Exploring the mechanisms behind HIV drug resistance in sub-Saharan Africa: conceptual mapping of a complex adaptive system based on multi-disciplinary expert insights

HIV drug resistance (HIVDR) remains a threat to the effectiveness of antiretroviral therapy (ART). Over the last decade, major efforts have been made to achieve global 90-90-90 goals by 2020 and to end the HIV epidemic as a public health threat by 2030 [1]. However, levels of HIVDR are rising, compromising the effectiveness of ART and potentially also the efforts to attain the last 90 goal [2]. In 2017, mathematical modeling predicted that if left unchecked, excess levels of pretreatment HIVDR to the NNRTI drug class could directly lead to 890 000 AIDS deaths, 450 000 new infections, and 6.5 billion USD extra ART costs by 2030 in sub-Saharan Africa (SSA) [3]. Recently several cases of multi-drug class resistant HIV have been reported [4, 5].

Several causes of both pre-treatment HIVDR and acquired HIVDR have been described in the literature. Due to the high genetic variability of the virus, selective pressure stemming from a combination of incomplete adherence (defined here in the broader sense of not taking ART as prescribed, which can be influenced by a multitude of factors which are both within and out of the control of the clients themselves) and a low genetic barrier of ART may lead to the emergence of HIVDR [6]. In addition to biological and pharmacokinetic factors influencing the selection and emergence of HIVDR lie other, indirectly related factors. In a meta-analysis Shubber et al. identified diverse barriers to adherence such as forgetfulness, traveling, medication toxicity, stigmatization, food insecurity, alcohol or substance misuse [7]. Other crucial aspects to prevent HIVDR are for example sufficient ART availability and a well-functioning ART supply system [8]. These and other factors described in literature relate to several fields of science and in some cases also to other complex problems. For example, ART drugs have been reported to be used in a mixture of recreational drugs called whoonga in South Africa [9‐11]. The complex problem of drug abuse, is therefore linked to HIVDR as this exposure to ART may have consequences for pre-treatment drug resistance.

Despite the fact that most of the factors contributing to HIVDR are presumed to be known, and that models to mitigate these causes have been built, pre-treatment HIVDR, especially in SSA, is still increasing [12].

As the factors influencing the emergence of HIVDR are numerous, have roots in different fields of science and are interconnected with other complex problems, we argue that HIVDR should be approached as a complex adaptive system (CAS), combining knowledge of diverse experts and stakeholders. Such systems have been defined by Plsek et al. as ‘a collection of individual agents with the freedom to act in ways that are not always totally predictable, and whose actions are interconnected so that one agent’s actions changes the context for other agents’ [13]. A successful intervention on one element of the system does not guarantee resolving the core problem. Rather, interventions should be planned keeping in mind the entire system, its particular dynamics and possible feedback loops and with the aim of reshaping the system in a favorable way [13, 14]. Feedback loops can be reinforcing or balancing, meaning that a change in a certain direction will either evolve into more change or balance itself out by propagating an opposite effect. CAS have been studied in several other contexts such as ecosystem management, healthcare management and obesity [15‐17]. Moreover, the importance of using systems thinking in health care has been widely described in the literature [13, 14, 18‐20]. In 2017, Rutter et al. described the need of approaching public health problems as complex systems in order to identify, implement and evaluate effective interventions [14]. Such interventions should be done at leverage points in the systems. These are points where a small intervention can have a large impact on the system [21]. Identifying leverage points is difficult and sometimes counterintuitive. Gaining insights in subsystems or feedback loops may therefore facilitate the identification of leverage points [22].

With this study, we aimed to make a first attempt at understanding the complexity behind HIVDR by combining the expertise and viewpoints from different HIVDR experts. In this article we describe how we identified three interconnected feedback loops influencing HIVDR by developing a systems map that represents the CAS of HIVDR in SSA based on the insights of international HIVDR experts from different disciplines. We discuss the insights gained from these feedback loops and possible applications for quantitative modelling, complexity-informed intervention design and policy development [23, 24].

The systems map was designed based on semi-structured interviews with international experts from or with experience in SSA. For the purpose of this study, international experts were defined as stakeholders from diverse disciplines and institutions, working at an international level on HIVDR related to SSA and with a minimum of five years of experience. The participants were selected based on their expertise concerning HIVDR and with the aim of creating a mix of backgrounds and institutions covering all aspects of HIVDR. Purposive sampling was done starting from the expertise and connections of the Rega Institute and the Institute for the Future in Leuven, Belgium. This was supplemented with snowball sampling, using the expertise and connections of participants, and theoretical sampling, looking for the missing perspectives based on the emergent findings. They were contacted through email or in person when an opportunity presented itself, for example at international conferences. The interviews were held face to face (n = 6) or online over Skype or Zoom (n = 9) and were conducted in English. Semi-structured interviews of approximately 60 min were conducted until data saturation was reached, aiming to cover all possible factors influencing HIVDR in SSA. For the purpose of this study we describe data saturation as the point at which no new elements were uncovered in new interviews and no new connections which significantly changed the final conceptual model, were uncovered.

An interview guide was designed with the input of several HIVDR and social science experts and was adapted according to insights developed through analysis (Additional file 2). The guide contained three sections: the first section entailed sociodemographic questions concerning the interviewees gender, age and educational background. The questions of section two related to the interviewees professional and personal experience with HIV or HIVDR in SSA. The third and main section covered their perspectives on the factors influencing HIVDR. All experts were asked what, in their experience, were the main causes of HIVDR. As a general guideline, the interviewer aimed to cover the following four areas: causes related to 1) availability of ART at the healthcare centre, 2) PLHIV’s ability to fetch ART, 3) PLHIV taking ART as prescribed and 4) ART suppressing the viral load. Additionally, when causes outside these four areas came up, they were also further discussed. Subsequently, depending on the expertise of the participant, follow-up questions such as “What do you think is causing the situation you just mentioned?” aimed to clarify the deeper reasons behind some of those initially indicated causes.

Analysis of the semi-structured interviews was inspired by the QUAGOL method and done simultaneously with the data collection [25]. After each interview a technical report was written describing relevant characteristics of the participant and interview context, helpful for understanding the data in their specific context. The interviews were transcribed verbatim by an external firm and the quality of each transcription was verified by listening to the audio tapes and correcting possible errors in the transcripts. Each transcript was (re-)read until a list of factors influencing HIVDR as well as connections between those factors, mentioned either explicitly or implicitly by the interviewee, was extracted. Connections were assigned a positive, negative or dual polarity. A positive connection indicates that the influencing and influenced element evolve both in the same direction (e.g. A—> B: when factor A increases, B increases too). A negative connection indicates that both elements evolve in the opposite direction (e.g. A- > B: when factor A increases, B decreases and vice versa). A dual connection indicates that both effects are possible. Take the following paraphrased quote as an example: “You know, sometimes people form peer support groups so that each month someone will pick up the medication for the whole group. This way people have to go only once every six months instead of on a monthly basis.” This would be translated into a negative arrow from “peer support group” to “required frequency of hospital visits”. Subsequently for each of the first six interviews separately, these factors were visualized in a small systems map while re-reading the interview again in order to visualize all the mentioned connections between these factors. Afterwards the separate systems maps of the first six interviews were merged together into one and from that point onwards data from the following interviews was added to the map. Throughout the analysis newly discovered insights were constantly compared with previous findings resulting in an iterative process of re-reading interviews and reviewing the detailed systems map.

The model was designed in Kumu, an online mapping tool which enables the user to save data such as interview quotes and memos for each element and connection [26]. In the first, confidential, version of the systems map, all interview quotes which mention a certain element or connection, are collected in the comment fields associated with the element or connection in the KUMU tool, facilitating our analysis. From this first draft systems map causal loops were identified manually as series of elements connected to each other in a circular way. Causal loops which contributed to the same mechanism were identified as a subsystem (this can be compared with a road map: all possible routs you could take to go from Brussels to Amsterdam would be classified together as the subsystem “routs from Brussels to Amsterdam”). Because the subsystems consisted of many elements and connections, they were conceptualized into one overall mechanism per subsystem which reflected the overall messages of interviews as well as possible. While each separate element and connection was mentioned in one or several interviews, the resulting feedback loops are based on the combination of knowledge from the different experts. The conceptualization of the subsystems was linked back to the original interviews, discussed with several stakeholders and strengthened with literature evidence.

In total 15 international experts were interviewed. Table 1 summarizes the scientific and institutional background of the interviewees. A diverse sample of experts with different expertise and institutional affiliation was reached, permitting us to gain insights in the various aspects of the CAS. Out of the 15 participants, 13 were researchers or had previous research experience in the field of HIVDR.

Data saturation for elements (factors influencing HIVDR) was reached after about nine interviews and for connections (pathways of influence between two elements) after 12 interviews (Fig. 1).

All elements and connections identified from the semi-structured interviews are represented in Table 2, Table 3 and Additional file 3. Based on this data, we visualized the system in two ways [27]. The first visualization divides the elements in five layers according to their relation to biology (elements and processes happening inside the body), individual factors (psychology, personal factors and behavior of adherence), social context (personal characteristics as a member of the community and baseline conditions in the community), healthcare system (treatment plan and healthcare organization), and ‘overarching’ factors (such as international policy, research and funding) (Additional file 1).

For the second visualization we grouped the same elements and connections in ten different thematic clusters (Fig. 2) ([27], page 1). The clusters represent elements belonging to the same themes identified in the interview data, being adherence and retention in care, biology, clinical manifestations, complex problems, genetic barrier of the medication, global effort to tackle HIVDR, health literacy and empowerment, health system resources, psychosocial factors and quality of care. The elements included in each cluster are presented in Table 3. When visualizing these clusters and the connections between them, three major feedback loops or sub-systems emerge, indicated by the three circles in Fig. 2.

The three subsystems described above exist on different societal levels (personal, programme and population level) and are intrinsically linked with each other. The alternating adherence subsystem takes place on the personal level until HIVDR emerges, at which point the individual will add to the burden of the healthcare system. The increased burden on the healthcare system may then impact the overall quality of care, which in turn may impact the adherence of PLHIV through a delayed switch in ART after detection of viral non-suppression, thus increasing the chances of personal- and population-level HIVDR emergence. Diminished quality of ART service delivery may also impact adherence counselling and support, thereby directly impacting the alternating adherence subsystem at the personal level. Both pathways will eventually lead to an increase in HIVDR, which is reacted upon at the population level by researching and developing new drugs that are more forgiving with respect to adherence (e.g. long-acting drugs) and that have higher genetic barriers to resistance. Policy makers overly relying on these new ART regimens may shift focus away from high quality service delivery and HIVDR prevention measures. As described above, decreased quality of care may then impact the healthcare system burden at the population level and/or alter personal-level adherence.

The HIVDR system is influenced by several other complex problems at different points in the three subsystems. Food insecurity for example, may negatively affect adherence considering PLHIV have to take the ART with a meal each day. Other examples are political instability and disease outbreaks (such as the COVID-19 pandemic), which may destabilize the healthcare system, increase the burden on healthcare personnel and may cause PLHIV to have priorities other than adherence to ART.

Our study illustrates the added value of qualitative methodology to visualize the complexity and dynamics of a system. This may help decision makers to gain insight into the systems complexity and to identify leverage points in order to design targeted and complexity-informed interventions. This methodology can be transferred to study HIVDR in specific settings or could be used to gain insights into other complex problems. Moreover, the content of the model presented in this study may (partially) be extrapolated to other chronic diseases such as diabetes or obesity in order to understand their drivers and feedback loops.

The conceptual model presented here also lays the basis for quantitative mathematical modelling of the factors influencing HIVDR. This will allow quantitative modelers to collect data on relevant parameters in the system to monitor any changes, desired or not, in the entire system. An important advantage of basing a quantitative model on this conceptual map lies in the multidisciplinary manner this map was developed, therefore identifying mechanisms which might not have been identified using a monodisciplinary approach.