Trends in all-cause mortality during the scale-up of an antiretroviral therapy programme: a cross-sectional study in Lusaka, Zambia

Abstract Objective To follow the trends in all-cause mortality in Lusaka, Zambia, during the scale-up of a national programme of antiretroviral therapy (ART). Methods Between November 2004 and September 2011, we conducted 12 survey rounds as part of a cross-sectional study in Lusaka, with independent sampling in each round. In each survey, we asked the heads of 3600 households to state the number of deaths in their households in the previous 12 months and the number of orphans aged less than 16 years in their households and investigated the heads’ knowledge, attitudes and practices related to human immunodeficiency virus (HIV). Findings The number of deaths we recorded – per 100 person–years – in each survey ranged from 0.92 (95% confidence interval, CI: 0.78–1.09) in September 2011, to 1.94 (95% CI: 1.60–2.35) in March 2007. We found that mortality decreased only modestly each year (mortality rate ratio: 0.98; 95% CI: 0.95–1.00; P = 0.093). The proportion of households with orphans under the age of 16 years decreased from 17% in 2004 to 7% in 2011. The proportions of respondents who had ever been tested for HIV, had a comprehensive knowledge of HIV, knew where to obtain free ART and reported that a non-pregnant household member was receiving ART gradually increased. Conclusion The expansion of ART services in Lusaka was not associated with a reduction in all-cause mortality. Coverage, patient adherence and retention may all have to be increased if ART is to have a robust and lasting impact at population level in Lusaka.


Introduction
The individual health benefits associated with antiretroviral therapy (ART) are well established 1,2 and treatment of human immunodeficiency virus (HIV) infections is now recommended worldwide. 3 Over the past decade, governments and donors have invested heavily in control and treatment of HIV and acquired immunodeficiency syndrome (AIDS). ART programmes have been expanded under the assumption that -given sufficient coverage -ART can reduce HIV-related mortality and reverse the rise in all-cause mortality that has been frequently associated with increasing prevalence of HIV infection. 4,5 In several studies in sub-Saharan Africa, reductions in population-level mortality have followed the introduction of ART services. [6][7][8][9][10][11][12] While encouraging, these studies only followed the short-term impact of the initiation of an ART programme and none was conducted solely in an urban setting where HIV prevalence was high.
The prevalence of HIV among adults living in Zambia was estimated to be 12.5% in 2011. 13,14 Since 2004, the Zambian Ministry of Health has being scaling up a programme for HIV care and treatment across the country's nine provinces. By 2011, 80% of the Zambians who were eligible for ART -more than 400 000 people -were receiving treatment. 15 In the capital city of Lusaka, which has a population of about 1.7 million, 16 HIV treatment in the public sector is currently available at 19 primary health centres and a tertiary care hospital. Between November 2004 and June 2011 -as public sector ART ser-vices were scaled up -we implemented a population-based, repeated cross-sectional study in Lusaka district. Our primary hypothesis was that, as the expanding public ART programme covered an ever larger proportion of the population, we would observe a reduction in all-cause mortality. We also sought to describe the trends in HIV-related knowledge, attitudes and practices and the trends in the presence of orphans under the age of 16 in the city's households.

Methods
The implementation of our study, which involved 12 rounds of household surveys, has already been described in detail. 17 Briefly, we divided Lusaka district into 24 clinic catchment areas, each comprising a varying number of standard enumeration areas. For each survey round we selected three enumeration areas within each catchment area using probability-proportional-to-size sampling. Within each selected enumeration area, interviewers first located the approximate centre of the area. Then, working outward in a randomly selected direction, they selected households at regular intervals -such as every fifth household they encountered -until they had selected 50 households. If interviewers were unable to contact a respondent in a selected household after three attempts -or if consent for that household's participation could not be obtained -they selected a neighbouring replacement household. In each of the 12 survey rounds, interviewers sampled 3600 households over a period of about three weeks.
Objective To follow the trends in all-cause mortality in Lusaka, Zambia, during the scale-up of a national programme of antiretroviral therapy (ART). Methods Between November 2004 and September 2011, we conducted 12 survey rounds as part of a cross-sectional study in Lusaka, with independent sampling in each round. In each survey, we asked the heads of 3600 households to state the number of deaths in their households in the previous 12 months and the number of orphans aged less than 16 years in their households and investigated the heads' knowledge, attitudes and practices related to human immunodeficiency virus (HIV). Findings The number of deaths we recorded -per 100 person-years -in each survey ranged from 0.92 (95% confidence interval, CI: 0.78-1.09) in September 2011, to 1.94 (95% CI: 1.60-2.35) in March 2007. We found that mortality decreased only modestly each year (mortality rate ratio: 0.98; 95% CI: 0.95-1.00; P = 0.093). The proportion of households with orphans under the age of 16 years decreased from 17% in 2004 to 7% in 2011. The proportions of respondents who had ever been tested for HIV, had a comprehensive knowledge of HIV, knew where to obtain free ART and reported that a non-pregnant household member was receiving ART gradually increased. Conclusion The expansion of ART services in Lusaka was not associated with a reduction in all-cause mortality. Coverage, patient adherence and retention may all have to be increased if ART is to have a robust and lasting impact at population level in Lusaka.

Research
Mortality trends during scale-up of antiretroviral therapy in Zambia Sujit D Rathod et al.
Interviewers asked a member of each selected household to identify the household heads. If an adult male and an adult female were identified as joint heads of a selected household, the female head was selected for interview. This preferential selection of female heads of households was to facilitate comparison of our data with data collected in the Zambia Demographic and Health Surveys. The selected head was interviewed in English, Nyanja or Bemba. The interviewer described the objectives of the survey to the head and the head was asked for their written informed consent. Once consent had been given, the interviewer used a standardized questionnaire to collect detailed demographic and medical information about all members of the household -including the timing of any deaths that had occurred among household members within the previous 12 months. Interviewees were asked to state how many orphans under the age of 16 years were living in their households. They were also asked about their knowledge, attitudes and practices relating to HIV. As in the 2007 Zambia Demographic and Health Survey, 13 interviewees were considered to have comprehensive knowledge of HIV if they correctly answered five questions about HIV transmission risk and so indicated that they knew that: HIV cannot be transmitted through mosquitoes, HIV cannot be transmitted by witchcraft, HIV transmission risk can be reduced through condom use, HIV transmission can be reduced by having one HIV-negative sex partner, and a healthy-looking person can have HIV.
Interviewees were asked if they had ever been tested for HIV and whether any non-pregnant members of their households were taking ART. The Institutional Review Board of the University of Zambia, the University of North Carolina at Chapel Hill and the University of Alabama at Birmingham approved the study protocol.
When planning our analysis, we assumed that most inhabitants of a specific catchment area would seek medical care at their local community clinic and that study households would only have access to ART when ART became available at that clinic. Our initial aim was therefore to compare trends in all-cause mortality at community level and to determine whether mortality trends remained the same before and after ART became available at the local clinic. However, a substantial proportion of interviewees in the first survey round reported that they had received medical care at clinics that were located outside their own communities. 17 We therefore decided to focus on district-wide trends. As accurate cause-of-death information could not be collected from our interviewees, we studied all-cause mortality -rather than HIV-related mortality -as our primary outcome measure. To relate such mortality to the scale-up of the ART programme, we report the number of people actively enrolled in the ART programme in Lusaka district at the time of each survey round, as well as the cumulative enrolment. 18,19 We tabulated the demographic characteristics of the study households and interviewees. For each survey round, we calculated rates of all-cause mortality as the numbers of deaths per 100 person-years -using the 12 months before the interview as the reference period. In these calculations, for each household member who had died in the 12 months before the interview and for each member who was reported to be less than 12 months of age at the time of the interview, we used the number of months that that member had been alive in the previous 12 months and 6 months as survival times, respectively.
To determine trends in mortality, we used a Poisson regression to estimate the mortality rate ratio -i.e. an estimate of the relative annual change in mortality -for the period 2004-2011. Next, we investigated this ratio for trend heterogeneity by sex and age group. The age groups we used -under 5, 5-14, 15-49 and over 50 years -were selected to facilitate comparison of our mortality data with those collected in Zambian Demographic and Health Surveys. We also calculated age-standardized mortality rates, using the INDEPTH Network's population structure for sub-Saharan Africa for reference. 20 In secondary analyses, we described trends in the respondents' HIV-related knowledge, attitudes and practices. For each question on such knowledge, attitudes and practices, we calculated the proportion in each survey round of each possible response -i.e. yes, no or do not know. We then calculated the proportion of comprehensive HIV knowledge among the interviewees in each survey round. We also described trends in the proportions of households that included orphans and households with a non-pregnant member taking ART. We analysed prevalence trends for each question using separate generalized linear regression models that provided estimates of the mean yearly change between 2004 and 2011.
The values we report have been adjusted for the complex sampling design. We calculated standard errors and corresponding 95% confidence intervals (CI) using Taylor series linearization. 21 Data were analysed using Stata 11.2 (StataCorp LP, College Station, United States of America).

Results
Of the 43 200 households included in our analysis, 3800 (9%) were replacement households. Of the replacement households, 2650 (70%) were selected because a household head could not be contacted and 1088 (29%) because a household head declined to participate. Interviewees in the 43 200 households provided information on 207 798 household members. Table 1 contains details of the timing of the surveys, along with contextual information about the number of clinics providing ART and the number of patients receiving ART at the time of each survey.
We only observed minor betweensurvey differences in the respondents' sex, marital status and -for female interviewees -parity. Food security and employment increased over the study period whereas immigration and denial of medical care -because of the patients' inability to pay for care -decreased ( Table 2). The overall mean age of the respondents were 33.2 years (range between surveys: 32.2-34.3), they had attained education for 8.6 years (range between surveys: 8.0-9.5) and had an average of three living children (range between surveys: 2.9-3.2).
Across all survey rounds, household members contributed 204 263 personyears of survival and 2537 deaths. Allcause mortality was found to decrease modestly over time (mortality rate ratio: 0.98; 95% CI: 0.95-1.00; P = 0.093; Fig. 1) and a similarly small decrease was seen after the age-standardization of the data (mortality rate ratio: 0.98; 95% CI: 0.95-1.00; P = 0.083). Although there was clustering of mortality by enumeration area, the effective sample size for the analysis of mortality trend remained extremely large.
As there was no evidence of mortality trend heterogeneity by age group Mortality trends during scale-up of antiretroviral therapy in Zambia Sujit D Rathod et al.
(P = 0.31) or sex (P = 0.50), we calculated stratum-specific mortality rates across the study period ( Over the first seven surveys, interviewees were increasingly likely to know where to get free ART. The proportion of interviewees who reportedly knew where to obtain such free therapy had approached 100% by the eighth survey and remained very high for the remain-der of the study period. Across the whole study period, interviewees were increasingly likely to report that a non-pregnant household member was taking ART; to report that they had been tested for HIV; to have comprehensive HIV knowledge; or to know someone with HIV or someone who had died of AIDS. Conversely, the proportions of interviewees who reportedly believed that HIV/AIDS was "a punishment from God for promiscuity" and who reported the presence in their household of at least one orphan aged less than 16 years decreased during the study ( Fig. 2; P < 0.001 for the trend in each measure).

Discussion
Between 2004 and 2011, we documented positive trends in HIV-related knowledge, attitudes and practices across Lusaka district. These results align with the corresponding steady increase in individuals enrolled into HIV care and actively receiving ART. Although our study was timed to coincide with the roll-out of one of the largest urban ART programmes in Africa, 18,22 these encouraging trends did not correlate with a substantial decline in all-cause mortality in our study area. We observed measurable increases in employment, food security, knowledge about HIV and health-care access and decreases in the proportion of households with orphan members. Taken together, these trends indicate an increase in the general wellbeing of the study population.
Our failure to observe a substantial reduction in mortality contrasts with the results of several studies conducted elsewhere in Africa. In rural South Africa, for example, researchers investigated the effects of an ART programme's expansion over a 3-year period and observed reductions in mortality of more than 20% among adults and about 49% among young children. 6,7 Similar reductions in adult mortality were observed in rural Malawi, in the 4 years after the expansion of an ART programme. 8,9 In urban Ethiopia, Reniers et al. used data from burial site registries and reported reductions in adult mortality of more than 20% in the 3 years after ART had been introduced. 10 Burial site surveillance was also used in Malawi, which showed a 37% reduction in all-cause mortality in the 5 years after ART scale-up. 11 Finally, investigation of sibling survival data from Demographic

Research
Mortality trends during scale-up of antiretroviral therapy in Zambia Sujit D Rathod et al.
and Health Surveys conducted in 27 African countries, including nine in which an HIV treatment programme funded by the United States President's Emergency Plan for AIDS Relief was operating, found that the odds of death from any cause were 16% lower in these nine countries than in the 18 other countries. 12 Our results also appear inconsistent with country-wide data from the World Bank and the Joint United Nations Programme on HIV and AIDS, which indicate that, in Zambia, there have been large declines in all-cause mortality since 2001 23 and in HIV-related mortality since 2003. 14 Mathematical models have demonstrated that low or delayed ART up-take will attenuate reductions in AIDS mortality [24][25][26][27] and this may be applicable in Lusaka. However, there are several other possible reasons why we failed to observe a more substantial reduction in all-cause mortality. First, it is possible that increases in non-HIV-related mortality offset the expected declines in HIV-related mortality. We believe this to be unlikely, however, given the apparent improvements in interviewees' socioeconomic status -e.g. in terms of employment and food security -and in their access to health care. Further, in Zambia the risk of death attributable to HIV is known to be extremely high. 28 Second, our survey began shortly after a rapid scale-up began and without an opportunity to measure the mortal-ity rate experienced immediately before the introduction of free ART services. The introduction of free ART may have rapidly led to substantial reductions in HIV-related mortality that we simply missed because they occurred before our first survey.
Third, the threshold of ART coverage necessary to produce major community-wide reductions in all-cause mortality may not have been reached yet. Early screening and entry into care are needed to ensure optimal survival for HIV-infected individuals. It was only early in 2010 that the Zambian ART programme adopted more inclusive eligibility criteria for ART, including a higher CD4+ lymphocyte threshold of 350 cells per mm 3 . In addition, greater programmatic focus is needed in the areas of patient adherence and retention. Attrition is high in the Lusaka ART programme, 18,29 as in other ART programmes in sub-Saharan Africa. [30][31][32] As mathematical models show, when many patients are lost to follow-upand, therefore, discontinue therapy prematurely -the public health benefits of ART programme expansion are substantially reduced.
One encouraging result from our study was that the proportion of study households that held young orphans decreased from 17% in November 2004 to 7% in September 2011. This apparent decline in orphanhood is supported by the figures reported in Zambia's Demographic and Health Surveys for 2001-2002 and 2007. 13, 33 Orphanhood has been shown to be associated with adult AIDS mortality. 34,35 We also observed encouraging trends in HIV-related knowledge, attitudes and practices. This observation demonstrates the successful penetration of health communication messages about HIV among the residents of Lusaka. As comprehensive knowledge about HIV and of the location of ART services approaches universality among the adult residents of Lusaka, those residents are, presumably, increasingly prepared to access ART themselves -if needed -or to facilitate the entry of other residents into HIV care. The stigma associated with HIV appears to be on the wane in Lusaka. The proportion of interviewees who reported that a non-pregnant member of their household was receiving ART was only 0.8% in November 2004 but had risen more than 9-fold, to 7.4%, by September 2010.   Mortality trends during scale-up of antiretroviral therapy in Zambia Sujit D Rathod et al.

Survey date
The strengths of our study include the use of a standardized questionnaire, the large sample size and a study period that extended beyond the first few years after ART scale-up. Our study also had some important limitations. First, the study relied on data reported by selfidentified heads of households and both the reporting and self-identification may be prone to bias. Any such bias is, however, likely to have been similar in each survey round and, reassuringly, the mean adult mortality that we recorded per 1000 person-years -13.3 deathsis similar to the corresponding values recorded in Zambia's Demographic and Health Surveys for 2001-2002 and 2007: 14.1 and 12.5 deaths, respectively. Second, we made no attempt to investigate those members of the study households who died more than 12 months before the interviews or those who left study households -and, possibly, died -in the 12 months before the interviews. HIV-related immigration into a community and HIV-related emigration out of a community -as described in South Africa 36 -can create unquantifiable bias in the estimation of mortality. Third, although HIV-related mortality would have been a more useful primary outcome in our study, we were unable to collect information regarding the cause or circumstances surrounding each death of interest. Fourth, although only 9% of the heads of households included in the initial selections of households for each survey round refused to participate in the study, such a level of non-participation could have been sufficient to alter our main findings -if the non-participating households had been systematically different from the participating households in terms of our primary and secondary outcomes. Finally, our sampling frame was based on the most recent and available census data for Lusaka, which were collected in 2000. These data do not necessarily reflect a population that has been highly dynamic and growing since the year 2000.
In conclusion, despite encouraging increases in comprehensive HIV knowledge and improved HIV-related attitudes and practices, the expansion of ART services in Lusaka between 2004 and 2011 coincided with only a modest reduction in all-cause mortality. While this finding was unexpected, it empha-sizes several critical factors for improving the population-level impact of ART. Further expansion of coverage with ART services requires close monitoring and investment, particularly for patients who qualify for HIV treatment but have not yet become ill. Greater coverage needs to be accompanied by a more systematic application of effective measures to increase patient adherence and retention. Monitoring population-level outcomes -using resource-appropriate methods such as the one described here -should be implemented, to provide ongoing feedback on programme performance. Finally, investments in ART programmes -and, more broadly, in health-systems -must continue. In settings such as Lusaka, which have clearly benefited from successful education and outreach programmes, a realignment of the ART programme's priorities may be needed to ensure maximal public health benefit. ■