We found that over a quarter of biomedical publications do not discuss any limitations. In abstracts, limitations are rarely mentioned. Industry-supported publications appear to express less uncertainty after controlling for factors that justify confidence in the study results. Finally, we found major differences between journals in how uncertainty is expressed and limitations acknowledged.
Complete reporting of study design, success of execution and, if appropriate, statistical analysis give meaning to the results of empirical studies. That is why, in science, we value (detailed) methods sections in research publications 
. In addition, methods sections facilitate attempts at replication by others. It is often easier to plan
a flawless study than to execute
one. This distinction is important since readers can increasingly read the investigators' intentions through, for example, trial registration websites or separate publications dealing solely with a study's rationale, design and protocol details 
. Access to details on study execution or log books is still rare 
. In principle, one could place an unabridged list of all important differences between the study plan and actual execution in an appendix to a manuscript. However, deciding which events during a study's execution count as protocol violations and which violations cause limitations requires judgment 
. Consider two examples: If the actual time interval between blood sampling and refrigerating the samples was 2 percent longer than stated in the protocol in 29 out of 1,000 samples collected for later determination of some compound, should one report that? By contrast, if in a randomized trial, a post-hoc chemical analysis shows that the interventional drug contained only 10% of the intended amount of the active compound, should that be acknowledged as a study limitation 
? Our findings indicate that many authors appear not to be aware of limitations or are reluctant to admit them even after peer review.
Obviously, the present work would be quite incomplete if we did not address its limitations. First, for sub-study 1, the body of papers we assessed is five years old. However, we think that is unlikely that the awareness about acknowledging limitations has changed much, if at all, in recent years. Second, in sub-study 2, we lost 10, and could not analyze six of our 300 publication sample due to administrative errors or technical problems. Third, the software application used to calculate the hedging scores was not perfect. In particular, its accuracy is only about 93% 
. Fourth, the weights assigned to the types of hedges are somewhat arbitrary. Fifth, it would have been interesting to determine an article's hedging score separately for each of the most relevant sections such as the title, abstract, results and discussion sections and thus be able to fine-tune the normalization using the word count, which was now summed across the whole publication. Sixth, the binary coding of industry support will have caused some misclassification. If misclassification were random, we may have underestimated the lower degree of hedging in industry-supported publications. A finer scale of industry support would have been attractive, but difficult to achieve.
As far as we are aware, research on the use of limitations is scarce. Previously, Ioannidis reported that 17% of publications mentioned any limitations. However, he used an automated search strategy on the texts that is likely to have missed acknowledgments of limitations 
. Goodman et al., in a masked before-after study on manuscripts submitted to the Annals of Internal Medicine, found that newly submitted manuscripts scored worst on reporting about blinding, non-inclusions, drop-outs, multivariate methods, generalization of results and study limitations. Of those, reporting on limitations and generalization benefited most from peer review by physicians and epidemiologists trained in research methods, although even after peer review, the section on limitations remained among the weaker sections 
. Our work does not shed light on the question of why industry support should lead to the expression of less uncertainty, all other factors being equal? However, there is a vast field specialized in writing texts for the industry and Carl Elliott wrote on the more cynical aspects thereof 
. More recently, Medtronic, a US medical device company, has been accused of manipulating 13 journal articles and paying large amounts of money to authors 
Although the STROBE statement encourages authors to “Discuss limitations [..], taking into account sources of potential bias or imprecision. Discuss both direction and magnitude of any potential bias”, “Give a cautious overall interpretation of results considering objectives, limitations, [..] etc.” and the CONSORT statement urges authors to address “Trial limitations, addressing sources of potential bias, imprecision, and, if relevant, multiplicity of analyses.” 
, it will probably not be easy to curb the tendency of many authors to only sparingly admit issues that were imperfect by design, execution or both. Until a major change in attitudes occurs, ways to improve the reporting of all serious limitations in biomedical research reports may require a structured approach at the editorial offices: Explicit journals' instructions for authors, more structured support and dedicated checklists for peer reviewers, more structured approaches at the editorial offices, perhaps involving automatic comparison of a manuscript's hedging score to reference values for hedging scores conditional on the study's design and outcome. Finally, checks on the number and nature of limitations acknowledged and if acknowledged limitations are reflected in the strength of the conclusions drawn may be useful. After publication, web-based rapid response facilities and acknowledgment for those critics who enlighten readers as to serious limitations may be considered. On the other hand, the authors may sometimes be the only ones who know about a limitation due to problems during study execution and have to decide whether it is important enough to be mentioned as a limitation. We believe that in general, editors may encourage authors to write about their methods used more extensively at the expense of the length of the discussion sections that are sometimes filled with much speculation. Spending at least a fixed proportion of the discussion section on limitations may also be explored 
. We would also welcome if more journals followed the Annals of Internal Medicine's rule that the abstract should mention at least one limitation. In the meantime perhaps the National Library of Medicine may consider adding limitations to the abstracts in PubMed.
Replication of this work is needed. Our work may be improved by calculating hedging scores per section. Sensitivity analyses with the hedging-weights may shed light on the role of the weights assignment in the hedging software. Some of the ideas we suggested above may be carried out as research projects at the editorial offices. For example, we may try to assess if more structured approaches are implementable, what the challenges involved are with such approaches, and what effects can be measured after implementation on for example transparent reporting, time investment needed, authors avoiding journals with a strict approach in place, and effects on journals' impact factors. Long-term effects may involve the quality of clinical practice guidelines and patient care and its outcomes. And finally, larger studies may try to replicate and investigate in depth how by-journal hedging score variation is related to editorial policies.
In conclusion, our data show that reporting of limitations to original biomedical research is probably incomplete. Either directly, through clinical decision-making by evidence-based clinicians or indirectly through its effects on systematic reviews and clinical guidelines, optimal patient care may be jeopardized. Finally, scientific progress may be slowed down: reporting limitations more completely would aid the design and implementation of future studies. An appropriate amount of hedging given these limitations could further guide future scientific inquiry.