Reconstructing and Evaluating Antibiotic Treatment Episodes for Antimicrobial Stewardship Using the MAPS-Toolkit
The paper introduces MAPS, an open-source Python toolkit that reconstructs and evaluates antibiotic treatment episodes from minimal electronic health record data to enable scalable, episode-based antimicrobial stewardship monitoring that captures critical clinical decisions like stopping, switching, and narrowing therapy, which are invisible to traditional aggregate metrics.
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
Hospitals are the front lines in the battle against superbugs, infections that no longer respond to standard medicines. To win this fight, doctors must use antibiotics with extreme precision: giving them only when necessary, choosing the right drug, and stopping the treatment as soon as it is safe. This careful management is called antimicrobial stewardship. The goal is to prevent bacteria from learning how to survive these drugs, a process that happens when antibiotics are used too long or too broadly. For years, hospitals have tracked their antibiotic use by counting pills or measuring total days of treatment. While these numbers tell a hospital how much medicine it is using, they miss the most important part of the story: the decisions doctors make as a patient gets better. They cannot see if a doctor switched a patient from a strong intravenous drug to a milder pill, or if they stopped the medicine early because the infection cleared up. Without seeing these specific choices, it is hard to know if a hospital is truly managing its antibiotics well.
A team of researchers at Amsterdam University Medical Centers has built a new digital tool to solve this problem. They created a software package called MAPS, which stands for Modeling Antibiotic Prescriptions for Stewardship. Instead of just counting doses, this tool looks at the sequence of prescriptions a patient receives and stitches them together into a single "episode" of care. Imagine a patient who starts with an intravenous antibiotic for two days, then switches to a pill for three more days. Old counting methods would see two separate treatments. MAPS sees one continuous story, allowing researchers to spot exactly when the switch happened and how long the total treatment lasted. The software is designed to be flexible; it can take raw data from almost any hospital computer system and organize it into these clear stories without needing complex, custom coding for each new location.
To test if this approach worked, the researchers applied MAPS to two very different sets of medical records. The first came from their own hospital in the Netherlands, covering over 150,000 antibiotic prescriptions given to patients between 2020 and 2024. The second came from a massive public database of intensive care patients in the United States, containing over 550,000 prescriptions. By running the software on both, they generated more than 250,000 distinct treatment episodes. This allowed them to look closely at three critical decisions that doctors make: when to stop a treatment, when to switch from a vein to a pill, and when to narrow the scope of the drug to target a specific bug.
The results revealed that these decisions vary wildly depending on which department is treating the patient. In the Dutch hospital, the median length of an antibiotic treatment was about three days, but this changed significantly by specialty. Emergency medicine teams tended to stop treatments very quickly, often within a single day, while cardiac surgery and intensive care units kept patients on antibiotics for longer, with median durations around four days. In the American database, the patterns were similar but with different timing; the median time to switch from an intravenous line to an oral pill was about three days, though some departments switched much faster than others. The researchers also tracked how often doctors reduced the strength of the antibiotic once they knew more about the infection. This "de-escalation" happened in anywhere from 4% to nearly 30% of cases, depending entirely on the medical specialty involved.
These variations do not necessarily mean some doctors are doing a better job than others. The researchers emphasize that these numbers reflect the different types of patients each department sees and the unique workflows they follow. A patient in intensive care is often much sicker than one in a general ward, requiring longer and more complex treatment. The value of the MAPS tool is that it makes these decision patterns visible for the first time on a large scale. Before this, such detailed information was hidden inside the medical records, accessible only if a human reviewer manually read through thousands of charts. Now, hospitals can see exactly when and how often they are stopping, switching, or narrowing treatments.
The study confirms that it is possible to reconstruct these detailed treatment stories using only the basic prescription data that hospitals already collect every day. The software successfully identified over 250,000 episodes from more than 700,000 prescriptions across 129,000 patients. It found that while the overall volume of antibiotic use is important, the quality of care lies in the transitions between drugs. By turning raw data into clear, sequential episodes, the tool provides a foundation for hospitals to monitor their performance continuously rather than relying on occasional, manual reviews. This shift allows stewardship teams to spot trends, such as a specific department that consistently delays switching to pills, and offer timely feedback to improve care. Ultimately, the work suggests that with the right digital tools, hospitals can move beyond simple counting to a deeper understanding of how antibiotics are actually used, helping to slow the rise of drug-resistant infections.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.