CoChat

High-throughput target trial emulation for Alzheimer’s disease drug repurposing with real-world data

Open access

The DOI matches its Crossref record; 69 of the 69 references it cites were found in a registry or an index.

Abstract

Target trial emulation is the process of mimicking target randomized trials using real-world data, where effective confounding control for unbiased treatment effect estimation remains a main challenge. Although various approaches have been proposed for this challenge, a systematic evaluation is still lacking. Here we emulated trials for thousands of medications from two large-scale real-world data warehouses, covering over 10 years of clinical records for over 170 million patients, aiming to identify new indications of approved drugs for Alzheimer’s disease. We assessed different propensity score models under the inverse probability of treatment weighting framework and suggested a model selection strategy for improved baseline covariate balancing. We also found that the deep learning-based propensity score model did not necessarily outperform logistic regression-based methods in covariate balancing. Finally, we highlighted five top-ranked drugs (pantoprazole, gabapentin, atorvastatin, fluticasone, and omeprazole) originally intended for other indications with potential benefits for Alzheimer’s patients.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-023-43929-1

Cites

69 of 69 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 System overview. a Target trial emulations were conducted for all drugs in two large-scale and longitudinal real-world data: the OneFlorida electronic health records data and MarketScan administrative claims. The contrast was made between individuals exposed to the target drug versus different comparison drugs (a random drug or a similar drug). b Machine learning-based propensity score models and inverse probability of treatment re-weighting were used for adjusting high-dimensional baseline covariates (e.g. age, gender, disease comorbidities, medications, etc) or covariates selected based on causal diagrams and knowledge. A model selection framework tailored for ML-PS models was proposed to better balance baseline covariates on the train, unseen test, and combined sets. c Adjusted survival analysis per endpoint was computed for each drug. Top-ranked repurposing hypotheses were selected. RWD Real-World Data, EHR Electronic Health Records, MCI Mild Cognitive Impairment, AD Alzheimer’s Disease, ATC Anatomical Therapeutic Chemical classification, DAGs Directed Acyclic Graphs, CV Cross Validation, ML-PS Machine Learning-based Propensity Score modeling, IPTW Inverse Probability of Treatment Weight.
Fig. 1 System overview. a Target trial emulations were conducted for all drugs in two large-scale and longitudinal real-world data: the OneFlorida electronic health records data and MarketScan administrative claims. The contrast was made between individuals exposed to the target drug versus different comparison drugs (a random drug or a similar drug). b Machine learning-based propensity score models and inverse probability of treatment re-weighting were used for adjusting high-dimensional baseline covariates (e.g. age, gender, disease comorbidities, medications, etc) or covariates selected based on causal diagrams and knowledge. A model selection framework tailored for ML-PS models was proposed to better balance baseline covariates on the train, unseen test, and combined sets. c Adjusted survival analysis per endpoint was computed for each drug. Top-ranked repurposing hypotheses were selected. RWD Real-World Data, EHR Electronic Health Records, MCI Mild Cognitive Impairment, AD Alzheimer’s Disease, ATC Anatomical Therapeutic Chemical classification, DAGs Directed Acyclic Graphs, CV Cross Validation, ML-PS Machine Learning-based Propensity Score modeling, IPTW Inverse Probability of Treatment Weight.CC BY · PubMed Central
Fig. 2 Balance performance of the regularized logistic regression-based propensity score models (LR-PS) selected by different model selection strategies, OneFlorida database, 2012–2020. a The proportion of successfully balanced trials by LR-PS selected using different model selection strategies. b–d The average number of unbalanced baseline covariates before and after re-weighting on (b) train and test combined set, (c) train set, and (d) unseen test set. Three model selection strategies are (i) the AUC score on the validation fold during the cross-validation procedure, (ii) the cross-entropy loss on the validation fold, and (iii) our proposed strategy, which leverages balance performance on the training and validation combined folds and generalization performance on the validation fold. We reported drugs with ≥10% balanced trials among 100 emulations. A covariate is assumed balanced if its standardized mean difference (SMD) of its prevalence between exposure groups is at most 0.1 and a trial is assumed balanced if the ratio of unbalanced features among all covariates before/after re-weighting is ≤2%. The error bars indicate 95% confidence intervals by 1000-times bootstrapping. Welch’s t test (two-sample, two-sided) is used for testing the means of binary indicators for balanced trials, and p values and their associated significance marks are shown. *p < 0.05; **p < 0.01; ***p < 0.001; not significant with p ≥ 0.05 were not marked; AUC, area under the receiver operating characteristic curve. Source data are provided as a Source Data file.
Fig. 2 Balance performance of the regularized logistic regression-based propensity score models (LR-PS) selected by different model selection strategies, OneFlorida database, 2012–2020. a The proportion of successfully balanced trials by LR-PS selected using different model selection strategies. b–d The average number of unbalanced baseline covariates before and after re-weighting on (b) train and test combined set, (c) train set, and (d) unseen test set. Three model selection strategies are (i) the AUC score on the validation fold during the cross-validation procedure, (ii) the cross-entropy loss on the validation fold, and (iii) our proposed strategy, which leverages balance performance on the training and validation combined folds and generalization performance on the validation fold. We reported drugs with ≥10% balanced trials among 100 emulations. A covariate is assumed balanced if its standardized mean difference (SMD) of its prevalence between exposure groups is at most 0.1 and a trial is assumed balanced if the ratio of unbalanced features among all covariates before/after re-weighting is ≤2%. The error bars indicate 95% confidence intervals by 1000-times bootstrapping. Welch’s t test (two-sample, two-sided) is used for testing the means of binary indicators for balanced trials, and p values and their associated significance marks are shown. *p < 0.05; **p < 0.01; ***p < 0.001; not significant with p ≥ 0.05 were not marked; AUC, area under the receiver operating characteristic curve. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 3 Generated drug repurposing hypotheses for AD with adjusted hazard ratios and 95% confidence intervals in the five-year follow-up period. Target Trial emulations of these drugs (a–e) were performed on OneFlorida and MarketScan data separately. For each drug, treated groups consisted of patients who were exposed to the trial drug, and control groups were built by either: (i) randomly selecting alternative drug groups, or (ii) using drug groups under the same second-level Anatomical Therapeutic Chemical classification codes (ATC-L2) as the trial drug. The primary analysis emulated 100 trials consisting of 50 random control groups and 50 ATC-L2 control groups (FL-All and MS-All), and two sensitivity analyses using only random controls (FL-Rand and MS-Rand) or only ATC-L2 controls (FL-ATC and MS-ATC). The error bars indicate 95% 1000-time-bootstrapped confidence intervals of aHR from balanced trials. The aHR was calculated by the Cox proportional hazard model for each balanced trial after re-weighting. The average number, denoted by #, of patients in treated and control arms was also shown. FL OneFlorida, MS MarketScan, aHR adjusted hazard ratio, CI confidence interval, ATC Anatomical Therapeutic Chemical classification. Source data are provided as a Source Data file.
Fig. 3 Generated drug repurposing hypotheses for AD with adjusted hazard ratios and 95% confidence intervals in the five-year follow-up period. Target Trial emulations of these drugs (a–e) were performed on OneFlorida and MarketScan data separately. For each drug, treated groups consisted of patients who were exposed to the trial drug, and control groups were built by either: (i) randomly selecting alternative drug groups, or (ii) using drug groups under the same second-level Anatomical Therapeutic Chemical classification codes (ATC-L2) as the trial drug. The primary analysis emulated 100 trials consisting of 50 random control groups and 50 ATC-L2 control groups (FL-All and MS-All), and two sensitivity analyses using only random controls (FL-Rand and MS-Rand) or only ATC-L2 controls (FL-ATC and MS-ATC). The error bars indicate 95% 1000-time-bootstrapped confidence intervals of aHR from balanced trials. The aHR was calculated by the Cox proportional hazard model for each balanced trial after re-weighting. The average number, denoted by #, of patients in treated and control arms was also shown. FL OneFlorida, MS MarketScan, aHR adjusted hazard ratio, CI confidence interval, ATC Anatomical Therapeutic Chemical classification. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 4 Sensitivity analysis of generated drug repurposing hypotheses when adjusting for baseline covariates selected by using both existing knowledge and causal discovery algorithms. The adjusted hazard ratios and 95% confidence intervals in the five-year follow-up period were reported. Trial emulations of these drugs (a–e) were performed on OneFlorida and MarketScan data separately. For each drug, treated groups consisted of patients who were exposed to the trial drug, and control groups were built by either: (i) randomly selecting alternative drug groups, or (ii) using drug groups under the same second-level Anatomical Therapeutic Chemical classification codes (ATC-L2) as the trial drug. The overall analysis emulated 100 trials consisting of 50 random control groups and 50 ATC-L2 control groups (FL-All and MS-All), and two separate analyses using only random controls (FL-Rand and MS-Rand) or only ATC-L2 controls (FL-ATC and MS-ATC). The error bars indicate 95% 1000-time-bootstrapped confidence intervals of aHR from balanced trials. The aHR was calculated by the Cox proportional hazard model for each balanced trial after re-weighting. The average number of patients, denoted by #, in treated and control arms was also shown. FL OneFlorida, MS MarketScan, DAG directed acyclic graphs, aHR adjusted hazard ratio, CI confidence interval, ATC Anatomical Therapeutic Chemical classification. Source data are provided as a Source Data file.
Fig. 4 Sensitivity analysis of generated drug repurposing hypotheses when adjusting for baseline covariates selected by using both existing knowledge and causal discovery algorithms. The adjusted hazard ratios and 95% confidence intervals in the five-year follow-up period were reported. Trial emulations of these drugs (a–e) were performed on OneFlorida and MarketScan data separately. For each drug, treated groups consisted of patients who were exposed to the trial drug, and control groups were built by either: (i) randomly selecting alternative drug groups, or (ii) using drug groups under the same second-level Anatomical Therapeutic Chemical classification codes (ATC-L2) as the trial drug. The overall analysis emulated 100 trials consisting of 50 random control groups and 50 ATC-L2 control groups (FL-All and MS-All), and two separate analyses using only random controls (FL-Rand and MS-Rand) or only ATC-L2 controls (FL-ATC and MS-ATC). The error bars indicate 95% 1000-time-bootstrapped confidence intervals of aHR from balanced trials. The aHR was calculated by the Cox proportional hazard model for each balanced trial after re-weighting. The average number of patients, denoted by #, in treated and control arms was also shown. FL OneFlorida, MS MarketScan, DAG directed acyclic graphs, aHR adjusted hazard ratio, CI confidence interval, ATC Anatomical Therapeutic Chemical classification. Source data are provided as a Source Data file.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 69 references

Research works printed but not linked: 2

  1. Ke, G. et al. LightGBM: A Highly Efficient Gradient Boosting Decision Tree. in Advances in Neural Information Processing Systems 30 (Curran Associates, Inc., 2017).
  2. Kingma, D. P. & Ba, J. Adam: A Method for Stochastic Optimization. in 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7–9, 2015, Conference Track Proceedings (2015).

Cited by

See all 80

Metadata

Authors
Zang, C., Zhang, H., Xu, J., Zhang, H., Fouladvand, S., Havaldar, S., Cheng, F., Chen, K., Chen, Y., Glicksberg, B. S., Chen, J., Bian, J., & Wang, F.
Affiliations
  • Weill Cornell Medicine (US)
  • University of Florida Health (US)
  • University of Kentucky (US)
  • Icahn School of Medicine at Mount Sinai (US)
  • Cleveland Clinic (US)
  • University of Connecticut (US)
  • University of Pennsylvania (US)
Corresponding author
Wang, Fei
Venue
Nature Communications
Published
2023-12-11
Volume
14
Issue
1
Pages
8180-8180
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4389549989
Cited by
69
References
63
Citations by year
2026: 22 · 2025: 33 · 2024: 13
FWCI
13.82
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Science Foundation (1750326)
  • International Business Machines Corporation
  • National Institutes of Health (1rf1ag084178-01, RF1G077820, 1rf1ag077820-01a1)
  • National Institute on Aging (R56AG074001, R56AG069880, R01AG066707, R56AG074604, RF1AG082211, RF1 AG084178, RF1AG072449, RF1AG059319, R21AG083003, RF1AG077820, R21AG068717, K08AG065463, R01AG083039, R01AG084236, R01AG084250, R01AG080991, R01AG080624, R01AG082118, R01AG073435, P30AG072946, R01AG076448, R01AG076234, U01AG073323)
  • National Institute of Mental Health (R01MH124740)
  • National Institute on Drug Abuse (UM1DA049406)
  • National Cancer Institute (R01CA246418, R21CA245858)
  • National Institute of General Medical Sciences (P20GM109025)
  • National Institute of Allergy and Infectious Diseases (R01AI172875)
  • National Institute of Neurological Disorders and Stroke (RF1NS133812)
  • U.S. National Library of Medicine (R01LM013519, R01LM014344)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Alzheimer's disease research and treatments
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
target trial emulation, Alzheimer's disease, drug repurposing, real-world data, propensity score matching, inverse probability of treatment weighting, confounding adjustment, covariate balancing, model selection, pantoprazole, gabapentin, atorvastatin, fluticasone, omeprazole
MeSH terms
Atorvastatin, Alzheimer Disease, Humans, Propensity Score, Drug Repositioning

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: High-throughput target trial emulation for Alzheimer’s disease drug repurposing with real-world data. https://app.cochat.ai/papers/10.1038/s41467-023-43929-1
The paper
Zang, C., Zhang, H., Xu, J., Zhang, H., Fouladvand, S., Havaldar, S., Cheng, F., Chen, K., Chen, Y., Glicksberg, B. S., Chen, J., Bian, J., & Wang, F. (2023). High-throughput target trial emulation for Alzheimer’s disease drug repurposing with real-world data. Nature Communications. https://doi.org/10.1038/s41467-023-43929-1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.