Improving Cause-of-Death Classification from Verbal Autopsy Reports

Abstract

In many lower-and-middle income countries including South Africa, data access in health facilities is restricted due to patient privacy and confidentiality policies. Further, since clinical data is unique to individual institutions and laboratories, there are insufficient data annotation standards and conventions. As a result of the scarcity of textual data, natural language processing (NLP) techniques have fared poorly in the health sector. A cause of death (COD) is often determined by a verbal autopsy (VA) report in places without reliable death registration systems. A non-clinician field worker does a VA report using a set of standardized questions as a guide to uncover symptoms of a COD. This analysis focuses on the textual part of the VA report as a case study to address the challenge of adapting NLP techniques in the health domain. We present a system that relies on two transfer learning paradigms of monolingual learning and multi-source domain adaptation to improve VA narratives for the target task of the COD classification. We use the Bidirectional Encoder Representations from Transformers (BERT) and Embeddings from Language Models (ELMo) models pre-trained on the general English and health domains to extract features from the VA narratives. Our findings suggest that this transfer learning system improves the COD classification tasks and that the narrative text contains valuable information for figuring out a COD. Our results further show that combining binary VA features and narrative text features learned via this framework boosts the classification task of COD.

Multi-step Transfer Learning in Natural Language Processing for the Health Domain

“…On more recent works that incorporate context and character information, Yan et al [58] achieved recall scores of 0.6990 while Manaka et al [79] gave 0.6000. Considering the improvement added by character information on improving COD classification, Manaka et al [49] added features from multiple domains and reported a score of 0.8755. These findings suggest that our proposed transfer learning methodology can adapt to VA datasets across various demographics as the works compared against used VA datasets collected in other developing countries, including Ghana, India and Tanzania.…”

“…Similar to the work by Manaka et al [49], the initial step of the Multi-Step Transfer Learning framework involves an exploratory search for models across a number of domains to identify the one that best represents the VA corpus. Three sets of ELMo language models were trained in three different domains, with the objective of identifying the optimal text representations for the VA language modeling task.…”

“…This hybrid deep neural network comprises four components: Embedding, CNN, Transformer, and a Multilayer Perceptron (MLP) with Fully Connected layers for the prediction of CRISPR/Cas9 single guide RNA cleavage efficiency. MSDA has also been shown to enhance the COD classification task in VA reports [49]. In their study, Manaka et al [49] demonstrated that better performance could be achieved by combining the English and biomedical domains for learning representations from the VA corpus.…”

“…MSDA has also been shown to enhance the COD classification task in VA reports [49]. In their study, Manaka et al [49] demonstrated that better performance could be achieved by combining the English and biomedical domains for learning representations from the VA corpus. They incorporated character-level information for learning VA embeddings in the English domain and word-level information for learning VA embeddings in the biomedical domain, resulting in improved results compared to using embeddings from these domains separately.…”