This paper is in the following e-collection/theme issue:

Reviews in Medical Informatics (121) New Methods (1173) Machine Learning (2212) Tools, Programs and Algorithms (405) Clinical Information and Decision Making (2051) Decision Support for Health Professionals (1624)

Published on in Vol 14 (2026)

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/82499, first published .
Causal Discovery in Observational Medical Research: Scoping Review

Causal Discovery in Observational Medical Research: Scoping Review

Causal Discovery in Observational Medical Research: Scoping Review

Authors of this article:

Zuting Liu1, 2 Author Orcid Image ;   Tian Luo1, 2 Author Orcid Image ;   Hailin Ma1, 2 Author Orcid Image ;   Jiali Mo1, 2 Author Orcid Image ;   Xia Yang1, 2 Author Orcid Image ;   Zhenglong Huang1, 2 Author Orcid Image ;   Jingkun Li3 Author Orcid Image ;   Jie Kuang1, 2 Author Orcid Image
Article Authors Cited by Tweetations (3) Metrics

    Review

    1Department of Epidemiology, School of Public Health, Jiangxi Medical College, Nanchang University, Nanchang, China

    2Jiangxi Provincial Key Laboratory of Disease Prevention and Public Health, Nanchang University, Nanchang, China

    3Department of Biostatistics, School of Public Health, Harbin Medical University, Harbin, China

    *these authors contributed equally

    Corresponding Author:

    Jie Kuang, PhD

    Department of Epidemiology

    School of Public Health, Jiangxi Medical College

    Nanchang University

    No. 1299 Xuefu Avenue

    Nanchang, 330019

    China

    Phone: 86 15807008572

    Email: kuangjie@ncu.edu.cn


    Background: Observational data are fundamental to medical research but present formidable challenges for causal inference. Machine learning–based causal discovery algorithms have emerged as a promising solution to identify causal structures directly from such data. However, the current literature is skewed toward theoretical and methodological innovations, with a critical gap in systematic assessments of performance in medical research settings and a lack of practical guidance for clinicians and researchers on selecting and applying these algorithms in specific medical contexts.

    Objective: This study aimed to systematically map and synthesize the application of causal discovery methods in observational medical research, detailing the methodologies used, their application domains, the robustness of the findings, and the practical challenges encountered.

    Methods: Following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines, we conducted a systematic search of Scopus, Web of Science, PubMed, MEDLINE, Embase, and CINAHL from inception to May 2025. We included studies that applied any causal discovery algorithms within a medical research context, encompassing both analyses of real-world observational data and method-validation studies using synthetic or benchmark datasets with a clear medical focus. Purely methodological papers and studies based solely on experimental data were excluded. Data were extracted and synthesized using a descriptive analysis focused on study characteristics, algorithm types, application domains, reported numerical results, and implementation challenges.

    Results: Out of 2296 identified publications, 72 (3.1%) met the inclusion criteria. Our synthesis revealed three key themes. The first theme was methodological landscape, where constraint-based algorithms were the most prevalent (38/72, 52.8%), with the fast causal inference (10/72, 13.9%) and Peter-Clark algorithms (9/72, 12.5%) being most common. Score-based (19/72, 26.4%) and hybrid (14/72, 19.4%) methods also represented significant and growing segments (methods were not mutually exclusive). The second theme was application domains and findings, where the majority of studies (54/72, 75%) were in clinical research, with a strong focus on mental health (19/72, 26.4%; eg, identifying symptom networks in schizophrenia and posttraumatic stress disorder) and chronic diseases (19/72, 26.4%; eg, elucidating progression pathways in Alzheimer and diabetes). Etiological research was the primary objective (28/72, 38.9%). Public health applications (18/72, 25%) frequently assessed the causal impacts of behavioral interventions. The third theme was implementation challenges and innovations, where common challenges included pervasive unmeasured confounding, limited sample sizes (noted in more than 20% of studies), and reliance on unvalidated causal assumptions. Emerging innovations focused on longitudinal data frameworks and the integration of multimodal data sources to strengthen causal claims.

    Conclusions: This review underscores the growing application of causal discovery algorithms in medical research while also highlighting challenges such as the lack of standardized validation frameworks and persistent confounding. Future efforts must focus on developing evaluation standards and fostering interdisciplinary collaboration to translate these powerful computational techniques into reliable tools for medical research and practice.

    JMIR Med Inform 2026;14:e82499

    doi:10.2196/82499

    Keywords

    causal discovery; medical research; observational data; algorithm; scoping review 


    The advent of big data has driven significant advances in machine learning (ML), yet current methodologies often focus on identifying correlations rather than uncovering underlying causal relationships [1]. This limitation is particularly critical in medicine, where understanding causality is essential for effective clinical decision-making. Contemporary causal research addresses 2 interrelated challenges: causal discovery (identifying causal structures and directions between variables) and causal inference (quantifying effects within predefined causal frameworks) [2-5]. Causal discovery is concerned with deducing the causal structure and directionality among variables from data, with the primary objective of elucidating direct and indirect causal pathways by constructing causal graphs. Typically, it relies on statistical principles such as conditional independence tests to discern potential causal structures from correlational patterns. In contrast, causal inference focuses on quantifying the magnitude of the effect of specific interventions on an outcome variable, presuming a known or assumed causal structure. Therefore, inferring causal relationships, that is, causal discovery, from observational medical data is critical for advancing precision and personalized medicine. A central class of computational methods developed for causal discovery, such as the Peter-Clark (PC) algorithm and fast causal inference (FCI), aims to learn a causal graph that represents the underlying data-generating mechanism. The growing scale and accessibility of diverse medical data resources, encompassing real-world records, such as surgical, therapeutic, and diagnostic information, as well as genetic, lifestyle, and environmental data, have become invaluable for such research [6]. However, these data are often affected by confounding, selection bias, and missing data, posing substantial challenges to deriving reliable causal conclusions. Although randomized controlled trials (RCTs) remain the gold standard for medical decision-making [7], the limits of RCTs and the growing abundance of observational data underscore an urgent need for more robust causal discovery methodologies.

    Observational datasets, particularly those derived from real-world settings, offer significant advantages for medical research, including large sample sizes, accessibility, and the ability to reflect real-world clinical scenarios. Previous studies have demonstrated their potential to elucidate disease etiology [8-10], optimize treatment strategies [11,12], and refine prognostic models [13,14]. However, the specific application of causal discovery methods to harness these datasets remains particularly challenging and fragmented. Key barriers include unmeasured confounding, complex data structures, and a lack of consensus on methodological best practices. As a result, there is an insufficient synthesis of how these methods can be robustly applied, validated, and translated into clinical insights. This scoping review aims to fill this gap by systematically analyzing causal discovery methods applied to observational medical studies. Specifically, we critically assess: (1) The evolving methodological landscape and algorithm suitability across clinical contexts; (2) High-impact application domains and translational outcomes; (3) Persistent implementation challenges, such as unmeasured confounding and sample limitations. By mapping these dimensions, we provide a foundation for advancing causal discovery in digital health and precision medicine.


    Study Design

    This scoping review adhered to Arksey and O’Malley’s framework for scoping review [15] and the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines [16] (Multimedia Appendix 1). Our primary research question was: “What are the methodologies, clinical applications, and implementation challenges of causal discovery in observational medical research?” The aim was to synthesize evidence to guide future methodological development and clinical translation. Given the rapidly evolving nature of this field, no temporal restrictions were applied.

    Search Strategy

    To comprehensively review studies of causal discovery algorithms in medical research and its subtypes, 2 researchers (XY and ZH) performed a systematic literature search of Scopus, Web of Science, PubMed, MEDLINE, Embase, and CINAHL in May 2025 and was restricted to studies published in English, using the following search strategies: “causal discovery/causal structure discovery/causal structure learning/causal graph discovery/learning causal models.” The search was focused on biomedical databases (PubMed, Embase, etc) as this review aims to map applications in medical research. While computer science–focused databases, such as IEEE Xplore or ACM Digital Library, include methodological developments, they primarily emphasize algorithmic innovation; seminal methodological papers with medical applications are typically also indexed within the selected biomedical sources. The search strategy was designed in consultation with an information specialist. The entire search strategy for the 6 databases is displayed in Multimedia Appendix 2.

    Eligibility Criteria

    Two researchers (HM and JM) independently screened studies based on predefined inclusion and exclusion criteria by examining their titles, abstracts, and full texts. Studies met the following inclusion criteria: (1) application focus: empirical implementation of causal discovery algorithms in medical or health contexts. (2) Data and methodology: the study applied a causal discovery approach within the observational research paradigm. This included analyses of real-world observational medical data as well as method-validation studies that used synthetic or benchmark datasets to address a medical research question or simulate a medical data environment. (3) Output: the study generated an explicit causal graph or causal structure. (4) Language: English-language publications. The exclusion criteria were: (1) The study purely focused on methodological or theoretical papers without medical applications. (2) Experimental (RCT) or basic science studies. (3) Duplicate publications, brief abstracts, or conference summaries lacking full methodological and results sections. (4) Studies where causal discovery was not the primary methodology. Discrepancies in study eligibility were resolved through consensus between 2 independent reviewers (JK and JL).

    Data Extraction and Synthesis

    Screening was performed using Rayyan [17], with dual-reviewer verification at all stages. A standardized electronic form was used to extract the following data items: (1) study characteristics: year of publication, country or region of corresponding author affiliation, study design, and sample size; (2) causal discovery methods: specific algorithms, methodological category, and the primary software or package used for implementation; (3) data properties: data type, dimensionality, and temporal features; (4) application domain and specific clinical or public health objectives; and (5) validation approaches and key reported limitations.

    Data synthesis was conducted in 2 complementary ways to align with the scoping review’s descriptive aims. First, a descriptive quantitative analysis summarized structured characteristics (eg, causal discovery algorithms, software tools, and application domains) using frequencies and percentages. Second, for textual data on reported methodological considerations and challenges, an inductive content summary was performed. Two reviewers (ZL and TL) independently identified recurrent issues and patterns, which were then discussed, consolidated, and organized into overarching descriptive themes (eg, methodological landscape and challenges). This process ensured the results reflected both the quantitative distribution and qualitative substance of the included literature. While this scoping review did not undertake a formal critical appraisal of individual studies using standardized quality assessment or risk-of-bias tools, we systematically extracted study-reported methodological limitations and implementation challenges as a prespecified data item during full-text screening. These author-reported limitations were subsequently subjected to inductive qualitative synthesis, alongside other extracted textual data, to identify recurrent patterns and to develop the descriptive themes presented in the “Challenges” section. This approach was intended to map commonly reported methodological issues in the literature rather than to appraise or grade the quality of individual studies.


    Article Selection

    After the removal of duplicates, 2296 unique records were identified for screening. Of these, 72 studies [8-14,18-82] (3.1%) met the inclusion criteria and were included in the final synthesis (Figure 1). After eliminating 2548 duplicates (52.6%), title or abstract screening excluded 1909 records (39.4%). A full-text review of 387 articles led to the exclusion of 315 (81.4%) due to reasons such as a methodological focus (201/315, 63.8%), duplicate publications or abstracts (50/315, 15.9%), and basic research or experimental designs (64/315, 20.3%).

    ‎
    Figure 1. PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) flowchart for the study selection process. RCT: randomized controlled trial.

    Characteristics of the Articles

    The characteristics of the included studies (N=72) [8-14,18-82] are summarized in Table 1, covering publication year, country or region, and document type. The number of annual publications progressively increased from 2017 to 2025, with the years 2023-2025 accounting for the majority (46/72, 63.9%). Publications peaked in 2023 and 2024. The majority of studies were peer-reviewed journal articles (64/72, 88.9%), with the remaining studies being conference proceedings. The studies originated from 18 distinct countries or regions, with the United States contributing the largest proportion (25/72, 34.7%), followed by China (8/72, 11.1%), the Netherlands (6/72, 8.3%), and Japan (4/72, 5.6%), accounting for a combined total of 59.7% of publications. Three primary themes emerged from the analysis: (1) methodological landscape, that is, dominant causal discovery approaches and typologies in medical research; (2) application domains, that is, multidimensional implementation scenarios across observational studies; and (3) challenges, that is, methodological limitations and barriers to clinical translation.

    Table 1. Characteristics of the studies used in the review, including the year, type, and country of publication (N=72).
    FeaturesStudies, n (%)References
    Year of publication

    		202514 (19.4)[8-10,13,18-26,82]

    		202416 (22.2)[11,14,27-40]

    		202316 (22.2)[41-56]

    		20228 (11.1)[57-64]

    		20215 (6.9)[12,65-68]

    		20207 (9.7)[69-75]

    		20192 (2.8)[76,77]

    		20181 (1.4)[78]

    		20173 (4.2)[79-81]
    Type of publication

    		Journal article64 (88.9)[8-14,18-31,33,34,37,38,41-48,50,51,53-63,65-82]

    		Conference article8 (11.1)[32,35,36,39,40,49,52,64]
    Country of publication

    		United States25 (34.7)[18,20,22,23,26,28,36,38,43,46,47,51,56-58,60,61,63,65-67,70,73,74,79]

    		China8 (11.1)[19,29,33,42,52,69,77,78]

    		Netherlands6 (8.3)[9,27,59,71,80,81]

    		Japan4 (5.6)[39,40,49,75]

    		Other countries29 (40.3) [8,10-14,21,24,25,30-32,34,35,37,41,44,45,48,50,53-55,62,64,68,72,76,82]

    Themes

    Theme 1: Methodological Landscape

    Causal discovery methodologies are broadly divided into several major categories: constraint-based, score-based, continuous optimization, functional causal models, and hybrid methods. It should be noted that these categories are not mutually exclusive; a single study could use methods from more than one category. Thus, the percentages reported below are calculated based on the total number of included studies (N=72) [8-14,18-82] and may sum to more than 100%. Among the 72 studies analyzed, constraint-based methods constituted the predominant approach (38/72, 52.8%) [8-10,12,14,18,24,25,27,29-32,34-38,43, 45,48,52,53,55,57-59,63,64,68,70,71,73,74,77,79-81], followed by score-based (19/72, 26.4%) [8,11,18-20,23,24,26,28,32,33,35,39,42,61,67,73,74,82], hybrid (14/72, 19.4%) [8,13,22,23,33,41,46,47,51,56,60,63,65,69], continuous optimization (9/72, 12.5%) [29,33,36,40,44,49,50,54,62], and functional causal models (5/72, 6.9%) [21,32,66,75,78]. At the algorithmic level, FCI (10/72, 13.9%) [10,18,29,34,38,43,58,63,73,74] and PC (9/72, 12.5%) [12,18,25,27,29,35,36,53,70] algorithms were the most implemented constraint-based methods, whereas greedy fast causal inference (GFCI) (10/72, 13.9%) [13,22,23,41,46,47,51,60,63,65] and Non-combinatorial Optimization via Trace Exponential and Augmented Lagrangian for Structure learning (6/72, 8.3%) [36,40,49,50,54,62] demonstrated notable dominance within the hybrid and continuous optimization categories, respectively. Additionally, time-series causal discovery (eg, Peter-Clark momentary conditional independence) was actively explored in 11 studies [32,44,48,55,64,66,68,72,76,77,82]. These findings highlight a clear trajectory toward the adoption of more integrated, scalable, and nonparametric causal discovery frameworks. The reviewed studies used methodologies from 2 distinct conceptual frameworks. The majority used structural causal discovery algorithms (eg, PC, FCI, and NOTEARS) to learn causal graphs or networks. An analysis of implementation tools revealed the software landscape for causal discovery (Table 2). Of the 72 included studies, 45 (62.5%) [8-10,12,13,18-22,24,26-28,30, 32,36-39,41,43,44,46-48,50,51,54,55,57,60-63,65,68-71,74,75,77,79] explicitly specified the software or package used. Among these, R was the most prevalent environment, used in 22 studies [8-10,12,18,20,22,24,26-28,30,37,38,48,55,57,60,61,68,74,82] (22/45, 48.9% of reporting studies, or 22/72, 30.6% of all studies), with the pcalg package being the most frequently cited. The standalone Tetrad suite was used in 10 studies [13,18,41,43,46,47,51,63,65,74] (10/45, 22.2% of reporting studies, or 10/72, 13.9% overall). Out of 45 studies, Python and MATLAB were implemented in 7 [9,21,32,36,50,54,62] (15.5%) and 8 [32,44,69-71,75,77,79] (17.7%) of the software-reporting studies, respectively. A notable finding regarding transparency was that 27 studies [11,14,23,25,29,31,33-35,40,42,45,49,52,53, 56,58,59,64,66,67,72,73,76,78,80,81] (37.5% of the total) did not report any software details, posing a significant barrier to reproducibility. For details, refer to Table 2.

    Table 2. Characteristics of the studies used in the review, including author, causal discovery method, method type, software, disease or area, and type of data.
    AuthorsCausal discovery methodMethod typeSoftwareDisease or areaType of dataValidation methods
    Gururaghavendran and Murray [18]PCa, FCIb, FGESc, GRaSPdConstraint-based; Score-basedR (pcalg), TetradCardiovascular mortalityCohortInternal stability assessment; Expert review
    Petrungaro et al [8]PC-stablee, Inter-IAMBf, HCg, Tabuh, MMHCi, H2PCjConstraint-based; Score-based; HybridR (bnlearn)SepsisCross-sectionalNone reported
    Li et al [19]GESkScore-basedPython (DoWhy, Causal Discovery Toolbox)Alzheimer diseaseCohortNone reported
    Korvink et al [20]GESScore-basedR (pcalg, dagitty)Public healthCross-sectionalNone reported
    Penson et al [9]BCCDlConstraint-basedR (lavaan, RUcausal)Chronic fatigueCross-sectionalNone reported
    Noh et al [21]LiNGAMmFunctional-basedPython (DoWhy)DiabetesCross-sectionalStatistical verification
    Tseng et al [22]GFCInHybridR (lavaan)AutismCohortNone reported
    Yu et al [10]FCI and MGMoConstraint-basedR (rCausalMGM)DementiaCohortNone reported
    Bronstein et al [23]GFCI, GraSP, GraSP-FCIpScore-based; HybridNot specifiedClinical high risk for psychosisCohortInternal stability assessment
    Giuliani et al [13]GFCIHybridTetradSchizophreniaCross-sectionalNone reported
    Colineaux et al [24]Hill climbing, Inter-IAMB, ARACNEqConstraint-based; Score-basedR (bnlearn)Health care useCohortExpert review
    Ribeiro-Dantas et al [14]iMIICrConstraint-basedNot specifiedBreast cancerCohortExpert review; Internal stability analysis
    Verveen et al [27]PCConstraint-basedR (pcalg)COVID-19CohortNone reported
    Ren et al [28]GGESsScore-basedR (pcalg)Esophageal cancerCohortNone reported
    Guo et al [29]PC, FCI, and GAEtConstraint-based; Continuous optimization-basedNot specifiedNonsuicidal self-injuryCross-sectionalNone reported
    Foraita et al [30]Modified PCuConstraint-basedR (pcalg, tPC, micd)Childhood obesityCohortNone reported
    Ribeiro et al [31]AnchorFCIvConstraint-basedNot specifiedCardiometabolic diseasesCross-sectionalInternal stability analysis
    Sabzevar et al [32]RDCMw, MVGCx, PCMCIy, VarLINGAMzConstraint-based; Score-based; Functional-basedMATLAB, PythonSchizophreniaCross-sectional (fMRI time series)None reported
    Zhang et al [33]MMHC, H2PC, CBAMNaaScore-based; Continuous optimization-based; HybridNot specifiedAcute kidney injuryCohortInternal stability analysis; Statistical verification; Independent dataset validation
    Bernasconi et al [11]SEMabScore-basedNot specifiedBreast cancerCohortExpert review
    Fay et al [25]PCConstraint-basedNot specifiedAortic MorphologyCross-sectionalNone reported
    Gąsior et al [41]GFCIHybridTetradCongenital heart diseaseCross-sectionalNone reported
    Gao et al [42]RLac-based causal discovery (IIEad strength)Score-basedNot specifiedType 2 diabetesCross-sectionalNone reported
    Lee et al [43]FCIConstraint-basedTetradCardiac surgery (CABGae and AVRaf)CohortNone reported
    Miley et al [51]GFCIHybridTetradSchizophreniaCohortIndependent dataset validation
    Thomas et al [44]DYNOTEARSagContinuous optimization-basedMATLABSleep disordersCross-sectionalNone reported
    Vowels et al [45]SAMahConstraint-basedNot specifiedMental health (depression and anxiety)Cross-sectional and cohortNone reported
    Bao et al [57]Modified PC, MMPC, IAMBai, GSajConstraint-basedR (bnlearn)HIVCross-sectionalStatistical verification; Internal stability assessment
    Singh et al [58]FCIConstraint-basedNot specifiedIn-hospital mortalityCross-sectionalNone reported
    Pierce et al [46]GFCIHybridTetradPTSDakLongitudinal time-seriesNone reported
    Langwerden et al [59]BCCDConstraint-basedNot specifiedMental disordersCross-sectionalNone reported
    Bronstein et al [60]GFCIHybridR (rcausal)Eating disordersCross-sectionalNone reported
    Ganopoulou et al [12]PCConstraint-basedR (MXM)Coronary chronic total occlusionsCross-sectionalNone reported
    Rawls et al [65]GFCIHybridTetradAlcohol use disorderCross-sectionalNone reported
    Li et al [66]ANMalFunctional-basedNot specifiedCOVID-19Longitudinal time-series and cross-sectionalNone reported
    Wang et al [69]PKCLamHybridMATLABT2DMan and osteoporosisCross-sectionalExpert review
    Saxe et al [70]GLL-PCao and TPCapConstraint-basedMATLABPTSDCohortNone reported
    Lee et al [47]GFCIHybridTetradNursing home hospitalizationCohortExpert review
    Andersen et al [48]TPCConstraint-basedR (causalDisco)Mental health (stress and depression)Cross-sectional and cohortIndependent dataset validation
    Shen et al [67]FGES, Modified FGES (temporal constraints)Score-basedNot specifiedT2DMCohortIndependent dataset validation
    Petersen et al [68]TPCConstraint-basedR (causalDisco)DepressionCohortInternal stability assessment
    Schoenmacker et al [71]BCCDConstraint-basedMATLABADHDaqCohortNone reported
    Ahangaran et al [72]PCDSDarProbabilistic graphical modelNot specifiedAmyotrophic lateral sclerosisLongitudinal time-seriesNone reported
    Wang et al [73]FCIb and LSTMasConstraint-based; Score-basedNot specifiedKnee osteoarthritisCohortNone reported
    Chen et al [78]McDSLatFunctional-basedNot specifiedAcute kidney injuryCohortNone reported
    Galatzer-Levy et al [79]SVMau, LGMMav, and HITON-PCawConstraint-basedMATLABPTSDCohortNone reported
    Sokolova et al [80]BCCDConstraint-basedNot specifiedADHD and ASDaxCross-sectionalNone reported
    van Dijk et al [81]BCCDConstraint-basedNot specifiedThe Wechsler Adult Intelligence Scale-IVCross-sectionalNone reported
    Li et al [49]NOTEARSayContinuous optimization-basedNot specifiedGeneral health statusLongitudinal time-seriesNone reported
    Shen et al [74]FCI, FGES, and SEMConstraint-based; Score-basedTetrad, R (lavaan)Alzheimer diseaseCohortExpert review
    Mouches et al [62]NOTEARSContinuous optimization-basedPython (CausalNex)Brain aging and CVDazCross-sectionalNone reported
    Shen et al [61]FGESScore-basedR (rcausal)White matter hyperintensitiesCohortNone reported
    Zhang et al [63]FCI and GFCIConstraint-based; HybridTetradCOVID-19-associated AKIbaLongitudinal time-seriesNone reported
    Park et al [50]NOTEARSContinuous optimization-basedPython (CausalNex)Quality of life and mental healthCross-sectionalNone reported
    Fu et al [26]GESScore-basedR (package unspecified)Progressive supranuclear palsyCohortNone reported
    Nogueira et al [64]ItsPCbbConstraint-basedNot specifiedICUbc survival analysisCohortNone reported
    Li et al [34]FCI and PPCITbdConstraint-basedNot specifiedHospital-acquired pressure injury and spinal cord injuryCohortStatistical verification
    Kazemi et al [35]PC, GES, and GIESbeConstraint-based; Score-basedNot specifiedMaternal quality of life (obesity and stress-related)CohortStatistical verification
    Naik et al [36]NOTEARS and PCConstraint-based; Continuous optimization-basedPython (CausalNex, gcastle, pgmpy)Non–small cell lung cancerCross-sectionalNone reported
    Ribeiro Santiago et al [82]Bayesian Structure Learning (partition MCMCbf)Score-basedR (BiDAG, blavaan)Adolescent emotional disorders (depression or anxiety)CohortExpert review
    McCormick et al [37]GGMbg, PC-stable, and SEMConstraint-basedR (bnlearn, lavaan, EGAnet)Oral healthCross-sectionalNone reported
    Zeng et al [38]FCIConstraint-basedR (rCausalMGM)Childhood Sjögren’sCohortNone reported
    Li et al [52]PCConstraint-basedNot specifiedPrimary liver cancerCohortNone reported
    Nagaraj et al [53]PCConstraint-basedNot specifiedStress-related disordersLongitudinal time-seriesNone reported
    Vigneshwaran et al [54]NOTEARSContinuous optimization-basedPython (CausalNex)Brain aging and gray matter atrophyCross-sectionalNone reported
    Petersen et al [55]TPCConstraint-basedR (causalDisco)CVD and depressionCohortExpert review
    Zhang et al [40]NOTEARSContinuous optimization-basedNot specifiedPsychological stress and obesity-related risksLongitudinal time-seriesExpert review
    Tosaki et al [39]Causal Bayesian networkScore-basedINGORAtherosclerosis, hypertension, diabetes, dyslipidemia, osteopenia, chronic kidney disease, COPDbh, and obesityCohortNone reported
    Fung et al [56]Causal Bayesian networkHybridNot specifiedCOVID-19Cross-sectionalStatistical verification
    Ahangaran et al [76]PCDSDProbabilistic graphical modelNot specifiedAmyotrophic lateral sclerosisLongitudinal time-seriesNone reported
    Yang et al [77]CD-SFbi and CD-SU-SFbjConstraint-basedMATLABLung cancerCross-sectionalNone reported
    Itahashi et al [75]LiNGAMbkFunctional-basedMATLABAdolescent mental issuesCohortNone reported

    aPC: Peter-Clark algorithm.

    bFCI: fast causal inference.

    cFGES: fast greedy equivalence search.

    dGRaSP: greedy relaxation of the sparsest permutation.

    ePC-stable: Peter-Clark stable.

    fInter-IAMB: interleaved incremental association Markov blanket.

    gHC: hill climbing.

    hTabu: tabu search.

    iMMHC: max-min hill climbing.

    jH2PC: hybrid of HPC and PC.

    kGES: greedy equivalence search.

    lBCCD: Bayesian Constraint-based Causal Discovery.

    mLiNGAM: linear non-Gaussian acyclic model.

    nGFCI: greedy fast causal inference.

    oMGM: mixed graphical model.

    pGraSP-FCI: greedy relaxation of the sparsest permutation fast causal inference.

    qARACNE: algorithm for the reconstruction of accurate cellular networks.

    riMIIC: interpretable multivariate information-based inductive causation.

    sGGES: greedy generalized equivalence search.

    tGAE: graph autoencoder.

    uModified PC: modified Peter-Clark algorithm.

    vAnchorFCI: anchor fast causal inference.

    wRDCM: regression dynamic causal modeling.

    xMVGC: multivariate Granger causality.

    yPCMCI: Peter-Clark momentary conditional independence.

    zVarLINGAM: vector autoregressive linear non-Gaussian acyclic model.

    aaCBAMN: cycle-breaking algorithm based on modified NOTEARS.

    abSEM: structural expectation-maximization.

    acRL: reinforcement learning.

    adIIE: inverse information entropy strength.

    aeCABG: isolated coronary artery bypass grafting.

    afAVR: isolated aortic valve replacement.

    agDYNOTEARS: dynamic noncombinatorial optimization via trace exponential and augmented lagrangian for structure learning.

    ahSAM: structural agnostic modeling.

    aiIAMB: incremental association Markov blanket.

    ajGS: grow-shrink.

    akPTSD: posttraumatic stress disorder.

    alANM: additive noise model.

    amPKCL: prior-knowledge-driven local causal structure learning.

    anT2DM: type 2 diabetes mellitus.

    aoGLL-PC: global-local learning Peter-Clark algorithm.

    apTPC: temporal Peter-Clark algorithm.

    aqADHD: attention-deficit/hyperactivity disorder.

    arPCDSD: probabilistic causal discovery in sequential datasets.

    asLSTM: long short-term memory.

    atMcDSL: multiple-cause discovery combined with structure learning.

    auSVM: support vector machine.

    avLGMM: latent growth mixture modeling.

    awHITON-PC: HITON-PC algorithm.

    axASD: autism spectrum disorder.

    ayNOTEARS: noncombinatorial optimization via trace exponential and augmented Lagrangian for structure learning.

    azCVD: cardiovascular disease.

    baAKD: acute kidney injury.

    bbItsPC: irregular time-series PC.

    bcICU: intensive care unit.

    bdPPCIT: predictive permutation conditional independence tests.

    beGIES: greedy interventional equivalence search.

    bfMCMC: Markov chain Monte Carlo.

    bgGGM: Gaussian graphical model.

    bhCOPD: chronic obstructive pulmonary disease.

    biCD-SF: causal discovery based on the streaming feature.

    bjCD-SU-SF: causal discovery with symmetrical uncertainty based on the streaming feature.

    bkLiNGAM: linear non-Gaussian acyclic model.

    Theme 2: Application Domains

    Examination of the 72 studies [8-14,18-82] indicated that causal discovery applications were dominated by clinical research (54/72, 75%), with public health comprising the remainder (18/72, 25%). Etiological investigation emerged as the predominant focus within clinical research, followed by predictive modeling, highlighting the method’s central role in elucidating disease mechanisms and enabling clinical prognosis. This application profile demonstrated the adaptability of causal discovery methodologies across the spectrum from individual-level clinical research to population health studies.

    Causal Discovery Methodological Approaches to Disease Categories

    Among the 54 clinical studies [8-14,18,19,21-23,25-29, 31-34,36,38,39,42-46,51-54,58-61,63-65,67,69-81], mental health represented the most extensively examined area (19/72, 26.4%) [13,22,23,27,29,32,45,46,51,53, 59,60,65,70,71, 75,79-81], with significant attention to schizophrenia (3/72, 4.2%) [13,32,51], posttraumatic stress disorder (3/72, 4.2%) [46,70,79], autism spectrum disorder (2/72, 2.8%) [22,80], and attention-deficit/hyperactivity disorder (2/72, 2.8%) [71,80]. Chronic disease investigations equally comprised 26.4% (19/72) of studies [9-12,14,19,21,26,28,31,36,38,42,52,69,72-74,76], spanning neurodegenerative diseases [10,19,26,72-74,76] (particularly Alzheimer disease [10,19,26,74] and amyotrophic lateral sclerosis [72,76]), malignancies [11,14,28,36,52], diabetes [21,42], and other chronic conditions [9,12,31,38,69]. Additional clinical applications included acute illnesses [8,33,78], pharmacological research [18,63], and other various specialized clinical scenarios [25,34,39,43,44,54,58,61, 64,67,77], collectively demonstrating the extensive penetration of causal discovery methodologies in clinical research. For details, refer to Table 2.

    Research Focus and Clinical Applications

    The 54 clinical studies [8-14,18,19,21-23,25-29,31-34,36,38,39, 42-46,51-54,58-61,63-65,67,69-81] showed a distinct concentration in research aims. Etiological exploration represented the predominant focus (28/72, 38.9%) [8-10,18,22,23,25,27,29,31,39,42-46,51,54,60,61,63,65,69-71,74,75,80], establishing causal discovery as a core methodology for investigating disease mechanisms. Predictive modeling constituted the second major application area (13/72, 18.1%) [19,21,26,33,34,58,64,72,73,76-79], while prognostic evaluation [13,14,28,52] and disease diagnosis [32,36,38,53] each accounted for 4 studies respectively. Collectively, these 4 domains accounted for 49 investigations (49/72, 68.1%), underscoring the method’s primary use in addressing fundamental clinical questions from causation to outcome assessment.

    For example, in a study by Zhang et al [63], causal discovery methods were applied to construct causal networks of COVID-related acute kidney injury using longitudinal electronic health record (EHR) data. Specifically, the FCI and GFCI algorithms were used to automatically identify the development of COVID-19–related acute kidney injury with minimal prior assumptions about pathway connectivity. The results provided a more refined understanding of how remdesivir and other risk factors contribute to acute kidney injury, identifying critical time points for potential intervention and offering valuable insights for clinical decision-making [63]. Similarly, Sokolova et al [80] demonstrated the application of the Bayesian constraint-based causal discovery algorithm to cross-sectional phenotypic data, which revealed directional relationships between attention-deficit/hyperactivity disorder and autism spectrum disorder symptom domains that traditional correlational methods failed to elucidate, generating novel hypotheses regarding the etiological mechanisms underlying their comorbidity [80].

    However, the limited applications in clinical decision support development (1/72, 1.4%) [67] and measurement scale implementation (2/72, 2.8%) [59,81] suggested that the translation of causal discovery into practical clinical tools remained underdeveloped, highlighting substantial potential for future research and implementation in real-world health care settings.

    Public Health Research Applications

    Public health domains were addressed in 25% (18/72) of studies [20,24,30,35,37,40,41,47-50,55-57,62,66,68,82], demonstrating the expanding relevance of causal discovery beyond clinical settings. Health behavior interventions constituted the largest subgroup (8/72, 11.1%) [30,35,37,40,41,49,56,62], primarily focusing on lifestyle quantification and health promotion. Disease management decision-making and health care optimization strategies were examined in 9.7% (7/72) of studies [20,24,47,50,66,68,82], encompassing social determinants of health, health care use patterns, and nursing home hospitalization reduction. The remaining studies addressed infectious disease transmission research [57,66] and life-course epidemiology [55]. These applications collectively demonstrated the methodological versatility of causal discovery in tackling population-level health challenges and informing public health policy.

    Theme 3: Challenges
    Common Observational Study Challenges

    Medical observational studies using causal discovery methods consistently faced several fundamental limitations. Unmeasured confounding represented the most prevalent challenge (32/72, 44.4%) [11,12,18,20,21,25-27,29,37,38,41,43,44,52-57,59,62, 63,65,66,70,71,73,75,82-84], potentially introducing spurious discoveries and biased effect estimates. Sample size limitations were reported in 29.2% (21/72) of studies [9,10,23,24,27,30,31,35,38,40,41,49,55,61,63,69,71,73,78,81,85], restricting statistical power and generalizability. Data quality issues emerged across multiple investigations, including substantial missing data and discretization-induced biases [10,12,30,33,47,49,54,77]. The absence of longitudinal designs in several studies further hindered the verification of temporal directionality in causal relationships [10,13,19,29,54,74,86].

    Causal Discovery-Specific Methodological Issues

    The application of causal discovery algorithms revealed distinct methodological challenges. Unvalidated algorithmic assumptions were identified in 18.1% (13/72) of studies [8,9,37,43,44,48,52,57,61,71,82,87,88], compromising validity when core assumptions such as causal sufficiency and the Markov property were violated. Limited cross-disease validation [21] and insufficient external validation [59] constrained the broader applicability of findings. Additionally, overreliance on prior knowledge introduced subjectivity into the discovery process, potentially influencing causal structure learning and interpretation [24,36,74,89,90].

    Proposed Methodological Improvements

    To address these challenges, future research should focus on the following. First, developing confounder-robust algorithms, as demonstrated by AnchorFCI in cardiometabolic disease research [31], and integrating domain knowledge in neurodegenerative studies [10]. Second, advancing longitudinal causal discovery frameworks for irregularly sampled temporal data in EHRs, as seen in nephrotoxicity and intensive care unit studies [63,64]. Third, enabling multimodal data integration strategies, such as coupling neuroimaging with EHRs or synthesizing wearable data with clinical assessments to explore lifestyle diseases [40,49,61]. Finally, incorporating artificial intelligence (AI) architectures like large language models, deep learning, and reinforcement learning to enhance clinical interpretability and model transferability across institutions [21,25,36,42,58,67].


    Principal Findings

    This scoping review represents the first systematic mapping of causal discovery algorithms in observational medical research. It synthesizes key methodological, translational, and implementation gaps across clinical and public health domains. Despite the field’s nascent stage, research momentum has accelerated markedly, reflecting growing translational potential. The scarcity of qualifying studies (72 of 2296 screened) underscores both the emergent nature of these methods and the novelty of this review. Our integration of fragmented evidence establishes a foundational framework for interdisciplinary refinement, advancing beyond traditional observational research limitations and supporting the causal discovery’s role in medical settings.

    Translational Gaps in Application Domains

    Clinical research concentrates on etiological exploration, leveraging causal graphs to deconstruct complex diseases (eg, mental health and chronic conditions). However, a severe imbalance exists between mechanism discovery and clinical decision support development (1/72, 1.4%), revealing a critical mechanism-application dissociation. This translational gap primarily stems from 4 interconnected barriers: most methods remain at the proof-of-concept stage without large-scale clinical validation, creating a fundamental trust deficit; existing tools lack seamless integration with clinical workflows and EHR systems; real-world data quality issues and computational demands hinder practical implementation; and crucially, statistically robust findings often fail to translate into clinically actionable recommendations that clinicians can readily understand and apply at the point of care. Public health applications focus on behavior interventions (eg, lifestyle-outcome linkages) and system optimization (eg, health care use patterns), but face domain-specific confounders (eg, policy shifts). Cross-domain translation requires context-adapted methodological frameworks.

    Methodological Challenges to Practical Translation

    Causal discovery in observational medical research remains dominated by constraint-based (eg, PC and FCI) and score-based methods (eg, GES and FGES), valued for their reliability in exploratory analysis and high-dimensional data, as well as their inherent interpretability; their graphical outputs and testable assumptions align well with clinical reasoning. However, their limitations in identifying causal directions, handling unmeasured confounding, and computational efficiency have spurred new approaches: continuous optimization (eg, NOTEARS) enhances scalability, although often at the cost of transparency; hybrid methods (eg, GFCI) improve robustness; functional models (eg, LiNGAM) enable direction identification; and time-series approaches (eg, PCMCI) capture dynamic causality. This interpretability-scalability trade-off critically guides model selection in clinical settings, where trust and actionable insight depend on understandable outputs. This evolution marks a transition toward more robust, scalable, and temporally-aware causal inference.

    Unmeasured confounding remains fundamental, distorting effect estimates and causal structures. Partial ancestral graph algorithms (eg, FCI and RFCI [91]) and negative control designs [92] offer partial solutions but lack universal robustness. Small-sample limitations undermine statistical power and generalizability, particularly in rare diseases or early-phase studies. Transfer learning [93], multimodal data fusion (eg, Cov-Pneum dataset [94]), and cross-disease validation mitigate data scarcity [95], while embedding expert knowledge (eg, in Bayesian networks) reduces model uncertainty. Beyond these challenges, translating causal discovery into practice requires concerted action on 2 fronts: establishing benchmark datasets and implementing a rigorous, explainable, multitiered validation framework. We strongly recommend the curation of medical benchmark datasets derived from gold-standard sources, such as large-scale RCTs or well-validated longitudinal cohorts, to provide a critical foundation for fairly evaluating and comparing algorithms. For instance, the study by Gururaghavendran and Murray [18] used a zero placebo effect as the benchmark to compare multiple causal discovery algorithms, demonstrating that guidance from expert insight and prior knowledge significantly enhances their performance.

    Explainable artificial intelligence techniques hold significant potential for improving the causal discovery process. Hasan and Gani [96] introduced a knowledge-guided causal AI system that integrates domain-specific prior knowledge as structural constraints with observational data to refine causal graph discovery. The results show that this approach also lowers computational demands while improving reliability [96]. One study introduced a new causal discovery approach, called REX, leveraging ML models coupled with interpretability techniques, showing REX’s effectiveness and robustness in accurately recovering true causal structures, along with its applicability to real-world problems [97]. In addition, moving beyond theoretical sensitivity analyses, a robust validation strategy must integrate internal validation (eg, using bootstrap to assess graph stability), external validation on independent datasets, and most importantly, rigorous evaluation of biological and clinical plausibility against established knowledge or new experimental findings. This comprehensive approach is essential for bridging the current trust gap and generating clinically actionable insights.

    Future Directions

    To propel the field beyond current methodological constraints and accelerate clinical translation, the following strategic advancements warrant prioritized investigation:

    1. Advancing temporal causal inference: integrate time-series data (eg, EHRs) with target trial emulation frameworks to resolve temporal ambiguity, mitigate time-varying confounding, and strengthen causal directionality inference [98].
    2. Enabling multimodal data integration: leverage causal ML methodologies to unify heterogeneous data sources (eg, EHRs, genomics, and behavioral data), thereby facilitating robust estimation of individualized treatment effects and bridging critical evidence gaps [10,36,99].
    3. Developing AI-enhanced causal discovery: synthesize deep learning (for high-dimensional feature extraction), reinforcement learning (for adaptive decision-making), and causal representation learning to enhance the accuracy, scalability, and interpretability of causal inference in complex biomedical systems [21,25,42,100].
    4. Accelerating clinical translation: it is imperative to establish reproducible benchmarking frameworks and standardized reporting guidelines. We encourage journals to adopt policies that actively promote code and data availability, thereby enhancing transparency and reproducibility. The integration of sensitivity analyses into routine methodological reporting should be considered essential for evaluating the robustness of inferred causal relationships. Furthermore, the development of cross-institutional validation protocols and standardized tools that facilitate the incorporation of clinical domain knowledge into causal discovery pipelines will be critical for fostering broader clinical adoption.
    5. Enhancing interpretability: improving the interpretability of causal discovery outputs and building clinical trust requires heightened methodological transparency and consistent reporting standards. The field should move toward developing domain‑specific reporting guidelines for causal discovery in medicine, which would detail the algorithms used, parameter settings, and evaluation metrics. Systematically embedding interpretability‑focused techniques, such as feature importance analysis or causal pathway explanation, can help bridge inferred causal structures with established clinical evidence.

    Ultimately, the real‑world impact of causal discovery hinges on its ability to deliver comprehensible and actionable insights that resonate with clinical researchers and practitioners. However, whether causal discovery can genuinely drive scientific progress, generate novel insights, and inform scholarly discourse in fields can ultimately be determined only through its repeated application in practice, coupled with rigorous and transparent reporting of the results [101].

    Strengths and Limitations

    This review provides a timely, comprehensive synthesis of causal discovery algorithms across diverse medical domains. Nevertheless, limitations exist: First, restricting the search to English-language publications may have excluded valuable non-English evidence. Second, our review reveals a pronounced geographic concentration of research, with the majority of studies originating from the United States and China. This likely reflects disparities in access to large-scale digital health infrastructure, advanced computational resources, and targeted research funding. Consequently, the current evidence base and its associated methodological trends may not be fully representative of global research activity or directly generalizable to health care settings in other regions. Third, the search was conducted within biomedical databases and relied on a core set of terms centered on “causal discovery.” While this approach efficiently captured studies self-identifying with this paradigm, it is possible that some relevant applications published under different terminology (eg, “Bayesian network structure learning”) in computational science venues were not retrieved. Finally, while we mapped application breadth, a formal assessment of reporting quality or bias risk was not conducted; future work should implement standardized appraisal tools to systematically evaluate methodological rigor.

    Conclusions

    Causal discovery is emerging as a pivotal approach in medical observational research, offering transformative potential for etiological exploration and clinical decision-making. However, critical barriers impede clinical translation, notably the absence of standardized validation frameworks and persistent confounding. Future progress necessitates rigorous evaluation standards and AI-causal methodology integration to enable robust, generalizable real-world decision support.

    Acknowledgments

    In accordance with journal policy, we confirm that no generative artificial intelligence tools were used in the writing of this manuscript or in the generation of its scientific content. Artificial intelligence–assisted tools were used solely for minor language polishing and proofreading. All conceptualization, analysis, intellectual interpretation, and authorship of the final manuscript are solely attributable to the human authors.

    Data Availability

    All data generated or analyzed during this study are included in this published article and Multimedia Appendices 1-2.

    Funding

    This study was supported by the National Natural Science Foundation of China (grant numbers 82360667 and 82160645), the Natural Science Foundation of Jiangxi Province (grant number 20212BAB206091), and the National Undergraduate Training Program for Innovation and Entrepreneurship (grant number 202510403077).

    Authors' Contributions

    JK and JL were involved in the concept and design of the review. ZL, TL, HM, JM, XY, and ZH made major contributions to literature search, data collection, data analysis, and data interpretation. ZL and JK wrote and edited the manuscript. All authors read and approved the final manuscript.

    JK and JL are co-corresponding authors of this paper. JL can be reached at ljk545464@163.com.

    Conflicts of Interest

    None declared.

    Multimedia Appendix 1

    PRISMA-ScR checklist.

    DOCX File , 72 KB
    Multimedia Appendix 2

    Queries used for the database search.

    DOCX File , 19 KB
    1. Hernán MA. The c-word: scientific euphemisms do not improve causal inference from observational data. Am J Public Health. 2018;108(5):616-619. [CrossRef] [Medline]
    2. Nogueira AR, Pugnana A, Ruggieri S, Pedreschi D, Gama J. Methods and tools for causal discovery and causal inference. WIREs Data Min Knowl Discov. 2022;12(2):e1449. [CrossRef]
    3. Spirtes P, Zhang K. Causal discovery and inference: concepts and recent methodological advances. Appl Inform (Berl). 2016;3:3. [FREE Full text] [CrossRef] [Medline]
    4. Peters J, Janzing D, Schölkopf B. Elements of Causal Inference: Foundations and Learning Algorithms. In: Computer Reviews. Cambridge. The MIT Press; 2017.
    5. Spirtes P, Glymour C, Scheines R. Causation, Prediction, and Search. Cambridge, MA. MIT Press; 2000.
    6. Kaufman H, Rappoport N, Gilad A, Linial M. Advancing causal inference in medicine using biobank data. J Biomed Inform. 2025;171:104903. [CrossRef] [Medline]
    7. Bothwell LE, Greene JA, Podolsky SH, Jones DS. Assessing the gold standard--lessons from the history of RCTs. N Engl J Med. 2016;374(22):2175-2181. [CrossRef] [Medline]
    8. Petrungaro B, Kitson NK, Constantinou AC. Investigating potential causes of sepsis with bayesian network structure learning. Appl Intell. 2025;55(7):17. [CrossRef]
    9. Penson A, Bucur IG, Walraven I, Grootenhuis MA, Maurice-Stam H, van der Heiden-van der Loo M, et al. et al. Structural equation modeling to explore putative causal factors for chronic fatigue in childhood cancer survivors: a DCCSS LATER study. J Cancer Surviv. 2025. [CrossRef] [Medline]
    10. Yu X, Lophatananon A, Holmes V, Muir KR, Guo H. Investigating causal networks of dementia using causal discovery and natural language processing models. NPJ Dement. 2025;1(1):4. [CrossRef] [Medline]
    11. Bernasconi A, Zanga A, Lucas PJF, Scutari M, Trama A, Stella F. A causal network model to estimate the cardiotoxic effect of oncological treatments in young breast cancer survivors. Prog Artif Intell. 2024. [CrossRef]
    12. Ganopoulou M, Kangelidis I, Sianos G, Angelis L. Causal models for the result of percutaneous coronary intervention in coronary chronic total occlusions. Applied Sciences. 2021;11(19):9258. [CrossRef]
    13. Giuliani L, Sanmarchi F, Mucci A, Rucci P, Caporusso E, Bucci P, et al. et al. Investigating the causal pathways among psychopathological variables, cognitive impairment, and real-life functioning in people with schizophrenia. Schizophrenia (Heidelb). 2025;11(1):1. [FREE Full text] [CrossRef] [Medline]
    14. Ribeiro-Dantas MDC, Li H, Cabeli V, Dupuis L, Simon F, Hettal L, et al. et al. Learning interpretable causal networks from very large datasets, application to 400,000 medical records of breast cancer patients. iScience. 2024;27(5):109736. [FREE Full text] [CrossRef] [Medline]
    15. Arksey H, O'Malley L. Scoping studies: towards a methodological framework. Int J Soc Res Methodol. 2005;8(1):19-32. [CrossRef]
    16. Tricco AC, Lillie E, Zarin W, O'Brien KK, Colquhoun H, Levac D, et al. et al. PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation. Ann Intern Med. 2018;169(7):467-473. [FREE Full text] [CrossRef] [Medline]
    17. Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan-a web and mobile app for systematic reviews. Syst Rev. 2016;5(1):210. [FREE Full text] [CrossRef] [Medline]
    18. Gururaghavendran R, Murray EJ. Can algorithms replace expert knowledge for causal inference? A case study on novice use of causal discovery. Am J Epidemiol. 2025;194(5):1399-1409. [CrossRef] [Medline]
    19. Li C, Wang S, Xia Y, Shi F, Tang L, Yang Q, et al. et al. Risk factors and predictive models in the progression from MCI to alzheimer's disease. Neuroscience. 2025;565:312-319. [CrossRef] [Medline]
    20. Korvink M, Biondolillo M, Van Dijk JW, Banerjee A, Simenz C, Nelson D. Detection of potential causal pathways among social determinants of health: a data-informed framework. Soc Sci Med. 2025;373:118025. [CrossRef] [Medline]
    21. Noh MJ, Kim YS. Diabetes prediction through linkage of causal discovery and inference model with machine learning models. Biomedicines. 2025;13(1):124. [FREE Full text] [CrossRef] [Medline]
    22. Tseng A, Francis SM, Rawls E, Conelea C, Grissom NM, Kummerfeld E, et al. et al. Integrating causal discovery and clinically-relevant insights to explore directional relationships between autistic features, sex at birth, and cognitive abilities. Psychol Med. 2025;55:e89. [CrossRef] [Medline]
    23. Bronstein MV, Kummerfeld E, Bearden CE, Cornblatt BA, Walker EF, Woods SW, et al. et al. Delineating empirically plausible causal pathways to suicidality among people at clinical high risk for psychosis. J Psychopathol Clin Sci. 2025;134(3):239-250. [CrossRef] [Medline]
    24. Colineaux H, Lepage B, Chauvin P, Dimeglio C, Delpierre C, Lefèvre T. Contribution of structure learning algorithms in social epidemiology: application to real-world data. Int J Environ Res Public Health. 2025;22(3):348. [FREE Full text] [CrossRef] [Medline]
    25. Fay L, Hepp T, Winkelmann MT, Peters A, Heier M, Niendorf T, et al. et al. Determinants of ascending aortic morphology: cross-sectional deep learning-based analysis on 25 073 non-contrast-enhanced NAKO MRI studies. Eur Heart J Cardiovasc Imaging. 2025;26(5):895-907. [CrossRef] [Medline]
    26. Fu M, Sankararaman S, Vossel KA, Chang TS. Identifying common disease trajectories of progressive supranuclear palsy with electronic health records. Mov Disord Clin Pract. 2025;12(10):1528-1538. [CrossRef] [Medline]
    27. Verveen A, Nugroho FA, Bucur IG, Wynberg E, van Willigen HDG, Davidovich U, et al. et al. Long-term sequelae of SARS-CoV-2 two years following infection: exploring the interplay of biological, psychological, and social factors. Psychol Med. 2024;54(15):1-11. [CrossRef] [Medline]
    28. Ren S, Beeche CA, Iyer K, Shi Z, Auster Q, Hawkins JM, et al. et al. Graphical modeling of causal factors associated with the postoperative survival of esophageal cancer subjects. Med Phys. 2024;51(3):1997-2006. [CrossRef] [Medline]
    29. Guo X, Wang L, Li Z, Feng Z, Lu L, Jiang L, et al. et al. Factors and pathways of non-suicidal self-injury in children: insights from computational causal analysis. Front Public Health. 2024;12:1305746. [FREE Full text] [CrossRef] [Medline]
    30. Foraita R, Witte J, Börnhorst C, Gwozdz W, Pala V, Lissner L, et al. et al. A longitudinal causal graph analysis investigating modifiable risk factors and obesity in a European cohort of children and adolescents. Sci Rep. 2024;14(1):6822. [FREE Full text] [CrossRef] [Medline]
    31. Ribeiro AH, Crnkovic M, Pereira JL, Fisberg RM, Sarti FM, Rogero MM, et al. et al. AnchorFCI: harnessing genetic anchors for enhanced causal discovery of cardiometabolic disease pathways. Front Genet. 2024;15:1436947. [FREE Full text] [CrossRef] [Medline]
    32. Sabzevar S, Masoudnia S, Araabi BN, Nazemzadeh M, Tavakoli H. Enhanced diagnosis of schizophrenia using integrated node embedding and causal discovery on imbalanced fMRI data. 2024. Presented at: Proceedings of the 10th International Conference on Signal Processing and Intelligent Systems (ICSPIS) 2024; December 25-26, 2024:234-239; Shahrood, Iran. [CrossRef]
    33. Zhang M, Zhang X, Dai M, Wu L, Liu K, Wang H, et al. et al. Development and validation of a multi-causal investigation and discovery framework for knowledge harmonization (MINDMerge): A case study with acute kidney injury risk factor discovery using electronic medical records. Int J Med Inform. 2024;191:105588. [CrossRef] [Medline]
    34. Li Y, Scheel-Sailer A, Riener R, Paez-Granados D. Mixed-variable graphical modeling framework towards risk prediction of hospital-acquired pressure injury in spinal cord injury individuals. Sci Rep. 2024;14(1):25067. [FREE Full text] [CrossRef] [Medline]
    35. Kazemi K, Ryhta I, Azimi I, Niela-Vilen H, Axelin A, Rahmani AM, et al. et al. Impact of physical activity on quality of life during pregnancy: a causal ML approach. Annu Int Conf IEEE Eng Med Biol Soc. 2024;2024:1-6. [CrossRef] [Medline]
    36. Naik N, Khandelwal A, Joshi M, Atre M, Wright H, Kannan K. Applying large language models for causal structure learning in non small cell lung cancer. 2024. Presented at: Proceedings of the IEEE 12th International Conference on Healthcare Informatics (ICHI); June 3-6, 2024:688-693; Orlando, FL. [CrossRef]
    37. McCormick KM, Ribeiro Santiago PH, Jamieson L. The impact of COVID-19 on the oral health self-care practices of Australian adults. J Public Health (Berl.). Jun 05, 2024;34(3):525-534. [CrossRef]
    38. Zeng W, Thatayatikom A, Winn N, Lovelace TC, Bhattacharyya I, Schrepfer T, et al. et al. The Florida scoring system for stratifying children with suspected Sjögren's disease: a cross-sectional machine learning study. Lancet Rheumatol. 2024;6(5):e279-e290. [CrossRef] [Medline]
    39. Tosaki T, Uchino E, Harada Y, Sakuragi M, Koyanagi Y, Okajima S. Subgrouping causal networks of disease onset in large-scale health and medical data using supercomputer Fugaku. 2024. Presented at: Proceedings of the IEEE International Conference on Bioinformatics and Biomedicine (BIBM); December 3-6, 2024:4384-4391; Lisbon, Portugal. [CrossRef]
    40. Zhang J, Cong R, Deng O, Li Y, Lam K, Jin Q. Analyzing lifestyle and behavior with causal discovery in health data from wearable devices and self-assessments. 2024. Presented at: Proceedings of the IEEE International Conference on E-Health Networking, Application and Services, HealthCom 2024; November 18-20, 2024; Nara. [CrossRef]
    41. Gąsior JS, Młyńczak M, Williams CA, Popłonyk A, Kowalska D, Giezek P, et al. et al. The discovery of a data-driven causal diagram of sport participation in children and adolescents with heart disease: a pilot study. Front Cardiovasc Med. 2023;10:1247122. [FREE Full text] [CrossRef] [Medline]
    42. Gao XE, Hu JG, Chen B, Wang YM, Zhou SB. Causal discovery approach with reinforcement learning for risk factors of type II diabetes mellitus. BMC Bioinformatics. 2023;24(1):296. [FREE Full text] [CrossRef] [Medline]
    43. Lee JJR, Srinivasan R, Ong CS, Alejo D, Schena S, Shpitser I, et al. et al. Causal determinants of postoperative length of stay in cardiac surgery using causal graphical learning. J Thorac Cardiovasc Surg. 2023;166(5):e446-e462. [FREE Full text] [CrossRef] [Medline]
    44. Thomas A, Niranjan M, Legg J. Causal analysis of physiological sleep data using granger causality and score-based structure learning. Sensors (Basel). 2023;23(23):9455. [FREE Full text] [CrossRef] [Medline]
    45. Vowels LM, Vowels MJ, Carnelley KB, Millings A, Gibson-Miller J. Toward a causal link between attachment styles and mental health during the COVID-19 pandemic. Br J Clin Psychol. 2023;62(3):605-620. [FREE Full text] [CrossRef] [Medline]
    46. Pierce B, Kirsh T, Ferguson AR, Neylan TC, Ma S, Kummerfeld E, et al. et al. Causal discovery replicates symptomatic and functional interrelations of posttraumatic stress across five patient populations. Front Psychiatry. 2022;13:1018111. [FREE Full text] [CrossRef] [Medline]
    47. Lee K, Kummerfeld E, Robinson E, Anderson L, Rantz M. Data-driven analytics to discover APRN's impact on nursing home hospitalization: causal discovery analysis. J Am Med Dir Assoc. 2023;24(11):1746-1754. [CrossRef] [Medline]
    48. Andersen TO, Sejling C, Jensen AK, Drews HJ, Ritz B, Varga TV, et al. et al. Nighttime smartphone use, sleep quality, and mental health: investigating a complex relationship. Sleep. 2023;46(12):zsad256. [CrossRef] [Medline]
    49. Li YX, Deng O, Ogihara A, Nishimura S, Jin Q. Causal discovery of health features from wearable device and traditional Chinese medicine diagnosis data. 2023. Presented at: Proceedings of the 25th International Conference on Human-Computer Interaction, HCI International 2023; July 23, 2023:556-569; Copenhagen, Denmark. [CrossRef]
    50. Park CH, Kwon J, Lee JT, Ahn S. Impact of criterion versus norm-referenced assessment on the quality of life in Korean medical students. J Korean Med Sci. 2023;38(17):e133. [FREE Full text] [CrossRef] [Medline]
    51. Miley K, Meyer-Kalos P, Ma S, Bond DJ, Kummerfeld E, Vinogradov S. Causal pathways to social and occupational functioning in the first episode of schizophrenia: uncovering unmet treatment needs. Psychol Med. 2023;53(5):2041-2049. [FREE Full text] [CrossRef] [Medline]
    52. Li X, Qian XL, Liang LT, Kong LJ, Dong QL, Chen JJ. Causally-aware intraoperative imputation for overall survival time prediction. 2023. Presented at: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); June 17-24, 2023:15681-15690; Vancouver, BC. [CrossRef]
    53. Nagaraj S, Goodday S, Hartvigsen T, Boch A, Garg K, Gowda S, et al. Dissecting the heterogeneity of "in the wild" stress from multimodal sensor data. NPJ Digit Med. Dec 20, 2023;6(1):237. [CrossRef] [Medline]
    54. Vigneshwaran V, Wilms M, Forkert ND. The causal link between cardiometabolic risk factors and gray matter atrophy: an exploratory study. Heliyon. 2023;9(11):e21567. [FREE Full text] [CrossRef] [Medline]
    55. Petersen AH, Ekstrøm CT, Spirtes P, Osler M. Constructing causal life-course models: comparative study of data-driven and theory-driven approaches. Am J Epidemiol. 2023;192(11):1917-1927. [FREE Full text] [CrossRef] [Medline]
    56. Fung H, Sgaier SK, Huang VS. Discovery of interconnected causal drivers of COVID-19 vaccination intentions in the US using a causal Bayesian network. Sci Rep. May 16, 2023;13(1):6988. [CrossRef] [Medline]
    57. Bao L, Li C, Li R, Yang S. Causal structural learning on MPHIA individual dataset. J Am Stat Assoc. 2022;117(540):1642-1655. [FREE Full text] [CrossRef] [Medline]
    58. Singh H, Mhasawade V, Chunara R. Generalizability challenges of mortality risk prediction models: a retrospective analysis on a multi-center database. PLOS Digit Health. 2022;1(4):e0000023. [FREE Full text] [CrossRef] [Medline]
    59. Langwerden RJ, Van der Heijden PT, Claassen T, Derksen JJL, Egger JIM. The structure of dimensions of psychopathology in normative and clinical samples: applying causal discovery to MMPI-2-RF scales to investigate clustering of psychopathology spectra and -factors. Front Psychiatry. 2022;13:1026900. [FREE Full text] [CrossRef] [Medline]
    60. Bronstein MV, Everaert J, Kummerfeld E, Haynos AF, Vinogradov S. Biased and inflexible interpretations of ambiguous social situations: associations with eating disorder symptoms and socioemotional functioning. Int J Eat Disord. 2022;55(4):518-529. [FREE Full text] [CrossRef] [Medline]
    61. Shen X, Raghavan S, Przybelski SA, Lesnick TG, Ma S, Reid RI, et al. et al. Causal structure discovery identifies risk factors and early brain markers related to evolution of white matter hyperintensities. Neuroimage Clin. 2022;35:103077. [FREE Full text] [CrossRef] [Medline]
    62. Mouches P, Wilms M, Bannister JJ, Aulakh A, Langner S, Forkert ND. An exploratory causal analysis of the relationships between the brain age gap and cardiovascular risk factors. Front Aging Neurosci. 2022;14:941864. [FREE Full text] [CrossRef] [Medline]
    63. Zhang J, Kummerfield E, Hultman G, Drawz PE, Adam TJ, Simon G, et al. et al. Application of causal discovery algorithms in studying the nephrotoxicity of remdesivir using longitudinal data from the EHR. AMIA Annu Symp Proc. 2022;2022:1227-1236. [FREE Full text] [Medline]
    64. Nogueira AR, Ferreira CA, Gama J. Temporal nodes causal discovery for in intensive care unit survival analysis. 2022. Presented at: Proceedings of the EPIA Conference on Artificial Intelligence; August 31, 2022:587-598; Lisbon, Portugal. [CrossRef]
    65. Rawls E, Kummerfeld E, Zilverstand A. An integrated multimodal model of alcohol use disorder generated by data-driven causal discovery analysis. Commun Biol. 2021;4(1):435. [FREE Full text] [CrossRef] [Medline]
    66. Li Z, Xu T, Zhang K, Deng HW, Boerwinkle E, Xiong M. Causal analysis of health interventions and environments for influencing the spread of COVID-19 in the United States of America. Front Appl Math Stat. 2021;6:611805. [CrossRef]
    67. Shen X, Ma S, Vemuri P, Castro MR, Caraballo PJ, Simon GJ. A novel method for causal structure discovery from EHR data and its application to type-2 diabetes mellitus. Sci Rep. 2021;11(1):21025. [FREE Full text] [CrossRef] [Medline]
    68. Petersen AH, Osler M, Ekstrøm CT. Data-driven model building for life-course epidemiology. Am J Epidemiol. 2021;190(9):1898-1907. [CrossRef] [Medline]
    69. Wang W, Hu G, Yuan B, Ye S, Chen C, Cui Y, et al. et al. Prior-knowledge-driven local causal structure learning and its application on causal discovery between type 2 diabetes and bone mineral density. IEEE Access. 2020;8:108798-108810. [CrossRef]
    70. Saxe GN, Ma S, Morales LJ, Galatzer-Levy IR, Aliferis C, Marmar CR. Computational causal discovery for post-traumatic stress in police officers. Transl Psychiatry. 2020;10(1):233. [FREE Full text] [CrossRef] [Medline]
    71. Schoenmacker GH, Groenman AP, Sokolova E, Oosterlaan J, Rommelse N, Roeyers H, et al. et al. Role of conduct problems in the relation between attention-deficit hyperactivity disorder, substance use, and gaming. Eur Neuropsychopharmacol. 2020;30:102-113. [CrossRef] [Medline]
    72. Ahangaran M, Jahed-Motlagh MR, Minaei-Bidgoli B. A novel method for predicting the progression rate of ALS disease based on automatic generation of probabilistic causal chains. Artif Intell Med. 2020;107:101879. [CrossRef] [Medline]
    73. Wang Y, You L, Chyr J, Lan L, Zhao W, Zhou Y, et al. et al. Causal discovery in radiographic markers of knee osteoarthritis and prediction for knee osteoarthritis severity with attention-long short-term memory. Front Public Health. 2020;8:604654. [FREE Full text] [CrossRef] [Medline]
    74. Shen X, Ma S, Vemuri P, Simon G, Alzheimer’s Disease Neuroimaging Initiative. Challenges and opportunities with causal discovery algorithms: application to Alzheimer's pathophysiology. Sci Rep. 2020;10(1):2975. [FREE Full text] [CrossRef] [Medline]
    75. Itahashi T, Okada N, Ando S, Yamasaki S, Koshiyama D, Morita K, et al. et al. Functional connectomes linking child-parent relationships with psychological problems in adolescence. Neuroimage. 2020;219:117013. [FREE Full text] [CrossRef] [Medline]
    76. Ahangaran M, Jahed-Motlagh MR, Minaei-Bidgoli B. Causal discovery from sequential data in ALS disease based on entropy criteria. J Biomed Inform. 2019;89:41-55. [FREE Full text] [CrossRef] [Medline]
    77. Yang J, Li N, Fang S, Yu K, Chen Y. Semantic features prediction for pulmonary nodule diagnosis based on online streaming feature selection. IEEE Access. 2019;7:61121-61135. [CrossRef]
    78. Chen W, Hu Y, Zhang X, Wu L, Liu K, He J, et al. et al. Causal risk factor discovery for severe acute kidney injury using electronic health records. BMC Med Inform Decis Mak. 2018;18(Suppl 1):13. [FREE Full text] [CrossRef] [Medline]
    79. Galatzer-Levy IR, Ma S, Statnikov A, Yehuda R, Shalev AY. Utilization of machine learning for prediction of post-traumatic stress: a re-examination of cortisol in the prediction and pathways to non-remitting PTSD. Transl Psychiatry. 2017;7(3):e0. [FREE Full text] [CrossRef] [Medline]
    80. Sokolova E, Oerlemans AM, Rommelse NN, Groot P, Hartman CA, Glennon JC, et al. et al. A causal and mediation analysis of the comorbidity between attention deficit hyperactivity disorder (ADHD) and autism spectrum disorder (ASD). J Autism Dev Disord. 2017;47(6):1595-1604. [FREE Full text] [CrossRef] [Medline]
    81. van Dijk MJAM, Claassen T, Suwartono C, van der Veld WM, van der Heijden PT, Hendriks MPH. Evaluating WAIS-IV structure through a different psychometric lens: structural causal model discovery as an alternative to confirmatory factor analysis. Clin Neuropsychol. 2017;31(6-7):1141-1154. [CrossRef] [Medline]
    82. Ribeiro Santiago PH, Smithers L, Townsend M, Quintero A, Sawyer A, Soares G, et al. et al. The longitudinal network of peer problems and emotional symptoms among Australian adolescents: Bayesian structure learning of directed acyclic graphs. Dev Psychol. 2025;61(8):1479-1494. [CrossRef] [Medline]
    83. Cheek C, Zheng H, Hallstrom BR, Hughes RE. Application of a causal discovery algorithm to the analysis of arthroplasty registry data. Biomed Eng Comput Biol. 2018;9:1179597218756896. [FREE Full text] [CrossRef] [Medline]
    84. Bilgel F, Karahasan BC. Understanding covid-19 mobility through human capital: a unified causal framework. Comput Econ. 2023:1-41. [FREE Full text] [CrossRef] [Medline]
    85. Młyńczak M, Krysztofiak H. Discovery of causal paths in cardiorespiratory parameters: a time-independent approach in elite athletes. Front Physiol. 2018;9:1455. [FREE Full text] [CrossRef] [Medline]
    86. Constantinou A, Kitson NK, Liu Y, Chobtham K, Amirkhizi AH, Nanavati PA, et al. et al. Open problems in causal structure learning: a case study of COVID-19 in the UK. Expert Syst Appl. 2023;234:121069. [CrossRef]
    87. Huie JR, Vashisht R, Galivanche A, Hadjadj C, Morshed S, Butte AJ, et al. et al. Toward a causal model of chronic back pain: challenges and opportunities. Front Comput Neurosci. 2022;16:1017412. [FREE Full text] [CrossRef] [Medline]
    88. Anderson LM, Lim KO, Kummerfeld E, Crosby RD, Crow SJ, Engel SG, et al. et al. Causal discovery analysis: a promising tool in advancing precision medicine for eating disorders. Int J Eat Disord. 2023;56(11):2012-2021. [CrossRef] [Medline]
    89. Tao Z, Chi M, Chen L, Ban T, Tu Q, Gao F, et al. et al. Clinical causal analysis via iterative active structure learning. Memetic Comp. 2025;17(1):13. [CrossRef]
    90. Jin W, Ni Y, Spence AB, Rubin LH, Xu Y. Directed cyclic graphs for simultaneous discovery of time-lagged and instantaneous causality from longitudinal data using instrumental variables. J Mach Learn Res. 2025;26:22. [Medline]
    91. Colombo D, Maathuis MH, Kalisch M, Richardson TS. Learning high-dimensional directed acyclic graphs with latent and selection variables. Ann Statist. 2012;40(1):294-321. [CrossRef]
    92. Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology. 2010;21(3):383-388. [FREE Full text] [CrossRef] [Medline]
    93. Taghados Z, Azimifar Z, Monsefi M, Jahromi MA. CausalCervixNet: convolutional neural networks with causal insight (CICNN) in cervical cancer cell classification-leveraging deep learning models for enhanced diagnostic accuracy. BMC Cancer. 2025;25(1):607. [FREE Full text] [CrossRef] [Medline]
    94. Agarwal S, Arya KV, Meena YK. MultiFusionNet: multilayer multimodal fusion of deep neural networks for chest X-ray image classification. Soft Comput. 2024;28(19):11535-11551. [CrossRef]
    95. Tang X, Guo R, Mo Z, Fu W, Qian X. Causality-driven candidate identification for reliable DNA methylation biomarker discovery. Nat Commun. 2025;16(1):680. [FREE Full text] [CrossRef] [Medline]
    96. Hasan U, Gani M. A knowledge-guided framework to enhance causal reasoning and human-AI collaboration. IEEE; 2025. Presented at: Proceedings of the IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops); March 7-21, 2025; Washington DC, DC, USA. [CrossRef]
    97. Renero J, Maestre R, Ochoa I. ReX: causal discovery based on machine learning and explainability techniques. Pattern Recognition. 2026;172:112491. [CrossRef]
    98. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758-764. [FREE Full text] [CrossRef] [Medline]
    99. Weberpals J, Feuerriegel S, van der Schaar M, Kehl KL. Opportunities for causal machine learning in precision oncology. NEJM AI. 2025;2(8):AIp2500277. [CrossRef]
    100. Jiao L, Wang Y, Liu X, Li L, Liu F, Ma W, et al. et al. Causal inference meets deep learning: a comprehensive survey. Research (Wash D C). 2024;7:0467. [FREE Full text] [CrossRef] [Medline]
    101. Petersen AH, Ekstrøm CT, Spirtes P, Osler M. Causal discovery and epidemiology: a potential for synergy. Am J Epidemiol. 2024;193(10):1341-1342. [CrossRef] [Medline]

    ‎
    AI: artificial intelligence
    EHR: electronic health record
    FCI: fast causal inference
    GFCI: greedy fast causal inference
    ML: machine learning
    PC: Peter-Clark algorithm
    PRISMA-ScR: Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews
    RCT: randomized controlled trial

    Edited by A Coristine; submitted 15.Aug.2025; peer-reviewed by JMI Arockiasamy, C Ma; comments to author 26.Oct.2025; revised version received 16.Feb.2026; accepted 19.Feb.2026; published 13.Mar.2026.

    Copyright

    ©Zuting Liu, Tian Luo, Hailin Ma, Jiali Mo, Xia Yang, Zhenglong Huang, Jingkun Li, Jie Kuang. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 13.Mar.2026.

    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 use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on https://medinform.jmir.org/, as well as this copyright and license information must be included.