Markus Haug

Background Routinely collected health data can reveal recorded care around diagnosis, but disease-specific curation does not scale across thousands of conditions. We scaled a previously developed automated framework that reduces heterogeneous clinical events into standardised, diagnosis-centred summaries of real-world care patterns. Methods In this retrospective observational study, we used EST-Health-30, a pseudonymised 30% random sample of Estonian residents with health-care records from 2012 to 2024. We constructed cohorts for ICD-10 three-character diagnostic categories using the first eligible observed diagnosis after a 3-year diagnosis-free observable lookback. For each category, we summarised recorded clinical events from a 90-day pre-index window, a 30-day post-index window, and a 365-day post-index window. An enrichment-based workflow compared earlier self-comparator periods and matched population controls, then filtered and aggregated concepts across clinical domains. Diagnosis-concept relationships were assessed by rubric-guided language-model classification, and atlas plausibility and utility by ten experts. Findings Of 1645 observed diagnostic categories, 1080 met inclusion criteria, yielding 3240 diagnosis-window summaries across 509,856 individuals. Each summary characterised diagnosis-associated events by prevalence, enrichment, timing, co-occurrence, and variation between patient groups after reducing candidate concepts by 97-98%. Among 296,303 retained diagnosis-concept pairs, the language model classified 24% as directly related, 55% as indirectly related, and 21% as noisy. Experts rated the atlas highly for generating research questions (median 6.5 of 7 [IQR 6-7]), providing information difficult to obtain conventionally (6 [5-7]), and supporting observational study design (5.5 [5-6]). Interpretation The atlas makes multidimensional diagnosis-centred patterns in longitudinal health records inspectable at population scale in order to support clinical orientation, hypothesis generation, and observational study planning. Outputs represent recorded care rather than individual patient trajectories or causal effects. Funding Estonian Research Council; European Union; Estonian Ministry of Education and Research; Innovative Medicines Initiative 2 Joint Undertaking; EFPIA.

Background: Drug adherence is crucial for chronic disease management, yet treatment discontinuation remains common due to factors such as side effects, inefficacy, or cost. These reasons are often recorded only in free-text clinical notes, making large-scale analysis difficult. While large language models (LLMs) can interpret such unstructured data more effectively than traditional natural language processing methods, few studies have systematically categorized reasons for discontinuation or identified whether the decision was initiated by the patient or the clinician, especially in low-resource languages such as Estonian. Objective: This study aimed to assess the ability of LLMs to extract and classify reasons for drug discontinuation and identify who initiated it using Estonian electronic health records and characterize the observed discontinuation patterns and initiators for statins and antidiabetic medications. Methods: We combined prescription data with free-text anamneses from a 10% sample of the Estonian population (2012-2019). LLMs (Llama 3.1-70B and GPT-4o) were applied to extract discontinuation phrases and reasons, classify them into a clinician-developed taxonomy, and identify who discontinued the treatment. Performance was evaluated on 100 randomly chosen cases per drug group. Results: Extraction yielded 625 antidiabetic drug and 233 statin discontinuation cases. Validation confirmed a precision of 0.93 to 0.98 for extracting phrases and 0.95 to 0.96 for extracting reasons. Classification of discontinuation reasons achieved weighted F1-scores of 0.81 to 0.84, whereas classification of who initiated discontinuation achieved weighted F1-scores of 0.64 to 0.78. Adverse reactions were the most frequent reason overall, accounting for 70% (163/233) of statin discontinuations and 44.8% (280/625) of antidiabetic drug discontinuations. Regarding antidiabetic drugs, treatment inefficacy and contraindications were more common. Patients more often stopped due to adverse reactions or nonmedical reasons, whereas physicians more often initiated discontinuation for contraindications. Conclusions: LLMs can accurately extract and classify medication discontinuation reasons and show variable performance in identifying discontinuation initiators in Estonian clinical narratives. Both local and proprietary models showed promising results, enabling scalable analyses that complement structured health records. This demonstrates the potential of LLMs to unlock information from clinical notes, turning this underused electronic health record component into a valuable resource for monitoring treatment patterns and detecting adverse event signals.

Objective: To address the unresolved bottleneck of selecting cohort-relevant clinical concepts for treatment trajectory analysis in observational health data, we introduce CohortContrast, an OMOP-compatible R package for enrichment-based concept identification, temporal and semantic noise reduction, and concept aggregation, enabling cohort-level characterization and downstream trajectory analysis. Materials and Methods: We developed CohortContrast and applied it to OMOP-mapped observational data from the Estonian nationwide OPTIMA database, which includes all cases of lung, breast, and prostate cancer, focusing here on lung and prostate cancer cohorts. The workflow combines target-control statistical enrichment, temporal/global noise filtering, hierarchical concept aggregation and correlation-based merging, with optional patient clustering for downstream trajectory exploration. We validated the approach with a clinician-based plausibility assessment of extracted diagnosis-concept pairs and evaluated a large language model (LLM) as an auxiliary filtering step. Results: We analyzed 7,579 lung cancer and 11,547 prostate cancer patients. The workflow reduced concept dimensionality from 5,793 to 296 concepts (94.9%) in lung cancer and from 5,759 to 170 concepts (97.0%) in prostate cancer, and identified three exploratory patient subgroups in both cohorts. In a plausibility assessment of 466 diagnosis-concept pairs, validators rated 31.3% as directly linked and 57.5% as indirectly linked. Discussion: CohortContrast reduces manual concept curation by prioritizing and aggregating cohort-relevant concepts while preserving clinically interpretable treatment patterns in OMOP-based real-world data. Conclusion: CohortContrast enables scalable reduction of broad OMOP concept spaces into clinically interpretable, cohort-specific representations for exploratory trajectory analysis and real-world evidence research.

Identifies clinically relevant concepts in Observational Medical Outcomes Partnership Common Data Model cohorts using an enrichment-based workflow. Defines target and control cohorts and extracts medical interventions that are over-represented in the target cohort during the observation period. Users can tune filtering and selection thresholds. The workflow includes chi-squared tests for two proportions with Yates continuity correction, logistic tests, and hierarchy and correlation mappings for relevant concepts. The results can be optionally explored using the bundled graphical user interface. For workflow details and examples, see .

BACKGROUND Drug adherence is crucial for chronic disease management, yet treatment discontinuation remains common due to factors such as side effects, inefficacy, or cost. These reasons are often recorded only in free-text clinical notes, making large-scale analysis difficult. While large language models (LLMs) can interpret such unstructured data more effectively than traditional natural language processing methods, few studies have systematically categorized reasons for discontinuation or identified whether the decision was initiated by the patient or the clinician, especially in low-resource languages such as Estonian. OBJECTIVE This study aimed to assess the ability of LLMs to extract and classify reasons for drug discontinuation and identify who initiated it using Estonian electronic health records and characterize the observed discontinuation patterns and initiators for statins and antidiabetic medications. METHODS We combined prescription data with free-text anamneses from a 10% sample of the Estonian population (2012-2019). LLMs (Llama 3.1-70B and GPT-4o) were applied to extract discontinuation phrases and reasons, classify them into a clinician-developed taxonomy, and identify who discontinued the treatment. Performance was evaluated on 100 randomly chosen cases per drug group. RESULTS Extraction yielded 625 antidiabetic drug and 233 statin discontinuation cases. Validation confirmed a precision of 0.93 to 0.98 for extracting phrases and 0.95 to 0.96 for extracting reasons. Classification of discontinuation reasons achieved weighted F1-scores of 0.81 to 0.84, whereas classification of who initiated discontinuation achieved weighted F1-scores of 0.64 to 0.78. Adverse reactions were the most frequent reason overall, accounting for 70% (163/233) of statin discontinuations and 44.8% (280/625) of antidiabetic drug discontinuations. Regarding antidiabetic drugs, treatment inefficacy and contraindications were more common. Patients more often stopped due to adverse reactions or nonmedical reasons, whereas physicians more often initiated discontinuation for contraindications. CONCLUSIONS LLMs can accurately extract and classify medication discontinuation reasons and show variable performance in identifying discontinuation initiators in Estonian clinical narratives. Both local and proprietary models showed promising results, enabling scalable analyses that complement structured health records. This demonstrates the potential of LLMs to unlock information from clinical notes, turning this underused electronic health record component into a valuable resource for monitoring treatment patterns and detecting adverse event signals.

Background and objectives With the proliferation of real-world or observational health data, there is increasing interest in studying treatment trajectories. The real-life treatment trajectories can be complex, and one has to simplify the patterns to draw any conclusions; however, oversimplification will cause the loss of essential details. Thus, the visualization challenge is to strike a balance between the two extremes. Methods We have implemented the observation of treatment trajectories starting from cohort definitions in cooperation with medical specialists, data processing, and then generating the interactive visualizations and detailed data tables within an open-source R package as a Shiny dashboard. The created R package called TrajectoryViz (https://github.com/HealthInformaticsUT/TrajectoryViz) enables reproducible visual analysis and visual content generation for various data investigations. Results We illustrate the use of the tool by assessing the sequence of events present within the data of cervical cancer prevention pathways, as well as the proportions of timely follow-up procedure events. Conclusion Building a toolset to access, manage, and analyze observational health data enables more accessible visual analysis of complicated data, adding time dimension to otherwise simplified event sequences that make up trajectories.

Objective: To introduce 2 R-packages that facilitate conducting health economics research on OMOP-based data networks, aiming to standardize and improve the reproducibility, transparency, and transferability of health economic models. Materials and Methods: We developed the software tools and demonstrated their utility by replicating a UK-based heart failure data analysis across 5 different international databases from Estonia, Spain, Serbia, and the United States. Results: We examined treatment trajectories of 47 163 patients. The overall incremental cost-effectiveness ratio (ICER) for telemonitoring relative to standard of care was 57 472 €/QALY. Country-specific ICERs were 60 312 €/QALY in Estonia, 58 096 €/QALY in Spain, 40 372 €/QALY in Serbia, and 90 893 €/QALY in the US, which surpassed the established willingness-to-pay thresholds. Discussion: Currently, the cost-effectiveness analysis lacks standard tools, is performed in ad-hoc manner, and relies heavily on published information that might not be specific for local circumstances. Published results often exhibit a narrow focus, central to a single site, and provide only partial decision criteria, limiting their generalizability and comprehensive utility. Conclusion: We created 2 R-packages to pioneer cost-effectiveness analysis in OMOP CDM data networks. The first manages state definitions and database interaction, while the second focuses on Markov model learning and profile synthesis. We demonstrated their utility in a multisite heart failure study, comparing telemonitoring and standard care, finding telemonitoring not cost-effective.

Background There is a lack of knowledge on how patients with asthma or chronic obstructive pulmonary disease (COPD) are globally treated in the real world, especially with regard to the initial pharmacological treatment of newly diagnosed patients and the different treatment trajectories. This knowledge is important to monitor and improve clinical practice. Methods This retrospective cohort study aims to characterise treatments using data from four claims (drug dispensing) and four electronic health record (EHR; drug prescriptions) databases across six countries and three continents, encompassing 1.3 million patients with asthma or COPD. We analysed treatment trajectories at drug class level from first diagnosis and visualised these in sunburst plots. Results In four countries (USA, UK, Spain and the Netherlands), most adults with asthma initiate treatment with short-acting ß2 agonists monotherapy (20.8%–47.4% of first-line treatments). For COPD, the most frequent first-line treatment varies by country. The largest percentages of untreated patients (for asthma and COPD) were found in claims databases (14.5%–33.2% for asthma and 27.0%–52.2% for COPD) from the USA as compared with EHR databases (6.9%–15.2% for asthma and 4.4%–17.5% for COPD) from European countries. The treatment trajectories showed step-up as well as step-down in treatments. Conclusion Real-world data from claims and EHRs indicate that first-line treatments of asthma and COPD vary widely across countries. We found evidence of a stepwise approach in the pharmacological treatment of asthma and COPD, suggesting that treatments may be tailored to patients’ needs.