Tracking clusters of patients over time enables extracting information from medico-administrative databases.

CONTEXT: Identifying clusters (i.e., subgroups) of patients from the analysis of medico-administrative databases is particularly important to better understand disease heterogeneity. However, these databases contain different types of longitudinal variables which are measured over different follow-up periods, generating truncated data. It is therefore fundamental to develop clustering approaches that can handle this type of data. OBJECTIVE: We propose here cluster-tracking approaches to identify clusters of patients from truncated longitudinal data contained in medico-administrative databases. MATERIAL AND METHODS: We first cluster patients at each age. We then track the identified clusters over ages to construct cluster-trajectories. We compared our novel approaches with three classical longitudinal clustering approaches by calculating the silhouette score. As a use-case, we analyzed antithrombotic drugs used from 2008 to 2018 contained in the Échantillon Généraliste des Bénéficiaires (EGB), a French national cohort. RESULTS: Our cluster-tracking approaches allow us to identify several cluster-trajectories with clinical significance without any imputation of data. The comparison of the silhouette scores obtained with the different approaches highlights the better performances of the cluster-tracking approaches. CONCLUSION: The cluster-tracking approaches are a novel and efficient alternative to identify patient clusters from medico-administrative databases by taking into account their specificities.

Predicting veal-calf trading events in France.

Global trade has been ranked as one of the top five drivers of infectious disease threat events. More specifically, livestock trade is known to increase the speed at which infectious diseases circulate and to facilitate their dissemination over large distances Therefore, predicting animal movements arising from trade is crucial for assessing epidemic risk and the impact of preventive measures. In this study, we developed a statistical framework for predicting trading events using predictors accessible from routinely collected data. We focused on veal calves, a category of animals with significant commercial value; the dataset considered the veal calf trade in France between January 2011 and June 2019. A subset of farms with consistent trade behaviour over time was built to be used throughout the study. To predict sale or purchase event occurrences, our predictive framework was based on random forests as a binary classification tool, an approach that allows a large number of potential predictors. We explored the robustness of model predictions with respect to the delay in data acquisition and prediction lag time. Overall, sales were more accurately predicted than purchasing events. Unsurprisingly, a delay in data acquisition led to a decrease in the performance of indicators, whereas prediction lag time had little impact. Sale-related predictors mostly reflected past trading events, whereas purchase-related predictors were associated with past trading events, farm management and general farm characteristics. The model outputs also suggested that the veal calf trading network is driven by sales rather than by purchases. Regardless of the length of the delay in data acquisition and prediction lag, the random forest approach fitted on data with municipality as trading unit and a 28-day trading period provided better performance scores (F1-score, positive predictive value and negative predictive value) than scenarios with finer temporal and spatial aggregation units. Predicted trade events can therefore be used to reconstruct the entire veal calf trading network and transfers between selling and purchasing units for each period. This predicted network could be further used to simulate the spread of pathogens via animal trade.

Tracking Temporal Clusters from Patient Networks.

Creating homogeneous groups (clusters) of patients from medico-administrative databases provides a better understanding of health determinants. But in these databases, patients have truncated care pathways. We developed an approach based on patient networks to construct care trajectories from such truncated data. We tested this approach on antithrombotic treatments prescribed from 2008 to 2018 contained in the échantillon généraliste des bénéficiaires (EGB). We constructed a patient network for each patients' age (years from birth). We then applied the Markov clustering algorithm in each network. The care trajectories were finally constructed by matching clusters identified in two consecutive networks. We calculated the silhouette score to assess the performance of this network approach compared to three existing approaches. We identified 12 care trajectories that we were able to associate with pathologies. The best silhouette score was obtained for the network approach. Our approach allowed to highlight care trajectories taking into account the longitudinal, multidimensional and truncated nature of data from medico-administrative databases.