 47, 88–92, 124 ordinal, 35, 46, 52–58, 100–102, 113–116,
118–119, 124
188 Index
value labels (contd) normal, 53, 56, 84, 101 reorienting, 48–49 transformation, 83, 125
variance, 53, 54, 85 equal, 79, 85
washout period 19 Wilcoxon test, 86, 106
Wilcoxon signed rank test, 86, 106, 113, 114, 115, 117, 118, 119
Wilcoxon rank sum test, see Mann-Whitney test