This study is the first to report longitudinal prevalence of depression in older PLWH in SSA alongside biological, psychosocial, and functional outcomes. We report a high depression prevalence (20.9% DSM-IV, 16.6% GDS-15) but were unable to identify other SSA studies using comparable diagnostic criteria.

One South African study (n = 422) reported prevalence of 14.8% by ICD-10 criteria in a demographically similar rural cohort (median age 60 vs 57, % primary educated 52.4 vs 64.0) but with substantially lower cART treatment rates (95.5% vs 49.3%) (Nyirenda et al. 2013). ICD-10 includes more somatic symptoms than DSM-IV, potentially resulting in over-reporting in chronic disease (Almeida and Almeida 1999).

Other high-income (USA) (Balderson et al. 2013; Grov et al. 2010) and middle-income settings (Brazil) (Filho et al. 2012) report prevalence of 39.1% (CES-D), 14.0% (DSM-IV), and 34.6% (GDS), respectively. These are socio-demographically different to this Tanzanian cohort (USA median age 54–55.8, %female 28–28.9, completed high school 78.7–82%) (Balderson et al. 2013; Grov et al. 2010) (Brazil mean age 57.6, %female 44.2, %literate 76.9, cART-treated 94.2%) (Filho et al. 2012).

These studies, like ours, report a high depression prevalence (highest in higher-income settings) despite differing in population demographics and proportion receiving cART. This is surprising as we might expect a higher depression prevalence among cART-untreated individuals given the biological inflammatory pathway hypothesis (Bhatia and Munjal 2014).

Baseline depression did not correlate with the HIV disease severity variables we investigated in contrast to cross-sectional studies which frequently report worse HIV disease outcomes and cART adherence. An East African study (Tanzania, Kenya, and Uganda, n = 2307, 58.3%female, median age 34–48) reported that depression was associated with 50% lower cART adherence and increased HIV viral load (Meffert et al. 2019). This older cohort may be unusual given that almost all received cART (vs 68% in the East African study), and self-reported adherence was high (Meffert et al. 2019). We may be observing a ‘healthy survivor’ effect which may be present in other older PLWH cohorts.

Prevalence of other psychiatric disorders

There are few studies reporting prevalence of non-mood psychiatric disorders in HIV, particularly in low- and middle-income countries, and we were unable to identify other studies of psychiatric disorders in older PLWH in SSA. Comparisons with other studies are therefore challenging.

Psychiatric disorders other than depression were low which is consistent with similar findings in the Global Burden of Disease survey (anxiety disorders 3.4% alcohol disorders 1.4% (The Institute for Health Metrics and Evaluation 2016)). SSA studies of younger PLWH report higher levels of both anxiety and alcohol disorders. A South African study (n = 65, md age 30) reported a MINI DSM-IV generalised anxiety disorder prevalence of 6.7% and 10.1% alcohol dependence(Olley et al. 2006), whereas 21.7% met anxiety disorder criteria in a larger Nigerian study (n = 300, median age 37) (Olagunju et al. 2012).

Though our cohort included individuals in ‘middle age’, the skew towards older age may be relevant. Anxiety disorders peak in middle age and decrease in older age (Bandelow and Michaelis 2015). Similarly, alcohol misuse peaks at age 25–34 in South African general population data (Trangenstein et al. 2018).

Social circumstances may also be important. Much of our cohort were employed (88.5%) and lived with others (83%) compared to the younger South African cohort where a third lacked family support and unemployment rates were high (24.7%) (Olagunju et al. 2012).

The prevalence of mental disorders tends to be higher in those with chronic disease such as obesity, diabetes, heart disease, cancer, and COPD and increases with the number of chronic diseases (Prince et al. 2007; The Institute for Health Metrics and Evaluation 2016). A meta-analysis investigating mental disorders in chronic disease reported a 36.6% prevalence of anxiety and/or depression in chronic disease, compared to 4.4% in the general population, and that chronic disease increased the risk of anxiety and/or depression by 310% using DSM-V and ICD-10 criteria (Dare et al. 2019). The population prevalence of comorbid mental disorders (depression, anxiety, bipolar disorders, schizophrenia) with chronic diseases is estimated at 45.8% in SSA (Prince et al. 2007; The Institute for Health Metrics and Evaluation 2016). It is surprising therefore that this older cohort with a chronic disease burden appears to have lower levels of both depression and other mental disorders (Dare et al. 2019).

Outcomes of depression at baseline

This is the only study we are aware of which investigates longitudinal outcomes of depression in older adults with HIV in SSA. The prevalence of depression appeared to decrease between baseline and follow-up though this was only significant by GDS and not DSM-IV criteria. Longitudinal reductions in depression have been reported in other studies of PLWH (supplementary Table 4), though these are majority male populations in high-income countries and/or intravenous drug user (IVDU) settings. One SSA study reported 8.9% reduction over 6 months, in the context of young PLWH before and after commencing cART (Olley et al. 2006). However, comparing existing data to our majority female cohort of older people stable on cART is challenging.

Persistent depression and chronic inflammation hypothesis

At follow-up, individuals depressed at baseline were more likely to ‘screen-positive’ on the GDS (depressive symptoms) but not to meet case-level DSM-IV. Depression is an episodic disorder and may remit over the 2-year follow-up. However, prior episodes may increase the risk of subsequent ones, in part due to the ‘kindling’ hypothesis where the level of stressor resulting in depression is thought to have less of an effect (Rabkin et al. 1997; Johnson et al. 1999). It is unclear whether these GDS findings represent subthreshold depression.

Chronic ‘subthreshold’ depressive symptoms are common in other chronic inflammatory diseases and also in conditions resulting in damage to frontal pathways (e.g. cerebrovascular disease) commonly reported to occur in chronic HIV infection (Harrison 2017). Increased CNS and peripheral inflammatory biomarkers are reported in depression and may reduce with antidepressant medication (Hegdahl et al. 2016).

HIV disease severity

We hypothesised that depression could be associated with ongoing HIV-related inflammation in this cohort, but baseline depression was not significantly associated with HIV disease severity including viral suppression at follow-up. In contrast, other cross-sectional SSA studies report associations with poor medication adherence, high HIV viral load, and low CD4 count (Ammassari et al. 2004; Olisah et al. 2015). This again may reflect the ‘healthy survivor’ effect and good HIV management seen in this cohort where cART adherence and HIV disease outcomes improved. HIV viral load testing introduction in 2017 may have contributed to this finding, as proportionally, more individuals receiving second-line therapy (indicating treatment failure) increased, and cART adherence also improved, potentially due to improved disease monitoring and clinician feedback.

Lack of association with HIV viral load does not exclude ongoing chronic inflammation as shown by an increased proportion receiving second-line therapy at follow-up. Chronic inflammation resulting from HIV infection may lead to depression (as with other chronic inflammatory conditions) despite a peripherally suppressed HIV viral load, given that the CNS is a ‘reservoir site’ for the HIV virus, due to limited CNS cART penetration (Gray et al. 2014). This potential ongoing CNS inflammation and damage might explain the association of depression with poorer neurological outcomes but not the measured HIV disease outcomes.

There is a high percentage of viraemic individuals at follow-up despite good reported cART adherence, and a larger proportion classified WHO stage 3 or 4 (advanced disease). Several factors may explain this finding. Some evidence suggests older age is hypothesised to increase risk of higher viral load and incomplete immune recovery (Pirrone et al. 2013). Although self-reported medication adherence is high, this may not represent objective adherence. Current guidelines suggest enhanced adherence counselling for low-level viraemia, and since viral load testing became available only a short while before our follow-up phase, many individuals may have undergone counselling prior to a switch to second-line therapy. Additionally, treatment failure may have contributed to this outcome, but we lack data on local resistance patterns.

Neurological impairment

Baseline depression was significantly associated with self- reported neurological and functional impairments at follow-up. It is unclear whether this represents a true increase.

Somatic manifestations of depression are common in older people and in SSA (Grover et al. 2019; Lee et al. 2008). Limited access to mental health services in SSA may result in misdiagnosis of depression as physical illness (Lee et al. 2008). Self-reported ‘slow hand movements’ could represent residual depression symptoms (psychomotor retardation, anergia), somatic representations of distress, or neurological impairment.

Similarly, ‘negative cognitions’ are well-recognised in depression and may persist despite remission, potentially leading depressed individuals to over-report symptoms and impairment.

Self-reported neurological symptoms were associated with depression but the (potentially more objective) clinical diagnosis of HAND. In PLWH, depressive symptoms, but not objective neuropsychological performance, may explain the variance between self-reported and clinical findings (Rourke et al. 1999). Furthermore, self-reported functional impairment may be inaccurate in those with cognitive impairment (Thames et al. 2011). However, over-reporting is unlikely, as baseline neurological and functional impairments were not associated with depression, and other psychiatric diagnoses were uncommon. The significant follow-up associations were self-reported and clinician-rated and included the objective measure of employment. This suggests that negative cognitive biases associated with depression are not the major explanation for our findings.

Psychosocial factors

Given that we found no association between depression and HIV disease factors, psychosocial factors may be relevant. Multiple factors such as isolation, discrimination, stigma, and living with chronic disease have been linked to depression in HIV (Munoz-Moreno et al. 2009). At baseline, we found no association between living alone, unemployment, educational background, and depression, but depression symptoms (GDS-15 ≥ 5) were associated with unemployment at follow-up.

Self-reported significant negative life events at follow-up were more likely in those depressed at baseline though overall depression prevalence reduced. Significant life stress is well-recognised to predict depression (Munoz-Moreno et al. 2009). Our findings suggest a possible bidirectional relationship. As discussed for neurological impairment, events may have been viewed more negatively due to negative cognitive biases (Wenzlaff and Bates 1998).

HIV is well-recognised to negatively impact long-term social and economic outcomes for families (Talman et al. 2013). Kenyan and South African data suggest earnings are lower in PLWH and that reduced earning potential results in sale of valuable assets such as livestock causing ongoing health and economic vulnerability (Talman et al. 2013). PLWH with depression may also be more likely to experience disability and reduced work productivity. In India, depression in HIV was associated with poorer environmental and social quality of life indicators (unemployment, low-income, poor diet, inadequate housing, and lack of life partner) (Charles et al. 2012). Similarly, a Ugandan study reported PLWH with depression (PHQ-9 and MINI) to be more likely to be unemployed and/or have lower weekly income than those without depression (Wenzlaff and Bates 1998). Therefore, depressed PLWH may be at higher risk of negative socioeconomic life events (Deshmukh et al. 2017).

Individuals with depression may be more likely to experience difficult social situations, negative interactions, and selectively focus on negative emotional stimuli (Steger and Kashdan 2009). This may result in fewer intimate relationships, greater negative responses from others, and greater emotional distress in response to social stressors (Steger and Kashdan 2009). Consequently, PLWH with depression may be more likely to experience stressful life events and also respond more negatively to them, creating a double burden of disability, whilst having lower ‘reserve’ due to lower socioeconomic status (Wagner et al. 2012). Conversely other studies report lower subsequent negative life events and stressors at follow-up in those with depression (Olley et al. 2006).

In summary, though comparable longitudinal data are lacking, it seems plausible that both HIV and depression may interact to result in greater self-reported negative life events and worse functional outcomes (‘objective’ and self-reported). Our data indicates that inflammation from HIV onto the CNS and physical stressors work together to manifest depression in PLWH.

Limitations

We selected a local translation of the GDS-15 based on extensive use in SSA but recognise that limitations in transcultural settings are well-recognised (Howorth et al. 2019). Similarly, we used a local translation of the MINI, previously used in the same clinic (Sumari-de Boer et al. 2018), but the low observed prevalence of some psychiatric disorders could be attributable to cultural differences in understanding of the MINI questions. We nevertheless could not identify any robustly validated cultural alternative. Though we report on psychosocial outcomes, we did not obtain information on self-perceived stigma and socioeconomic background, risk, and outcome factors which may have been relevant.

At baseline, psychiatric screening was completed by registered Tanzanian nurses with mental health research experience (one at doctoral level). Despite efforts to develop rapport, sensitivity of the questions and well-recognised stigma associated with mental illness may have led to under-reporting of psychiatric symptoms. Similarly, self-reported negative life events may have been under- or over-reported depending on rapport. Due to scarcity of specialist mental health services in Tanzania, previous psychiatric disorders are likely to be undiagnosed and subsequently under-reported. Similarly, since depression is episodic, depression which was not present during the study period will have been missed.

In 2018, the psychiatric interview was abbreviated to prevent participant fatigue and retain as many participants as possible. Since there were low rates of other psychiatric disorders, we decided to focus on depression. This may have impacted longitudinal findings.

Neurological symptoms were self-reported. A neurological examination was completed at baseline and follow-up but was insufficiently comprehensive to fully corroborate symptoms (e.g. peripheral neuropathy). The lack of HIV viral load testing at baseline limited assessment of HIV disease severity.

We had a relatively high rate of loss to follow-up. However, we were able to verify from clinic records that almost all not evaluated (72/91) remained under active HIV clinic follow-up and only 8/91 were recorded to have died. Though those seen and not seen at follow-up did not differ in recorded sociodemographic factors, it is not clear whether the follow-up cohort was representative in terms of outcome. Individuals with more severe depression may be less likely to engage with medical services, potentially skewing our results (DiMatteo et al. 2000; Dixon et al. 2016).

Study data collection corresponded with the crop-planting (rainy) season. Those failing to attend clinic may have been fitter and prioritising the annual community agricultural activities or, conversely, less able to travel in difficult road conditions.