HIV and the growing health burden from noncommunicable diseases in Botswana: modelling study

Background The “greying of AIDS” – the aging of the population living with HIV who benefit from antiretroviral treatment (ART) and the emergence of age-related non-communicable diseases (NCDs) – has been well documented. The emerging health systems challenges – eg, the implications of HIV on the disease burden from NCDs on the population level, and the evolving role of HIV as a co-morbidity or co-existing disease of various NCDs – are less well understood. The paper elucidates these challenges by providing a quantitative analysis of HIV-NCD interactions for Botswana. Methods We projected the prevalence of HIV and of selected NCDs in Botswana using demographic and HIV-specific estimates building on data on the state and the dynamics of the HIV epidemic, using the Spectrum modelling software, and extrapolating on estimates of the prevalence of NCDs from the 2015 global burden of disease (GBD). Results HIV has slowed down overall population aging and thus has attenuated the growing burden of many NCDs so far, because cohorts reaching old age have been decimated by AIDS-related mortality in the 1990s and early 2000s. Aging and the rise in the prevalence of NCDs, however, will accelerate rapidly from about 2030 because of reduced attrition of cohorts living with HIV since the start of the ART scale-up in Botswana. While HIV prevalence will decline over time, the health needs of people living with HIV will become more complex. HIV prevalence among the growing populations affected by various important NCDs will not decline for decades, because of the aging of the population living with HIV and interactions between HIV, ART and NCDs. Conclusions Even though HIV prevalence is projected to decline steeply to 2030 because of reduced HIV incidence, the prevalence of HIV among people affected by many of the most important NCDs will increase or barely change. While the health care needs of people living with HIV will increase and become more complex, HIV will also emerge as a key factor complicating the management of the growing burden of NCDs. Health systems will need to prepare for the challenge of large numbers of patients living with both HIV and NCDs.


II. Appendix S2. Separate Projections for Men and Women
While we report estimates and projections for the population overall in the paper, the underlying analysis was done for men and women separately, in light of the straightforward availability of the relevant data differentiated by sex. Overall, the results come out similarly for the male and female populations, which is one reason why the paper focuses on the population overall. As an example for the separate results for men and women, Annex Tables 1A and 1B show estimates of the prevalence of NCDs among people living with HIV differentiated by sex.

Appendix S3 Alternative Projections of HIV-NCD Intersection
The results presented in the paper (Tables 1 to 4) are based on the assumption that the agespecific prevalence of NCDs will remain constant from 2017 -a crude assumption which is not implausible as estimated prevalence rates of NCDs (unlike mortality) in the Global Burden of Disease database change only slowly over time. One advantage of this approach is that the results can be attributed clearly to changes in the age structure of the relevant populations.
It would be similarly plausible to extrapolate based on trends in the prevalence of NCDs (although it is important to bear in mind that the scaling-up of antiretroviral therapy occurred from 2004, and the estimates would reflect any direct effects on this on the prevalence of NCDs). For this reason, and as a robustness check of the results reported in the paper, we reproduce key results (Tables 1 and 3) based on extrapolation of trends in 2010-2017 (Annex Tables 2 and 3). The sensitivity check excludes diabetes and kidney diseases, because we our estimates of the prevalence of diabetes is based on cross-sectional data from the 2014 STEPS survey rather than the GBD 2017 estimates of prevalence of NCDs over time. Based on extrapolation of trends in prevalence of NCDs, rather than assuming that age-and sex-specific prevalence remains constant from 2015.

Appendix S4. Alternative Estimates for Constant HIV Incidence
As a robustness check of our results with regard to the national HIV policies implemented or their effectiveness, we consider an alternative scenario in which HIV incidence remains constant from 2017 (at 1.3. percent annually). Results are summarized for prevalence of NCDs among people living with HIV (Annex Table 4, corresponding to Table 1), and HIV prevalence among people affected by various NCDs (Annex Table 5, corresponding to Table 3).
We do not provide a sensitivity analysis with respect to the pace of scaling up treatment, because coverage is already high at the outset. This means that most people access treatment relatively early, before reaching a stage of the disease characterized by high mortality. In this context, the most important outcome of faster scaling up -for the purposes of our projections - Based on extrapolation of trends in prevalence of NCDs, rather than assuming that age-and sex-specific prevalence remains constant from 2015.

A-7
is reduced HIV incidence through viral suppression captured in the sensitivity analysis on incidence.
Annex  Table shows all "level 2" NCD categories from GBD 2017, and a selection of more specific diseases (indented). The relative increase is calculated as the ratio of prevalence in 2040 and 2015, respectively, minus one, multiplied by 100.

Appendix S5Estimating and Projecting HIV-NCD Comorbidities
One important factor in our estimates is the form of the link between HIV status and the prevalence of NCDs. To the extent that these are positively correlated within the categories by sex and age we are using, for behavioural reasons (if an NCD risk factor is more common among PLWH than otherwise) or reflecting direct effects of HIV or long-term treatment on NCDs, our estimates of the prevalence of NCDs among PLWH and of HIV among people affected by NCDs would be underestimates.
There is a considerable literature on HIV-NCD Comorbidities, including a comprehensive survey of the state of knowledge in the special issue of the Journal of Acquired Immune Deficiency Syndromes on "HIV Noncommunicable Disease Comorbidities in Low-and Middle-Income Countries in the ART Era" of September 2014. However, evidence across NCDs on such

A-9
HIV-NCD linkages is uneven, and many studies do not provide estimates for matching populations not affected by HIV/AIDS, and the link between HIV and NCDs is likely endogenous, depending on types of drugs and factors like treatment initiation and duration (Althoff, Smit, Reiss, and Justice, 2016), and changing over time (Vachiat and others, 2017). Moreover, the bulk of the evidence is based on data from high-income countries, and their results may not translate well into the context of Southern Africa (Levitt, Steyn, Dave, and Bradshaw (2011), Mosepele and Botsile (2018)).
In light of these uncertainties, and considering the possibility that the link between HIV and NCDs may change over the next decades as drug regimens are optimized to reduce adverse consequences of long-term ART (plausible in light of the evidence on risk of cardiovascular disease and different types of antiretroviral drugs used so far, see Islam, Wu, Jansson, and Wilson (2012);Martin-Iguacel, Llibre, and Friis-Moller (2015); Vachiat and others (2017)), we do not include an elevated risk of developing NCDs among PLWH in most of our results. However, we report alternative estimates for ischemic heart disease and diabetes mellitus (see Table 4 of main paper), two diseases for which the evidence on HIV-NCD links is relatively strong, and which are important contributors to the burden of disease in Botswana, ranking 2nd and 5th, respectively, among causes of death (with HIV still being the no. 1 cause).
For ischemic heart disease, we assume in the alternative specification that prevalence is elevated by a factor of 1.5 among people with HIV, controlling for sex and age. This factor is consistent with and at the lower end of results reported in comprehensive surveys specifically on HIV and ischemic heart disease (Vachiat and others (2017) or HIV and cardiovascular diseases more generally (Islam, Wu, Jansson, and Wilson (2012);Martin-Iguacel, Llibre, and Friis-Moller (2015), which report a relative risk of cardiovascular disease among PLWH (compared to HIVnegative people) of between 1.5 and 2.0.
For diabetes, sex-and age-specific prevalence among PLWH is elevated by a factor of 1.6, relative to HIV-negative people in that population group. This estimate follows Prioreschi and others (2017), who synthesize available evidence from sub-Saharan Africa (but point to a high variation among the estimates from the underlying studies).
While there is a considerable empirical literature on links between HIV/AIDS and cancers, we did not include cancers here because most of the literature regards the declining role of AIDS-defining cancers following introduction of antiretroviral therapy (which is not important here because treatment coverage is already high at the beginning of the projection) and the subsequent increase in the role of non-AIDS defining cancers, largely in line with the aging of the population living with HIV (Chinula,Moses, and Gopal, 2017), a factor we do capture. Moreover, the evidence is heterogeneous with regard to types of cancer and different types of antiretroviral therapy (Borges and others (2017)).
The sex-and age-specific prevalence of the selected NCDs (ischemic heart disease and diabetes mellitus), among PLWH and otherwise, has been estimated with the following equation: The implied assumption that NCD M is constant across sex and age groups may be restrictive.
To introduce variations in NCD M by sex or age, we would require corresponding empirical estimates, or need to adopt an explicit model of NCD incidence and progress with and without NCDs. Empirical studies differentiating M by sex or age are rare, one such study (Althoff and others (2015), focusing on incidence of myocardial infarction, renal disease, and non-AIDSdefining cancer) suggests that the factor M is even across ages. One study adopting an more explicit modeling approach (Smit and others (2018), on Zimbabwe) concludes that results are primarily driven by the changing age profile of people living with HIV rather than the cumulative exposure to HIV and ART, and points to the limitations imposed by sparse data availability for age-specific NCD incidence or prevalence estimates.