Trends in mortality among ART‐treated HIV‐infected adults in the Asia‐Pacific region between 1999 and 2017: results from the TREAT Asia HIV Observational Database (TAHOD) and Australian HIV Observational Database (AHOD) of IeDEA Asia‐Pacific

Abstract Introduction AIDS‐related deaths in people living with HIV/AIDS have been decreasing in number since the introduction of combination antiretroviral treatment (cART). However, data on recent causes of death in the Asia‐Pacific region are limited. Hence, we analysed and compared AIDS‐related and non‐AIDS–related mortality in high‐ and low‐income settings in the region. Methods Patients from the TREAT Asia HIV Observational Database (TAHOD) and Australian HIV Observational Database (AHOD) receiving cART between 1999 and 2017 were included. Causes of death verification were based on review of the standardized Cause of Death (CoDe) form designed by the D:A:D group. Cohorts were grouped as AHOD (all high‐income sites), TAHOD‐high (high/upper‐middle income countries) and TAHOD‐low (lower‐middle income countries). TAHOD sites were split into high/upper‐middle income and lower‐middle income country settings based on World Bank classifications. Competing risk regression was used to analyse factors associated with AIDS and non‐AIDS–related mortality. Results Of 10,386 patients, 522 died; 187 from AIDS‐related and 335 from non‐AIDS–related causes. The overall incidence rate of deaths during follow‐up was 0.28 per 100 person‐years (/100 PYS) for AIDS and 0.51/100 PYS for non‐AIDS. Analysis indicated that the incidence rate of non‐AIDS mortality decreased from 0.78/100 PYS to 0.37/100 PYS from year groups 2003 to 2007 to 2013 to 2017 (p < 0.001). Similarly, incidence rates of AIDS‐related deaths decreased from 0.51/100 PYS to 0.09/100 PYS from year groups 2003 to 2007 to 2013 to 2017 (p < 0.001). More recent years of follow‐up were associated with reduced hazard for non‐AIDS mortality (2008 to 2012: aSHR (adjusted sub‐hazard ratio) 0.72, 95% confidence interval (CI) 0.54 to 0.96, p = 0.027; 2013 to 2017: aSHR 0.64, 95% CI 0.47 to 0.87, p = 0.004) compared to years 2003 to 2007. The AHOD cohort had almost twice the hazard of non‐AIDS mortality compared to TAHOD‐low (lower‐middle income sites) (aSHR 1.72, 95% CI, 1.20 to 2.46, p = 0.003); there were no differences between cohorts for AIDS‐related mortality (p = 0.834). Conclusion AIDS and non‐AIDS–related mortality rates have decreased over the past years in the Asia‐Pacific region. There is a greater risk for non‐AIDS–associated deaths in the AHOD cohort compared to lower‐middle income settings in TAHOD.

cancers [5]. According to the Data collection on Adverse events of anti-HIV Drugs (D:A:D) study, trends over time in all-cause mortality in people with HIV demonstrated a decrease in AIDS-related death rates [1]. Moreover, the proportion of non-AIDS cancers increased from 9% in 1999 to 2000 to 23% in 2009 to 2011 [1]. A Swiss HIV cohort study demonstrated similar results with 84% of deaths between 2005 and 2009 due to non-AIDS-related causes [6]. Improvement in CD4 cell count was accountable for the recent reductions in rates of AIDS-related deaths [1].
A previous study conducted in the Asia-Pacific region examined factors predictive of AIDS-related and non-AIDS-related mortality, as well as the cumulative incidences of each cause of death before the year 2007: 215 deaths (89 from AIDS, 97 from non-AIDS and 29 unknown) were identified, immune deficiency defined by lower CD4 cell counts was predictive of increased risk of AIDS-related and non-AIDS-related deaths, and older age (≥50 years) predicted non-AIDS mortality [7]. Other studies in the region have identified risk factors for developing AIDS-defining cancers [8,9].
We hypothesize that the causes of AIDS-related and non-AIDS-related mortality have changed over the past decade as HIV treatment programs in the Asia-Pacific region have matured, and that rates of AIDS mortality have decreased, followed by a relative increase in non-AIDS-related mortality. The objective of this study was to investigate the incidence rates of AIDS-related and non-AIDS-related deaths, and identify factors associated with different causes of death (COD) across country income levels in the Asia-Pacific region during the cART era.

| Study population and definitions
This analysis includes both Australian HIV Observational Database (Australian HIV Observational Database) and the TREAT Asia HIV Observational Database (TAHOD) enrolled patients who started cART and attended at least one subsequent follow-up visit. The AHOD and TAHOD are observational clinical cohort studies of patients with HIV infection in Australia and New Zealand (AHOD) and 12 countries (Cambodia, China and Hong Kong SAR, India, Indonesia, Japan, Malaysia, the Philippines, Singapore, South Korea, Taiwan, Thailand, and Vietnam) in Asia and the Pacific region (TAHOD), whose methods have previously been described [7][8][9].
The start date (baseline) for cART was defined as the date patients started their first triple regimen. Those who used mono/dual therapy prior to starting cART were excluded. Cohorts were grouped as AHOD (all high-income sites), TAHOD-high (high/upper-middle income countries) and TAHOD-low (lower-middle income countries). TAHOD sites were split into high/upper-middle income and lower-middle income settings (there are no low-income sites in TAHOD) based on World Bank classifications [10]. Deaths were classified as AIDS-related where the death was a direct consequence of any AIDS-defining condition (as defined by the Centers for Disease Control and Prevention [11]). Non-AIDS deaths were further classified as being related to cardiovascular disease, liver disease, cancer, accidental/overdose/violence, infectious disease, other and unknown [1]. Causes of death were based on review of the standardized Cause of Death (CoDe) form designed by the D:A:D group [12].

AIDS-related and non-AIDS-related mortality
Proportions of AIDS-related and non-AIDS-related mortality were analysed descriptively. Calendar years were categorized into ≤2007, 2008 to 2012 and 2013 to 2017. The denominator used to calculate each proportion was the total number of deaths in each year group. Mortality rates for different causes of deaths were also examined. The proportions and mortality rates were plotted across calendar years of follow-up. Unknown causes of death were plotted as a separate group to allow for visual inspection of the reported cases of non-AIDS causes of death.

| Factors associated with AIDS-related and non-AIDS-related mortality
Fine and Grays' competing risk regression model was used to analyse factors associated with AIDS-related and non-AIDSrelated mortality. The resulting sub-hazard ratio (SHR) can be interpreted in a similar way to the hazard ratio. Follow-up time began from cART initiation and left truncated at cohort entry for those who entered the cohort after starting cART. Left truncation minimizes survival bias by taking into account survival time prior to cohort entry in those who had initiated cART before enrolment into the cohort. The earliest year of follow-up started in 1999 for AHOD and 2002 for TAHOD. While followup time ended at date of death for patients who experienced mortality, all other patients were censored on their date of last follow-up. Both AHOD and TAHOD cohorts had their last follow-up date in year 2017. Other causes of death and loss to follow-up (LTFU) [13][14][15] were defined as competing events. Time-fixed covariates were age, sex, HIV mode of exposure, and cohort groups. Patients were considered hepatitis B virus (HBV) or hepatitis C virus (HCV) positive if they were ever tested positive for HBV surface antigen or positive for HCV antibody respectively. Covariates which varied with time were diabetes, hypertension, body mass index (BMI), HIV RNA, CD4 cell count and calendar year (≤2002, 2003 to 2007, 2008 to 2012, and 2013 to 2017). These variables were time-updated from risk time which began at antiretroviral treatment (ART) initiation or cohort entry for those who had initiated cART prior to cohort entry, till the date of last follow-up for those who were alive and date of death for those who experienced mortality. Time-updated variables contribute different risk time in different categories. The missing category in the time-updated variables refers to missing baseline values or patients who never had a test in the entire follow-up time. Diabetes was defined as documentation of one fasting blood glucose measurement ≥7 mmol/L. Hypertension was not included as blood pressure was not recorded by the AHOD cohort. BMI was defined as underweight (<18.5), normal range (18.5 to 24.9) and overweight (≥25). LTFU was defined as not having been seen in clinic for >12 months without evidence of transfer.
With the exception of cohort groups, covariates in the univariate analysis with p < 0.10 were fitted in the multivariate model using backward stepwise selection process. Covariates with p < 0.05 in the multivariate model were considered significant. A sensitivity analysis was performed by including age as a time-updated covariate, and adjusting for age, sex and CD4 cell count in the final multivariate model.
Ethics approvals for TAHOD were obtained from the respective local ethics committees of all participating sites, the data management and biostatistical centre at the Kirby Institute (The University of New South Wales (UNSW) Human Research Ethics Committee), and the coordinating centre at TREAT Asia/amfAR. Written informed consent was sought in TAHOD if required by a site's local institutional review board. All TAHOD data transfers are anonymized prior to submission to the Kirby Institute. Ethics approval for AHOD was obtained from the respective local ethics committees of all participating sites and the UNSW Human Research Ethics Committee; written informed consent was obtained from all participants. All data management and statistical analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA) and Stata software version 14.2 (Stata Corp., College Station, TX, USA).

| RESULTS
A total of 10,386 patients who had initiated cART and were in follow-up between years 1999 and 2017 were included in the study. With a median follow-up time of 6.0 years, 522 patients (155 from AHOD, 208 from TAHOD-high and 159 from TAHOD-low) experienced AIDS or non-AIDS mortality: 187 died from AIDS-related (incidence rate of 0.28 per 100 person-years (PYS)) and 335 died from non-AIDS-related causes (0.51/100 PYS) ( Table 1). In AHOD, the mortality rate for AIDS-related death was 0.15/100 PYS and for non-AIDS-related death it was 0.66/100 PYS, with a median follow-up time of 6.3 years. The mortality rates for TAHODhigh were 0.29/100 PYS for AIDS and 0.42/100 PYS for non-AIDS deaths with a median follow-up of 7.1 years, and for TAHOD-low the mortality rate was 0.40/100 PYS for AIDS and 0.49/100 PYS for non-AIDS with a median follow-up of 5.5 years.
The incidence rate of non-AIDs-related mortality during follow-up was 0.51 per 100 PYS (  Tables S1 and S2 show that by including time-updated age, sex and CD4 in the final multivariate model, there was a similar increasing trend in the hazard for mortality with increasing age, which has now shown significant associations with both AIDS and non-AIDS mortality. Other risk factors and their effects are similar to those observed in the main analyses.

| DISCUSSION
In our cross-regional observational cohorts who received cART, the overall incidence rate of mortality over a median follow-up time of 6.0 years was 0.28/100 PYS from AIDSrelated and 0.51/100 PYS from non-AIDS-related causes. Moreover, incidence rates for AIDS-related and non-AIDSrelated mortality have decreased over the years with the AHOD cohort showing higher hazard for non-AIDS deaths compared to TAHOD-low sites.
With the introduction of cART, AIDS-related death has decreased followed by an increase in non-AIDS death [4].   [1]. AIDS-related deaths demonstrated similar decreases in death rates over the same period from 5.9/1000 PYS to 2.0/1000 PYS [1]. In our study, incidence rates of AIDS-related deaths decreased from 0.51/100 PYS to 0.09/100 PYS from year groups 2003 to 2007 to 2013 to 2017. AIDS-related COD was the most common COD in the TAHOD cohort in year group ≤2007, and over the years there was a decrease in the overall incidence of AIDS deaths for all cohorts. These findings are thought to be associated with the introduction of effective cART in resource-limited settings, which slows the decline of CD4 cell counts, extending the time to AIDS-related death and the total number of AIDS deaths itself [3,16].
In a Swiss HIV cohort study, AIDS-related death was highest in 1992 (11.0/100 PYS, 95% CI 9.94 to 12.1) and decreased to 0.211/100 PYS (95% CI 0.122 to 0.363) in 2010 [6]. Non-AIDS-related mortality decreased from 1.74/ 100 PYS (95% CI 1.36 to 2.23) in 1993 to 0.86/100 PYS (95% CI 0.657 to 1.13) in 2010 [6]. The results of our study also demonstrated a declining trend in non-AIDS deaths in the later calendar years. Additionally, we found that there has been an increase in the proportion of unknown deaths recorded in the AHOD cohort. This increase in unknown deaths in AHOD was due to delay in reporting of the causes of death by the site. Mortality rates for AIDS and non-AIDS causes of death have declined over the years, with the incidence rate of non-AIDS-related death remaining high in AHOD, but declining, compared to other causes of death. Overall, there are higher incidences of non-AIDS deaths than AIDS-related mortality in later years in the Asia-Pacific region, similar to those demonstrated in other regions [1,3].
HIV-infected patients are prone to increased risk of cancer, and an estimated 30% to 40% of patients will develop cancer during their lifetime [17]. According to the D:A:D multicohort collaboration data, the incidence of non-AIDS cancers increased from 1.6/1000 PYS (1999 to 2000) to 2.1/1000 PYS (2009 to 2011) [1]. As people with HIV have a longer lifespan with potent cART, more patients develop malignancies including cancers not associated with AIDS [18,19], and this might also explain the higher proportion of non-AIDS deaths found in our cohorts. However, in this study, mortality rates for non-AIDS-related malignancies decreased from 0.16/100 PYS in year groups before 2007 to 0.07/100 PYS in year groups 2013 to 2017. Early treatment due to aggressive screening may be accountable for the decrease in non-AIDS malignancy mortality. Moreover, more detection of cancer due to aggressive screening in the AHOD cohort may be the reason for the higher risk of non-AIDS-related deaths in the AHOD cohort. The number of cardiovascular-related deaths was also higher in the AHOD cohort. Although there were no differences between cohorts for AIDS-related mortality, the AHOD cohort was at higher risk of non-AIDS mortality compared to both TAHOD cohorts with almost twice the hazard of non-AIDS mortality compared to TAHOD-low. The non-AIDS mortality highlights the importance for healthcare professionals to provide more comprehensive healthcare that includes screening for malignancy or cardiovascular disease.
Cardiovascular events associated with lower CD4 cell counts were suggested in both the Strategies for Management of Antiretroviral Therapy (SMART) and D:A:D studies [20,21]. Furthermore, immunodeficiency, identified by lower CD4 cell counts, was also strongly associated with liver-related non-AIDS death [22]. In a prospective cohort study of HIVinfected patients diagnosed at the age of 50 years or more, the baseline CD4 cell count was lower and presented shorter survival [23]. Moreover, continuing improvements in CD4 cell count over time were attributable to the reduction in AIDSrelated deaths [1]. Our findings are similar to results of these studies, demonstrating lower CD4 cell counts and older age as independent risk factors for both AIDS and non-AIDS mortality. Our study was also comparable to studies conducted throughout South America and the Caribbean, which have shown that low BMI and low haemoglobin were predictive of mortality, especially in the early phases of cART initiation [24]. Moreover, one prospective cohort study which followed up 1119 HIV patients for a median number of 8.2 years demonstrated that HIV-infected persons with food insecurity, categorized by low BMI, were about two times more likely to experience mortality [25]. In our study, low BMI categorized as underweight was also a risk factor for AIDS and non-AIDS mortality, putting emphasis on the need to incorporate nutritional support in HIV treatment programs. There are several limitations to our study. Information on patient adherence data is not collected in AHOD and it is possible that poor patient compliance or adherence may have been a reason for AIDS death. The delay in reporting in the causes of death in AHOD may lead to incomplete COD ascertainment and underestimation of the AIDS and non-AIDS causes of death. However, the majority of the unknown causes of death would be traced at a later date. While TAHOD represents a primarily urban referral patient population, crossing from low to high country income categories, it is less representative of national-level data. There is also the risk of unascertained mortality among those with LTFU.

| CONCLUSIONS
We observed that the rates of AIDS-related and non-AIDSrelated mortality in our Asia-Pacific cohorts have decreased over the past years. Factors associated with increased risk of AIDS-related mortality were higher viral load, high blood glucose, low CD4 cell count and low BMI. Older age, hepatitis B and C co-infection, high blood glucose, low CD4 cell count and BMI were associated with increased non-AIDS death. Blood glucose levels, CD4 cell count and being underweight were risk factors for both categories of deaths. The higher risk for non-AIDS-associated deaths in the AHOD cohort compared to the lower-middle income setting in TAHOD may be related to older age and higher per cent of reporting number of deaths. The slower rate in the decline of non-AIDS deaths and the higher proportions of non-AIDS deaths still persisting over AIDS-related deaths demonstrate that longer lifespans contribute to increased risk of developing cancers and other conditions more common with older age, resulting in more deaths from non-AIDS causes.

A U T H O R S ' C O N T R I B U T I O N S
JYC contributed to conceptualization and design of the study; IYJ wrote the first draft of the manuscript and interpreted the results; DR contributed to acquisition of data, data analysis and interpretation; IW, CO, MG and IA contributed to editing of the article; all authors have read and approved the final manuscript.

A C K N O W L E D G E M E N T S
unconditional grants from Merck Sharp & Dohme; Gilead Sciences; Bristol-Myers Squibb; Boehringer Ingelheim; Janssen-Cilag; and ViiV Healthcare. The Kirby Institute is funded by the Australian Government Department of Health, and is affiliated with the Faculty of Medicine, UNSW Sydney. The content of this publication is solely the responsibility of the authors and does not necessarily represent the official views of any of the governments or institutions mentioned above.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article: Table S1. Factors associated with AIDS mortality in the TREAT Asia HIV Observational Database and Australian HIV Observational Database cohorts with age as a time-dependent covariate including age, sex and CD4 in the multivariate model a Table S2. Factors associated with non-AIDS mortality in the TREAT Asia HIV Observational Database and Australian HIV Observational Database cohorts with age as a time-dependent covariate including age, sex and CD4 in the multivariate model a