Causal inference and counterfactual prediction in machine learning for actionable healthcare

Abstract

Big data, high-performance computing, and (deep) machine learning are increasingly becoming key to precision medicine—from identifying disease risks and taking preventive measures, to making diagnoses and personalizing treatment for individuals. Precision medicine, however, is not only about predicting risks and outcomes, but also about weighing interventions. Interventional clinical predictive models require the correct specification of cause and effect, and the calculation of so-called counterfactuals, that is, alternative scenarios. In biomedical research, observational studies are commonly affected by confounding and selection bias. Without robust assumptions, often requiring a priori domain knowledge, causal inference is not feasible. Data-driven prediction models are often mistakenly used to draw causal effects, but neither their parameters nor their predictions necessarily have a causal interpretation. Therefore, the premise that data-driven prediction models lead to trustable decisions/interventions for precision medicine is questionable. When pursuing intervention modelling, the bio-health informatics community needs to employ causal approaches and learn causal structures. Here we discuss how target trials (algorithmic emulation of randomized studies), transportability (the licence to transfer causal effects from one population to another) and prediction invariance (where a true causal model is contained in the set of all prediction models whose accuracy does not vary across different settings) are linchpins to developing and testing intervention models.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Fig. 1: Conditional versus interventional probabilities.
Fig. 2: Examples of confounding bias and collider bias.
Fig. 3: An example of M-bias.
Fig. 4: A selection diagram for illustrating transportability.

References

  1. 1.

    Norgeot, B., Glicksberg, B. S. & Butte, A. J. A call for deep-learning healthcare. Nat. Med. 25, 14–15 (2019).

    Google Scholar 

  2. 2.

    Wiens, J. et al. Do no harm: a roadmap for responsible machine learning for health care. Nat. Med. 25, 1337–1340 (2019).

    Google Scholar 

  3. 3.

    Silver, D. et al. Mastering the game of Go without human knowledge. Nature 550, 354–359 (2017).

    Google Scholar 

  4. 4.

    Jin, P., Keutzer, K. & Levine, S. Regret minimization for partially observable deep reinforcement learning. In 35th Int. Conf. Machine Learning 80, 2342–2351 (ICML, 2018).

  5. 5.

    Pearl, J. & Mackenzie, D. The Book of Why: The New Science of Cause and Effect (Basic Books, 2018).

  6. 6.

    Chouldechova, A. Fair prediction with disparate impact: a study of bias in recidivism prediction instruments. Big Data 5, 153–163 (2017).

    Google Scholar 

  7. 7.

    Kusner, M., Loftus, J., Russell, C. & Silva, R. Counterfactual fairness. In Advances in Neural Information Processing Systems Vol. 31, 4069–4079 (MIT Press, 2017).

  8. 8.

    Christodoulou, E. et al. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J. Clin. Epidemiol. 110, 12–22 (2019).

    Google Scholar 

  9. 9.

    Bian, J., Buchan, I., Guo, Y. & Prosperi, M. Statistical thinking, machine learning. J. Clin. Epidemiol. 116, 136–137 (2019).

    Google Scholar 

  10. 10.

    Baker, R. E., Peña, J. M., Jayamohan, J. & Jérusalem, A. Mechanistic models versus machine learning, a fight worth fighting for the biological community? Biol. Lett. 14, 20170660 (2018).

    Google Scholar 

  11. 11.

    Gulshan, V. et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 316, 2402–2410 (2016).

    Google Scholar 

  12. 12.

    Winkler, J. K. et al. Association between surgical skin markings in dermoscopic images and diagnostic performance of a deep learning convolutional neural network for melanoma recognition. JAMA Dermatol. 155, 1135–1141 (2019).

    Google Scholar 

  13. 13.

    Komorowski, M., Celi, L. A., Badawi, O., Gordon, A. C. & Faisal, A. A. The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care. Nat. Med. 24, 1716–1720 (2018).

    Google Scholar 

  14. 14.

    Lewis, D. K. Causation J. Philos. 70, 556–567 (1973).

    Google Scholar 

  15. 15.

    Mackie, J. L. The Cement of the Universe (Clarendon, 1974).

  16. 16.

    Pearl, J. Causality: Models, Reasoning and Inference (Cambridge Univ. Press, 2009).

  17. 17.

    Rothman, K. J., Greenland, S. & Lash, T. Modern Epidemiology 3rd edn (Lippincott Williams & Wilkins, 2012).

  18. 18.

    Lehmann, E. L. Model specification: the views of Fisher and Neyman, and later developments. Stat. Sci. 5, 160–168 (1990).

    MathSciNet  MATH  Google Scholar 

  19. 19.

    Vansteelandt, S., Bekaert, M. & Claeskens, G. On model selection and model misspecification in causal inference. Stat. Meth. Med. Res. 21, 7–30 (2012).

    MathSciNet  MATH  Google Scholar 

  20. 20.

    Asteriou, D., Hall, S. G., Asteriou, D. & Hall, S. G. in Applied Econometrics 2nd edn 176–197 (Palgrave Macmillan, 2016).

  21. 21.

    Sackett, D. L. Bias in analytic research. J. Chronic Dis. 32, 51–63 (1979).

    Google Scholar 

  22. 22.

    Banack, H. R. & Kaufman, J. S. The ‘obesity paradox’ explained. Epidemiology 24, 461–462 (2013).

    Google Scholar 

  23. 23.

    Pearl, J. Causal diagrams for empirical research. Biometrika 82, 669–688 (1995).

    MathSciNet  MATH  Google Scholar 

  24. 24.

    Greenland, S., Pearl, J. & Robins, J. M. Causal diagrams for epidemiologic research. Epidemiology 10, 37–48 (1999).

    Google Scholar 

  25. 25.

    Westreich, D. & Greenland, S. The table 2 fallacy: Presenting and interpreting confounder and modifier coefficients. Am. J. Epidemiol. 177, 292–298 (2013).

    Google Scholar 

  26. 26.

    Wei, L., Brookhart, M. A., Schneeweiss, S., Mi, X. & Setoguchi, S. Implications of m bias in epidemiologic studies: A simulation study. Am. J. Epidemiol. 176, 938–948 (2012).

    Google Scholar 

  27. 27.

    Cooper, G. F. et al. An evaluation of machine-learning methods for predicting pneumonia mortality. Artif. Intell. Med. 9, 107–138 (1997).

    Google Scholar 

  28. 28.

    Ambrosino, R., Buchanan, B. G., Cooper, G. F. & Fine, M. J. The use of misclassification costs to learn rule-based decision support models for cost-effective hospital admission strategies. In Proc. Annual Symp. Computer Applications Medical Care 304–308 (AMIA, 1995).

  29. 29.

    Caruana, R. et al. Intelligible models for healthcare: predicting pneumonia risk and hospital 30-day readmission. in Proc. ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining 1721–1730 (ACM, 2015).

  30. 30.

    Lucero, R. J. et al. A data-driven and practice-based approach to identify risk factors associated with hospital-acquired falls: applying manual and semi- and fully-automated methods. Int. J. Med. Inform. 122, 63–69 (2019).

    Google Scholar 

  31. 31.

    Hernán, M. A. & Robins, J. M. Using big data to emulate a target trial when a randomized trial is not available. Am. J. Epidemiol. 183, 758–764 (2016).

    Google Scholar 

  32. 32.

    Petito, L. C. et al. Estimates of overall survival in patients with cancer receiving different treatment regimens: emulating hypothetical target trials in the surveillance, epidemiology, and end results (SEER)–Medicare linked database. JAMA Netw. Open 3, e200452–e200452 (2020).

    Google Scholar 

  33. 33.

    Pearl, J. Causal diagrams for empirical research. Biometrika 82, 669–688 (1995).

    MathSciNet  MATH  Google Scholar 

  34. 34.

    Westland, J. C. Structural Equation Models 1–15 (Springer, 2019).

  35. 35.

    Bollen, K. A. & Pearl, J. in Handbook of Causal Analysis for Social Research (ed. Morgan, S. L.) 301–328 (Springer, 2013).

  36. 36.

    Hernán, M. A. & Robins, J. M. Estimating causal effects from epidemiological data. J. Epidemiol. Commun. Health 60, 553 (2006).

    Google Scholar 

  37. 37.

    van der Laan, M. J. & Rubin, D. Targeted maximum likelihood learning. Int. J. Biostat. 6, 2 (2006).

    MathSciNet  Google Scholar 

  38. 38.

    Schuler, M. S. & Rose, S. Targeted maximum likelihood estimation for causal inference in observational studies. Am. J. Epidemiol. 185, 65–73 (2017).

    Google Scholar 

  39. 39.

    van der Laan, M. J. & Rose, S. Targeted Learning: Causal Inference For Observational And Experimental Data (Springer, 2011).

  40. 40.

    Naimi, A. I., Cole, S. R. & Kennedy, E. H. An introduction to g methods. Int. J. Epidemiol. 46, 756–762 (2017).

    Google Scholar 

  41. 41.

    Robins, J. M. & Hernán, M. A. in Longitudinal Data Analysis (eds Fitzmaurice, G. et al.) 553–599 (CRC, 2008).

  42. 42.

    Rosenbaum, P. R. & Rubin, D. B. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41–55 (1983).

    MathSciNet  MATH  Google Scholar 

  43. 43.

    Li, J., Ma, S., Le, T., Liu, L. & Liu, J. Causal decision trees. IEEE Trans. Knowl. Data Eng. 29, 257–271 (2017).

    Google Scholar 

  44. 44.

    Hahn, P. R., Murray, J. & Carvalho, C. M. Bayesian regression tree models for causal inference: regularization, confounding, and heterogeneous effects. Bayesian Anal. https://doi.org/10.1214/19-BA1195 (2020).

  45. 45.

    Lu, M., Sadiq, S., Feaster, D. J. & Ishwaran, H. Estimating individual treatment effect in observational data using random forest methods. J. Comput. Graph. Stat. 27, 209–219 (2018).

    MathSciNet  Google Scholar 

  46. 46.

    Schneeweiss, S. et al. High-dimensional propensity score adjustment in studies of treatment effects using health care claims data. Epidemiology 20, 512–522 (2009).

    Google Scholar 

  47. 47.

    Verma, T. & Pearl, J. in Machine Intelligence and Pattern Recognition Vol. 9 (eds Shachter, R. D. et al.) 69–76 (Elsevier, 1990).

  48. 48.

    Jaber, A., Zhang, J. & Bareinboim, E. Causal identification under Markov equivalence. In 34th Conf. Uncertainty in Artificial Intelligence (UAI, 2018).

  49. 49.

    Richardson, T. in Compstat (eds Dutter, R. & Grossmann, W.) 482–487 (Springer, 1994).

  50. 50.

    Heckerman, D., Meek, C. & Cooper, G. In Innovations in Machine Learning (eds Holmes, D. E. & Jain, L. C.) 1–28 (Sprigner, 2006).

  51. 51.

    Peter Spirtes, C. G. and R S. Causation, Prediction, and Search 2nd edn (MIT Press, 2003).

  52. 52.

    Glymour, C., Zhang, K. & Spirtes, P. Review of causal discovery methods based on graphical models. Front. Genet. 10, 524 (2019).

    Google Scholar 

  53. 53.

    Colombo, D. & Maathuis, M. H. Order-independent constraint-based causal structure learning. J. Mach. Learn. Res. 15, 3921–3962 (2014).

    MathSciNet  MATH  Google Scholar 

  54. 54.

    Shalit, U., Johansson, F. D. & Sontag, D. Estimating individual treatment effect: generalization bounds and algorithms. In Proc. 34th Int. Conf. Machine Learning Vol. 70 (eds Precup, D. & Teh, Y. W.) 3076–3085 (PMLR, 2017).

  55. 55.

    Hartford, J., Lewis, G., Leyton-Brown, K. & Taddy, M. Deep {IV}: a flexible approach for counterfactual prediction. In Proc. 34th Int. Conf. Machine Learning Vol. 70 (eds Precup, D. & Teh, Y. W.) 1414–1423 (PMLR, 2017).

  56. 56.

    Pearl, J. & Bareinboim, E. External validity: from do-calculus to transportability across populations. Stat. Sci. 29, 579–595 (2014).

    MathSciNet  MATH  Google Scholar 

  57. 57.

    Dahabreh, I. J., Robertson, S. E., Tchetgen, E. J., Stuart, E. A. & Hernán, M. A. Generalizing causal inferences from individuals in randomized trials to all trial‐eligible individuals. Biometrics 75, 685–694 (2019).

    MathSciNet  MATH  Google Scholar 

  58. 58.

    Bareinboim, E. & Pearl, J. Causal inference and the data-fusion problem. Proc. Natl Acad. Sci. USA 113, 7345–7352 (2016).

    Google Scholar 

  59. 59.

    Pearl, J. & Bareinboim, E. Transportability of causal and statistical relations: a formal approach. In Proc. IEEE Int. Conf. Data Mining (IEEE, 2011).

  60. 60.

    Lee, S., Correa, J. D. & Bareinboim, E. General identifiability with arbitrary surrogate experiments. In Proc. 35th Conf. Uncertainty in Artificial Intelligence (UAI, 2019).

  61. 61.

    Huang, J., Smola, A. J., Gretton, A., Borgwardt, K. M. & Schölkopf, B. Correcting sample selection bias by unlabeled data. In Advances in Neural Information Processing Systems Vol. 19 (eds Schölkopf, B. et al.) 601–609 (MIT Press, 2007).

  62. 62.

    Peters, J., Bühlmann, P. & Meinshausen, N. Causal inference by using invariant prediction: identification and confidence intervals. J. R. Stat. Soc. Ser. B Stat. Methodol. 78, 947–1012 (2016).

    MathSciNet  MATH  Google Scholar 

  63. 63.

    Subbaswamy, A., Schulam, P. & Saria, S. Preventing failures due to dataset shift: learning predictive models that transport. In Proc. 22nd Int. Conf. Artificial Intelligence and Statistics 3118–3127 (AiStats, 2019).

  64. 64.

    Hernán, M. A., Hsu, J. & Healy, B. A second chance to get causal inference right: a classification of data science tasks. CHANCE 32, 42–49 (2019).

    Google Scholar 

  65. 65.

    Wiens, J. et al. Do no harm: a roadmap for responsible machine learning for health care. Nat. Med. 25, 1337–1340 (2019).

    Google Scholar 

  66. 66.

    Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat. Mach. Intell. 1, 206–215 (2019).

    Google Scholar 

  67. 67.

    Kusner, M. J. & Loftus, J. R. The long road to fairer algorithms. Nature 578, 34–36 (2020).

    Google Scholar 

  68. 68.

    van Amsterdam, W. A. C., Verhoeff, J. J. C., de Jong, P. A., Leiner, T. & Eijkemans, M. J. C. Eliminating biasing signals in lung cancer images for prognosis predictions with deep learning. npj Digit. Med. 2, 122 (2019).

    Google Scholar 

  69. 69.

    Moons, K. G. M. et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration. Ann. Intern. Med. 162, W1–W73 (2015).

    Google Scholar 

Download references

Acknowledgements

J.B.’s, Y.G.’s and M.P.’s research for this work was in part supported by the University of Florida (UF)’s Creating the Healthiest Generation—Moonshot initiative, supported by the UF Office of the Provost, UF Office of Research, UF Health, UF College of Medicine and UF Clinical and Translational Science Institute. M.W.’s research for this work was supported in part by the Lanzillotti–McKethan Eminent Scholar Endowment.

Author information

Affiliations

Authors

Contributions

M.P., Y.G., J.B. and M.W. conceived the premise, wrote the paper, designed the figures and tables, and revised the paper. M.S., X.E. and S.R. contributed to specific sections, aided with the figures and tables, and with revision. J.K., I.B. and J.M. contributed to specific sections and helped with revisions.

Corresponding author

Correspondence to Mattia Prosperi.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Prosperi, M., Guo, Y., Sperrin, M. et al. Causal inference and counterfactual prediction in machine learning for actionable healthcare. Nat Mach Intell 2, 369–375 (2020). https://doi.org/10.1038/s42256-020-0197-y

Download citation

Further reading

Search

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing