The increasing availability of Electronic Health Records (EHRs) data presents novel opportunities to create data driven tools based on Artificial Intelligence for clinical purposes. While healthcare reimbursement data may not strictly constitute EHRs, they nonetheless offer valuable insights into patients’ medical pathways, including medical visits, procedures, and medication usage. However, the inherent complexity and dimensionality of this type of data pose significant hurdles for the direct application of Machine Learning techniques. Consequently, unsupervised methods for encoding patients’ medical data while reducing dimensionality are imperative. In this regard, representation learning emerges as a interesting approach for creating meaningful patient representations.Our contribution involves a comprehensive benchmarking assessment of three prominent representation learning paradigms. We aim to encode and cluster breast cancer patients’ medical pathways using reimbursement data extracted from the French Nationwide Healthcare Database (SNDS). Our findings underscore the limitations of solely evaluating representation learning approaches based on classical machine learning performance metrics. Thus, we advocate for evaluating the quality of the patients’ representation learning latent space through statistical analyses and developing new metrics that reflect the clinical reality of patients.

ReLU neural networks suffer from a problem of explainability because they partition the input space into a lot of polyhedrons. This paper proposes a constrained neural network model that replaces polyhedrons by orthotopes: each hidden neuron processes only a single component of the input signal. When the number of hidden neurons is large, we show that our neural network is equivalent to a logistic regression whose input is a non-linear transformation of the processed signal. Hence, the training of our neural network always converges to a unique solution. Numerical simulations show that the loss of performance with respect to state-of-the-art methods is negligible even though our neural network is strongly constrained on robustness and explainability.

This paper proposes an understandable neural network whose score function is modeled as an additive sum of univariate spline functions. It extends usual understandable models like generative additive models, spline-based models, and neural additive models. It is shown that this neural network can be approximated by a logistic regression whose inputs are obtained with a non-linear preprocessing of input data. This preprocessing depends on the neural network initialization but this paper establishes that it can be replaced by a non random kernel-based preprocessing that no longer depends on the initialization. Hence, the convergence of the training process is guaranteed and the solution is unique for a given training dataset.