Drug Safety Warnings in English, Russian, Arabic and Hebrew Wikipedia: A Cross-Sectional Audit of 33 Drugs Across Three Risk Domains
This cross-sectional audit of 33 drugs across English, Russian, Arabic, and Hebrew Wikipedia editions reveals that actionable drug-safety warnings are generally weaker in the non-English versions, with Russian articles often lacking content and Arabic articles frequently failing to escalate mentioned risks to explicit directives.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Millions of people turn to the internet when they need to understand a medication, and for many, the first place they look is Wikipedia. It is a massive, free encyclopedia written by volunteers, available in hundreds of languages. While doctors and pharmacists remain the primary source of medical advice, the sheer volume of people reading these online articles means that what is written there matters for public health. A critical piece of information for any medicine is its safety warning: a clear instruction telling a patient when they should not take a drug, such as if they have a specific genetic condition, if they are pregnant, or if the drug might cause a dangerous heart rhythm. For a warning to work, it must be present, it must be clear, and it must be written in the language the patient understands. If a warning exists in English but is missing or vague in another language, the safety net fails for those readers.
Researchers recently decided to test whether this safety net is equally strong across different languages. They focused on three specific types of serious drug risks: a severe blood reaction that can happen to people with a common inherited enzyme deficiency, harm to an unborn baby, and a dangerous heart rhythm problem. The team examined the Wikipedia articles for 33 different drugs that are known to carry these risks. They compared how these risks were described in the English version of Wikipedia against the Russian, Arabic, and Hebrew versions. To do this, they did not just read the articles themselves; they used advanced computer programs, known as large language models, to scan the text of every article. These programs were trained to identify three levels of information: whether the risk was not mentioned at all, whether it was mentioned but without a clear instruction, or whether the article gave an explicit directive telling readers to avoid the drug.
The study found a distinct pattern of inequality. In the vast majority of cases, the English articles provided the clearest, most direct warnings. When the researchers looked at the Russian, Arabic, and Hebrew articles, they found that these versions were significantly less likely to give readers a clear safety instruction. Out of nine possible comparisons between the three non-English languages and the three types of risks, eight showed that the non-English articles were weaker than the English ones. The English articles were more likely to say "do not take this drug" in a specific situation, whereas the other languages often just mentioned the risk without telling the reader what to do about it.
The reasons for this gap differed depending on the language. For the Russian articles, the problem was often that the articles simply did not exist or were too short to contain the necessary information. Once a Russian article did mention a risk, it was just as likely as an English article to turn that mention into a clear warning. However, the situation in Arabic and Hebrew was different. In these languages, the articles often did mention the risk, but they frequently stopped short of giving an explicit instruction to avoid the medicine. It was as if the article said, "this drug can cause this problem," but left out the crucial next sentence: "therefore, do not take it." This meant that even when the information was present, it was less actionable for the reader.
The researchers also looked at the underlying structure of Wikipedia to see if the problem could be fixed easily. They found that the database that powers Wikipedia, called Wikidata, has a specific field designed to record when a drug is forbidden for certain conditions. However, this field was almost completely empty; it had been used only a dozen times across the entire database, and none of the 33 drugs in the study had an entry there. This suggests that the gap in safety warnings is not just a matter of writing more text, but of failing to use the structured tools that could automatically generate these warnings across all languages.
The study concludes that while the English version of Wikipedia generally provides strong, clear safety instructions, readers using Russian, Arabic, or Hebrew are less likely to receive that same level of protection. The findings suggest that the problem is not just a lack of content in some languages, but a difference in how risks are communicated in others. As more people rely on artificial intelligence tools that pull information from Wikipedia to answer health questions, these gaps in the source material could lead to unequal safety outcomes for patients around the world. The researchers note that their study was limited to a small group of drugs and languages, so the results are not a final verdict on all of Wikipedia, but they do highlight a specific area where a simple, low-cost fix—filling in the missing structured data—could help close the gap.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.