**3. Discussion**

In this study, we gained insight into the complexity of HIVDR in the Ukonga and Gongolamboto areas of Dar es Salaam by developing a model representing the CAS of its interconnected factors, together with local actors and PLHIV. It is important to note that our aim was to understand the CAS of factors influencing HIVDR through the mental models of the people most affected by it. Therefore, the model does not represent one fixed reality but rather an interconnected network of elements influencing HIVDR, which are constantly evolving over time, and which are highly dependent on context.

Three leverage points were identified based on the insights provided by our systems map. The first, shallow, leverage point aims at reinforcing the motivation to adhere to therapy, for instance through the encouragemen<sup>t</sup> of positive health outcomes. The second also shallow one, aims at decreasing stigmatization by strengthening community education. The third identified leverage point is at a deeper level and requires the restructuring of certain aspects of care through combining microfinance and peer support groups for PLHIV. Our work provides valuable insights at the systems level which, after strengthening of the healthcare system viewpoint, can be used to design and test interventions at these leverage points.

In addition to the identified leverage points, we obtained some other system-level insights. First, our data clearly showed the impact of psychological wellbeing on the

dynamics of the HIVDR system as also described extensively by Zlati´c et al. [21]. In particular, stigmatization was found to be the driver of several important feedback loops. Second, at the healthcare system level, we found that some counsellors give very strict guidelines to their clients which are ill-adapted to their life circumstances. These are failing to convey their purpose, and therefore sometimes work counterproductively. Clients may refrain from taking their medication if they do not find the advised type of food or if they come home one hour late. Future seminars on HIVDR for healthcare workers may need to be revised to refocus on the objective of the counselling sessions (preventing HIVDR and ensuring good health of PLHIV) rather than on the individual rules they have to follow. Third, at the community level, we found a delayed reinforcing feedback loop, indicating that PLHIV openly disclosing and discussing their HIV status are conducting a type of community education. This can reduce community stigmatization over time, encouraging more PLHIV to disclose their status. Previous studies have shown the correlation between knowledge and HIV related stigma [22,23]. One study in South Africa found that a decrease in stigma was associated with an increase in knowledge over a period of four years [22]. To identify the tipping point at which this reinforcing loop is kicked into action additional research is needed.

To explore the contents of our systems map beyond the local level, we compared it with a systems map of factors influencing HIVDR for Sub-Saharan Africa, which was informed by experts and developed using the same methodology [8]. Overall, the content of the systems maps remains largely similar. As can be expected, however, the expert systems map contained more extensive information at the healthcare system level and the local map goes into more detail at the personal level. A notable difference is that, whereas in the expert map the economic factor food insecurity was considered to be important but external to the system, it became clear that at the local level those factors were at the very core of the system, forming daily barriers to adherence for PLHIV. This shows that in order to fully understand the CAS of HIVDR, the viewpoints of PLHIV, actors and experts, as well as those groups at the local and broader geographical level need to be integrated.

A shortcoming of this study is its timeframe as two important events happened: (1) at the time of data collection the healthcare centres in the study site had just switched their ART regimens from TLE to TLD, a therapy which evokes less side-effects and which has a lower chance of provoking mutations in the virus and (2) between the data collection and validation the world was hit by the COVID-19 pandemic which, for a period of time brought a number of changes to the system. From March until July 2020, all PLHIV in the study area were given ART for six months instead of the usual one or three months, wearing face masks was obligatory in the healthcare centre, which caused problems for clients who could not afford them and transportation fees increased due to strict rules for seat capacity of commuter buses. While further research is needed to clarify the impact of these interruptions on the HIVDR prevalence in the population, our systems map can help to understand how these measures may have impacted the adherence level of PLHIV. Moreover, the systems mapping method described can be used to study the impact of the COVID-19 pandemic on other aspects on the healthcare system, to study other public health problems, or to be transferred to study HIVDR in other study sites.

#### **4. Materials and Methods**

## *4.1. Study Design*

An iterative systems mapping design was used to visualize and analyse the CAS of factors associated with HIVDR in our case study site in Dar es Salaam, Tanzania. Qualitative methods were used for data collection and analysis. The systems analysis and identification of leverage points were based on a systems thinking inspired analysis guide [24].

#### *4.2. Study Site and Participants*

The study was conducted at the DUCS site in the Ukonga and Gongolamboto administrative wards, Illala district, Dar es Salaam region, Tanzania. The DUCS follows more

than 100,000 residents from more than 20,000 households and collects sociodemographic and other data on a six-monthly basis [17]. This study site was chosen because of the rich data available which may support future intervention designs. We included three types of stakeholders in this study, each representing a different perspective: local experts, local actors, and PLHIV. Local experts were people with professional expertise on HIVDR, based in Tanzania. For the purpose of this study, local actors are defined as people who have good insights in the daily lives of the local citizens and who, through their job, status or daily activities are able to make a positive impact in their society. The local actors were selected with the aim of including a range of people who could provide us with insights about HIV in the community from diverse angles in order to create an overview that is as comprehensive as possible. PLHIV in several stages of their treatment, on different therapy regimens and with varying treatment-adherence levels were selected purposefully and recruited by research assistants of the DUCS.

#### *4.3. Data Collection Procedures*

The systems map was developed in three phases (Table 2). During the preparation phase we organized a workshop with local experts to discuss factors influencing HIVDR in our study site. During this meeting we started from a Sub-Saharan systems map based on knowledge from international experts, developed in previous research and adapted this map to the local situation. This adapted map served as a basis to design the semistructured interview guides and was not used further in data analysis. This way, the CAS of HIVDR in our study site was constructed anew from the interview data, truly allowing the perspectives and mental models of the local inhabitants to form the map, without the influence of previous research.

**Table 2.** Overview of the different activities and participants in the project.


## *4.4. Semi-Structured Interviews*

The first draft of the systems map was designed based on semi-structured interviews with PLHIV and local actors at DUCS in the Ukonga and Gongolamboto areas in Dar es Salaam, Tanzania. Semi-structured interviews do not consist of a set of rigorous questions but rather use a set of common themes to be explored with all the participants. This type of interview allows new themes to come up and be explored, based on the interviewee's answers.

The participants were called on their cell-phone and invited for a face-to-face interview at the DUCS office in the local community centre located in the Ukonga area. This location is neutral and not linked to any activities involving PLHIV and was therefore chosen to avoid stigmatization of the participants. The interviews were held in Kiswahili by I.M., a local social scientist and participants were reimbursed for their transportation costs. Each interview session lasted for about forty-five minutes. The interviews were audio recorded after seeking consent from study participants, transcribed verbatim and translated into English.

The semi-structured interview guide was informed by the expert meeting and designed by A.K., A.V. and I.M. with the aim of capturing the deeper factors influencing

HIVDR in the DUCS area. After each interview day I.M., A.K. and A.V. met to debrief the interviews and the interview guide was adapted according to the insights gained. After a first analysis of the interviews, a selection bias was noted as only participants enrolled in care were interviewed. In order to have a more diverse perspective on the factors influencing HIVDR in the study area, two additional participants who had not been attending healthcare services regularly in the past months were recruited and interviewed during a phone conversation. Interviews were conducted until data saturation was reached. For the purpose of this study, data saturation was defined as the moment in which no new elements or connections are discovered in two consecutive interviews (Table S1).

## *4.5. Data Analysis*

The analysis of the semi-structured interviews was conducted by two researchers (L.Z.) and (A.K.) with a combined background in psychology, biomedical science and systems thinking. The method used was inspired by the QUAGOL method [25]. After each interview, a technical report was written, containing all the specifics needed for a full comprehension of the data in their specific context. In order to ascertain a correct interpretation and cultural understanding of the transcripts, they were each individually discussed in a series of meetings between A.K., L.Z. and I.M.

For each transcript, a respective systems map was made, visualizing the factors influencing HIVDR mentioned in the interview and the connections between those factors. Seven interviews with PLHIV and five local actor interviews were schematized and coded by A.K. The other five interviews with PLHIV and five local actor interviews were schematized by both A.K. and L.Z. and the interviews were coded by L.Z. Possible differences were discussed until a consensus was found. In a next phase the separate schemes were merged together into one comprehensive systems map containing all the codes extracted from the interviews. The systems map was designed with the online mapping tool KUMU, which facilitates the visualization and analysis of the map, as different types of data can be stored behind the elements and connections [26]. Though here described linearly, the coding and mapping was an iterative process in which the interviews were re-read at several points in time, codes were revised throughout discussions between the researchers and findings were constantly compared with insights from previously analysed interviews.
