RxMap: an LLM-assisted tool for medication normalization.

RxMap: an LLM-assisted tool for medication normalization - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice JAMIA Open . 2026 May 30;9(3):ooag085. doi: 10.1093/jamiaopen/ooag085 RxMap: an LLM-assisted tool for medication normalization Eero Korpela 1 , Leah H Rubin 2, 3, 4, 5 , Raha M Dastgheyb 6 , Yanxun Xu 7, 8, ✉ Author information Article notes Copyright and License information 1 Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD, United States 2 Departments of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, United States 3 Department of Psychiatry and Behavioral Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, United States 4 Department of Molecular and Comparative Pathobiology, Johns Hopkins University School of Medicine, Baltimore, MD, United States 5 Department of Epidemiology, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD, United States 6 Departments of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, United States 7 Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD, United States 8 Division of Quantitative Sciences, Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University School of Medicine, Baltimore, MD, United States ✉ Corresponding author: Yanxun Xu, PhD, Department of Applied Mathematics and Statistics, Johns Hopkins University Whiting School of Engineering, 3400 North Charles Street, Wyman N429, Baltimore, MD 21218, United States ( yanxun.xu@jhu.edu ) Roles Eero Korpela : MSE , Conceptualization, Data curation, Formal analysis, Validation, Visualization, Writing - original draft, Writing - review & editing Leah H Rubin : PhD , Conceptualization, Resources, Writing - review & editing Raha M Dastgheyb : PhD , Visualization, Writing - review & editing Yanxun Xu : PhD , Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Validation, Visualization, Writing - original draft, Writing - review & editing Received 2026 Feb 23; Revised 2026 Apr 1; Accepted 2026 May 12; Collection date 2026 Jun. © The Author(s) 2026. Published by Oxford University Press on behalf of the American Medical Informatics Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. PMC Copyright notice PMCID: PMC13221971  PMID: 42220340 Abstract Objective To develop a freely available, researcher-oriented system for accurate normalization of free-text medication strings to standardized RxNorm ingredient-level concepts. Methods RxMap implements a fully automated normalization pipeline that combines deterministic RxNorm candidate generation with large language model (LLM)–assisted lexical parsing and hierarchical ingredient-level reconciliation. Raw medication strings are normalized to RxNorm ingredient-level (IN/MIN) concepts. ATC codes are assigned post-normalization, and an optional review interface supports transparent human review. Results Evaluation on 22 624 unique medication strings from the IPUMS MEPS dataset demonstrated substantial improvements over deterministic RxNorm matching alone. The best-performing RxMap configuration improved RxCUI-level precision, recall, and F1-score from 0.865, 0.853, and 0.859 to 0.969, 0.964, and 0.966, respectively, with similar gains at the ingredient level. Conclusion RxMap provides accurate, scalable normalization of free-text medication data to RxNorm concepts. The system operates fully automatically, offers optional review for quality control, and enables reproducible batch processing for research workflows. Keywords: medical concept normalization, RxNorm, large language models, data harmonization Introduction The normalization of medication data to standardized drug terminologies is foundational for transforming heterogeneous healthcare data into analyzable and interoperable resources. 1 Without normalization, medication information recorded across disparate sources cannot be reliably aggregated, compared, or analyzed at scale. In real-world settings, including electronic health records, surveys, dispensing systems, and clinical databases, medication information is recorded using highly heterogeneous conventions reflecting local practices and data collection contexts. 2–5 These entries frequently combine brand and generic names, abbreviations, misspellings, truncated strings, and inconsistent expressions of strength, units, and multi-ingredient formulations. Aligning such inputs to standardized drug concepts is essential for reproducible research, cross-study comparison, and downstream analytics. Despite this need, medication normalization in many research workflows remains partially manual, making it slow, difficult to scale, and error-prone. 6 , 7 Standardized terminologies such as RxNorm and the Anatomical Therapeutic Chemical (ATC) classification system provide the semantic backbone for addressing this challenge. RxNorm offers a comprehensive, ingredient-centric vocabulary widely used in the United States to represent clinical drugs and their relationships, enabling normalization of heterogeneous medication strings to a common identifier space. 8 ATC annotates medications into therapeutic and pharmacologic classes, supporting downstream analyses such as drug utilization, similarity, and class-level comparisons. 9 Together, these resources enable medication data to be represented in consistent and analytically useful forms. However, the practical task of mapping messy, real-world medication strings to these standards remains difficult. This work focuses on the task of medication normalization, defined as mapping free-text medication name strings to standardized drug concepts. We assume medication strings are already available from sources such as surveys or dispensing records and seek to normalize these inputs to RxNorm ingredient-level (IN) or multi-ingredient (MIN) concepts. Where relevant for downstream analysis, normalized concepts are further annotated with ATC classes, with explicit handling of complex combination products. Existing computational approaches to medication normalization include lexicon- and rule-based matching against RxNorm, fuzzy string retrieval, and embedding-based methods that retrieve and rank candidate concepts. 10–12 More recently, large language models (LLMs) have been explored for abbreviation expansion and candidate generation from free-text inputs. 13 , 14 While these approaches improve coverage, several challenges persist for messy, non-clinical medication strings: brittleness to misspellings and truncated forms; difficulty selecting among many plausible candidates with overlapping attributes; uneven support for multi-ingredient formulations; and limited integration of ATC annotation within open, reproducible workflows. Widely used clinical NLP toolkits (eg, MedXN, MedEx, cTAKES) perform well at extracting medications from clinical narratives but are not designed for robust normalization, and comprehensive ATC integration remains uncommon. 11 , 15–17 To address these gaps, we present RxMap, a free, researcher-oriented web application and pipeline for medication normalization with optional ATC annotation ( https://www.rx-map.com/ ). RxMap combines deterministic RxNorm candidate generation with constrained LLM-assisted lexical parsing and resolves mappings at the RxNorm IN and MIN levels. The system explicitly supports complex combination products and integrates a RxNorm-to-ATC crosswalk to enable class-level analyses. An interactive review interface provides optional inspection and correction of automatically generated mappings and supports reproducible batch-processing workflows. Unlike proprietary normalization services embedded within specific electronic health record ecosystems, 18 RxMap is designed as an open, standalone tool for research use. Methods Overview RxMap implements a fully automated pipeline for normalizing raw medication strings to standardized RxNorm ingredient and multi-ingredient level concepts (IN/MIN), represented by RxNorm concept unique identifiers (RxCUIs). The pipeline comprises three stages: (i) input preprocessing, (ii) ingredient-level normalization via deterministic RxNorm matching augmented with constrained LLM-assisted parsing and hierarchical candidate selection, and (iii) downstream pharmacologic annotation using ATC classifications ( Figure 1 ). Figure 1. Open in a new tab Overview of RxMap pipeline. The pipeline integrates preprocessing, deterministic and LLM-assisted candidate generation, hierarchical candidate selection to RxNorm concepts, and automated ATC code annotation. Cached mappings and a stored database support reproducibility, while a frontend review interface allows for manual review and resolves ambiguous cases before producing finalized mappings for export. Input preprocessing Raw medication strings undergo minimal preprocessing to inputs for consistent querying and caching, including removal of parenthetical content, whitespace normalization, and uppercasing. No spelling correction, abbreviation expansion, or drug-specific heuristics are applied, preserving the original semantics for downstream matching. Hybrid normalization For each preprocessed medication string, RxMap generates candidate ingredient-level representations from two sources and reconciles them using a hierarchical selection strategy. A deterministic candidate is obtained via RxNorm’s approximate string-matching algorithm (getApproximateMatch function as described in Peters et al. 10 ) which performs fuzzy lexical matching based on Jaccard similarity scores between normalized token sets, and is then resolved by default paths between related concepts within RxNorm to an active generic concept (IN or MIN). After deterministic mapping, an LLM parses the raw string to produce a standardized list of generic active ingredient names. When a deterministic candidate is available, it is provided in the LLM prompt as supplementary context; however, the LLM is explicitly instructed to prioritize the raw medication string and to disregard the deterministic candidate if inconsistencies arise. These two sources provide complementary candidate ingredient sets. Further details on deterministic and LLM-assisted candidate generation are provided in Supplementary Materials S2 . Candidate ingredient sets are reconciled by first accepting exact agreement when available. We denote this perfect match between ingredient sets as a high confidence match. When lexical discrepancies occur between the two ingredient sets (eg, {ACETYLSALICYLIC ACID, CODEINE} vs {ASPIRIN, CODEINE PHOSPHATE}), both sets are normalized to canonical RxNorm ingredient names by independently applying the deterministic mapping procedure described above to each ingredient string. After normalization, both sets resolve to the same canonical representation (eg, {ASPIRIN, CODEINE}), and the sets are re-compared for exact equivalence. Equivalent sets after normalization constitute a normalized perfect match and are labeled as moderate confidence within RxMap. If agreement is still not achieved, subset matching is applied to resolve combination products in a structure-preserving manner. Specifically, the LLM-derived ingredient set is decomposed into candidate subsets evaluated in descending order of size. Each subset is deterministically matched to valid IN/MIN concepts when available. When a valid subset with exact ingredient match is identified, it is selected as a finalized mapping and its constituent ingredients are removed from the remaining set. This process continues iteratively until no further valid subsets can be identified. The final output may therefore consist of multiple RxNorm concepts (eg, a multi-ingredient combination plus one or more single ingredients) that collectively cover the LLM-derived ingredient set. Because subset matching operates exclusively on the LLM-generated set without deterministic reconciliation, these matches are designated as low confidence. Examples for the three mapping outcomes (high, moderate, and low confidence) are shown in Figure 2 . Figure 2. Open in a new tab Hierarchical candidate selection logic for three medication concept normalization pathways in RxMap: Perfect Match (high confidence match), Normalized Perfect Match (moderate confidence match), and Subset Match (low confidence match). As defined above, mapping confidence is a structured, rule-based categorization based on match type intended to support user interpretability and workflow efficiency in RxMap, rather than a computed probabilistic score. Empirical verification of these categories is provided in Supplementary Materials S2.2.1 . ATC classification (post-normalization) ATC classification is applied as a post-normalization annotation step on finalized RxNorm concepts. ATC codes are preferentially obtained directly from RxNorm when available; otherwise, RxClass is queried. 19 Assignments from RxClass are accepted only when the linked concepts share identical active ingredient sets with the normalized RxNorm concept, ensuring pharmacologic consistency. When combination-level ATC codes are unavailable, ATC annotations may be provided for classifiable subsets and/or individual ingredients and labeled as partial coverage. Implementation details of ATC assignment, ingredient-level validation, and partial coverage handling are provided in Supplementary Materials S3 . Batch processing, review interface, and reproducibility RxMap supports single-string queries and batch CSV uploads. In batch mode, each input medication string is represented as a single row with normalized RxNorm concepts displayed as structured annotations. Users can filter by mapping confidence and optionally edit automatically generated mappings prior to export. The system queries the live RxNorm API, uses deterministic decoding for LLM-assisted parsing, and applies caching to support reproducibility. Additional details on the user interface, batch review workflows, and session persistence are provided in Supplementary Materials S4 , and system architecture, caching, and reproducibility safeguards are described in Supplementary Materials S1 and S5 . Results We evaluated RxMap using medication data from two datasets: IPUM MEPS 20 and a harmonized dataset denoted CHARTER/NNTC/HNRP. 21–23 The datasets were comprised of 22,624 and 1,363 unique raw medication strings, respectively. Each dataset was mapped to establish a gold-standard RxCUI and corresponding ingredient set. Dataset details are provided in Supplementary Materials S6 . Table 1 provides randomly sampled examples with gold-standard mappings to illustrate the diversity and difficulty of the normalization task. Table 1. Representative examples of free-text medication strings from the IPUMS MEPS and CHARTER/NNTC/HNRP datasets and gold-standard RxNorm mapping. Dataset RXNAME Mapped medication string Mapped RxCUIs IPUMS MEPS PEDIOTIC HYDROCORTISONE/NEOMYCIN/POLYMYXIN B 217627 DIPHEN-ATROP DIPHENHYDRAMINE | ATROPINE 3498 | 1223 LO/OVRAL 28 ETHINYL ESTRADIOL/LEVONORGESTREL 214558 DDAVP NASAL SPRAY*5ML DESMOPRESSIN 3251 LOESTRIN FE 1/20 ETHINYL ESTRADIOL/NORETHINDRONE | FERROUS FUMARATE 384410 | 24941 GANI-TUSS-DM NR (A.F., S.F., RASPBERRY) DEXTROMETHORPHAN/GUAIFENESIN 214488 POTASS. CL 10MEQ @@ POTASSIUM CHLORIDE 8591 PPA/PYRIL/P-TLOX/PHENIR PHENIRAMINE/PHENYLPROPANOLAMINE/PHENYLTOLOXAMINE/PYRILAMINE 689741 ACETAM CODEINE #3 ACETAMINOPHEN/CODEINE 817579 RYNATAN/R-TENNATE CHLORPHENIRAMINE/PHENYLEPHRINE/PYRILAMINE 214416 CHARTER/NNTC/HNRP ADVIL PM DIPHENHYDRAMINE/IBUPROFEN 644895 ACETAMINOPHEN; COMB.; DEXTROMETHORPHAN HBR; DOXYLAMINE SUCCINATE; PSEUDOEPHEDRINE HCL ACETAMINOPHEN/DEXTROMETHORPHAN/DOXYLAMINE/PSEUDOEPHEDRINE 466524 NEOMYC/POLYMY/BACITR/LIDO BACITRACIN/LIDOCAINE/NEOMYCIN/POLYMYXIN B 1007611 PENICILLIN G, CLEMIZOL CLEMIZOLPENICILLIN/PENICILLIN G 1007623 GS-5885 (NS5A INHIBITOR) LEDIPASVIR 1591922 DIPHENHYDRAMINE HCL; COMB.; TETRACYCLINE (EENT) DIPHENHYDRAMINE | TETRACYCLINE 3498 | 10395 UNIVASC MOEXIPRIL 30131 PREMPRO ESTROGENS, CONJUGATED (USP)/MEDROXYPROGESTERONE 1006917 MORPHINE HYDROCHLORIDE MORPHINE 7052 PYLERA BISMUTH SUBCITRATE/METRONIDAZOLE/TETRACYCLINE 1007769 Open in a new tab Examples illustrate common challenges including brand/generic variation, abbreviations, misspellings, truncation, and multi-ingredient products. For combination drugs, the slash (/) denotes a multi-ingredient RxNorm concept, while the pipe symbol (|) separates distinct drug concepts when a single input string maps to multiple RxCUIs We assessed the performance of RxMap against two deterministic baselines and examined the effect of two different LLMs backends within the same framework. The four pipelines were: (1) RxNorm-OHDSI, a deterministic baseline developed for the Canada Vigilance database; 24 (2) RxNorm+, the deterministic matching component of RxMap; (3) RxMap (GPT-5-Nano), the full pipeline using OpenAI’s gpt-5-nano for candidate generation; and (4) RxMap (Gemini-2.0-Flash), the same pipeline using Google’s gemini-2.0-flash. Performance was measured using precision, recall, and F1 at both the RxCUI level (exact match) and ingredient level (constituent identification). Ingredient-level evaluation expands predicted and reference RxCUIs into ingredient sets, enabling partial correctness for multi-ingredient products. Table 2 presents the comparative performance metrics. The results for the IPUMS dataset demonstrate that the hybrid RxMap pipelines substantially outperform the deterministic baselines in both exact concept matching and ingredient identification. The Gemini-2.0-Flash variant achieved the highest overall scores, with an F1 of 0.966 at the RxCUI level and 0.972 at the ingredient level, representing a major improvement over the RxNorm+ baseline (F1 of 0.859 and 0.874, respectively). The baseline RxNorm-OHDSI pipeline achieved limited recall, primarily due to its inability to robustly map multi-ingredient formulations. RxMap’s standalone deterministic component (RxNorm+) captured a substantial portion of well-formatted entries but frequently failed on inputs with misspellings (eg, “BENZONATE”), truncations (eg, “SULFAMET”), or brand name drugs (eg, “LO-OVRAL-28”). In contrast, the proposed hybrid pipeline addressed these gaps: the LLM component successfully parsed and normalized such challenging strings, and the backend logic preserved the structure of combination products. Performance across the two LLM variants was comparable, with Gemini-2.0-Flash exhibiting modestly better performance, suggesting that RxMap’s performance is robust to the choice of underlying LLM. Similar results for CHARTER/NNTC/HNRP were observed. Table 2. Performance comparison of medication normalization methods on IPUMS MEPS and harmonized CHARTER/NNTC/HNRP dataset. RxCUI Ingredient Dataset Performance Metrics RxNorm-OHDSI RxNorm+ RxMap (gpt-5-nano) RxMap (gemini-2.0-flash) RxNorm-OHDSI RxNorm+ RxMap (gpt-5-nano) RxMap (gemini-2.0-flash) IPUMS MEPS F1 a 0.240 0.859 0.906 0.966 0.300 0.832 0.919 0.972 Precision b 0.244 0.865 0.911 0.969 0.288 0.853 0.929 0.976 Recall c 0.236 0.853 0.900 0.964 0.313 0.812 0.909 0.967 CHARTER/NNTC/HNRP F1 0.720 0.896 0.919 0.956 0.877 0.898 0.925 0.953 Precision 0.721 0.896 0.942 0.962 0.890 0.901 0.960 0.966 Recall 0.718 0.895 0.898 0.950 0.865 0.895 0.892 0.941 Open in a new tab a Harmonic mean of precision and recall. b Proportion of predicted RxCUI/expanded ingredient mappings that are correct. c Proportion of gold-standard RxCUI/expanded ingredient mappings correctly recovered. Metrics are reported at the RxCUI level (exact concept match) and the ingredient level (constituent ingredient match) for two deterministic baselines and two variants of the proposed RxMap hybrid pipeline. Discussion This paper presents RxMap, a researcher-oriented system for medication normalization that achieves high accuracy on noisy, real-world medication strings. Across large, heterogeneous evaluation sets, RxMap substantially outperforms deterministic RxNorm matching at both the RxCUI and ingredient levels, demonstrating practical gains in coverage without sacrificing precision. The observed gains stem from two design choices addressing common medication normalization failures. First, constrained LLM parsing recovers signal from misspellings, abbreviations, truncated brand names, and pharmacy-specific notations that often defeat deterministic methods. Second, hierarchical, ingredient-preserving selection prevents premature collapse of combination products, improving accuracy for multi-ingredient medications common in survey and dispensing data. Together, these components support reliable normalization in realistic settings. Analysis revealed that most RxMap errors occur during subset matching. Common patterns include resolution to generic ingredient forms rather than specific chemical variants (eg, vitamin D rather than cholecalciferol), mapping herbal supplements to active compounds rather than botanical sources, and selection of incorrect salt forms in branded products. These errors reflect a trade-off: LLM generalization improves robustness to misspellings but can collapse fine-grained chemical distinctions. Low-confidence labeling for subset matches enables users to prioritize review when chemical specificity is critical. Importantly, RxMap is automation-first: normalization and ATC annotation execute fully automatically at scale, while the review interface is optional and intended for quality control in low-confidence edge cases. This design supports both high-throughput batch processing and targeted auditability when needed, aligning with practical research use where reproducibility and traceability are essential. From a practical standpoint, RxMap introduces modest computational overhead relative to deterministic methods due to LLM inference. In our evaluation, processing averaged ∼3.7 seconds per medication string at ∼$0.0002 per mapping (Gemini-2.0-Flash), supporting feasibility for large-scale batch use. Medication-level caching avoids repeated LLM calls for recurring strings, substantially reducing runtime and cost while preserving reproducibility. Several limitations warrant further work. Performance depends partly on the underlying LLM, though deterministic validation and caching improve stability. Coverage is limited by RxNorm, leaving rare or newly introduced products challenging. ATC mappings are not always one-to-one, especially for combination therapies, motivating subset-level annotation. The system also focuses on English, U.S.-centric inputs; extending to additional languages, terminologies, and datasets is needed for broader generalizability. A natural extension of this work is integration with pipelines that extract medical concepts from unstructured clinical text. Such integration would enable an end-to-end processing pipeline for transforming unstructured clinical narratives into standardized, analyzable medication data suitable for pharmacoepidemiologic research and clinical decision support applications. Supplementary Material ooag085_Supplementary_Data ooag085_supplementary_data.zip (34.7MB, zip) Acknowledgements The authors gratefully acknowledge the contributions of the study participants for all studies (CHARTER, NNTC, HNRP) and dedication of the staff at the sites. Contributor Information Eero Korpela, Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD, United States. Leah H Rubin, Departments of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, United States; Department of Psychiatry and Behavioral Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, United States; Department of Molecular and Comparative Pathobiology, Johns Hopkins University School of Medicine, Baltimore, MD, United States; Department of Epidemiology, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD, United States. Raha M Dastgheyb, Departments of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, United States. Yanxun Xu, Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD, United States; Division of Quantitative Sciences, Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University School of Medicine, Baltimore, MD, United States. Author contributions Eero KORPELA (Conceptualization, Data curation, Formal analysis, Validation, Visualization, Writing—original draft, Writing—review & editing), Leah H. RUBIN (Conceptualization, Funding acquisition, Writing—review & editing), Raha M. DASTGHEYB (Conceptualization, Visualization, Writing—review & editing), Yanxun XU (Conceptualization, Funding acquisition, Data curation, Formal analysis, Validation, Visualization, Writing—original draft, Writing—review & editing) Supplementary material Supplementary material is available at [JAMIA Open] online. Conflicts of interest None declared. Funding This work was in part supported by the National Institute of Mental Health (R01MH128085, Xu), National Institute of Allergy and Infectious Diseases (R01AI197147, MPI: Xu, Rubin, Ni). Software availability RxMap application can be accessed from https://www.rx-map.com/ . Data availability All datasets analyzed in this paper are publicly accessible upon request from https://meps.ipums.org/meps/ and https://hnrp.hivresearch.ucsd.edu/ . References 1. Chen H, Li R, Cleveland A, et al.  Enhancing data quality in medical concept normalization through large language models. J Biomed Inform. 2025;165:104812. 10.1016/j.jbi.2025.104812 [ DOI ] [ PubMed ] [ Google Scholar ] 2. Peters L, Kapusnik-Uner JE, Bodenreider O.  Methods for managing variation in clinical drug names. AMIA Annu Symp Proc. 2010;2010:637-641. [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Le H, Chen R, Harris S, et al.  RxNorm for drug name normalization: a case study of prescription opioids in the FDA adverse events reporting system. Front Bioinform. 2023;3:1328613. 10.3389/fbinf.2023.1328613 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Newman-Griffis D, Divita G, Desmet B, et al.  Ambiguity in medical concept normalization: an analysis of types and coverage in electronic health record datasets. J Am Med Inform Assoc. 2021;28:516-532. 10.1093/jamia/ocaa269 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Lagerlund O, Strese S, Fladvad M, et al.  WHODrug: a global, validated and updated dictionary for medicinal information. Ther Innov Regul Sci. 2020;54:1116-1122. 10.1007/s43441-020-00130-6 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Ghosh R, Kempf D, Pufko A, et al.  Automation opportunities in pharmacovigilance: an industry survey. Pharmaceut Med. 2020;34:7-18. 10.1007/s40290-019-00320-0 [ DOI ] [ PubMed ] [ Google Scholar ] 7. Harpaz R, Callahan A, Tamang S, et al.  Text mining for adverse drug events: the promise, challenges, and state of the art. Drug Saf. 2014;37:777-790. 10.1007/s40264-014-0218-z [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. RxNorm API - APIs. Accessed January 8, 2025. https://lhncbc.nlm.nih.gov/RxNav/APIs/RxNormAPIs.html 9. ATCDDD - Home. Accessed January 6, 2026. https://atcddd.fhi.no/ 10. Peters L, Kapusnik-Uner JE, Nguyen T, et al.  An approximate matching method for clinical drug names. AMIA Annu Symp Proc. 2011;2011:1117-1126. [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Weeks HL, Beck C, McNeer E, et al.  medExtractR: a targeted, customizable approach to medication extraction from electronic health records. J Am Med Inform Assoc. 2020;27:407-418. 10.1093/jamia/ocz207 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Luo Y-F, Henry S, Wang Y, et al.  The 2019 n2c2/UMass lowell shared task on clinical concept normalization. J Am Med Inform Assoc JAMIA. 2020;27:1529-15e1. 10.1093/jamia/ocaa106 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Hüser M, Doole J, Pinho V, et al.  A machine learning approach for automating review of a RxNorm medication mapping pipeline output. J Biomed Inform. 2025;170:104909. 10.1016/j.jbi.2025.104909 [ DOI ] [ PubMed ] [ Google Scholar ] 14. Matos J, Gallifant J, Pei J, et al.  EHRmonize: a framework for medical concept abstraction from electronic health records using large language models. In: Wu S, Shabestari B, Xing L, eds. Applications of Medical Artificial Intelligence. Cham: Springer Nature Switzerland; 2025:210-20. [ Google Scholar ] 15. Sohn S, Clark C, Halgrim SR, et al.  MedXN: an open source medication extraction and normalization tool for clinical text. J Am Med Inform Assoc. 2014;21:858-865. 10.1136/amiajnl-2013-002190 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Xu H, Stenner SP, Doan S, et al.  MedEx: a medication information extraction system for clinical narratives. J Am Med Inform Assoc. 2010;17:19-24. 10.1197/jamia.M3378 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Savova GK, Masanz JJ, Ogren PV, et al.  Mayo clinical text analysis and knowledge extraction system (cTAKES): architecture, component evaluation and applications. J Am Med Inform Assoc. 2010;17:507-513. 10.1136/jamia.2009.001560 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Meldau E-L, Bista S, Rofors E, et al.  Automated drug coding using artificial intelligence: an evaluation of WHODrug koda on adverse event reports. Drug Saf. 2022;45:549-561. 10.1007/s40264-022-01162-7 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. RxClass API - APIs. Accessed January 9, 2026. https://lhncbc.nlm.nih.gov/RxNav/APIs/RxClassAPIs.html 20. Ruggles S, Flood S, Sobek M, et al. IPUMS USA: Version 16.0. 2025. 21. Heaton RK, Clifford DB, Franklin DR, et al.  HIV-associated neurocognitive disorders persist in the era of potent antiretroviral therapy: CHARTER study. Neurology. 2010;75:2087-2096. 10.1212/WNL.0b013e318200d727 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Welcome to the HNRP. Accessed January 8, 2025. https://hnrp.hivresearch.ucsd.edu/ 23. Morgello S, Gelman BB, Kozlowski PB, et al.  The national NeuroAIDS tissue consortium: a new paradigm in brain banking with an emphasis on infectious disease. Neuropathol Appl Neurobiol. 2001;27:326-335. 10.1046/j.0305-1846.2001.00334.x [ DOI ] [ PubMed ] [ Google Scholar ] 24. Chalabianloo N, Abdullah SS, Omrani MA, et al.  Enhancing adverse drug reaction data quality in Canada: a high-precision pipeline for medication name standardization and enrichment. Plos One,. 2025; 20: e0331940. 10.1371/journal.pone.0331940 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials ooag085_Supplementary_Data ooag085_supplementary_data.zip (34.7MB, zip) Data Availability Statement All datasets analyzed in this paper are publicly accessible upon request from https://meps.ipums.org/meps/ and https://hnrp.hivresearch.ucsd.edu/ . Articles from JAMIA Open are provided here courtesy of Oxford University Press ACTIONS View on publisher site PDF (798.6 KB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Optimizing Antibiotics Prescription#201808
• Computational intelligence approaches for classification of medical data: State-of-the-art, future challenges and research directions#348468
• Brief introduction of medical database and data mining technology in big data era#400290
• From algorithm to applications: Artificial intelligence – A future prospective in medicine#468193
• Charting a course for genomic medicine from base pairs to bedside#335741