Pith. sign in

REVIEW 4 major objections 5 minor 53 references

Cryptogenic stroke and migraine: using probabilistic independence and machine learning to uncover latent sources of disease from the electronic health record

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that latent causes of cryptogenic stroke in migraine patients can be recovered from electronic health record data as probabilistically independent sources, so that SHAP scores on those sources estimate causal effects…

desk verdict A promising EHR hypothesis-generation result undermined by a non-invertible mixing matrix that breaks the causal interpretation. read the letter →

arxiv 2505.04631 v2 pith:LF6SQEQR submitted 2025-04-22 stat.AP cs.LG

Joshua W. Betts , John M. Still , Thomas A. Lasko This is my paper · ORCID
classification stat.APcs.LG MSC 62P1062H25
keywords cryptogenicstrokemigraineelectronichealthrecordsindependentcomponentanalysislatentcausalsourcesSHAPvaluesLiNGAMriskprediction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to show that latent causes of cryptogenic stroke in migraine patients can be recovered from messy electronic health record data without human-curated features. It uses independent component analysis on 9000 clinical variables to extract 2000 probabilistically independent sources, then treats those sources as the independent error terms of a causal model, so that a random forest's SHAP scores on those sources estimate causal effects. On a held-out test set the model achieves an AUROC of 0.782 for 10-year cryptogenic stroke risk, with cross-validated AUROC 0.771. The top predictive sources are mostly protective medication signatures, and one source tied to allergic rhinitis is consistently associated with increased risk, which the authors interpret as a candidate root cause. If this interpretation is right, the method turns routine EHR data into a hypothesis generator for disease causes.

What carries the argument

The central object is the ICA decomposition $X = AS$, where $X$ is the stacked EHR data matrix, $S$ holds the 2000 probabilistically independent source expressions, and $A$ is the mixing matrix whose columns are clinical signatures. The machinery links these sources to the independent error terms of a LiNGAM structural equation model, then uses SHAP values computed on a random forest trained on $S$ to quantify each source's causal contribution to the cryptogenic stroke label.

What would settle it

Feed the same pipeline synthetic EHR data generated from a known causal graph with known cryptogenic-stroke incidence: if the top-SHAP sources do not match the planted causes, the causal interpretation fails. Clinically, a prospective study or natural experiment showing that treating allergic rhinitis or removing decongestants does not change cryptogenic stroke incidence would refute the allergic-rhinitis-as-cause claim.

Watch

Extended reading notes

Core claim

Under the LiNGAM assumptions, the paper claims that the 2000 probabilistically independent sources recovered by ICA from EHR data correspond to the exogenous error terms of the structural causal model that generated the observations. Because causal relations are transitive, these error terms exert a causal relationship onto the final cryptogenic stroke label, and the SHAP value of each source is therefore a quantitative estimate of its causal effect on a patient's predicted risk. Applying this to 72,876 migraine patients, the model found that the largest protective effects come from common preventive medications, while a source characterized by allergic rhinitis consistently increases cryptogenic stroke risk, albeit by a small amount.

Load-bearing premise

The load-bearing premise is that the EHR observation network is linear, non-Gaussian, acyclic, and free of unmeasured common causes outside the 2000 extracted sources, so that each ICA source really is an independent exogenous cause rather than an algebraic artifact.

Editorial extensions

If this is right

  • If the causal reading holds, the highest-ranked protective sources imply that common preventive medications—antiplatelet, lipid-lowering, and antihypertensive regimens—are the largest modifiable factors reducing cryptogenic stroke risk in migraine patients.
  • The allergic-rhinitis source, though small in effect, is consistently positive, which the paper identifies as a candidate root cause that should be investigated prospectively.
  • Patient-level SHAP waterfall explanations become actionable: a clinician could see which latent sources push a given patient's 10-year stroke risk up or down.
  • The same ICA-plus-SHAP pipeline can be applied to other disease labels, turning the EHR into a root-cause hypothesis generator.
  • The apparent protective effect of prior stroke is a sampling artifact of excluding post-stroke windows, so recurrent-stroke risk is not captured by this model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next experiment the paper does not run: compare cryptogenic stroke incidence in migraine patients with and without allergy immunotherapy or decongestant use; a difference would strengthen the allergic-rhinitis causal claim, while no difference would weaken it.
  • Because the 2000-source ICA was truncated by memory, some discovered signatures may merge several distinct causes; re-running with more sources could split the allergic-rhinitis signature into separate inflammation and decongestant components.
  • The causal interpretation depends on EHR documentation capturing the real causes; patients who use over-the-counter decongestants without recording them would bias the allergic-rhinitis source, a limitation the paper itself notes.
  • If the LiNGAM assumptions are violated in real EHR data, the SHAP ranks still describe predictive associations and could remain useful for risk stratification, but not for causal intervention.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes an unsupervised ICA-based decomposition of longitudinal EHR curves from roughly 310,000 neurology patients into 2000 latent 'disease sources', then projects a migraine cohort (N=72,876; 1,670 cryptogenic stroke cases) onto these sources and trains a random forest to predict 10-year cryptogenic stroke risk. The model achieves a test AUROC of 0.782. The authors interpret the ICA sources as the exogenous error terms of a LiNGAM structural causal model and interpret the SHAP values on the source expressions as quantitative causal effects, concluding that medication-related sources are protective and that an allergic rhinitis-related source is a potential cause of cryptogenic stroke in migraine patients. A secondary model for general ischemic stroke is used for comparison.

Significance. If the causal interpretation were valid, the paper would offer a scalable method for generating hypotheses about root causes of heterogeneous diseases from EHR data, and the specific findings would merit clinical follow-up. The study has genuine strengths: a large real-world cohort, a prospective sampling window relative to the stroke event, a described end-to-end pipeline, and a held-out test set. However, the central claim is not supported. The mixing matrix is non-square and non-invertible, the LiNGAM identifiability conditions are neither met nor tested, SHAP values are not causal effect estimators without additional assumptions, and the predictive evaluation lacks confidence intervals, calibration, and baseline comparators. As presented, the paper is best read as an exploratory predictive and associational study rather than a causal analysis, and the causal language would need to be substantially revised or removed.

major comments (4)
  1. [Model creation, S = A^{-1}E] The projection step is mathematically invalid as stated. The mixing matrix A is m×k with m=9000 and k=2000, so A^{-1} does not exist. If the authors instead used the FastICA unmixing matrix W or a pseudo-inverse, the resulting source expressions S_eval are a projection onto the learned subspace, not the exact exogenous error terms of a structural causal model. Since the label may depend on the omitted m-k components, the SHAP values computed on S_eval cannot be interpreted as causal effects of latent sources on cryptogenic stroke. This issue is load-bearing for the 'Latent causes' and 'Feature importance' sections and must be resolved before the causal claims can be considered.
  2. [Latent causes (LiNGAM assumptions)] The paper invokes LiNGAM to equate ICA sources with exogenous error terms, but LiNGAM requires a square, invertible mixing of as many non-Gaussian error terms as observed variables, together with linearity, acyclicity, and no unobserved common causes. With k=2000 and m=9000, the identifiability theorem does not apply, and the Discussion itself acknowledges that the dimension reduction may merge multiple disease sources into one signature. The assumptions of linearity, acyclicity, and causal sufficiency are asserted rather than tested, and the longitudinal, repeatedly sampled EHR structure with a future label makes these assumptions implausible. Consequently, the claim that the recovered sources 'do in fact exert a causal relationship onto the final label' is unsupported.
  3. [Feature importance (SHAP values)] SHAP values are additive feature attributions that decompose a model's prediction, and they do not generally estimate interventional causal effects. Even if the source expressions were true exogenous errors, a random forest trained on those expressions does not automatically provide causal effect estimates unless the causal structure, the functional form, and the absence of confounding are all correctly specified. The statement that 'the SHAP value of each input feature is a quantitative estimate of the causal effect of that latent source on the record's final label' is not justified by the cited literature and should be reframed as an associational measure.
  4. [Model performance and Discussion] The predictive evaluation is incomplete: the test AUROC of 0.782 is reported without confidence intervals or calibration measures, and no comparator is trained on raw EHR features or a standard clinical risk score such as ASCVD. The 'external validation' in the Discussion is a comparison with published literature, not validation on independent data. Additionally, the authors candidly report that their sampling window made prior stroke appear protective; this demonstrates that study-design artifacts can dominate the SHAP interpretations, which weakens confidence in the remaining causal-sounding conclusions such as the allergic rhinitis source.
minor comments (5)
  1. [Model creation] The word 'hyperparemeter' should be 'hyperparameter'.
  2. [Discussion and Conclusions] The phrase 'quantify the the causal effect' contains a duplicated article and should read 'quantify the causal effect'.
  3. [Abstract and Results] The Abstract reports ROC 0.771 while the Results report a test AUROC of 0.782 (Figure 4); the manuscript should clearly distinguish the cross-validated AUROC from the held-out test AUROC.
  4. [Table 1] The header 'N◦' appears to be a typographical artifact; it should read 'N'.
  5. [References] Reference [34], the custom FastICA implementation, should include a version number or an access date.

Circularity Check

2 steps flagged · score 6.0 of 10

The causal-effect findings are the fitted random forest's SHAP attributions renamed as causes, with the ICA-sources-as-error-terms bridge supplied by author-overlapping citations.

  1. fitted input called prediction [Methods, 'Feature importance']
    "Because the features used by our model that predict the probability P(label| S) are the mutually independent error terms of our structural causal model (Fig. 3(b)), the SHAP value of each input feature is a quantitative estimate of the causal effect of that latent source on the record's final label."

    The SHAP values are computed from the random forest trained on the same source projections S and the same EHR-derived labels that define the outcome; they are, by definition, the additive decomposition of that fitted model's predictions. No separate causal estimator, intervention, or external validation is used to obtain the reported 'inferred causes.' The paper's central findings (allergic rhinitis causative; medications protective) are exactly the ranking of SHAP attributions of the fitted model. Calling these attributions causal effects renames the fitted associations as discoveries; the empirical content of 'source X causes CS' is the SHAP value of source X in the fitted model, so the conclusion is the input by construction.

  2. self citation load bearing [Methods, 'Latent causes']
    "Under the LiNGAM model assumptions, 36 the latent sources whose expressions s′ j are captured by our ICA source matrix S represent unobserved latent causes of the target sink node (in this case, our CS label). 26, 27 Formally, these disentangled sources correspond to the exogenous independent error terms of the structural equation model that we assume to describe the underlying causal process of our problem (Fig. 3(b))."

    The decisive equivalence between ICA-disentangled sources and exogenous error terms of the EHR causal graph is not proved in this paper; it is assigned to references [26] and [27], prior works by the same research group (Lasko et al.; Strobl and Lasko). The paper then uses this equivalence to justify reading SHAP values as causal effects. Thus the central causal interpretation rests on an author-overlapping citation chain rather than on an independent, machine-checked, or externally falsified theorem. The problem is compounded by the paper's own stated dimensions: with m=9000 observed variables and k=2000 sources, the mixing matrix A is 9000x2000, so the projection S = A^{-1}E used in the paper does not have an inverse in the standard LiNGAM identification setting.

full rationale

The predictive component of the paper is self-contained: the random forest is trained on training projections and evaluated on a separate test set, yielding AUROC values (0.771 CV, 0.782 test) that are honest predictive results. That part is not circular. However, the paper's advertised contribution is causal, and the causal chain is not self-contained. The sources are constructed by ICA on the same EHR data that later supplies the labels, and the SHAP values are read from a random forest fitted to those same records. Because SHAP is by definition the fitted model's feature attribution, the 'inferred causes of CS' are the fitted associations renamed as root causes. The only non-empirical bridge, the claim that ICA source expressions are the exogenous error terms of a structural causal model, is imported from prior work by the same authors ([26], [27]) and from the LiNGAM assumption, which is asserted rather than tested. The paper itself flags a related weakness in the Discussion, noting that the k=2000 dimension reduction may merge multiple disease sources into one signature. The non-square mixing matrix (9000 x 2000) makes the stated S = A^{-1}E projection mathematically unavailable, further weakening the error-term identification. These issues do not invalidate the AUROC result, but they mean the central causal claims reduce, by construction, to a fitted model's SHAP attributions plus a self-cited interpretive assumption; hence the partial circularity score of 6.

Assumptions & free parameters 7 free parameters · 6 assumptions · 2 invented entities

The central causal claim rests on a long chain of assumptions and ad hoc choices: the ICA source count, the label thresholds, the imputation rules, and the LiNGAM plus causal sufficiency interpretation. The predictive model itself has few free parameters relative to the data, but the causal story depends on entities and assumptions for which no independent evidence is provided.

free parameters (7)
  • Number of ICA sources k = 2000
    Set by memory constraints, not by intrinsic dimensionality; limits granularity of disentangled sources and can merge distinct causes.
  • Record sampling density = 1 sample per record-year
    Arbitrary density at which longitudinal patient states are sampled; changes the size and composition of the training matrix X.
  • Prediction window = 10 years
    Defines the binary outcome horizon; a different horizon would change the labels and the model's performance.
  • Pre-stroke buffer = 1 month
    Data up to one month before the first stroke code are excluded; arbitrary and affects the temporal gap between predictors and outcome.
  • Stroke code coincidence threshold = 30%
    Codes coinciding with non-cryptogenic stroke codes in more than 30% of charts were removed from inclusion criteria; chosen by hand.
  • Single stroke code exclusion = Binary rule: exclude records with exactly one stroke code
    Records with only one stroke code instance were excluded based on chart review; this is a post hoc case definition choice.
  • Missing condition imputation baseline = 1 event per 20 years
    Missing condition curves were imputed with a constant annual event-arrival probability of once per 20 years; arbitrary baseline.
assumptions (6)
  • domain assumption The observed EHR data matrix X is generated as a linear, instantaneous mixture X = AS of independent non-Gaussian sources S.
    Required for ICA to recover the latent sources; introduced in the 'Clinical signature discovery' section with no empirical test of linearity or instantaneous mixing in this dataset.
  • domain assumption LiNGAM assumptions hold for the EHR observation network, so the ICA sources are the exogenous error terms (root causes) of the structural causal model.
    Invoked in the 'Latent causes' section with citation [36]; no validation of linearity, acyclicity, non-Gaussianity, or absence of latent confounders is provided.
  • domain assumption Causal sufficiency: the 2000 sources plus observed variables capture all common causes of cryptogenic stroke.
    Needed to interpret SHAP values as causal effects; the paper does not test whether unobserved common causes exist outside the model.
  • ad hoc to paper SHAP values on a random forest trained on independent error terms provide quantitative causal effect estimates.
    The paper asserts this via reference [27], but it is not established for nonlinear random forest models by a theorem or simulation in this paper.
  • domain assumption The ICD-10-based labeling algorithm with heuristic exclusions identifies true cryptogenic stroke events.
    Labels depend on billing codes and post hoc rules, such as the 30% coincidence threshold and single-code exclusion, without full adjudication of positive records.
  • domain assumption Medication list presence and imputed continuation curves reflect actual drug exposure.
    The protective medication findings depend on binary medication curves that assume continuation between visits and midpoint discontinuation; this is not validated against adherence data.
invented entities (2)
  • 2000 latent disease sources (independent components of EHR curves)
    purpose: Represent unobserved exogenous root causes of cryptogenic stroke in migraine patients; patient expressions are the inputs to the prediction model.
    The sources are inferred from the same EHR data used for labels; no independent measurement or prospective replication confirms that they are real disease mechanisms. The literature match of top signatures is post hoc.
  • Allergic rhinitis-associated latent source as a causative factor
    purpose: Postulated causative source of cryptogenic stroke in migraine patients, possibly reflecting chronic inflammation or decongestant use.
    Identified only through SHAP rankings on the fitted model; the paper itself frames it as a hypothesis needing testing and cites mixed prior literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cryptogenic stroke and migraine: using probabilistic independence and machine learning to uncover latent sources of disease from the electronic health record." pith.science (2026). https://pith.science/paper/LF6SQEQR

@misc{pith2026250504631,
  author       = {Pith},
  title        = {Pith review of: Cryptogenic stroke and migraine: using probabilistic independence and machine learning to uncover latent sources of disease from the electronic health record},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LF6SQEQR}},
  note         = {Machine review of arXiv:2505.04631}
}
read the original abstract

Migraine is a common but complex neurological disorder that doubles the lifetime risk of cryptogenic stroke (CS). However, this relationship remains poorly characterized, and few clinical guidelines exist to reduce this associated risk. We therefore propose a data-driven approach to extract probabilistically-independent sources from electronic health record (EHR) data and create a 10-year risk-predictive model for CS in migraine patients. These sources represent external latent variables acting on the causal graph constructed from the EHR data and approximate root causes of CS in our population. A random forest model trained on patient expressions of these sources demonstrated good accuracy (ROC 0.771) and identified the top 10 most predictive sources of CS in migraine patients. These sources revealed that pharmacologic interventions were the most important factor in minimizing CS risk in our population and identified a factor related to allergic rhinitis as a potential causative source of CS in migraine patients.

Figures

Figures reproduced from arXiv: 2505.04631 by the authors.

Figure 1
Figure 1. Preprocessing pipeline for EHR data. Longitudinal curves are inferred from the raw obser￾vational data and stacked, synchronizing the discrete data channels and allowing for all clinical variables to be sampled at any given point in time. Random instances of these curvesets records are then sampled and concatenated into the stacked data matrix X. Data pre-processing and continuation Instances of a patient’s health d… view at source ↗
Figure 2
Figure 2. Flowchart detailing the multi￾step inclusion and exclusion criteria (based on ICD-10 diagnosis) codes used to label patient records as either positive or negative for CS. Record labeling Due to CS’s varying definitions and status as a diagnosis of exclusion, there is no single ICD-10 code or even set of codes that neatly correspond to a diagnosis of CS. Additionally, some non-specific codes may apply to both cryptog… view at source ↗
Figure 3
Figure 3. Two views of the structural causal graph for an arbitrary patient, adapted from Mota et al.22 Nodes repre￾sent the model’s variables while edges (arrows) represent unidirectional and transitive causal relationships. The green rectangular node corresponds to the binary yes/no CS label; shaded blue nodes correspond to observations in the EHR; dashed red circles are latent nodes. Model creation The longitudinal curves … view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Above: ROC curve for test set evaluation (AUC = 0.782). Below: Waterfall chart showcasing the model’s decision-making for a random patient record, illustrating its ability to provide patient-level causal explanations for a record’s final label. Model performance The me…
Figure 6
Figure 6. Figure 6: Signature description diagrams for 4 of the most predictive sources, as shown in Fig.5. Bar length represents the normalized change in a clinical variable’s value for each unit of expression, shown in parenthesis. Inset log-scale histograms show source expression level…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 47 canonical work pages

  1. [1]

    Leading causes of death

    Centers for Disease Control and Prevention (CDC). Leading causes of death. Hyattsville (MD): Na- tional Center for Health Statistics; 2024. Available from: https://www.cdc.gov/nchs/fastats/ leading-causes-of-death.htm

    work page 2024

  2. [2]

    The top 10 causes of death

    World Health Organization (WHO). The top 10 causes of death. Geneva (CH): WHO; 2024. Available from: https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death

    work page 2024

  3. [3]

    Economic burden of stroke disease: a systematic review

    Rochmah TN, Rahmawati IT, Dahlui M, Budiarto W, Bilqis N. Economic burden of stroke disease: a systematic review. Int J Environ Res Public Health;18. doi:10.3390/ijerph18147552

    work page doi:10.3390/ijerph18147552

  4. [4]

    Classification of subtype of acute ischemic stroke

    Adams HP Jr, Bendixen BH, Kappelle LJ, Biller J, Love BB, Gordon DL, et al. Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment. Stroke. 1993 Jan;24(1):35-41. doi:10

    work page 1993

  5. [5]

    Etiology of stroke

    Mohr JP, Albers GW, Amarenco P, Babikian VL, Biller J, Brey RL, et al. Etiology of stroke. Stroke. 1997;28(7):1501-6. doi:10.1161/01.STR.28.7.1501

    work page doi:10.1161/01.str.28.7.1501 1997

  6. [6]

    Cryptogenic stroke

    Yaghi S, Bernstein RA, Passman R, Okin PM, Furie KL. Cryptogenic stroke. Circulation Research. 2017;120(3):527-40. doi:10.1161/CIRCRESAHA.116.308447

    work page doi:10.1161/circresaha.116.308447 2017

  7. [7]

    Stroke subtype classification to mechanism-specific and undetermined categories by TOAST, A-S-C-O, and causative classification system

    Marnane M, Duggan CA, Sheehan OC, Merwick A, Hannon N, Curtin D, et al. Stroke subtype classification to mechanism-specific and undetermined categories by TOAST, A-S-C-O, and causative classification system. Stroke. 2010;41(8):1579-86. doi:10.1161/STROKEAHA.109.575373

    work page doi:10.1161/strokeaha.109.575373 2010

  8. [8]

    An evidence-based causative classification system for acute ischemic stroke

    Ay H, Furie KL, Singhal A, Smith WS, Sorensen AG, Koroshetz WJ. An evidence-based causative classification system for acute ischemic stroke. Annals of Neurology. 2005;58(5):688-97. doi:https://doi.org/10.1002/ana.20617

    work page doi:10.1002/ana.20617 2005

Show all 53 references
  1. [9]

    StrokeClassifier: Ischemic stroke etiology classification by ensemble consensus modeling using electronic health records

    Lee HJ, Schwamm LH, Sansing L, Kamel H, de Havenon A, Turner AC, et al. StrokeClassifier: Ischemic stroke etiology classification by ensemble consensus modeling using electronic health records. Res Sq. 2023 Oct. doi:10.21203/rs.3.rs-3367169/v1

    2023 doi

  2. [10]

    Embolic stroke of undetermined source

    Hart RG, Catanese L, Perera KS, Ntaios G, Connolly SJ. Embolic stroke of undetermined source. Stroke. 2017;48(4):867-72. doi:10.1161/STROKEAHA.116.016414

    2017 doi

  3. [11]

    Ntaios G, Baumgartner H, Doehner W, Donal E, Edvardsen T, Healey JS, et al. Embolic strokes of undetermined source: a clinical consensus statement of the ESC Council on Stroke, the European Association of Cardiovascular Imaging and the European Heart Rhythm Association of the ...

    2024

  4. [12]

    Migraine and stroke: a complex association with clinical implications

    Kurth T, Chabriat H, Bousser MG. Migraine and stroke: a complex association with clinical implications. The Lancet Neurology. 2012 Jan;11(1):92-100. doi:10.1016/S1474-4422(11)70266-6

    2012 doi

  5. [13]

    The prevalence and impact of migraine and severe headache in the United States: figures and trends from government health studies

    Burch R, Rizzoli P, Loder E. The prevalence and impact of migraine and severe headache in the United States: figures and trends from government health studies. Headache: The Journal of Head and Face Pain. 2018;58(4):496-505. doi:10.1111/head.13281

    2018 doi

  6. [14]

    Nguyen AT, Curtis KM, Tepper NK, Kortsmit K, Brittain AW, Snyder EM, et al. U.S. medical eligibility criteria for contraceptive use. MMWR Recomm Rep. 2024 Aug;73(4):1-126

    2024

  7. [15]

    doi:10.1093/eurheartj/ehae150

    doi

  8. [16]

    Use of combined hormonal contraception and stroke: A case-control study of the impact of migraine type and estrogen dose on ischemic stroke risk

    Batur P, Yao M, Bucklan J, Soni P, Suneja A, Farrell R, et al. Use of combined hormonal contraception and stroke: A case-control study of the impact of migraine type and estrogen dose on ischemic stroke risk. Headache: The Journal of Head and Face Pain. 2023;63(6):813-21. doi:...

    2023 doi

  9. [17]

    Patent foramen ovale closure for treating migraine: A meta-analysis

    Zhang Y, Wang H, Liu L. Patent foramen ovale closure for treating migraine: A meta-analysis. Journal of Interventional Cardiology;2022(1):6456272. doi:https://doi.org/10.1155/2022/6456272

    2022 doi

  10. [18]

    Patent foramen ovale and cryptogenic stroke: a complex neuro–cardio–vascular problem

    Spies C, Wong M. Patent foramen ovale and cryptogenic stroke: a complex neuro–cardio–vascular problem. Expert Review of Cardiovascular Therapy. 2009;7(11):1455-67. doi:10.1586/erc.09.114

    2009 doi

  11. [19]

    off-label

    Goldsweig AM, Deng Y, Yao X, Desai NR, Cohen DJ, Aronow HD, et al. Approval, evidence, and “off-label” device utilization: The patent foramen ovale closure story. Circulation: Cardiovascular Quality and Outcomes. 2024;17(1):e010200. doi:10.1161/CIRCOUTCOMES.123.010200

    2024 doi

  12. [20]

    Computational phenotype discovery using unsupervised feature learning over noisy, sparse, and irregular clinical data

    Lasko TA, Denny JC, Levy MA. Computational phenotype discovery using unsupervised feature learning over noisy, sparse, and irregular clinical data. PLOS ONE. 2013 06;8(6):1-13. doi:10.1371/journal.pone.0066341

    2013 doi

  13. [21]

    Deep patient: an unsupervised representation to predict the future of patients from the electronic health records

    Miotto R, Li L, Kidd BA, Dudley JT. Deep patient: an unsupervised representation to predict the future of patients from the electronic health records. Scientific Reports. 2016 May;6(1):26094. doi:10.1038/srep26094

    2016 doi

  14. [22]

    Distilling knowledge from deep networks with applications to healthcare domain; 2015

    Che Z, Purushotham S, Khemani R, Liu Y. Distilling knowledge from deep networks with applications to healthcare domain; 2015

    2015

  15. [23]

    A data-driven approach to dis- cover and quantify systemic lupus erythematosus etiological heterogeneity from electronic health records

    Mota MB, Still JM, Gamboa JL, Strobl EV, Stein CM, Kawai VK, et al. A data-driven approach to dis- cover and quantify systemic lupus erythematosus etiological heterogeneity from electronic health records. 2025. arXiv:2501.07206 [cs.LG]

    2025 arXiv

  16. [24]

    Performance of breast cancer risk- assessment models in a large mammography cohort

    McCarthy AM, Guan Z, Welch M, Griffin ME, Sippo DA, Deng Z, et al. Performance of breast cancer risk- assessment models in a large mammography cohort. JNCI: Journal of the National Cancer Institute. 2019 09;112(5):489-97. doi:10.1093/jnci/djz177

    2019 doi

  17. [25]

    Predicting the risk of cancer in adults using supervised machine learning: a scoping review

    Abdullah Alfayez A, Kunz H, Grace Lai A. Predicting the risk of cancer in adults using supervised machine learning: a scoping review. BMJ Open. 2021;11(9). doi:10.1136/bmjopen-2020-047755

    2021 doi

  18. [26]

    Why do probabilistic clinical models fail to transport between sites

    Lasko TA, Strobl EV, Stead WW. Why do probabilistic clinical models fail to transport between sites. npj Digital Medicine. 2024 Mar;7(1):53. doi:10.1038/s41746-024-01037-4

    2024 doi

  19. [27]

    Unsupervised discovery of clinical disease signatures using probabilistic independence

    Lasko TA, Still JM, Li TZ, Mota MB, Stead WW, Strobl EV, et al. Unsupervised discovery of clinical disease signatures using probabilistic independence. 2024. arXiv:2402.05802 [cs.LG]

    2024 arXiv

  20. [28]

    Identifying patient-specific root causes with the heteroscedastic noise model

    Strobl EV, Lasko TA. Identifying patient-specific root causes with the heteroscedastic noise model. Journal of Computational Science. 2023;72:102099. doi:10.1016/j.jocs.2023.102099

    2023

  21. [29]

    From local explanations to global under- standing with explainable AI for trees

    Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From local explanations to global under- standing with explainable AI for trees. Nature Machine Intelligence. 2020 Jan;2(1):56-67. doi:10.1038/s42256- 019-0138-9

    2020 doi

  22. [30]

    Development of a large-scale de- identified DNA biobank to enable personalized medicine

    Roden D, Pulley J, Basford M, Bernard G, Clayton E, Balser J, et al. Development of a large-scale de- identified DNA biobank to enable personalized medicine. Clinical Pharmacology & Therapeutics;84(3):362-9. doi:https://doi.org/10.1038/clpt.2008.89

    doi

  23. [31]

    Random average shifted histograms

    Bourel M, Fraiman R, Ghattas B. Random average shifted histograms. Computational Statistics & Data Analysis. 2014;79:149-64. doi:https://doi.org/10.1016/j.csda.2014.05.004

    2014 doi

  24. [32]

    A Method for Constructing Local Monotone Piecewise Cubic Interpolants

    Fritsch FN, Butland J. A Method for Constructing Local Monotone Piecewise Cubic Interpolants. SIAM Journal on Scientific and Statistical Computing. 1984;5(2):300-4. doi:10.1137/0905021

    1984 doi

  25. [33]

    Scaling regression inputs by dividing by two standard deviations

    Gelman A. Scaling regression inputs by dividing by two standard deviations. Stat Med. 2008 Jul;27(15):2865-73. doi:10.1002/sim.3107

    2008 doi

  26. [34]

    Independent component analysis: algorithms and applications

    Hyv ¨arinen A, Oja E. Independent component analysis: algorithms and applications. Neural Networks. 2000;13(4):411-30. doi:10.1016/S0893-6080(00)00026-5

    2000 doi

  27. [35]

    Still JM. FastICA. ComputationalMedicineLab, GitHub; 2023. Available from: https://github.com/ ComputationalMedicineLab/fastica

    2023

  28. [36]

    Fast and robust fixed-point algorithms for independent component analysis

    Hyv ¨arinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE transactions on neural networks. 1999;10 3:626-34

    1999

  29. [37]

    A linear non-gaussian acyclic model for causal discovery

    Shimizu S, Hoyer PO, Hyvrinen A, Kerminen A. A linear non-gaussian acyclic model for causal discovery. Journal of Machine Learning Research. 2006;7(72):2003-30

    2006

  30. [38]

    A unified approach to interpreting model predictions

    Lundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. NIPS’17. Red Hook, NY, USA: Curran Associates Inc.; 2017. p. 4768–4777

    2017

  31. [39]

    External validation of the ACC/AHA ASCVD risk score in a Colombian population cohort

    Rodr ´ıguez-Ariza CD, Cabrera-Villamizar A, Rodr ´ıguez-Pulido AL, Callegari S, Ossa Rodr ´ıguez NA, Pinilla- Roncancio M, et al. External validation of the ACC/AHA ASCVD risk score in a Colombian population cohort. Scientific Reports. 2023 Apr;13(1):6139. doi:10.1038/s41598-0...

    2023 doi

  32. [40]

    Effectiveness of SCORE2 and ASCVD risk scores in combination with aortic arch calcification on chest radiograph in predicting significant coronary artery disease

    Cetin M, Maksutov C, Isakulov E, Hosoglu Y. Effectiveness of SCORE2 and ASCVD risk scores in combination with aortic arch calcification on chest radiograph in predicting significant coronary artery disease. Angiology. 2024;75(8):734-41. PMID: 37200508. doi:10.1177/00033197231177413

    2024 doi

  33. [41]

    Bohula EA, Wiviott SD, Giugliano RP, Blazing MA, Park JG, Murphy SA, et al. Prevention of stroke with the addition of ezetimibe to statin therapy in patients with acute coronary syndrome in IMPROVE-IT (IM- Proved Reduction of Outcomes: Vytorin Efficacy International Trial). Ci...

    2017 doi

  34. [42]

    Benazepril plus amlodipine or hy- drochlorothiazide for hypertension in high-risk patients

    Jamerson K, Weber MA, Bakris GL, Dahl ¨of B, Pitt B, Shi V, et al. Benazepril plus amlodipine or hy- drochlorothiazide for hypertension in high-risk patients. New England Journal of Medicine. 2008;359(23):2417-

    2008

  35. [43]

    doi:10.1056/NEJMoa0806182

    doi

  36. [44]

    Effect of amlodipine in stroke and myocardial infarction: A systematic review and meta-analysis

    Iyengar SS, Mohan JC, Ray S, Rao MS, Khan MY, Patted URH, et al. Effect of amlodipine in stroke and myocardial infarction: A systematic review and meta-analysis. Cardiol Ther. 2021 Sep;10(2):429-44

    2021

  37. [45]

    Effects of cholesterol-lowering with simvastatin on stroke and other major vascular events in 20536 people with cerebrovas- cular disease or other high-risk conditions

    Collins R, Armitage J, Parish S, Sleight P, Peto R, Heart Protection Study Collaborative Group. Effects of cholesterol-lowering with simvastatin on stroke and other major vascular events in 20536 people with cerebrovas- cular disease or other high-risk conditions. Lancet. 2004...

    2004

  38. [46]

    Antithrombotic therapy for primary and secondary prevention of ischemic stroke

    Greco A, Occhipinti G, Giacoppo D, Agnello F, Laudani C, Spagnolo M, et al. Antithrombotic therapy for primary and secondary prevention of ischemic stroke. JACC. 2023;82(15):1538-57. doi:10.1016/j.jacc.2023.07.025

    2023 doi

  39. [47]

    Long-term risk of recurrent stroke after a first-ever stroke

    Burn J, Dennis M, Bamford J, Sandercock P, Wade D, Warlow C. Long-term risk of recurrent stroke after a first-ever stroke. The Oxfordshire Community Stroke Project. Stroke. 1994;25(2):333-7. doi:10.1161/01.STR.25.2.333

    1994 doi

  40. [48]

    The burden of allergic rhinitis beyond allergies

    Alam R. The burden of allergic rhinitis beyond allergies. Immunology and Allergy Clinics. 2011 Aug;31(3):ix-x. doi:10.1016/j.iac.2011.05.013

    2011 doi

  41. [49]

    Re: The association between hay fever and stroke in a cohort of mid- dle aged and elderly adults

    Ng DK, Kwok Kl, Chan Ch. Re: The association between hay fever and stroke in a cohort of mid- dle aged and elderly adults. The Journal of the American Board of Family Medicine. 2008;21(5):477-8. doi:10.3122/jabfm.2008.05.080111

    2008 doi

  42. [50]

    Nonprescription medications for adults with allergic rhinitis

    Vordenberg SE. Nonprescription medications for adults with allergic rhinitis. JAMA. 2024 01;331(3):268-8. doi:10.1001/jama.2023.20171

    2024

  43. [51]

    Prescription and over-the-counter medication record integration: A holistic patient- centered approach

    Kebodeaux CD. Prescription and over-the-counter medication record integration: A holistic patient- centered approach. Journal of the American Pharmacists Association. 2019 Mar;59(2):S13-7. doi:10.1016/j.japh.2018.10.002

    2019 doi

  44. [52]

    Stroke associated With sympathomimetics contained in over-the-counter cough and cold drugs

    Cantu C, Arauz A, Murillo-Bonilla LM, L ´opez M, Barinagarrementeria F. Stroke associated With sympathomimetics contained in over-the-counter cough and cold drugs. Stroke. 2003;34(7):1667-72. doi:10.1161/01.STR.0000075293.45936.FA

    2003

  45. [53]

    Ischemic stroke associated with chronic xylometazoline nasal spray misuse: A rare avoidable adverse event

    Patel J, Patel I, Desai D, Desai S. Ischemic stroke associated with chronic xylometazoline nasal spray misuse: A rare avoidable adverse event. Annals of Indian Academy of Neurology. 2021;24(2)

    2021

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.