Jump to comment:

• Machine learning to predict death after cardiac surgery in patients with infective endacarditis: Why not! But, where are the (big) data please?
  Jerome Allyn, Nicolas Allou and Cyril Ferdynus
  Published on: 25 May 2017

• Published on: 25 May 2017

  Machine learning to predict death after cardiac surgery in patients with infective endacarditis: Why not! But, where are the (big) data please?

  • Jerome Allyn, Intensivist Réanimation Polyvalente Site Félix Guyon, Centre Hospitalier Universitaire La Réunion, Saint-Denis, France
  • Other Contributors:
    • Nicolas Allou, Intensivist
    • Cyril Ferdynus, Biostatistician

  The study of Olmos et al., on prediction of in-hospital mortality in patients with active infective endocarditis undergoing cardiac surgery, is of great interest.1
  Indeed, this topic is fascinating because it is complicated to make a choice in so dramatic and not so rare situation.
  To help with this decision-making, the authors proposed a model for predicting hospital mortality: a classic multivariate logistic regression model.
  However, the editorial published with this article evokes in the title a new method: machine learning.2 Machine learning, which is a field of artificial intelligence, has already been used for predicting hospital mortality after elective cardiac surgery.3 This study aimed at comparing a machine learning model, a classic logistic regression model and EuroSCORE II on a cohort including 6,520 patients. The comparison of these models was based on ROC curves and decision curve analysis (DCA).4 Whatever the method of comparison, machine learning model was more accurate than other models.
  Our experience in this area probably allows us to make some comments on this editorial. Considering, the increase of studies comparing machine learning with logistic regression, it is now known that supervised machine learning algorithms improve the prediction of post-operative mortality. However, the size of the cohort used in the present study makes it difficult to apply machine learning algorithms. Indeed, this cohort comprised 671 patients who...

  Show More

  The study of Olmos et al., on prediction of in-hospital mortality in patients with active infective endocarditis undergoing cardiac surgery, is of great interest.1
  Indeed, this topic is fascinating because it is complicated to make a choice in so dramatic and not so rare situation.
  To help with this decision-making, the authors proposed a model for predicting hospital mortality: a classic multivariate logistic regression model.
  However, the editorial published with this article evokes in the title a new method: machine learning.2 Machine learning, which is a field of artificial intelligence, has already been used for predicting hospital mortality after elective cardiac surgery.3 This study aimed at comparing a machine learning model, a classic logistic regression model and EuroSCORE II on a cohort including 6,520 patients. The comparison of these models was based on ROC curves and decision curve analysis (DCA).4 Whatever the method of comparison, machine learning model was more accurate than other models.
  Our experience in this area probably allows us to make some comments on this editorial. Considering, the increase of studies comparing machine learning with logistic regression, it is now known that supervised machine learning algorithms improve the prediction of post-operative mortality. However, the size of the cohort used in the present study makes it difficult to apply machine learning algorithms. Indeed, this cohort comprised 671 patients who had cardiac surgery, with a cohort of derivation containing only 424 patients (300 discharged alive and 124 died after surgery). This cohort was probably too small for efficient use of machine learning algorithms. However, it would have been interesting to fit a machine learning model in this field. Moreover, the fact that this database is spread over 18 years compromises the validity of this model in 2017.
  In conclusion, this article and the accompanying editorial very well illustrate the current problem of artificial intelligence in medicine: how to find reliable data in sufficient quantities?

  References
  1 Olmos C, Vilacosta I, Habib G, et al. Risk score for cardiac surgery in active left-sided infective endocarditis. Heart 2017; 10.1136/heartjnl-2016-311093.
  2 Donal E, Flecher E, Tattevin P. Machine learning to support decision-making for cardiac surgery during the acute phase of infective endocarditis. Heart Published Online First: 8 May 2017. doi:10.1136/heartjnl-2017-311512
  3 Allyn J, Allou N, Augustin P, et al. A Comparison of a Machine Learning Model with EuroSCORE II in Predicting Mortality after Elective Cardiac Surgery: A Decision Curve Analysis. PloS One 2017;12:e0169772. doi:10.1371/journal.pone.0169772
  4 Fitzgerald M, Saville BR, Lewis RJ. Decision curve analysis. JAMA 2015;313:409–10. doi:10.1001/jama.2015.37

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top