Reflection on modern methods: when worlds collide—prediction, machine learning and causal inference
Open accessThe DOI matches its Crossref record; 42 of the 43 references it cites were found in a registry or an index, and 1 are not checked yet.
Abstract
Causal inference requires theory and prior knowledge to structure analyses, and is not usually thought of as an arena for the application of prediction modelling. However, contemporary causal inference methods, premised on counterfactual or potential outcomes approaches, often include processing steps before the final estimation step. The purposes of this paper are: (i) to overview the recent emergence of prediction underpinning steps in contemporary causal inference methods as a useful perspective on contemporary causal inference methods, and (ii) explore the role of machine learning (as one approach to ‘best prediction’) in causal inference. Causal inference methods covered include propensity scores, inverse probability of treatment weights (IPTWs), G computation and targeted maximum likelihood estimation (TMLE). Machine learning has been used more for propensity scores and TMLE, and there is potential for increased use in G computation and estimation of IPTWs.
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.
Integrity signals
Registry
Cites
Statistics
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.
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Other innovations in confounding adjustment methods include Bayesian approaches for trajectory modeling [96, 97], machine learning to estimate weights for IPW and doubly robust estimators [98-100], and IPW approaches driven by latent variables or marginal structure models to facilitate a better understanding of selection into treatment [100-102].”
From Difference-in-Differences for Health Policy and Practice: A Review of Modern Methods (2025) · cites this paper as [99]
Experimental — accuracy not yet measured
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
The central role of the propensity score in observational studies for causal effects
ROSENBAUM, P. R., & RUBIN, D. B. (1983) · Biometrika · doi:10.1093/biomet/70.1.41 · Cited by 31,241 · cited 1×
“Propensity scores (PS) reduce information on multiple confounding covariates into one value: the propensity to be exposed or treated, 13 i.e. Pr(X ¼ 1 j Z) for a binary exposure X and a vector of covariates Z.”
BackgroundCausality: Models, Reasoning, and Inference
Pearl, J. (2009) · doi:10.1017/cbo9780511803161 · Cited by 11,653
Marginal Structural Models and Causal Inference in Epidemiology
Robins, J. M., Hernán, M. Á., & Brumback, B. (2000) · Epidemiology · doi:10.1097/00001648-200009000-00011 · Cited by 5,581 · cited 1×
“21 As with PS, (user-specified) logistic regression is the most common method to calculate IPTWs, but also as with PS the IPTWs have no causal interpretation themselves—making them natural quantities for estimation with machine learning.”
MethodStatistics and Causal Inference
Holland, P. W. (1986) · Journal of the American Statistical Association · doi:10.1080/01621459.1986.10478354 · Cited by 5,293
Causal Diagrams for Epidemiologic Research
Greenland, S., Pearl, J., & Robins, J. M. (1999) · Epidemiology · doi:10.1097/00001648-199901000-00008 · Cited by 3,728
Air Pollution and Hospital Admissions for Heart Disease in Eight U.S. Counties
Schwartz, J. (1999) · Epidemiology · doi:10.1097/00001648-199901000-00005 · Cited by 3,038
Estimation and Inference of Heterogeneous Treatment Effects using Random Forests
Wager, S., & Athey, S. (2018) · Journal of the American Statistical Association · doi:10.1080/01621459.2017.1319839 · Cited by 2,917 · cited 1×
“37 We anticipate increasing cross-over from computer science—including machine learning methods—into epidemiology for methods to address measurement error and missing data.”
MethodDoubly robust estimation in missing data and causal inference models
Bang, H., & Robins, J. M. (2005) · Biometrics · doi:10.1111/j.1541-0420.2005.00377.x · Cited by 1,998 · cited 1×
“Simulation studies find that under a range of circumstances, including with large, collinear covariate sets, double robust analyses may be more accurate (less systematic error or bias) than either outcome or exposure methods used in isolation.(32,36) Conversely, it is true that machine learning may not always be an im…”
BackgroundMaternal Pesticide Exposure from Multiple Sources and Selected Congenital Anomalies
Shaw, G. M., Wasserman, C. R., OʼMalley, C. D., Nelson, V., & Jackson, R. J. (1999) · Epidemiology · doi:10.1097/00001648-199901000-00011 · Cited by 1,784
Recursive partitioning for heterogeneous causal effects
Athey, S., & Imbens, G. (2016) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1510489113 · Cited by 1,586 · cited 2×
“Machine learning has been (and is likely to be increasingly) used to identify effect heterogeneity, 35 with recent methodological work (for example) demonstrating how random forests combined with the potential outcomes approach can robustly detect and estimate heterogeneity of treatment effects…”
BackgroundDoubly robust estimation of causal effects
Funk, M. J., Westreich, D., Wiesen, C., Stürmer, T., Brookhart, M. A., & Davidian, M. (2011) · American Journal of Epidemiology · doi:10.1093/aje/kwq439 · Cited by 1,158 · cited 3×
“32,36 Conversely, it is true that machine learning may not always be an improvement over traditional approaches used to estimate the outcome and exposure regressions.”
BackgroundMethodInfluentialTargeted Learning: Causal Inference for Observational and Experimental Data
van der Laan, M. J., & Rose, S. (2011) · Springer Series in Statistics · doi:10.1007/978-1-4419-9782-1 · Cited by 864 · cited 2×
“TMLEs possess many favourable statistical properties beyond their double robustness, including having a loss-based principle for dealing with multiple solutions.(33) Thus, the potential gains from double robust machine learning are 2-fold: we not only have two opportunities to obtain unbiased estimation of our final e…”
BackgroundMethodSelf-citationThe Impact of Residual and Unmeasured Confounding in Epidemiologic Studies: A Simulation Study
Fewell, Z., Davey Smith, G., & Sterne, J. A. C. (2007) · American Journal of Epidemiology · doi:10.1093/aje/kwm165 · Cited by 734 · cited 1×
“41”
Little, R. J., & Rubin, D. B. (2000) · Annual Review of Public Health · doi:10.1146/annurev.publhealth.21.1.121 · Cited by 721 · cited 1×
“Causal inference methods and best-prediction modelling have become less distinct in recent years due to the development of causal inference methods (often premised on a potential outcomes approach 7 or structural causal models 8 ) that harness predictive estimation in pre-final estimation steps.”
BackgroundGood practices for quantitative bias analysis
Lash, T. L., Fox, M. P., MacLehose, R. F., Maldonado, G., McCandless, L. C., & Greenland, S. (2014) · International Journal of Epidemiology · doi:10.1093/ije/dyu149 · Cited by 646 · cited 1×
“Methods such as regression calibration, 38 quantitative bias analysis 39 and multiple over-imputation 40 exist, albeit arguably under-utilized in epidemiology.”
MethodSnowden, J. M., Rose, S., & Mortimer, K. M. (2011) · American Journal of Epidemiology · doi:10.1093/aje/kwq472 · Cited by 511
Self-citationWestreich, D., Lessler, J., & Funk, M. J. (2010) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2009.11.020 · Cited by 502
Stacked generalization: an introduction to super learning
Naimi, A. I., & Balzer, L. B. (2018) · European Journal of Epidemiology · doi:10.1007/s10654-018-0390-z · Cited by 430 · cited 1×
“10,12 ) Rather, the purposes of this paper are: (i) to overview the recent emergence of prediction underpinning steps in contemporary causal inference methods as a useful perspective on contemporary causal inference meth-ods, and (ii) explore the role of machine learning (as one approach to ‘best…”
Measurement Error: Models, Methods, and Applications
Buonaccorsi, J. P. (2010) · Chapman & Hall/CRC eBooks · doi:10.1201/9781420066586 · Cited by 424
Targeted Maximum Likelihood Estimation for Causal Inference in Observational Studies
Schuler, M. S., & Rose, S. (2017) · American Journal of Epidemiology · doi:10.1093/aje/kww165 · Cited by 399 · cited 5×
“4,34 TMLEs possess many favourable statistical properties beyond their double robustness, including having a loss-based principle for dealing with multiple solutions.”
BackgroundInfluentialSelf-citationNaimi, A. I., Cole, S. R., & Kennedy, E. H. (2016) · International Journal of Epidemiology · doi:10.1093/ije/dyw323 · Cited by 332 · cited 1×
“3,25,29 Given that, for this simple example at least, we could use a standard regression model to estimate an effect size for the exposure (cid:2) outcome association to undertake prediction of potential outcomes, why bother?”
MethodBig Data in Public Health: Terminology, Machine Learning, and Privacy
Mooney, S. J., & Pejaver, V. (2018) · Annual Review of Public Health · doi:10.1146/annurev-publhealth-040617-014208 · Cited by 325
Measurement Error: Models, Methods, and Applications
Buonaccorsi, J. P. (2010) · Thammasat University Digital Collections · Cited by 272
Evaluating uses of data mining techniques in propensity score estimation: a simulation study
Setoguchi, S., Schneeweiss, S., Brookhart, M. A., Glynn, R. J., & Cook, E. F. (2008) · Pharmacoepidemiology and Drug Safety · doi:10.1002/pds.1555 · Cited by 271 · cited 1×
“15 comparing recursive partitioning and neural networks with logistic regression.”
MethodPylypchuk, R., Wells, S., Kerr, A., Poppe, K., Riddell, T., Harwood, M., Exeter, D., Mehta, S., Grey, C., Wu, B. P., et al. (2018) · The Lancet · doi:10.1016/s0140-6736(18)30664-0 · Cited by 269
A Unified Approach to Measurement Error and Missing Data: Overview and Applications
Blackwell, M., Honaker, J., & King, G. (2017) · Sociological Methods & Research · doi:10.1177/0049124115585360 · Cited by 221
Mortality Risk Score Prediction in an Elderly Population Using Machine Learning
Rose, S. (2013) · American Journal of Epidemiology · doi:10.1093/aje/kws241 · Cited by 193 · cited 1×
“11 The purpose of this paper is not to review in depth machine learning or causal inference.”
BackgroundSelf-citationThe Parametric g-Formula for Time-to-event Data: Intuition and a Worked Example
Keil, A. P., Edwards, J. K., Richardson, D. B., Naimi, A. I., & Cole, S. R. (2014) · Epidemiology · doi:10.1097/ede.0000000000000160 · Cited by 191 · cited 2×
“31 As ‘big data’ access improves, such approaches to answer policy-relevant questions are likely to increase.”
BackgroundImproving propensity score estimators’ robustness to model misspecification using super learner.
Pirracchio, R., Petersen, M. L., & van der Laan, M. (2015) · American Journal of Epidemiology · doi:10.1093/aje/kwu253 · Cited by 186
Targeted maximum likelihood estimation for a binary treatment: A tutorial
Luque-Fernández, M. Á., Schomaker, M., Rachet, B., & Schnitzer, M. E. (2018) · Statistics in Medicine · doi:10.1002/sim.7628 · Cited by 183 · cited 1×
“4,34 TMLEs possess many favourable statistical properties beyond their double robustness, including having a loss-based principle for dealing with multiple solutions.”
BackgroundPlasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study
Voight, B. F., Peloso, G. M., Orho-Melander, M., Frikke-Schmidt, R., Barbalic, M., Jensen, M. K., Hindy, G., Hólm, H., Ding, E. L., Johnson, T., et al. (2012) · The Lancet · doi:10.1016/s0140-6736(12)60312-2 · Cited by 148 · cited 1×
“high-density lipoproteins (HDLs) strongly predict CVD risk, but HDL itself is not causal of CVD as shown in Mendelian randomization studies.(2) Causal modelling rigorously tests hypotheses generated from theory and content knowledge external to the data with explicit attention to key assumptions such as consistency an…”
BackgroundUsing Super Learner Prediction Modeling to Improve High-dimensional Propensity Score Estimation
Wyss, R., Schneeweiß, S., van der Laan, M. J., Lendle, S., Ju, C., & Franklin, J. M. (2017) · Epidemiology · doi:10.1097/ede.0000000000000762 · Cited by 84 · cited 1×
“19 Recently, machine and ensemble learning methods have been applied to not only best prediction of exposure, but optimal selection and modelling of covariates in the propensity score algorithm based on optimizing the balance of confounding covariates between the exposed and unexposed.”
MethodCan We Train Machine Learning Methods to Outperform the High-dimensional Propensity Score Algorithm?
Karim, M. E., Pang, M., & Platt, R. W. (2017) · Epidemiology · doi:10.1097/ede.0000000000000787 · Cited by 72 · cited 1×
“Recently, machine and ensemble learning methods have been applied to not only best prediction of exposure, but optimal selection and modelling of covariates in the propensity score algorithm based on optimizing the balance of confounding covariates between the exposed and unexposed.(16,20)”
MethodSetodji, C. M., McCaffrey, D. F., Burgette, L. F., Almirall, D., & Griffin, B. A. (2017) · Epidemiology · doi:10.1097/ede.0000000000000734 · Cited by 68
Gruber, S., Logan, R. W., Jarrín, I., Monge, S., & Hernán, M. A. (2015) · Statistics in Medicine · doi:10.1002/sim.6322 · Cited by 64 · cited 1×
“g.(23)); this seems a fertile area to incorporate machine learning into epidemiological causal inference.”
BackgroundImputation approaches for potential outcomes in causal inference
Westreich, D., Edwards, J. K., Cole, S. R., Platt, R. W., Mumford, S. L., & Schisterman, E. F. (2015) · International Journal of Epidemiology · doi:10.1093/ije/dyv135 · Cited by 64 · cited 3×
“3,25,29 Given that, for this simple example at least, we could use a standard regression model to estimate an effect size for the exposure (cid:2) outcome association to undertake prediction of potential outcomes, why bother?”
BackgroundMethodBentley, R., Baker, E., Simons, K., Simpson, J. A., & Blakely, T. (2018) · International Journal of Epidemiology · doi:10.1093/ije/dyy116 · Cited by 52 · cited 2×
“For example, Bentley et al. 22 aimed to estimate the impact of cumulative exposure to social housing, and transitions in and out of social housing, on mental health.”
BackgroundSelf-citationAustin, P. C. (2012) · Multivariate Behavioral Research · doi:10.1080/00273171.2012.640600 · Cited by 51 · cited 2×
“14 However, examples to date are sparse (e.g. 30 ), although we anticipate this to be a growth area for epidemiology and related fields.”
BackgroundPirracchio, R., & Carone, M. (2016) · Statistical Methods in Medical Research · doi:10.1177/0962280216682055 · Cited by 45 · cited 1×
“16,20 Predicting weights for exposures: inverse probability of treatment weights (IPTWs) The PS (as stated above) can also be used to weight analyses with 1/PS for the exposed (or treated), and 1/(1-PS) for the unexposed (or untreated).”
MethodKreif, N., Tran, L., Grieve, R., Stavola, B. L. D., Tasker, R. C., & Petersen, M. L. (2017) · American Journal of Epidemiology · doi:10.1093/aje/kwx213 · Cited by 45
Chittleborough, C., Mittinty, M., Lawlor, D. A., & Lynch, J. W. (2014) · Child Development · doi:10.1111/cdev.12309 · Cited by 43 · cited 1×
“28 This estimation, or prediction, of potential outcomes (and potential mediators) is, again, a pre-final effect estimation step.”
BackgroundSelf-citationSocioeconomic and Tobacco Mediation of Ethnic Inequalities in Mortality over Time
Blakely, T. A., Disney, G., Valeri, L., Atkinson, J., Teng, A. M., Wilson, N., & Gurrin, L. C. (2018) · Epidemiology · doi:10.1097/ede.0000000000000842 · Cited by 24
Self-citationYou are smarter than you think: (super) machine learning in context
Keil, A. P., & Edwards, J. K. (2018) · European Journal of Epidemiology · doi:10.1007/s10654-018-0405-9 · Cited by 16 · cited 1×
“10,12 ) Rather, the purposes of this paper are: (i) to overview the recent emergence of prediction underpinning steps in contemporary causal inference methods as a useful perspective on contemporary causal inference meth-ods, and (ii) explore the role of machine learning (as one approach to ‘best…”
Background
Research works printed but not linked: 1
- Glymour M, Greenland S. Causal diagrams. In: Rothman K, Greenland S, Lash T (eds). Modern Epidemiology. 3rd edn. Philadelphia: Lippincott WIlliams & Wilkins, 2008, pp. 183–212. [[Google Scholar](https://scholar.google.com/scholar_lookup?title=Modern%20Epidemiology&publication_year=2008&)]
Cited by
Tennant, P. W. G., Murray, E. J., Arnold, K. F., Berrie, L., Fox, M. P., Gadd, S. C., Harrison, W. J., Keeble, C., Ranker, L. R., Textor, J., et al. (2021) · International Journal of Epidemiology · doi:10.1093/ije/dyaa213 · Cited by 1,052
“Without this information, many of the reported associations are uninterpretable, since estimating a specific causal effect requires conditioning on a specific set of variables that are determined by strong theoretical principles.(6,7) This is exemplified by the ‘Table 2 fallacy’, which occurs when the coefficients for…”
BackgroundSengupta, P. P., Shrestha, S., Berthon, B., Messas, E., Donal, E., Tison, G. H., Min, J. K., D’hooge, J., Voigt, J.-U., Dudley, J., et al. (2020) · JACC: Cardiovascular Imaging · doi:10.1016/j.jcmg.2020.07.015 · Cited by 221
“analyses methods such as propensity score matching or Bayesian inference, may be better suited; however, newer methods involving ML algorithms are being developed for causal inferences (20,21).”
Where Medical Statistics Meets Artificial Intelligence
Hunter, D. J., & Holmes, C. (2023) · New England Journal of Medicine · doi:10.1056/nejmra2212850 · Cited by 160
Critical appraisal of artificial intelligence-based prediction models for cardiovascular disease
van Smeden, M., Heinze, G., Van Calster, B., Asselbergs, F. W., Vardas, P. E., Bruining, N., de Jaegere, P., Moore, J. H., Denaxas, S., Boulesteix, A. L., et al. (2022) · European Heart Journal · doi:10.1093/eurheartj/ehac238 · Cited by 139
“For discussions on AI that are specifically designed to explore cause and effect, we refer the reader to Blakely et al.(77)”
Leist, A. K., Klee, M., Kim, J. H., Rehkopf, D. H., Bordas, S. P. A., Muniz-Terrera, G., & Wade, S. (2022) · Science Advances · doi:10.1126/sciadv.abk1942 · Cited by 127
“An overview of different papers that use ML for inverse probability weighting and propensity score matching is given in ( 102 ).”
Method
Metadata
- Authors
- Blakely, T., Lynch, J., Simons, K., Bentley, R., & Rose, S.
- Affiliations
- The University of Melbourne (AU)
- The University of Adelaide (AU)
- Harvard University (US)
- Corresponding author
- Blakely, Tony
- Venue
- International Journal of Epidemiology
- Publisher
- Oxford University Press (OUP)
- Published
- 2019-07-11
- Volume
- 49
- Issue
- 6
- Pages
- 2058-2064
- ISSN
- 0300-5771 (ISSN-L)
- 1464-3685
- PMC
- PMC8453374
- OpenAlex
- W2957103736
- Cited by
- 98
- References
- 43
- Citations by year
- 2026: 4 · 2025: 14 · 2024: 16 · 2023: 17 · 2022: 21 · 2021: 11 · 2020: 13 · 2019: 2
- FWCI
- 9.17
- Type
- journal article
- Language
- en
- Open access
- green
- europepmc.org
- Licence
- academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model (vor)
- Locations
- International Journal of Epidemiology (publishedVersion)
- Europe PMC (PubMed Central) (publishedVersion)
- PubMed Central (publishedVersion)
- PubMed (publishedVersion)
- Funders
- Health Research Council of New Zealand Programme (16/443)
- Australian Research Council
- Future Fellowships (FT150100131)
- NHMRC Centre of Research Excellence (GNT1099422)
- NIH Director's New Innovator Award (DP2-MD012722)
- NIH
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Epidemiology
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
- Keywords
- causal inference, predictive modeling, machine learning, propensity score, inverse probability of treatment weighting, g-computation, targeted maximum likelihood estimation, counterfactual modeling, potential outcomes
- MeSH terms
- Machine Learning, Humans, Causality, Propensity Score
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
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.