CTCécile Trottet

Cécile Trottet

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Cécile Trottet has a total of 37 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Federated modular clinical decision support networks for collaborative learning in resource-limited settings

    by Cécile Trottet, Jonathan Doenz, Pierrette M. Mastel, Victor P. Rwandarwacu, Lameck B. Luwanda, Chacha D. Mangu, Mark S. Iscoe, Rainer Tan, Michael Krauthammer, Alexandra V. Kulinkina, et al.

    Imperfect interoperability (IIO), where health facilities record different, often sparse subsets of clinical variables, remains a major barrier to deploying models trained with Federated Learning (FL) in global health settings. We introduce FedMoDN, a novel federated modular neural network architecture for collaborative learning across all features of an IIO distributed dataset, allowing healthcare facilities to use the full complement of their features without sharing, discarding, or imputing any data. We evaluate FedMoDN on a multi-site pediatric dataset comprising ~130,000 medical visits across 92 healthcare facilities in Tanzania and Rwanda. Across both internal and external validation health facilities, FedMoDN matches or surpasses models trained with centralized data sharing and competitive monolithic FL baselines, achieving a mean AUPRC of 0.80 versus 0.77 for the monolithic FL model on 18 external validation health facilities. Its relative advantage over a monolithic FL model rose from 4% (complete data) to 22% when 70% of test-time features were missing, and, unlike monolithic FL models, performance remained stable when health facilities contributed disjoint feature or label subsets. Furthermore, step-wise predictions provide clinically interpretable feature-attribution scores. By coupling IIO resilience with built-in interpretability, FedMoDN offers a promising decision support tool for resource-limited facilities sidelined by conventional FL.

  • Open Access

    Predicting interstitial lung disease progression in patients with systemic sclerosis using attentive neural processes - a EUSTAR study

    by Ahmed Allam, Aron N. Horvath, Matthias Dittberner, Cécile Trottet, Elise Siegert, Vincent Sobanski, Patricia Carreira Delgado, Dagna Lorenzo, Vanessa Smith, Serena Guiducci, et al.

    Background Systemic sclerosis (SSc) is an autoimmune disease with high mortality with lung involvement being the primary cause of death. Progressive interstitial lung disease (ILD) leads to a decline in lung function (forced vital capacity, FVC% predicted) with risk of respiratory failure. These patients could benefit from an early and tailored pharmacological intervention. However, up to date, tools for prediction of individual FVC changes are lacking. In this paper, we aimed at developing a trustworthy machine learning system that is able to guide SSc management by providing not only robust FVC predictions, but also uncertainty quantification (i.e. the degree of certainty of model prediction) as well as similarity-based explainability for any patient P (i.e. a list of past SSc patients with similar FVC trajectories like P). We further aimed to identify the key clinical factors influencing the model’s predictions and to use model-guided data representation to identify SSc patients with similar sequential FVC measurements. Methods We trained and evaluated machine learning (ML) models to predict SSc-ILD trajectory as measured by FVC% predicted values using the international SSc database managed by the European Scleroderma Trials and Research group (EUSTAR), which comprises clinical, laboratory and functional parameters. EUSTAR records patients’ data in annual assessment visits, and, given any visit, we aimed at predicting the FVC value of a patient’s subsequent visit, taking into account all available patient data (i.e. baseline and follow-up visit data up to the time point where we make the prediction). For training of our ML models, we included 2220 SSc patients that had at least 3 recorded visits in the EUSTAR database, were at least 18 years old, had confirmed ILD and sufficient clinical documentation. We developed sequential ML models implementing the attentive neural process formalism with either a recurrent (ANP RNN) or transformer encoder (ANP transformer) architecture. We compared these architectures with baseline sequential models including gated recurrent neural networks (RNNs) and multi-head self-attention transformer-based networks. Baseline non-sequential models included tree-based models such as gradient boosting trees, and regression-based models with varying regularization schemes. Our experiments used stratified 5-fold cross-validation to train and test the models using the average root mean squared error (RMSE), weighted RMSE, and mean absolute error (MAE) as performance metrics. We computed the coverage and Winkler score for uncertainty quantification, SHAP values for grading the input features importance and used the data embeddings of the ANP architectures for both similarity-based explainability and the identification of similar SSc patient journeys. Results Patients’ baseline FVC scores ranged from 22 to 150% predicted with a mean (SD) of 90.53% predicted (21.52). Our deep learning models showed better performance for FVC forecasting, compared to tree- and regression-based models. The top performing ANP RNN architecture was able to closely model future FVC values with average (SD) performance of 8.240 (0.168) weighted RMSE and 6.94 (0.190) MAE that was further used as feature generator for a logistic regression trained to predict a FVC% decline of at least 10% points achieving 0.704 AUC score. In comparison, a naïve baseline using the mean FVC value as a predictor achieved much lower FVC forecasting capabilities, with 18.718 (0.317) weighted RMSE, and 17.619 (0.599) MAE. SHAP value analysis indicated that prior FVC measurements, diffusion of carbon monoxide (DLCO) values, skin involvement, age, anti-centromere positivity, dyspnea and CRP-elevation contributed most to deep-learning-based FVC predictions. Regarding uncertainty quantification, ANP RNN achieved 79% coverage (i.e. the model would provide uncertainty estimates that included the true future FVC value in 79 out of 100 predictions) out of the box, and 90% using an additional conformal prediction module with an corresponding Winkler score of 892 (indicating the width of the uncertainty estimate plus penalty for mistakes), smaller than any other model at the same coverage level. We further demonstrate how the data abstraction provided by the ANP RNN model (embeddings) allows for deriving similar patient trajectories (for similarity-based explanation). Conclusions Our study demonstrates the feasibility of FVC forecasting and thus the ability to predict ILD trajectories in individual SSc patients using deep learning. We show that model predictions can be paired with uncertainty quantification and similarity-based model explainability, which are crucial elements for deploying trustworthy ML algorithms. Our study is thus an important first step towards reliable automated ILD trajectory (i.e. FVC%) prediction system with potential clinical utility.


  • Open Access

    Modeling Complex Disease Trajectories using Deep Generative Models with Semi-Supervised Latent Processes

    by Cécile Trottet, Manuel Schürch, Ahmed Allam, Imon Barua, Liubov Petelytska, Oliver Distler, Anna-Maria Hoffmann-Vold, Michael Krauthammer, The Eustar Collaborators

    In this paper, we propose a deep generative time series approach using latent temporal processes for modeling and holistically analyzing complex disease trajectories. We aim to find meaningful temporal latent representations of an underlying generative process that explain the observed disease trajectories in an interpretable and comprehensive way. To enhance the interpretability of these latent temporal processes, we develop a semi-supervised approach for disentangling the latent space using established medical concepts. By combining the generative approach with medical knowledge, we leverage the ability to discover novel aspects of the disease while integrating medical concepts into the model. We show that the learned temporal latent processes can be utilized for further data analysis and clinical hypothesis testing, including finding similar patients and clustering the disease into new sub-types. Moreover, our method enables personalized online monitoring and prediction of multivariate time series including uncertainty quantification. We demonstrate the effectiveness of our approach in modeling systemic sclerosis, showcasing the potential of our machine learning model to capture complex disease trajectories and acquire new medical knowledge.

  • Open Access

    Modular Clinical Decision Support Networks (MoDN) -- Updatable, Interpretable, and Portable Predictions for Evolving Clinical Environments

    by Cécile Trottet, Thijs Vogels, Martin Jaggi, Mary-Anne Hartley

    Data-driven Clinical Decision Support Systems (CDSS) have the potential to improve and standardise care with personalised probabilistic guidance. However, the size of data required necessitates collaborative learning from analogous CDSS's, which are often unsharable or imperfectly interoperable (IIO), meaning their feature sets are not perfectly overlapping. We propose Modular Clinical Decision Support Networks (MoDN) which allow flexible, privacy-preserving learning across IIO datasets, while providing interpretable, continuous predictive feedback to the clinician. MoDN is a novel decision tree composed of feature-specific neural network modules. It creates dynamic personalised representations of patients, and can make multiple predictions of diagnoses, updatable at each step of a consultation. The modular design allows it to compartmentalise training updates to specific features and collaboratively learn between IIO datasets without sharing any data.