Imputing Growth Snapshot Similarity in Early Childhood Development: A Tensor Decomposition Approach

In this paper, we discuss a tensor decomposition method for imputing similarity scores between individual clinical pictures at predefined patient age intervals in order to construct a dynamic similarity network of patients with respect to early childhood anthropomorphic development. The method leverages Canonical Polyadic Decomposition (or PARAFAC) to compute missing Euclidean similarity scores between pairwise growth pictures, made up of height and weight measurements. We construct a tensor made up of serial affinity matrices to model how the similarities between different patients change over different trajectory snapshots. We intend to use this method to aid Un Kilo de Ayuda (UKA), a non-governmental organization located in Mexico that is made up of facilitators seeking to identify children at risk for malnutrition and suboptimal development. This tensor completion strategy will assist UKA with determining pairs of children with similar clinical pictures, so that they can better assist with selecting treatment strategies and ultimately build better programs tailored to specific families’ needs.

I. Introduction

The rate of generated health data has grown exponentially in recent years due to the expansion of electronic medical records, wearable technologies, and health tracking applications. As a result, healthcare providers are investing their resources into building platforms that can leverage this data to improve patient health. The trend in healthcare is shifting from cure to prevention. Hospitals and healthcare systems house useful repositories of big data (like patient records, test reports, medical images, etc.) that can be leveraged to cut the costs of healthcare, to improve reliability and efficiency, and to provide more effective treatments to patients [19]. Applying data science methods to health data has been proven to assist with such advances. Some success stories include using wearable technologies to monitor and prevent health problems, advancing pharmaceutical research to help find cures for diseases, and reducing hospital readmissions to cut healthcare costs, among many others [2, 7, 20]. Leveraging patient-patient similarity is the backbone behind many of these models. However, patients do not always comply with appointment schedules, and occasionally measurements are missed during routine checkups. These events leave gaps in patient records, which hinder machine learning methods that take these values into consideration when making predictions.