Pilot RCT and Interview Study on an HIV Chatbot in Nigeria

The primary endpoint is the change of HIV knowledge from baseline to the 50-day follow-up assessed using the Patient's HIV Knowledge Questionnaire (PHKQ). The questionnaire consists of 15 items and includes the components (subscales) Transmission and Misconceptions, Causes and Treatment Outcomes, and Diet and Immunity (Jackson, Okonta, & Ukwe, 2020). Participants are required to evaluate statements such as "HIV infection is caused by evil spirits" using a three-point scale: "Yes," "No," and "I don't know." The correct answer is awarded one point, while incorrect responses or the option "I don't know" are each awarded zero points.

In addition to the Patient's HIV knowledge questionnaire (PHKQ), we will use another knowledge test, which includes 10 additional questions, to more compressively assess the participants' knowledge. The test measures additional knowledge areas, which:

• are geared to the study participants' specific situation, i.e. being new to the ART regime (such as how much time it takes until the virus may become undetectable )
• are very relevant but not covered by the PHKQ, such as knowing about Post-Exposure-Prophylaxis (PEP) to prevent further infections.

To assess whether the availability of an empathic digital agent can also increase people's resilience, i.e. their "positive adaptation within the context of significant adversity", we use the People Living with HIV (PLHIV) Resilience Scale (Gottert et al., 2019). The scale is particularly suited as it is tailored to the situation of people living with HIV, measured and psychometrically validated in three African countries including Uganda, Senegal and Cameroon. The 10-item PLHIV Resilience Scale covers topics such as self-confidence, self-respect etc. and invites participants to indicate whether these have been positively, negatively, or not affected by their HIV status. Higher resilience was associated with less depression in each country (Gottert et al., 2019).

Participants' psychological well-being (SRQ-20) - measured at 50 day follow up We will measure participants' psychological well-being with the WHO 20-item Self-Reporting Questionnaire (SRQ-20) (Beusenberg, 1994). SRQ-20 was designed to screen for general psychological distress, especially in developing countries. It contains 20 questions which have to be answered by 'yes' or 'no', such as "are you easily frightened" (Beusenberg, 1994). Scores range from 0 to 20, with higher scores indicating higher levels of distress. SRQ-20 is a widely used and validated measurement to assess psychological distress, including in the context of Nigeria and HIV care. For example, people living with HIV Nigeria were reported to have a thirty-fold increase of psychological distress in the first year on ART (Pierce et al., 2023).

The different ways in which users perceive the chatbot will be measured by assessing users' trust and their perceived relationship with the chatbot.

Trust is a multidimensional construct and thus a validated measurement tool which consists of three components will be used: competence, benevolence and integrity (Hu & Lu, 2021). Competence consists of 4 items. Benevolence and integrity consist of 3 items each. For example, for benevolence, one question is "whether the digital assistant is interested in my well-being".

Parasocial relationships questionnaires were originally developed to measure relationships between celebrities and people. They have been applied recently to measure the perceived relational connectedness between chatbots and their users (Youn & Jin, 2021). In this study, the interaction subscale, which consists of 4 items, will be applied. It requires users to evaluate their perceived relationship with the chatbot.

Another relevant aspect is to assess usage patterns of the chatbot, especially the intensity of chatbot use. These four main indicators will be extracted from the technical chatbot platform.

• Number of total interactions per user
• Number of questions asked per user
• Number of medication reminders sent to the user
• Number of conversation topics completed per user

The following relationships will be assessed:

• Relationship between users' gender, age and education status on the one hand side, and chatbot use and chatbot perception at the other hand side
• Relationship between chatbot use and outcomes (knowledge, resilience, and psychological well-being)

Given the novelty of the intervention in the context at hand, a relevant aspect is to explore the experience of the user to be able to generate an in-depth understanding of the ways in which they perceive the chatbot. This will be achieved through interviews with 15-25 participants from the intervention group. A particular goal of the interviews is to understand whether, and if so, how, female chatbot users perceive the chatbot differently as compared to male users. Trained interviewers will carry out interviews with selected users of the chatbot, i.e. with participants from the intervention group