1 Background The roll out of ART has led to dramatic reductions in - - PDF document

1 Beyond AIDS-related deaths: calculating the risk of non-communicable disease mortality attributable to HIV from verbal autopsy data C Calvert1, A Price1, 3, E Slaymaker1, G Reniers1, K Herbst3, D Michael4, S Clark5, B Zaba1, A Crampin1, 2

1London School of Hygiene and Tropical Medicine, London, UK 2Malawi Epidemiology and Intervention Research Unit, Malawi 3African Health Research Institute, UKZN, South Africa 3Malawi Epidemiology and Intervention Research Unit, Malawi 4National Institute for Medical Research, Mwanza, Tanzania 5The Ohio State University, Columbus, USA

2

Background

The roll out of ART has led to dramatic reductions in mortality in people living with HIV (PLHIV) in sub-Saharan Africa;(1, 2) however, mortality rates remain higher in PLHIV compared with HIV-negative people, despite evidence from high income settings suggesting that mortality rates in PLHIV with adequate viral suppression are the same as mortality rates in the general population.(3) Estimates from five studies in Eastern and Southern Africa indicate that, with wide availability of ART, the overall excess mortality attributable to HIV is between 22% and 46% amongst men living with HIV, with between 14% to 28% of mortality in HIV-positive women attributable to HIV.(4) This excess mortality attributable to HIV many be caused by: (1) the direct effect of HIV; (2) the effects of ART and; (3) other conditions, not traditionally classified as HIV-related, but which HIV increases the risk of (e.g. sepsis and malaria). Indeed, a study using information

• n deaths which occurred largely in the pre-ART era in Eastern and Southern Africa found

that HIV increased the risk of all other causes of death,(5) including over 10-fold increase in the risk of a number of non-communicable diseases (NCDs), e.g. digestive cancers. Quantifying the extent to which HIV is associated with NCDs is of particular importance, due to the considerable burden of NCDs in areas of high HIV prevalence. Recent estimates suggest that there has been a 46% increase in the numbers of deaths due to NCDs in sub- Saharan Africa since 1990.(6) Several pathways exist by which HIV may increase the risk of mortality from NCDs. Firstly, some malignant cancers arise from HIV opportunistic infections including Kaposi sarcoma and HIV-associated lymphoma. Secondly, biological effects of HIV infection may increase the risk

• f certain NCDs. For example, HIV is linked with detrimental reductions in high density

lipoproteins, important for transporting excess cholesterol to the liver.(7) Finally, some NCDs have been directly linked to ART use, including diabetes.(8) The increased risk of some NCDs, however, may be offset by lower body weights in PLHIV which lowers the risk of, for example, hypertension and diabetes. Methodological challenges, including limited data on causes of death in sub-Saharan Africa, the difficulty in ascertaining some NCD deaths and lack of information on HIV status, have hindered exploration of HIV as a risk factor for NCD mortality. The aim of our paper is to use standardised verbal autopsy data for individuals with known HIV status, to calculate rate ratios comparing NCD cause-specific mortality rates in HIV-positive and HIV-negative adults aged 15-59. To assess the impact of the HIV epidemic on NCD mortality at the population level, we will calculate population attributable fractions.

Data and Research Methods Data Sources

We used data from three study sites for this analysis: Karonga in Malawi, Kisesa in Tanzania, and uMkhanyakude in South Africa. These studies have all completed several rounds of demographic surveys, and conducted population-based HIV testing and verbal autopsies (VAs). Although similar data are collected, start dates, frequency of data collection and the level of detail of the information varies across studies. The methods of data collection for each study site are described in detail elsewhere.(9-11)

Data Preparation

Data analysts at each study site prepared the demographic surveillance site (DSS), HIV and VA data according to a standard specification. Each study extracted information including dates of entry to and exit from the study population, due to death or out-migration, and

3 information on the HIV test status of each study participant and signs and symptoms reported in a VA for the deceased. VA data were merged with the DSS data and only retained for analysis if linkage to a DSS death record was made. The VA and DSS data were also linked with HIV test data collected from sero-surveys. A number of assumptions were made when assigning HIV status which have been described in detail elsewhere.(12) Data from the time period before ART became available, which were only available for Kisesa, were excluded for this analysis, as were any data after the end of 2014. The cause of death attribution was done on the basis of VA interviews that were interpreted with the InSilicoVA tool.(13) Using InSilicoVA, estimates were generated for sub-populations defined by two broad age groups (below 60, and 60 or older), gender, and HIV status at the time of death. The individual-level cause-specific probabilities of death were then used to assign a main cause of death for each individual. This was assigned as the cause of death with the highest probability for each individual. Where no cause had greater than a 0.4 probability, the cause was assigned as indeterminate. For this analysis three main outcomes were defined: all NCD deaths; cancer-related deaths; and deaths related to cardiovascular

• disease. All NCD deaths comprised of deaths assigned to any of: cancer, severe anaemia,

severe malnutrition, diabetes mellitus, acute cardiac disease, sickle cell with crisis, stroke,

• ther and unspecified cardiac disease, chronic obstructive pulmonary disease, asthma, acute

abdomen, liver cirrhosis, renal failure, epilepsy, and other and unspecified NCD. Cardiovascular disease included acute cardiac disease, stroke, and other and unspecified cardiac disease.

Data Analysis

Analyses were carried out using Stata 14. Using survival analysis, total person years, numbers

• f deaths and cause-specific mortality rates were estimated by HIV status for adults aged 15-
• 59. Rate ratios, comparing cause-specific mortality rates in PLHIV and HIV-negative people,

were calculated. Where there was evidence that HIV increased the risk of NCD mortality, the percentage of deaths from that category of NCD (all, cancer or cardiovascular disease) which are attributable to HIV was subsequently calculated by subtracting the NCD mortality rate in the HIV-negative population from the NCD mortality rate in the total population with known HIV status. This difference was then divided by the overall NCD mortality rate in those with known HIV status, to give the percentage of NCD mortality attributable to HIV.

Preliminary findings

A total of 3,630 deaths with known HIV status to adults aged 15-59 were identified from the three studies. The majority of these deaths were from uMkhanyakude (N=2,765), with 496 deaths from Karonga and 369 from Kisesa. The percentage of the deaths identified in the DSS which received a VA varied across the sites with 100% VA coverage in Karonga, 95.0% in uMkhanyakude and 79.7% in Kisesa. There were limited data on the individual ART treatment status of these individuals who died; however, it has been estimated elsewhere that the percentage of adults 15-59 who had ever had ART in 2013 was 68.5% in Karonga, 64.6% in uMkhayakude and 30.7% in Kisesa.(12) The percentage of deaths attributed to NCDs varied from 8.5% in uMkhanyakude, to 23.1% and 26.4% in Kisesa and Karonga, respectively. Table 1 provides preliminary results, based

• n data up to 2014, of the NCD-related mortality rates by HIV status. In uMkhanykude, there

was no evidence that HIV increased the rate of mortality attributable to all NCDs (p=0.76), but strong evidence to suggest that there was double the rate of mortality attributable to cancer in

4 adults with HIV (rate ratio (RR):2.00, 95% confidence interval (CI):1.27-3.16, p-value=0.002). There was strong evidence for an increased rate of NCD mortality in PLHIV compared with HIV negative individuals in Karonga (RR: 2.30, 95% CI:1.50-3.53, p-value<0.001) and Kisesa (RR: 3.82, 95% CI:2.16-6.78, p-value<0.001).

Table 1: Non communicable disease cause-specific mortality rates by HIV status in adults (15-59), by study Cause & HIV status

• No. of

deaths Person years Rate per 10,000 (95% confidence interval) Crude rate ratio (95% confidence interval) p-value % of NCD mortality attributable to HIV uMkhanyakude All NCD Negative 126 93969 13.4 (11.3-16.0) 1 Positive 107 76629 14.0 (11.6-16.9) 1.04 (0.81-1.35) 0.76

• Cancer

Negative 30 93969 3.2 (2.2-4.6) 1 Positive 49 76629 6.4 (4.8-8.5) 2.00 (1.27-3.16) 0.002 31.0% Cardiovascular Disease Negative 40 93969 4.3 (3.1-5.8) 1 Positive 21 76629 2.7 (1.8-4.2) 0.64 (0.38-1.09) 0.10

• Karonga

All NCD Negative 105 77535 13.5 (11.2-16.4) 1 Positive 26 8357 31.1 (21.2-45.7) 2.30 (1.50-3.53) <0.001 11.2% Cancer Negative 17 77535 2.2 (1.4-3.5) 1 Positive 2 8357 2.4 (0.6-9.6) 1.09 (0.25-4.73) 0.91

• Cardiovascular

Disease Negative 4 77535 0.5 (0.2-1.4) 1 Positive 1 8357 1.2 (0.2-8.5) 2.32 (0.26-20.75) 0.44

• Kisesa

All NCD Negative 53 55170 9.6 (7.3-12.6) 1 Positive 15 4085 36.7 (22.1-60.9) 3.82 (2.16-6.78) <0.001 16.3% Cancer Negative 22 55169 4.0 (2.6-6.1) 1 Positive 5 4085 12.2 (5.1-29.4) 3.07 (1.16-8.11) 0.02 12.1% Cardiovascular Disease Negative 1 55170 0.2 (0-1.3)

• Positive

4085

• The percentage of cancer mortality attributable to HIV was estimated to be 31.0% in
• uMkhayakude. In Karonga, the percentage of NCD mortality attributable to HIV was 11.2%,

compared with 16.3% in Kisesa. These analyses have some limitations. Using models such as InSilicoVA to interpret VA data has well recognised limitations, including not using information provided in the narrative section where the caregiver/relative of the deceased gives an unprompted description of the

5 circumstances leading up to the death. We have therefore only relied on very broad groupings

• f NCDs, particularly as many deaths could only be classified as “other or unspecified NCDs”.

In future, incorporating information from physician review of the VAs will help us to understand better the association of HIV and, for example, more specific cancer diagnoses. A further limitation of VA analyses is that it is challenging to identify deaths caused by complications of ART, and many of these are likely to be classified as NCDs. Planned refinements to the InSilicoVA software in the next year, incorporation of data on ART usage, plus physician review of narratives should enable us to better distinguish deaths related to ART treatment. Our preliminary analyses provide unique insights into the contribution of HIV to NCD mortality across three study sites in Eastern and Southern Africa, suggesting that at least 10% of NCD mortality is attributable to HIV in the two East African studies. There was no evidence for an

• verall association between HIV and NCD mortality in the South African study, but there was

an increased risk for deaths from cancer. The observed variation between studies may arise from differences in detection and treatment of NCDs, highlighting the importance of integrating and/or improving links between HIV and general health services. In our ongoing analyses we will explore heterogeneity in these associations by sex and include data from at least one other Eastern Africa study, with the view to pooling data across the different study sites.

References

• 1. Kasamba I, Baisley K, Mayanja BN, Maher D, Grosskurth H. The impact of antiretroviral treatment on

mortality trends of HIV-positive adults in rural Uganda: a longitudinal population-based study, 1999–

• 2009. Tropical Medicine & International Health. 2012;17(8):e66-e73.
• 2. Reniers G, Araya T, Davey G, Nagelkerke N, Berhane Y, Coutinho R, et al. Steep declines in population-

level AIDS mortality following the introduction of antiretroviral therapy in Addis Ababa, Ethiopia. Aids. 2009;23(4):511-8.

• 3. Lewden C, Bouteloup V, De Wit S, Sabin C, Mocroft A, Wasmuth JC, et al. All-cause mortality in treated

HIV-infected adults with CD4 >/=500/mm3 compared with the general population: evidence from a large European observational cohort collaboration. Int J Epidemiol. 2012;41(2):433-45.

• 4. Slaymaker E, Todd J, Marston M, Calvert C, Michael D, Nakiyingi-Miiro J, et al. How have ART treatment

programmes changed the patterns of excess mortality in people living with HIV? Estimates from four countries in East and Southern Africa. Global Health Action. 2014;7:10.3402/gha.v7.22789.

• 5. Byass P, Calvert C, Miiro-Nakiyingi J, Lutalo T, Michael D, Crampin A, et al. InterVA-4 as a public health

tool for measuring HIV/AIDS mortality: a validation study from five African countries. Global Health

• Action. 2013;6:10.3402/gha.v6i0.22448.
• 6. Naghavi M, Forouzanfar MH. Burden of non-communicable diseases in sub-Saharan Africa in 1990 and

2010: Global Burden of Diseases, Injuries, and Risk Factors Study 2010. The Lancet.381:S95.

• 7. Rose H, Hoy J, Woolley I, Tchoua U, Bukrinsky M, Dart A, et al. HIV Infection and High Density

Lipoprotein Metabolism. Atherosclerosis. 2008;199(1):79-86.

• 8. Butt AA, McGinnis K, Rodriguez-Barradas MC, Crystal S, Simberkoff M, Goetz MB, et al. HIV Infection

and the Risk of Diabetes Mellitus. AIDS (London, England). 2009;23(10):1227-34.

• 9. Crampin AC, Dube A, Mboma S, Price A, Chihana M, Jahn A, et al. Profile: the Karonga Health and

Demographic Surveillance System. Int J Epidemiol. 2012;41(3):676-85.

• 10. Kishamawe C, Isingo R, Mtenga B, Zaba B, Todd J, Clark B, et al. Health & Demographic Surveillance

System Profile: The Magu Health and Demographic Surveillance System (Magu HDSS). Int J Epidemiol. 2015;44(6):1851-61.

• 11. Tanser F, Hosegood V, Barnighausen T, Herbst K, Nyirenda M, Muhwava W, et al. Cohort Profile:

Africa Centre Demographic Information System (ACDIS) and population-based HIV survey. Int J Epidemiol. 2008;37(5):956-62.

• 12. Slaymaker E, Wringe A, McLean E, Calvert C, Marston M, Reniers G, et al. Life and death on the HIV

care continuum: evidence from 7 population-based cohorts in sub-Saharan Africa. Forthcoming.

• 13. McCormick TH, Li ZR, Calvert C, Crampin AC, Kahn K, Clark SJ. Probabilistic cause-of-death assignment

using verbal autopsies. Journal of the American Statistical Association. 2016(just-accepted):1-38.