Reverse-Engineering Radiation Oncology: A Reproducible Pipeline for Affiliation Disambiguation and Author Localisation in the Medical Sciences
This paper presents a reproducible, field-agnostic pipeline that improves the accuracy of author location and seniority assignment in radiation oncology by deriving data directly from raw affiliation strings rather than relying on error-prone institutional tags, achieving significantly higher precision (F1 ≈ 0.92) than standard baselines.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine trying to map the world's greatest scientists by looking at a giant, messy address book. This is the challenge of "bibliometrics," the science of measuring scientific research. For years, researchers have relied on digital databases that act like a librarian who has already sorted the books for you, attaching neat, pre-labeled tags to every author's name. These tags tell you where a scientist works and what they study. But here's the catch: sometimes the librarian makes mistakes. If a scientist belongs to a giant team that has offices in ten different cities, the librarian might stamp all ten cities on every scientist's file, even if they only work in one. Or, if two scientists have the same name, the librarian might accidentally glue their careers together, making one person look like they have a double life. This paper, written by David Kaul, dives into the world of radiation oncology—the branch of medicine that uses high-energy rays to fight cancer—to see if we can fix these messy maps. Instead of trusting the librarian's pre-sorted tags, the author suggests we go back to the original, raw address written on the paper itself and do the sorting ourselves, using a set of clever, transparent rules.
The paper tackles a specific puzzle: How do we accurately figure out where radiation oncologists in German-speaking countries (Germany, Austria, and parts of Switzerland) actually work, based on the thousands of research papers they publish between the years 2000 and 2025? The author argues that simply trusting the database's "institution tags" is like trying to navigate a city using a map where every street is labeled with the name of the country's capital. It leads to confusion. To solve this, Kaul built a "reproducible pipeline," which is essentially a step-by-step recipe for cleaning up the data. He started by gathering a massive collection of 17,270 articles. Instead of just asking the database for "radiation oncology" papers, he used two different search strategies at once: one looking for specific topics and another scanning eight major medical journals, then combining them to make sure no important papers were missed.
Once the papers were collected, the real magic happened. The author ignored the database's pre-made location tags and instead looked directly at the raw text of the affiliations—the messy strings of words authors type when they submit their papers. He built a digital filter to spot keywords that actually mean "radiation oncology" (like "strahlenther" or "radiooncolog") and blocked words that sounded similar but meant something else (like "urology" or "diagnostic radiology"). He also created a special "city dictionary" that knows the difference between a street named "Heidelberglaan" in Utrecht and the actual city of Heidelberg, Germany, ensuring the computer didn't get tricked by partial matches.
The most critical part of the recipe was a "consortium guard." In radiation oncology, many researchers belong to big networks like the German Cancer Consortium (DKTK). The database often stamps the headquarters of this network onto every single member's profile, making it look like everyone works at the main office. Kaul's pipeline had a rule to spot these network mentions and strip them away, keeping only the author's actual local department. He also added a "seniority gate" to help find the true leaders of the field. If a famous professor's paper didn't list their specific department, the system would check if that person was a known expert; if so, it would look for their city in other parts of the paper, but only if they weren't an expert in a different field like general surgery.
When the author tested this new method against a manually checked "gold standard" of 100 papers, the results were clear. The old, naive method of trusting the database tags got the location right only about 64% of the time. In contrast, the new raw-string pipeline got it right about 92% of the time. The paper explicitly rules out the idea that the pre-made database tags are safe to use for finding where scientists actually work, showing they are full of "phantom locations" caused by network over-attribution and name mix-ups. The author is confident that this approach works because it is transparent and reproducible, meaning anyone can run the same code and get the same results. However, the paper also admits it's not perfect; it might miss some papers published outside the eight core journals, and it assumes that the first city listed in a network string is the author's home base, which is usually true but not guaranteed. Ultimately, this work suggests that to truly understand the landscape of a medical field, we need to stop trusting the librarian's pre-sorted tags and start reading the raw addresses ourselves.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.