ATC/DDD discordance exposes where national drug lists lose fidelity
Drug utilization research depends on clean classification, but medication databases are rarely as tidy as they look on paper. In this analysis of the World Health Organization ATC/DDD dataset updated in July 2026, investigators focused on medicines with multiple ATC codes and on records where the route of administration and defined daily dose diverged. The aim was practical: identify where national drug lists can drift away from the underlying WHO reference structure and where harmonization efforts may erase clinically meaningful differences in formulation, route, or dosing standardization.
The dataset contained 2,737 active pharmaceutical ingredients with assigned defined daily doses. From that base, 466 unique molecules met at least one of eight classification conditions, generating 913 distinct scenarios across the five ATC hierarchical levels. A decision-tree screening scheme and Python-based data-cleaning workflow were used to sort these cases systematically rather than by ad hoc review. The result is not just a cleaned database; it is a map of where classification breaks apart and where it remains stable.
One finding stood out immediately: 107 scenarios changed classification across the five ATC levels, while 806 remained stable.
Eight-pattern screening identifies where structure, route, and dose diverge
The eight-pattern system was designed to catch the combinations most likely to produce information loss when national formularies are harmonized against WHO data. Across the classification pairs, changes occurred in four pairs: 1/5, 2/6, 3/7, and 4/8. Variability was greatest when both DDD and route of administration differed, a reminder that dose standards and delivery route are not interchangeable technical details. In drug policy terms, a tablet swallowed at home and an injectable delivered in a monitored setting may share an active ingredient yet behave very differently in utilization datasets, exposure assumptions, and downstream comparisons.
The classification stabilized beyond ATC level 3. That matters because the upper levels of ATC coding are where broad therapeutic groupings live, while lower levels progressively sharpen distinctions among chemical subgroups, substances, and final code assignments. Once harmonization reaches the more granular levels, a medicine can appear fully reconciled even when the pathway used to get there involved collapsing discrepancies in route or dose. The current framework is built to prevent exactly that kind of silent flattening.
National drug lists often inherit these mismatches when local formularies are aligned to international references without a structured audit. A medicine may carry one ATC code in one setting, several in another, and a DDD that shifts depending on how the route or formulation is encoded. For drug utilization research, those differences can alter prevalence estimates, comparative consumption metrics, and the interpretation of prescribing intensity. For health systems, the same differences can propagate into clinical decision support systems, procurement dashboards, and reimbursement logic.
Why route and DDD mismatches matter to drug policy and research
The clinical logic behind this work is straightforward. The ATC system is meant to classify medicines by anatomical, therapeutic, and chemical properties, while the DDD standardizes assumed average maintenance exposure for adults. When a product’s route of administration changes, bioavailability, monitoring requirements, and clinical context can change with it. When the DDD changes, utilization calculations shift as well, even if the underlying molecule remains the same. A harmonization method that ignores one of those dimensions risks misclassifying the drug and distorting the scale of use.
This is especially relevant for medicines used in multiple forms or settings, where one formulation may be driven by acute care and another by chronic outpatient use. The investigators’ screening scheme preserves those differences rather than forcing premature collapse into a single label. In operational terms, that allows national agencies to flag multiclassified medicines before a local list is treated as definitive. In analytic terms, it creates a reproducible audit trail for why a record was assigned one way rather than another.
The workflow was also built around reproducibility. Instead of relying on manual reconciliation, the authors used a Python-based cleaning pipeline and a decision-tree approach to classify records according to the eight possible patterns across the five ATC levels. That design matters because harmonization projects frequently fail not on clinical complexity, but on inconsistency: two analysts may resolve the same record differently if the rules are not explicit. Here, the classification logic itself is part of the deliverable.
A reproducible audit tool for multiclassified medicines
The article’s practical contribution lies in reducing information loss during national list harmonization. When differences in ATC codes, DDDs, and routes of administration are preserved rather than discarded, the resulting database is better suited for comparative utilization analyses and policy review. That is especially important for countries or institutions maintaining national drug lists that must remain aligned with WHO standards while still reflecting local use patterns.
The authors position the method as a screening mechanism, not a substitute for expert review. That is the right boundary. Automated harmonization can identify likely conflicts at scale, but final adjudication still belongs to a domain expert who understands formulation nuances, regulatory labeling, and local prescribing conventions. The value here is speed with traceability: a way to surface records that deserve inspection before they contaminate broader analyses.
The study also provides a reminder that database harmonization is itself a clinical quality issue. Inaccurate ATC/DDD mapping can affect drug utilization research, comparative pharmacoepidemiology, formulary planning, and the interpretation of consumption trends over time. If the coding substrate is unstable, downstream conclusions can become unstable as well. In that sense, the work is less about classification as an abstract informatics exercise and more about protecting the validity of the evidence built on top of it.
No funding supported the project, and no competing interests were declared. The work used publicly available WHO classification data and required no ethical approval. Its main limitation is inherent to the task it set itself: a harmonization scheme can expose discordance, but it cannot resolve the clinical reason behind every discordant record without context from the underlying formulation, market authorization, or national coding practice. The next step is not broader rhetoric about standardization; it is external testing of the eight-pattern framework against real national drug lists to see where it prevents avoidable coding error and where local policy still requires manual adjudication.