Writing a publication summary and choosing its format
A plain language summary of a publication is written for a reader without specialist knowledge, in a format chosen for that reader, and an AI tool can draft it under human review.
Writing the summary
Understanding exactly what the public wants to learn, and why, is vital to drafting an effective summary or any other content meant for the public.
One communicator advises writing a summary of a publication about five levels of education below its readers' own. It should be written for an early adolescent, because short sentences in simple language are understood by a lot of people. Jargon should be avoided, with plain alternatives used wherever possible, and a term that cannot be avoided should be explained before it is used, in the opening sentences. Even a stand-alone summary should contain no more information than the main manuscript, though it should give some context so that a non-specialist can fully understand its messages. So it is important that the main manuscript is written well too.
Many medical writers are trained to write complex documents for specialist audiences, but writing for a non-specialist audience requires far more than translating difficult vocabulary into simpler terms: it needs a completely different skill set, which takes training and practice. Medical writers are uniquely placed to bring data to life and to help non-specialists visualise them and put them in context. Review teams, most of them far more used to reviewing highly complex documents for regulatory agencies, also have to know a non-specialist audience's needs and be skilled in meeting them for their review to be meaningful and helpful.
As of 2018, EMA guidelines required a plain language summary to be posted for every clinical study, and it could include patient-reported outcome data, but the EMA template was not particularly suited to explaining the technical and methodological aspects of studies of patient-reported outcomes. No regulatory guideline then mandated a summary for other types of patient-reported outcome study. So a publication of patient-reported outcomes should come with a specifically tailored summary that describes the study clearly and engagingly, including how the instrument works and what changes in its score mean. For a journal article, the summary can often be a peer-reviewed supplementary document associated with the article , and ideally it should be.
Formats
Summaries come as text only, from short text summaries that can be indexed alongside the abstract on PubMed to full-text plain language publications, as text with visuals, as infographics, and as multimedia such as videos, animations and podcasts. In a study of patients' format preferences, infographic-style summaries were the first choice in every patient group, followed by summaries of medium complexity with a reading age of 14 to 17. In one editor's experience, infographics are a fantastic way to get information across and should be encouraged, and for plain language summaries they are invaluable. A comparison of formats found video abstracts and plain language summaries better than original or graphical abstracts at being understood, at giving a feeling of understanding and at being enjoyed. Visual summaries, however, are more expensive and harder to produce, and search engines, which search text, have trouble finding them.
Drafting summaries with an AI tool
Springer Nature generated its plain language summaries of articles with the help of an AI tool while keeping people in the loop, because some parts of the work can be done by AI and some cannot. The tool produced the initial draft, simplified the medical terminology and the language, and handled pattern recognition and the consistency of the writing and the terminology. Whatever it generated then passed through people, who checked the content for accuracy and checked whether it could be understood, because the output often still carried technical jargon that the human experts had to simplify. AI cannot take responsibility for errors, so whatever needs correcting can be done only by people.
Once the tool had created a summary, Springer Nature's internal editors ran the initial fact check and quality evaluation, and the summary was then shared with the article's authors, who curated it and returned it to be published with the article. At first the output carried many mistakes, but by 2024 its mistakes were very minor, with some exaggerated terminology.
Springer Nature made its plain language summaries in two versions: one for healthcare professionals, technical readers and pharmaceutical companies, and one for patients and a general audience without knowledge of the technical terms. The general-public version used everyday language, visual explanations and outcomes relevant to patients. The technical version carried the clinical implications, the diagnostics and treatments, the statistical analysis and any practical guidelines, which may interest the general reader less. In 2024 Springer Nature was moving to open access, and an open-access article full of complex technical terms would be very difficult for a researcher, an early-career researcher, a patient or an editor to understand. So a plain language version was attached alongside the article, and a reader who wants to go further can then read the article itself. Springer Nature made these summaries only from articles that were open access or published within Springer Nature.
In one worked example, the original text reported that the study demonstrated a statistically significant reduction in HbA1c level, with a mean decrease of 1.8% in the treatment group, against the placebo group, a sentence very difficult for a non-medical reader to understand. The general-public version said instead that the new medicine helped people control their blood sugar, a term everyone can understand, and that their blood sugar levels improved about four times more than in those in the comparison group.
Springer Nature had first thought of plain language summaries as meant only for healthcare professionals and medical journals, but extended its pilot to two of its business journals, where the response was good and the authors liked the quality of the output. For Experimental and Molecular Medicine, it generated plain language versions of all the articles under the name of an editorial summary, and captured the authors' feedback on the quality of the output to improve the summaries through a reinforcement learning process. So far the authors were happy with the quality of the output.
Springer Nature sold plain language summaries to pharmaceutical companies , producing them from articles that a company selected and shared on certain diseases or clinical trial processes. Each summary went through two fact checks, first by Springer Nature's own medical writers and then by the company's medical writers, who checked the quality of the output before the company published the summary and distributed it to healthcare professionals along with its drugs.