Patient-centred care and good communication between patients and healthcare providers has been shown to improve medical adherence. However, it is costly and difficult to provide, especially in Global Health settings where patient volumes are high and costs must be kept low. As chat tools such as WhatsApp and WeChat are becoming widely adopted, provider-patient chat offers exciting opportunities for patient outreach. In particular we are seeing an increase in chat groups of patient peers moderated by a medical professional as a way of extending patient support, improving medical adherence and so on. Whilst these groups are promising in terms of patient support, they can be burdensome for already overworked medical providers.<\/p>\n
For example, an ethnographic study of an IVF clinic in China, conducted by Ding Wang, found that two nurses see around 170 patients each morning, spending a maximum of five minutes with them. To help improve medical adherence and the patient experience, the clinic set up nurse-facilitated chat groups on WeChat. It is clear that these chat groups add-value for both providers and patients \u2013 providing both medical advice and peer support \u2013 however they create extra work for nurses, who receive 1000\u2019s of messages per day, making it hard to manage them effectively. Using chat to support patient care is a growing trend in Global Health (e.g. groups for youth living with HIV (Kenya), maternal and infant health (India), TB patients (Africa)). However, all parties face similar challenges.<\/p>\n
In Project EPOCh we are exploring the possibility of building and deploying NLP and ML solutions to help facilitators manage chat streams, and eventually support medical adherence.<\/p>\n
In collaboration with a University of Washington team led by Keshet Ronen, we are conducting a qualitative study on two WhatsApp group chats from the Vijana-SMART study. These groups provided support for youth living with HIV in Kenya. The groups contained 27 and 28 members and lasted for 6 months. All messages from youth and their facilitator were downloaded, deidentified, translated into English, and are included in the dataset. The aim of this analysis is to categorise the types of messages, the patterns of sending and responses, and to outline which NLP (and potentially other cognitive services) might be appropriate to apply to this data set.<\/p>\n
Team members:<\/p>\n