Framework to prioritize health outcomes of particulate matter exposure using national claims data.

Air pollution, particularly particulate matter (PM) exposure, has long been recognized as a critical public health issue because of its pervasive impact on human health [ 1 , 2 ]. Both short- and long-term exposure to PM are linked to severe health outcomes, contributing to increased rates of mortality, hospitalization, and emergency room visits [ 1 – 4 ]. Consequently, much of the research on air pollution has focused on the general population, associating PM exposure with well-known “environmental diseases” such as cardiovascular, respiratory, autoimmune diseases/disorders, and cancer [ 5 ]. However, PM can infiltrate the systemic circulation, reach multiple organs, and trigger inflammatory responses that contribute to the development or progression of a wide range of diseases [ 6 ]. This suggests that the health impacts of PM may extend beyond the traditionally recognized conditions and environmental diseases. However, few studies have systematically examined a broad spectrum of health outcomes, while integrating multidisciplinary clinical expertise and accounting for the practical constraints of administrative claims data. Moreover, prior studies have primarily used a data-driven approach with population-based cohorts, often without considering the underlying pathological mechanisms or clinical feasibility [ 7 ]. As a result, our understanding of how PM exposure contributes to disease onset or influences the progression of pre-existing conditions across diverse clinical areas remains limited. In Korea, there is the Korea National Health Insurance Service (K-NHIS) database, which contains comprehensive, population-wide health insurance claims data from South Korea’s single-payer healthcare system. Covering nearly the entire Korean population, the K-NHIS includes information on inpatient and outpatient visits, procedures, and prescriptions, enabling large-scale investigations into the effects of PM across diverse clinical conditions [ 8 ]. Identifying diseases that are particularly susceptible to the effects of PM exposure is a critical first step in mitigating its health burden. By clarifying which conditions are most influenced by PM and which populations are most vulnerable, targeted interventions can be implemented to reduce exposure and prevent disease onset or progression. However, given the broad scope of possible analyses, it is essential to prioritize research questions by specifying relevant exposure periods and outcome types to focus efforts on the most clinically meaningful and feasible investigations. This study aims to identify and prioritize research questions for evaluations of PM-related health effects by implementing a structured collaborative process with a multidisciplinary team.

A total of 99 experts from 16 specialized areas participated in the study. The mean (standard deviation, SD) age was 46.1 (7.0) years, and 61.6% were male. Approximately 68.7% worked at tertiary hospitals, and the mean (SD) years of professional experience was 20.5 (7.4) years ( Table 1 ). We hosted 23 seminars covering topics such as environmental and clinical epidemiology, air pollution research methods, PM mitigation strategies, animal models, and recent trends in PM-related studies ( S2 Table ). Values are expressed as mean (SD) or N (%). Research questions were categorized into four groups based on the PM exposure duration and outcome type: short-term exposure and disease incidence, short-term exposure and disease progression, long-term exposure and disease incidence, and long-term exposure and disease progression. As part of Round 1, we conducted a literature review to identify diseases previously studied in relation to PM exposure. This review revealed that only 24 and 58 diseases were investigated for disease incidence following short- and long-term PM exposure, respectively, while only 14 and 6 diseases were investigated for disease progression under the same exposure durations ( Fig 2 ). This spider chart shows the number of diseases associated with short-term PM exposure and disease incidence, as identified in the literature (blue) and proposed during expert discussions (red). Notable increases in areas such as ophthalmology, dermatology, and psychiatry highlight underexplored domains in environmental health research. In Round 2, clinical experts were asked to specify research questions for each candidate disease, including the relevant exposure period and outcome type. As a result of this expert-driven process, the number of diseases considered increased substantially. Specifically, 108 and 124 diseases were identified as potentially associated with disease incidence following short- and long-term PM exposure, respectively. Similarly, 79 and 81 diseases were identified as potentially associated with disease progression for short- and long-term PM exposure, respectively. This expansion included disease areas that were underrepresented in the literature. For example, 18 psychiatric, 17 ophthalmologic, 10 dermatologic, and 10 otolaryngologic diseases were potentially associated with short-term exposure ( Fig 2 , panel A ). Additionally, 12 gynecological and immunological diseases were associated with long-term exposure in the context of disease incidence ( Fig 2 , panel C ). The oncology group concluded that cancer incidence was associated only with long-term PM exposure. Following Round 3 prioritization, 80 diseases were selected for further evaluation using the K-NHIS data ( Fig 3 ). A total of 211 research questions were defined, including 54 related to disease incidence following short-term PM exposure, 47 related to disease progression following short-term PM exposure, 64 related to disease incidence following long-term PM exposure, and 46 related to long-term disease progression ( Table 2 ). Among the 79 diseases that were not prioritized, the primary reasons for exclusion were limited feasibility based on K-NHIS data (35.4%), low causal evidence (32.9%), and low clinical impact (31.6%). Cell colors represent the proportion of “High” responses: ( (≤ 30%), (31%−60%), (61%−90%), and (91%−100%). This figure illustrates the diseases prioritized for further investigation based on structured expert discussion. Diseases are grouped by clinical specialty and reflect those identified as having potential associations with PM exposure, either in terms of disease incidence or progression.

This study has several limitations. First, despite our efforts to recruit a diverse range of experts, certain clinical specialties were underrepresented, which may have influenced the breadth of the diseases considered. However, to mitigate this, we employed a snowball sampling strategy that allowed all invited participants to recommend additional experts without specialty restrictions. In addition, although our expert panel reflected multidisciplinary expertise to some extent, it did not include specialists whose primary focus is on artificial intelligence, machine learning, or multimodal data analysis. The inclusion of such experts may have enhanced the methodological diversity and depth of our findings. Future research will engage a broader range of experts from these rapidly advancing fields. Second, although expert-driven prioritization offers valuable clinical insight, the rating process may be influenced by individual judgement. To enhance consistency and transparency, we applied predefined and standardized definitions for pathological causality, clinical impact, and data feasibility, which were provided to all experts as guidance for assigning high/medium/low scores. Third, the use of a uniform 1-month threshold to distinguish short-term and long-term exposure across all disease outcomes. Although this choice was informed by standardized exposure-duration categories in environmental health research, including those defined by the U.S. EPA, disease latency and progression patterns vary substantially. Thus, a single threshold may oversimplify exposure-response relationship between PM exposure and health outcomes. Future studies should consider outcome-specific exposure windows and where feasible, evaluate alternative thresholds. Fourth, diseases were grouped by clinical systems based on the structured adopted in the U.S. EPA Integrated Science Assessment (ISA). While this provided a practical and standardized framework for organizing expert input and ensuring comparability with prior literature, it may not fully account for distinct biological mechanisms through which PM affects different diseases within the same category. Future work should consider mechanism-informed or phenotype-based groupings (e.g., inflammation-, autonomic-, or oxidative stress-mediated pathways) to better reflect within-system heterogeneity. Fifth, this study was conducted in the context of the South Korean healthcare system using national claims data; therefore, the findings may reflect country-specific clinical practices, environmental exposure patterns, and coding systems. Therefore, the generalizability to other settings should be approached with caution. Nonetheless, this study was strengthened by its comprehensive and structured approach and by the engagement of nearly 100 experts with extensive clinical and research experience. By integrating expert input with a systematic literature review and emphasizing data feasibility, including consideration of claim codes and reimbursement practices, we developed a robust and replicable framework for prioritizing PM-related health research using administrative claims data. Importantly, the incorporation of multidisciplinary expert perspectives enabled the identification of clinically meaningful conditions that may not be fully captured in the existing literature. This approach allowed us to account for biological plausibility and real-world clinical relevance, complementing the evidence-based findings of the literature review. We also recognize a temporal gap between the literature synthesis and expert assessment: the review reflects evidence only up to the search end date and may lag due to publication/indexing delays, whereas expert input incorporates more current clinical observations. In addition, potential publication bias may limit the visibility of outcomes with null or non-significant findings, further underscoring the value of integrating evidence-based and expert-informed perspectives.

This study presents a scalable, multidisciplinary framework for setting research agendas that align environmental health inquiries with the clinical relevance and practical constraints of large-scale administrative data systems. This approach is particularly valuable for countries with national claims databases or health registries, providing a transferable approach for supporting evidence generation to inform policy and public health actions. As the health impacts of environmental exposures become increasingly complex and context-specific, the adoption of transparent, interdisciplinary, and data-integrated strategies is essential for advancing clinically meaningful and data-feasible environmental health research.

This study employed a comprehensive approach with three components: a literature review, multidisciplinary discussion, and knowledge-sharing ( Fig 1 ). The structured prioritization process was conducted between April and November 2024 as part of the AIR Impacted Research (AIR) study. Expert recruitment was conducted from April to June 2024. We invited clinical experts from departments associated with PM-related health outcomes, as well as leading researchers in environmental and clinical epidemiology in South Korea. To broaden the pool of experts, all invitees were asked to recommend colleagues with relevant expertise or research experience in PM exposure. We also invited presenters from major academic conferences, including the 2024 Sapporo Exposome Symposium and the 2024 Korean Society of Environmental Epidemiology. All invited experts provided informed consent to participate in this study. Consent was informed and obtained via email. Experts were given a detailed explanation of the study objectives, scope, and anticipated roles. Email responses confirming participation served as documentation of verbal consent. Additionally, we reviewed national disease statistics and insights from a national conference on PM and health to reflect the public and patient perspectives [ 9 ]. This study was approved by the Institutional Review Board of the Samsung Medical Center (No.2024-11-018). This conceptual framework summarizes the AIR team’s collaborative process, including literature review, expert discussions, and stakeholder engagement, to generate and prioritize research questions for evaluating PM-related health effects. To systematically prioritize diseases to evaluate the health impacts of PM exposure, we conducted a structured process consisting of a literature review, multidisciplinary discussion, and knowledge sharing. Between May and September 2024, we held 23 weekly seminars to facilitate interdisciplinary knowledge exchange and identify gaps in PM-related research. The seminars were open to the AIR team and external researchers interested in the health impacts of PM. A comprehensive literature review was conducted to examine the pathological mechanisms of PM and its associated health impacts. First, we reviewed a systematic report by the U.S. Environmental Protection Agency (EPA) that synthesized studies related to PM up to December 2019 [ 6 ]. We also reviewed the 2021 abstract collection on PM and air pollution jointly published by the Korea Disease Control and Prevention Agency and the Korean Academy of Medical Science, which summarizes the literature from 2010 to 2019 [ 10 ]. To ensure the inclusion of recent findings, we replicated the search strategy of the 2021 abstract collection to identify publications from January 2020 to March 2024 ( S1 Table ). Ongoing studies were identified using ClinicalTrials.gov (the search was conducted on April 17, 2024). The relevant literature on newly identified diseases was also reviewed through multidisciplinary discussions. We conducted three structured rounds of discussion to identify and prioritize the diseases. Preliminary meetings with environmental and clinical epidemiology experts were held to review the limitations of previous studies and refine the process. The AIR team included specialists in environmental epidemiology, data science, clinical epidemiology, various clinical departments, and government agencies. It included 14 clinical departments, oncology, respirology, cardiology, endocrinology, gastroenterology, immunology, nephrology, urology, gynecology, dermatology, otolaryngology, ophthalmology, psychology, and neurology, which participated in the disease prioritization process. Each group received a summary of the literature review and used both the summary and their clinical expertise to compile a list of diseases potentially influenced by PM exposure. Experts were asked to specify research questions for each candidate disease, including the relevant exposure period (short- or long-term) and health outcomes (disease incidence or progression). Short-term exposure was defined as <1 month and long-term exposure as ≥1 month consistent with exposure-duration categories in the U.S. EPA’s human health risk assessment framework [ 11 ]. In this study, disease progression was defined as a deterioration in the clinical course of a pre-existing condition. This included worsening of symptoms or severity (progression), recurrence after remission (relapse), acute worsening of a chronic condition (exacerbation), and development of secondary conditions attributable to the pre-existing disease (complication). This definition encompasses any clinical event that may negatively affect patient prognosis or quality of life [ 12 ]. Responses were collected via email to encourage thoughtful and unprompted contributions tailored to disease-specific characteristics. Each expert independently evaluated the candidate diseases using a structured survey with predefined definitions for three criteria—causality, clinical impact, and feasibility—rating each criterion as high, medium, or low. Causality was defined as the strength and coherence of evidence linking PM exposure to the outcome, considering biological plausibility and the quality of epidemiologic/experimental study designs. Clinical impact referred to the public health burden of the disease, including its incidence, morbidity, and mortality, and expected impact on clinical decision making. Feasibility denoted the likelihood of identifying and analyzing the outcome using the Korean National Health Insurance Service (K-NHIS) database (e.g., International Classification of Diseases, 10th revision (ICD-10) coding accuracy, data completeness, and data availability). Ratings were submitted individually and collected electronically, without group discussion at this stage, to minimize bias. Individual assessments were subsequently aggregated for final prioritization. Prioritization was determined based on expert ratings. For departments involving four or more experts, a disease was prioritized if 30% or more of the experts rated either causality or clinical impact as high. For departments with three or fewer experts, prioritization required at least two experts to rate either causality or clinical impact as high. All prioritized hypotheses were then evaluated for feasibility.

(DOCX) (DOCX)