Main

Syndemic theory integrates two concepts: disease concentration and disease interaction1. The concept of disease concentration emphasizes how and where multiple epidemics cluster together as a result of large-scale political, economic, ecological and social forces (for example, systemic racism, gender inequities, structural violence, drought and heat intensification)2. The concept of disease interaction emphasizes the ways in which overlapping epidemics have mutually reinforcing effects on worsening health and disease via biological and social processes3. In this study, we evaluate which stressors interact, and how they interact, with convergent chronic conditions to influence quality of life in a population-based sample of adults living in a large, urban community in South Africa. We developed a locally defined measure of stress on the basis of two ethnographic studies investigating how people understand stress on their own terms amid living with chronic illness. We argue that disentangling ‘what’ drives disease concentrations from ‘how’ they interact is crucial for explaining how history and context shape the conditions of disease epidemics and determining when non-medical social interventions should be prioritized over (or augment) clinical interventions.

Anthropologists have long been concerned with the conventional practice of using standardized scales of stress and mental illness without considering local ways in which people experience stress and psychiatric disorders and communicate distress4,5,6,7. While the use of standardized instruments facilitates comparative studies of population mental health across contexts8, building locally relevant tools to evaluate social impact in large-scale studies has become increasingly relevant and critical for interrogating syndemics. Weaver and Kaiser argue that a “study designed to assess a presumptive syndemic” should “begin with freelists, ethnographic interviews, observation, and/or focus group discussions to identify common elements” that shape disease conditions across multiple valences of influence9. For example, Brewis and colleagues conducted a combined analysis of data from an epidemiological survey and qualitative interviews to study how the consequences of chronic social inequality (crime, hunger and discrimination) drive health disparities across three low-resource but heterogeneous communities in Haiti. They analysed epidemiological survey data to understand differences in exposures across communities and textual data from focus groups and one-on-one interviews to understand “the nuance, context, and local embeddedness of core themes as they emerged from respondents’ own words”10. This work emphasizes the need to focus on what they call “syndemic localization”, a process by which social, political or ecological factors—defined and measured within and in relation to a local context—drive disease interactions differently within and between geographic areas.

Mixed-methods scholarship like this is increasingly needed to counter the idea of the “global syndemic”11,12, a concept that threatens to erase local histories of inequity and oppression from contemporary accounts of disease morbidity and mortality. For instance, many researchers have demonstrated that the relationship between diabetes and depression is bidirectional13 and is intensified by economic hardship around the world14 in wealthy and poor countries alike15,16. In contrast, clinical work tends to gloss over how local identities and power relations contribute to how people experience the chronicity of illness as well as recommended clinical care17. One reason why this disconnect may occur is that risk is conceived in individual terms (for example, self-control) rather than social terms (for example, what conditions and intersectional identities shape experience), which embodies a broader framing of what drives diabetes in the first place18,19. Sangaramoorthy warns, in the context of HIV, that clinicians and counselors “are trained to be experts in the mediation of disease-specific risk, transforming individual client’s perceptions of external risk into internal risk and obscuring other non-HIV/AIDS threats to well-being” (p. 303)20. For these reasons, a rigorous examination of how and why social dimensions of stress fuel diabetes and its comorbid companions (such as depression, hypertension and HIV), particularly in settings of historically engrained racism and inequality such as South Africa, requires a mixed-methods (that is, combined anthropological and epidemiological) approach.

The research we present was originally based on two qualitative studies in Soweto, South Africa, that illustrated how people perceived social and personal stress to be more challenging than disease diagnoses21,22,23. The preliminary anthropological work illustrated how structural and social factors may impede people’s abilities to manage their own care for chronic illnesses, including diabetes, cancer, depression, physical pain and infectious diseases. For example, women described how reconstructing their families and raising grandchildren after losing children to AIDS not only posed substantial psychological burdens but also affected how they ate and how they accepted and managed their diabetes. Many related diabetes treatment to shared AIDS nosologies, referring to diabetes as “the same” or “worse”21. A further analysis of survey data from 1,000 middle-aged women in Soweto found a 40% prevalence of elevated psychological distress; women who reported two diseases had increased rates of psychological distress, and this upward trend continued with each additional physical disease reported24. Yet, in a study of breast cancer survivors in Soweto, we found that women relied on radical acceptance of their disease diagnoses and illness prognoses, as well as on family support and the public health system, to cope and foster their own well-being25.

Acknowledgment of the manifold ways that social and biological stress interact is particularly important in Soweto, as multiple comorbidity is an increasing public health concern. South Africa maintains (1) the highest number of people living with HIV globally, of which many also experience tuberculosis26 and, increasingly, diabetes27; (2) elevated rates of automobile accidents, intimate partner violence, rape and murder28,29; (3) elevated rates of infant and maternal mortality, despite a high level of wealth in the aggregate compared with other countries in the region30; and (4) a massive rise in non-communicable diseases, including diabetes31. Focusing on social and economic factors that affect diabetes alone and together with other medical conditions thus provides a more realistic understanding of people’s experiences with sickness and health.

A clinical study in Khayelitsha, a peri-urban settlement near Cape Town, South Africa, found that 45% of adults sought prescriptions for at least one of the following diseases: HIV, tuberculosis, diabetes and hypertension32. The increases in longevity among those living with HIV have led to increased risks of developing type 2 diabetes33. Additionally, one in four patients had multiple comorbidities, a phenomenon that generally increased with age, while those receiving antiretroviral therapy were more likely to develop diabetes at a younger age32. Cohort studies in Uganda and South Africa were some of the first to document the convergence of HIV with non-communicable diseases in sub-Saharan Africa34,35,36,37,38,39. These cohort studies suggested that having multiple conditions increases the likelihood of depression and that non-communicable diseases are less common among those without HIV than among people who are living with HIV40. Studies also point to the increasing salience of diabetes and tuberculosis41, which is of concern in South Africa given that the country has one of the largest concentrations of tuberculosis worldwide26. The demand for chronic care associated with any combination of diabetes, HIV and tuberculosis poses extraordinary public health and health care challenges.

This article investigates how our locally constructed measure of stress interacts with multiple medical conditions among people residing in six different neighbourhoods in Soweto, an urban settlement in Johannesburg, South Africa. We first used ethnographic methods to shape the study questions and design locally valid measures, which we then applied to a large population-based study of Soweto residents. Finally, we tested the theory derived from our quantitative analysis by conducting a follow-up qualitative study of illness experiences among people with multiple comorbidities. In what follows, we describe the co-occurrence of these medical conditions and consider how these conditions interact with our locally designed measure of stress and other measures of psychological distress and well-being. In doing so, we discuss what interactions among medical and social conditions tell us about people’s experiences in Soweto and how this informs the study of syndemics more broadly.

Results

Epidemiological findings

Among the study participants who completed surveys and had complete data available (N = 783), there were 541 women and 242 men (Table 1). The mean age was 46.1 years (standard deviation, 12.7). Quality of life was slightly higher among men than among women (60.1 versus 57.6; t = 1.67; P = 0.10). Most participants reported no chronic medical comorbidities (428 (55%)), while 236 (30%) reported one comorbidity, 89 (11%) reported two comorbidities and 30 (3.8%) reported three or more comorbidities. On the emic measure of stress, women reported considerably higher levels of stress than men did (48.9 versus 44.5; t = 4.50; P < 0.001), differing by more than 0.3 standard deviation units. On the emic measure of coping, no gender-based differences were observed.

Table 1 Characteristics of the sample (N = 783)
Full size table

Table 2 shows the results of the multivariable regression models. In the fully adjusted multivariable regression model, the multimorbidity sum score (β = −3.86; 95% confidence interval (CI), −5.39 to −2.33; P < 0.001) and stress (β = −0.58; 95% CI, −0.67 to −0.48; P < 0.001) both had statistically substantial negative associations with quality of life. The disaggregated model in Supplementary Table 1 suggests that the multimorbidity estimates were primarily driven by diabetes (β = −9.06; 95% CI, −14.1 to −4.05; P < 0.001) and cancer (β = −12.8; 95% CI, −23.9 to −1.76; P = 0.02). When the multimorbidity and stress product term was added to the model, the product term was statistically substantial, suggestive of an interaction in which the negative association between multimorbidity and quality of life was amplified in the presence of high stress (β = −0.16; 95% CI, −0.27 to −0.05; P = 0.005). Sensitivity analyses yielded estimates that were substantively similar to the primary analysis (Table 3): the binary measures of caseness had statistically substantial associations with quality of life, although the interaction between caseness on the 28-item General Health Questionnaire (GHQ-28) and multimorbidity was not statistically substantial; and the quintiles of the stress scale showed increasingly stronger associations, and stronger interactions with multimorbidity, with increasing levels of stress.

Table 2 Correlates of quality of life as assessed using the 26-item World Health Organization Quality of Life-BREF (N = 783)
Full size table
Table 3 Correlates of quality of life as assessed using the 26-item World Health Organization Quality of Life-BREF, with alternative specifications for the measurement of stress (N = 783)
Full size table

Qualitative findings

Table 4 describes key themes and sub-themes that emerged from the interviews, along with exemplar quotations, for each sub-group. Study participants with diabetes, hypertension and high levels of stress (Group 2) often described a constant fear of having a debilitating medical complication (for example, amputation). They also described financial burdens associated with paying for medications and food, and social burdens such as those due to family conflict. Study participants with diabetes and HIV or tuberculosis (Group 3) reported similar concerns over access to care, the importance of self-care and financial stressors.

Table 4 Primary themes from the qualitative interviews (N = 88)
Full size table

In contrast, study participants with diabetes and hypertension but low levels of stress (Group 1) commonly described more social support, less trouble accessing or managing medication and care-seeking, acceptance of their illness, and a more positive outlook on their illness and future. This perspective was more aligned with that most commonly described among study participants who reported no medical comorbidities, who rarely sought care or focused on their health (Group 4).

Nearly everyone reported feelings of stress about financial difficulties. Most described finding comfort in being able to access health care through the public system (even when voicing concerns about stockouts or long waits). Although few relied on traditional herbal remedies to care for physical illness, most people described how they coped with psychological distress through individual religious practices (for example, prayer and reading the Bible) and group or social religious practices (for example, small-group Bible study, attending services and church-based counselling).

Discussion

Developing methods to evaluate syndemic theory poses challenges and opportunities as more scholars adopt a syndemic orientation for understanding and developing interventions for communities facing multiple clustered social and health conditions. Syndemic theory is predicated on the idea that social and structural factors precipitate disease concentration and disease interaction, and that local phenomena may differentially affect disease interactions and disease experiences across contexts. In previous work anthropological work, we have argued that structural violence, social trauma and chronic distress all have important roles to play in shaping syndemic experiences. In this article, we evaluated, through a combined ethnographic and epidemiological lens, how such experiences cluster with multiple convergent conditions and therefore become syndemic.

First, we argue that our theoretical postulates hold up for stress and multimorbidity. Our strongest finding in this study reveals a robust interaction between a locally designed stress scale and multimorbidity. This finding was consistent with our ethnographic findings, which showed that stress was associated with medical complications, financial difficulties, family discord and an unsettled future, while people doing well were more likely to describe social and emotional well-being—even when financial difficulties were common. Taken together, these mixed-methods findings support the important interplay between stress and living with multiple chronic illnesses. The high burden of physical and mental illness in this population substantiates this point.

Second, the study reveals the importance of grounding epidemiological work in detailed ethnographic study42. Constructing a locally relevant scale revealed the complex roles of various stressors (such as financial stress, which is embedded in the local political economy), as defined by participants, in conditioning the associations between multimorbidity and quality of life. Similarly, the coping scale emphasizes the fundamental importance of religious practices, social cohesion and caring for others in this community—thereby underscoring how ubuntu, or thinking about the self in relation to others, may play a role in reducing stress and fostering quality of life43. Using a generic life events scale, however useful, could have missed what people in this context themselves define as most critical for determining quality of life. The priority that our interlocutors put on these life stressors would probably have been less fully understood in a ‘rapid’ or strictly quantitative study.

Third, the qualitative data enriched our understanding of the epidemiological data by explaining what types of social stresses emerged within each group and how those social stressors clustered together and in relation to multiple morbidities. The qualitative data show how interlocking stresses produced undue burden on our study participants and affected their quality of life in more severe or enduring ways, or, in some cases, in ways that were mutually reinforcing with their co-occurring health conditions. People faced different challenges depending on their previous diagnoses, their outlook on those illnesses, the level of social support available to them and their financial security. In other words, the effects of multimorbidity on quality of life differed for people who had the same co-occurring diagnoses in part because of non-medical social and structural factors such as family stress and fear. We emphasize that, while the negative association between multimorbidity and quality of life is amplified by high levels of stress, it is not wholly explained by and cannot be reduced to that variable. People with diabetes and hypertension may perceive their illnesses differently if they report more or less psychological morbidity. Recognizing how people live well with multiple illnesses therefore requires critical attention to the non-medical factors that shape living with chronic illnesses, especially when they overlap and cause multiple burdens of medication, care-seeking and living well. Individual and group religious practices (such as prayer, small-group gathering and attending services) featured in many people’s narratives of what non-medical factors are crucial to good health43,44. Moreover, many people without previous medical diagnoses tended to avoid clinics and hospitals, even for preventive care, which substantiates the point that people with multiple conditions are often diagnosed only when severe symptoms force them to seek urgent or acute care45. These qualitative data thus demonstrate how social and medical conditions are not isolated experiences but instead are interactive and contingent with social experiences.

Interpretation of our findings is subject to several important limitations. First, we had planned on surveying a much larger sample of study participants, but data collection was stopped prematurely due to the first surge of the COVID-19 pandemic. Second, very few people in our sample reported both diabetes and an infectious disease (either HIV or tuberculosis). This finding may have resulted from our study design: 121 people refused to test for HIV, which is not uncommon in this context46,47. Third, and related to the previous limitation, the data on medical comorbidities (along with the data on stress and coping) were self-reported. While there is no practical way of understanding stress and coping without using self-reported measures, it is likely that some of the medical comorbidities, particularly HIV and tuberculosis, were subject to underreporting given the stigma that has been attached to HIV and tuberculosis in this context48. Such underreporting could have biased our estimates of the association between quality of life and HIV. More generally, however, if people with higher quality of life were more likely to underreport medical comorbidities, this would have biased our estimates of the association between medical comorbidity and quality of life toward the null rather than away from the null. Fourth, the cross-sectional design prevented us from assessing both disease and coping trajectories, which could have provided a more nuanced understanding of living with multimorbidity. Indeed, such an approach could change how syndemics are framed: rather than focusing on individuals as subjects of syndemics, it would recentre their agency as individuals who respond to, cope with and make sense of their illness, despite structural violence and social challenges.

This study illustrates the importance of grounding an epidemiological analysis of a syndemic in long-term ethnographic work. We argue that there is a need for more mixed-methods studies that draw from knowledge situated within contexts and developed with multidisciplinary teams, so that the field can better understand how and why syndemics emerge, given local structural and social conditions. Our data emphasize the role of non-medical factors in explaining how people live well with or suffer from multiple chronic conditions. Although many people described some satisfaction with their care in the public system (despite common critiques of wait times for clinicians and drug stock-outs), it was very clear that not all health and healing could come from the public health care system. Moving some of this care from the clinic to the church or community, at scale, may be an effective way to promote social well-being, good mental health and more effective management of physical conditions such as diabetes and hypertension in Soweto and other similar contexts in urban South African neighbourhoods.

How scholars measure syndemics will probably continue to change. Syndemics inherently differ from place to place. The roles of historical, ecological, political-economic and socio-cultural factors in shaping or perpetuating syndemics should be central to any investigation into what constitutes a syndemic. Untangling what factors are most relevant to disease concentration and disease interaction matters a great deal for a more precise and contextually relevant understanding of overlapping disease epidemics and future social interventions for public health, and can provide important contributions to future scholarship on syndemics.

Methods

Setting

We conducted this study in collaboration with the Developmental Pathways for Health Research Unit (DPHRU), a research unit associated with the South African Medical Research Council and the University of the . and based at Chris Hani Baragwanath Hospital in Soweto, South Africa. Research assistants were based at the DPHRU research station and fluent in multiple languages spoken in Soweto. The surveys for the Phase 1 epidemiological study were administered in people’s homes. The interviews for the Phase 2 qualitative study were conducted at the research station. All research participants were residents of Soweto.

Soweto is an urban settlement in Johannesburg, the largest city in South Africa. More than one million people reside in Soweto; most are Black South Africans, representing various ethnic identities, including Zulu, Sotho, Tswana, Tsonga and others. We use the term ‘Black’ to describe the study participants while acknowledging a problematic history of this identity as a political category instituted by apartheid to distinguish ‘Black’ from ‘Coloured’ and ‘White’49. Soweto is economically diverse, with middle-class neighbourhoods, working-class communities and informal settlements. The marginalization of Black South Africans and other non-white communities during apartheid and the decades afterwards have contributed to poor housing, lack of sanitation, unhealthy food access and deficient educational opportunities in the present day. These problems have been associated with the unequal burden of HIV and tuberculosis among Black compared with white South Africans, compounded by costly health care services in the private sector and systemic barriers in the public sector49.

Sampling

The Phase 1 epidemiological study was embedded within the infrastructure of a larger study being conducted through the DPHRU. No statistical method was used to predetermine sample size. Given the size of Soweto (200 km2), we sampled study participants in clusters based on churches, which are widely distributed throughout Soweto. Starting with a list of geolocations of each church structure, fieldworkers visited each church and verified its existence. The churches were used to identify 30 community clusters, each with a one-kilometre radius. For the purposes of our study, six clusters were randomly selected and then enumerated. Within each cluster, the research team walked down the streets, engaged potential participants and interviewed available people in their homes who were willing to participate in the study. If the person approached did not fit the inclusion criteria (described in more detail below), another member of the household who did meet these criteria was then approached. The Phase 1 epidemiological study participants were interviewed in their homes and were not provided with any compensation or study incentive. The University of the Witwatersrand Human Research Ethics Committee approved this study (M180544). All participants provided written informed consent before participating and were free to stop the survey at any time.

Phase 1: epidemiological survey data collection and analysis

For the Phase 1 study, we visited the six neighbourhood clusters over a period of one year (April 2019–March 2020). We finished 783 complete surveys before the study was shut down due to the COVID-19 pandemic. The response rate was 86%. Measurements were taken from this sample at a single time point. No data were excluded from the analyses. We enrolled participants 25 years of age or older who lived within each identified cluster and who considered themselves to be regular members of the household (that is, who had spent most nights in the home during the three months preceding the interview). Participation was limited to people 25 years of age and older because of our focus on chronic multimorbidity and because we wished to avoid interfering with recruitment for a concurrent study that was enrolling young adults. The exclusion criteria included people younger than 25 years of age; people who did not consider themselves to be residents of Soweto; and individuals who could not meaningfully communicate with the study team, such as people with cognitive impairments, people who were acutely intoxicated upon approach, people who were too ill or people who threatened our team with harassment or violence.

Our field teams collected survey data using tablets programmed with Research Electronic Data Capture. The primary outcome was quality of life, which we measured using the 26-item World Health Organization Quality of Life-BREF50. The primary explanatory variables of interest were multimorbidity (namely, the sum score of the most commonly reported medical comorbidities, including type 2 diabetes, hypertension, chronic pain, high cholesterol and cancer) and stress (measured using the 21-item Soweto Stress Scale, a locally developed and validated emic scale based on our ethnographic work conducted in Soweto over the past decade51).

Model 1 specifies a multivariable linear regression model to estimate stress and medical comorbidities as correlates of quality of life. We then added a vector of additional covariates (Model 2): age; sex; household asset wealth, measured using a 13-item checklist of assets in the household; perceived lack of neighbourhood safety, measured using two questions about feeling safe during the day and night; perceived neighbourhood social cohesion52; HIV status, measured by an at-home rapid test kit; coping, measured using the 14-item Soweto Coping Scale, an emic scale designed to measure different aspects of problem/emotion-focused and religious coping, also based on our ethnographic work conducted in Soweto; and neighbourhood cluster. In the final regression model, we added a product term to assess for an interaction between multimorbidity and stress (Model 3).

We used multiple specifications to probe for this hypothesized interactive relationship. First, we treated the stress scale as binary, with caseness denoted as a stress scale value greater than or equal to the 75th percentile. Second, because an arbitrary 75th-percentile threshold for the locally derived stress scale has no empirical precedent, we substituted for the Soweto Stress Scale the GHQ-28 (refs. 53,54) in the regression model. The GHQ-28 is a non-specific measure of psychological distress but has been used in global health studies for decades with well-established thresholds for caseness. Third, because a dichotomous variable may mask variability in quality of life at more granular levels of stress, we examined the interaction between multimorbidity and the stress scale split into quintiles, where each group represented 20% of the sample, ranging from the least stressed (first quintile) to the most stressed (fifth quintile). Fourth, we eliminated possible high leverage points to assess whether the estimated associations were dependent on extreme values. Last, to compare the estimates associated with the multimorbidity sum score variable versus the individual conditions that comprise it, we disaggregated the sum score and analysed the individual conditions separately55.

The statistical analyses were conducted using R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria). Two-sided tests were used throughout.

Phase 2: qualitative data collection and analysis

We then conducted in-depth, semistructured qualitative interviews with 88 participants from the epidemiological survey. The aim of these qualitative interviews was to explore major life events, disease-related stress, challenges associated with living with one of the comorbidities of focus, major barriers to or facilitators of health, challenges associated with care-seeking and comorbidity, systemic barriers to or facilitators of health care, and self-care regimens. These individuals were purposively sampled on the basis of their membership in one of several comorbidity clusters: Group 1 (diabetes, hypertension and low stress; N = 15), Group 2 (diabetes, hypertension and high stress; N = 19), Group 3 (diabetes and either HIV or tuberculosis; N = 7) and Group 4 (people living healthy lives without any medical diagnoses; N = 47). Phase 2 qualitative interviews were conducted at the DPHRU research station, and each participant was reimbursed 150 South African Rand (approximately US$12 at the time the study was conducted) for transportation to the research station. A handful of in-home qualitative interviews were conducted for participants who could not travel.

We transcribed all interviews verbatim. Audio from vernacular languages was transcribed and translated into English, while maintaining consistency with their original meaning. We used an inductive method that involved reading and rereading the transcripts and field notes while comparing the two to ensure that no data were misinterpreted. The study team designed a codebook on the basis of this inductive analysis, which included 30 main codes. These codes were well defined and collectively agreed on, and were reflected in the interview guide, field notes, selected transcripts and in-depth discussions. Each code was identified, defined, applied, revised and discussed among five core members of the research team. We attached the codes to each transcript using Dedoose software (version 7.0.23, SocioCultural Research Consultants, Los Angeles, Calif.), with a primary coder and two secondary coders reviewing and applying codes to each transcript. Further information about the methods and findings of the qualitative study are described elsewhere44.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.