For HIV-infected patients with indications for antiretroviral therapy (ART) in Africa, timely ART initiation during this period of high clinical risk is essential to effective and efficient care. Initiating ART, however, even for patients who have already been tested and enrolled into treatment programs, requires overcoming significant psychological, sociostructural, and operational barriers. First, patients must first accept their diagnosis to make rational choices about prioritizing care, but stigma and denial may take time to overcome.1 Second, disclosure of HIV status to family and friends can ease the burden of taking a life-long medication, but patients also fear that disclosure may lead to hostility or rejection.2 Third, requirements for multiple adherence counseling sessions before ART initiation and for “treatment supporters” present at many clinics may improve medication adherence after initiation, but these processes may inadvertently magnify barriers to ART initiation for patients unable to meet these requirements.3 Fourth, patients often reside far away from the clinic, and therefore, travel time, travel costs, and competing priorities from work and childcare may delay completion of pre-ART evaluation.4,5
The magnitude and consequences of failures or delays in ART initiation are incompletely understood because loss to follow-up between eligibility and ART initiation (ie, unknown outcomes) is high—from 15% to more than 50%—in many African clinics.6–12 “Loss to follow-up,” however, is not in and of itself a patient outcome but rather an artifact of what the clinics can conveniently observe and yields limited actionable public health information. Meaningful outcomes, hidden by loss to follow-up, include deaths before ART initiation; “silent” transfers (including possibly ART initiation at new sites); and disengagement from care with failure to start ART. Understanding outcomes in the group classified as “lost to follow-up” among patients already identified as eligible for ART is needed to assess the magnitude of delays and failures in ART initiation, the rate of death during the waiting period, and reasons for disengagement from care before ART initiation.
In this article, we examine delays and failures of ART initiation in a clinic-based cohort of ART-eligible patients over a 4-year period in rural Uganda. To address the selection bias potentially introduced by loss to follow-up, we identified a random sample of lost patients, intensively sought their outcomes in the community, and used these outcomes to reclassify outcomes among all lost patients. We have previously applied this approach to evaluate outcomes among patients who had already initiated ART.5,13 In the present article, we apply this approach to recover valid estimates of the magnitude and determinants of ART initiation, deaths, and disengagement from a population of ART-eligible patients.
We evaluated HIV-infected adults with a CD4 count–based indication for ART (ie, CD4+ T cell ≤250 cells/μL in contemporaneous Uganda National Guidelines) between October 1, 2007 and January 27, 2011 at the Immune Suppression Syndrome (ISS) Clinic in Mbarara District in southwestern Uganda. Of note, the study population includes both patients who were enrolled in the clinic before October 1, 2007 but who were not ART eligible until after this date. The ISS Clinic is a “prototypical” scale-up ART clinic in that the patient volume is high (up to 150 visits a day is common), no routine HIV RNA testing is available, and it is supported by the Uganda Ministry of Health through implementing partners of the US President's Emergency Fund for AIDS Relief. In this analysis, observation began on the date of first CD4 ≤250 cells/μL, and patients were followed until time of death, loss to follow-up (defined as 60 days late for last scheduled appointment), ART initiation, or closure of database on January 27, 2011. Among those lost to follow-up, a random sample was selected for tracking in the community to obtain updated data on vital status, ART use, and HIV care either through contact with the patient directly or a close relation of the patient when the patients could not be found in person.5 The size of the sample was determined by operational capacity: we tracked as many patients as we could with the available resources.
Sociodemographic and clinical information were collected in the course of routine care by providers using standardized paper forms from the Uganda Ministry of Health and supplemented by forms developed at the clinic. Data are then subsequently entered by hand into an OpenMRS database. Diagnosis of tuberculosis at time of eligibility was defined as any documentation of clinical or microbiological tuberculosis in the 9 months before or 30 days after ART eligibility. Malnourishment was defined by standardized World Health Organization (WHO) criterion of a body mass index (BMI) <18.5 kg/m2. For the sample of lost patients, a patient tracker went into the community to ascertain outcomes using a standardized questionnaire. Key information sought included vital status and date of death, if applicable. Patients found in person were asked whether they had seen a doctor or nurse for HIV care in the proceeding 90 days, whether they were on ART (defined as any ART within the last 14 days), and if so, when they started ART.
In the clinical population, as originally observed (ie, without incorporating outcomes obtained through tracking a sample of the lost patients), we estimated the incidence of loss to follow-up through 1 year after ART eligibility (defined as 60 or more days late for an appointment). New patients not on ART are typically asked to return in 2–4 weeks after their first visit, and established patients with CD4 levels above treatment thresholds and not on ART are customarily asked to return in 1–4 months. For this estimate, we used the cumulative incidence approach in which ART initiation and deaths were considered competing events.14–16 We also analyzed the group of patients who were lost to follow-up and, subsequently, successfully tracked to estimate the cumulative incidence of mortality, from time of last clinical visit using the Kaplan–Meier approach; the hazard of mortality after last clinic visit; and the predictors of mortality using a proportional hazards regression.
We then analyzed the entire clinical population in which we used the outcomes discovered through tracking a random sample of lost patients to represent outcomes in all lost patients. Loss to follow-up that is differential on either the outcome alone or on outcomes and exposures of interest is a well-recognized cause of bias. Conceptually, incorporating outcomes of the sample of successfully tracked patients addresses the effects of differential loss to follow-up by replacing unknown outcomes with known outcomes in a representative sample, which allows recovery of unbiased estimates. Statistically, lost patients with outcomes ascertained through tracking were given a weight inverse to the sampling probability, lost patients without updated outcomes were assigned a weight of zero, and all patients who remain under observation at the clinic receive a weight of 1.13,17 Operationally, consider a hypothetical clinical population of 10,000 patients, in which 1000 patients are lost to follow-up, and in which the ascertainer seeks a 100 of the lost patients, and successfully finds outcomes in 90 patients. In this example, after tracking, 910 patients have outcomes that remain unknown (900 of whom were lost and never sought, and 10 of whom were unsuccessfully sought). They are given a weight of zero and effectively removed from the analyses. The 90 patients who were initially lost to follow-up but then successfully sought (and who therefore now have known outcomes) are used to represent the 910 without known outcomes through a weight of 10.1 (910/90). The patients who remain under observation are assigned a weight of 1 (thus representing only themselves), as their outcomes are by definition completely known.
In this sample-weighted population, we used the cumulative incidence method to estimate the incidence of ART initiation (at the original or any other clinic), mortality before ART initiation, and disengagement from care (defined as 90 days without a visit to any facility) over the first year after ART eligibility for the entire clinic population. For the specific outcome of disengagement from care (and its complement “retention in care”), we assumed that care utilization among those interviewed in person represented all lost patients who were alive. Therefore, patients who were lost and interviewed in person were additionally weighted in inverse proportion to the probability that a living patient was interviewed in person. Confidence intervals for weighted descriptive estimates were obtained through bootstrapping.
We conducted 2 proportional hazards regressions of the reweighted data: one to identify factors associated with ART initiation and the second to identify factors associated with mortality—both in the entire clinical population. Predictor selection was guided by the desire to evaluate the influence of as many predictors as possible but avoid colliders and overfitting. Missing predictor data was addressed with multiple imputation18,19 under the “missing at random” assumption. Of note, multiple imputation was not used to address missing outcome data, a problem that we addressed with the sampling-based approach. All analyses were conducted in Stata version 11.1 (StataCorp, College Station, TX). Ethical approval was granted by the University of California, San Francisco, the Uganda National Counsel of Science and Technology, Kampala, Uganda, and the Mbarara University of Science and Technology, Mbarara, Uganda.
Between October 1, 2007 and January 27, 2011, 2633 patients first became eligible for ART as defined by immunological criteria (ie, a CD4 count ≤250 cells/μL). The median age was 32 years [interquartile ratio (IQR): 27–39 years] and 1070 were men (40.6%). The median CD4 count at eligibility was 131 cells/μL (IQR: 53–198), 645 (24.5%) patients with WHO stage 4, and 115 (4.4%) patients had a diagnosis of tuberculosis at eligibility. In 2247 (85.3%) patients, the first CD4 count taken at the ISS Clinic was <250 cells/μL, and we considered these patients eligible at the time of clinic enrollment (Table 1).
Tracking a Sample of the Patients Lost to Follow-Up
The cumulative incidence of loss to follow-up at 1 year was 21.3% (95% CI: 19.6–23.0). Of the 490 patients lost to follow-up, we sought a random sample of 132 (27%) patients for tracking (Fig. 1). In 111/132 (84.0%) patients, updated information about vital status or care status was obtained. Among the 111 successfully tracked patients, 42 had died and the cumulative incidence of mortality 1 year after last clinic visit was 36.8% (95% CI: 28.2–47.0). The hazard, or instantaneous rate, of death was highest in the 60 days after last visit at 38/100 person-years, falling to 11/100 person-years from 60 to 180 days and to 8/100 person-years, from 180 to 365 days after last visit (Fig. 2). Of the 69 successfully tracked patients who were alive, in 22 (32%) cases, the patient him or herself was interviewed and in 47 (68%), a sibling, spouse, close friend, or other close acquaintance acted as the informant. Of 22 patients who were alive and found in person, 14 (64%) patients reported having seen a doctor or nurse for HIV care at a different facility within the last 90 days. The remaining 8 (36%) of 22 patients reported no recent contact with health care. In a proportional hazards regression of predictors of survival among those lost to follow-up, nearer district of residence, lower enrollment CD4 level, lower BMI, and earlier calendar date of last visit were associated with higher mortality (see Table S1, Supplemental Digital Content, http://links.lww.com/QAI/A402).
Sample-Weighted Outcomes After ART Eligibility
After incorporating the outcomes among the lost with probability weights for the entire clinic population, the cumulative incidence of ART initiation at 30, 90, and 365 days after eligibility was 16.0% (95% CI: 14.2 to 17.7), 64.5% (95% CI: 60.9 to 68.1), and 81.7% (95% CI: 77.7 to 85.6). The cumulative incidence of death before ART initiation at 30, 90, and 365 days after ART eligibility was 1.6% (95% CI: 0.05 to 2.5), 5.5% (95% CI: 2.8 to 8.1), and 7.7% (95% CI: 5.6 to 10.1). One year after eligibility, 10.0% of eligible patients were alive but had never started ART (Fig. 3). The large majority (69.5%) of these patients not on ART after 1 year were lost to follow-up from the ISS Clinic.
In multivariable analyses, higher age at enrollment and, notably, being unemployed were associated with faster ART initiation after adjustment for other sociodemographic and clinical factors. Male sex, higher CD4 count levels at presentation, having no formal education, and being new to care at the time of eligibility were all associated with a decreased rate of ART initiation. In multivariate analysis, to identify factors associated with mortality, malnutrition, lower CD4 level at clinical enrollment, greater travel time from residence to clinic, and district of residence were associated with higher risk of mortality (Table 2).
Using a sampling-based approach to understand the experience of ART-eligible patients under routine program conditions in Uganda where loss to follow-up is high, we found that about 1 in 5 patients eligible for ART by CD4 count did not initiate ART 1 year after eligibility, and that 1 in 12 patients died while awaiting ART initiation. We identified a number of patient (CD4 level and male sex), systems (whether or not you were new to care at the time of ART eligibility), and structural factors (travel time), as well as, in a separate analysis, mortality to be associated with the rate of ART initiation, which shed light on the multidimensional determinants of barriers to effective care. These observations were made through application of a sampling-based approach in a clinic-based cohort to address biases presented by loss to follow-up. This approach strengthens both internal validity and external validity because the data come from a “real world” setting.
Although there is no absolute criterion about how soon eligible patients should start ART, our estimate of time to ART initiation demonstrates 2 concerning features. First, ART initiation is not rapid—at 60 days only 50% of eligible patients had initiated. In particular, a large fraction of eligible patients were classified as having WHO stage 3 or 4 and, therefore, likely had active opportunistic infections. These patients would likely benefit from ART initiation within 14 days, as suggested by ACTG protocol 5164.20 Second, uptake of ART remained incomplete 1 year after eligibility. The ISS Clinic, like most clinics in Africa, typically required 2–3 counseling sessions before ART initiation and also a treatment supporter. In resource-limited settings, these requirements may magnify barriers to ART initiation although they may strengthen adherence. Failure to achieve this last step of ART initiation is tantamount to a failed investment in testing and linkage.
We identified patient, structural, and operational factors associated with the rate of ART initiation that point to actionable gaps in public health ART delivery. Previous cross-sectional analyses have found that male sex is a consistent predictor of late stage presentation among HIV-infected persons in Africa,21 and in this analysis, men seem to start ART almost 30% more slowly than women after adjusting for presentation CD4 level. Further research on the psychological aspects of male health care behavior in Africa is needed. Patients with no education, most of whom are likely to be illiterate, started ART at a slower rate. Patients without education should be flagged as a vulnerable group who may require additional assistance to successfully navigate the health care system. Unemployed patients initiated ART faster. Work responsibilities have been identified as a barrier to retention in previous work and, therefore, likely explain diminished ART initiation rates.5 Exploration of alternative initiation strategies that are partly community based, using peer educators or “community adherence groups” are needed to enable access health care without jeopardizing the livelihoods of patients.22,23 Finally, we found an interesting programmatic effect: patients who are new to care at eligibility are initiated on ART faster than patients who become eligible after already being in care. This is likely because patients already in care receive longer return visits because they appear to be “stable.” Yet, this introduces a longer interval to their next visit when they do become eligible by CD4 count criteria. Randomized trails of both community-based peer advocates and Short Message Service technology have been shown to reduce HIV RNA rebound in Africa for patients already on ART. These evidence-based strategies should be explored as methods to facilitate ART initiation newly eligible patients as well.
This analysis identified 4 factors associated with mortality that can be used to guide clinical attention to high-risk patients. The CD4 level at ART eligibility is a consistent predictor of mortality and likely reflects opportunistic infections already in progress at the time of ART initiation. Rolling out enhanced diagnostics for smear-negative tuberculosis, Cryptococcus neoformans, and other common infections is critical in preventing these deaths.24,25 A BMI <18.5 kg/m2 conferred a 3-fold rise in the rate of death even after adjusting for CD4 level, TB diagnosis, and other sociodemographic factors. These patients likely represent a mix of individuals with undiagnosed infections and persons with macronutritional deficiency. Food supplementation in this subset of patients may improve survival and wider dissemination of nutritional interventions may improve outcomes.26–28 Travel time from residence to clinic was associated in a dose response relationship to survival. This finding suggests that addressing geographic and physical accessibility through decentralization may merit greater prioritization as a strategy to make treatment delivery more effective.29,30
This study has certain limitations. First, although our sample was formally a random subset of the lost to follow-up, not all outcomes were ascertained, and therefore the reweighted estimates may be biased. We believe given more than 80% of outcomes were ascertained, the opportunity for bias is not large. Second, although we believe the ISS Clinic is representative of large ART programs in Africa, processes are likely better than the “average clinic” because of its affiliation with Mbarara University and membership in the IeDEA consortium. If true, this would imply that the shortcomings with uptake of ART are even more concerning than those we observed here.
In summary, we used a sampling-based approach, which allows us to maximize validity with a highly generalizable patient population, to understand uptake of ART among eligible patients at a scale-up clinic in Uganda. This approach is broadly feasible. On-site start-up costs for 3 months to track 120 patients was approximately US $5000. This includes US $1200 dollars for a motorbike, US $600 per month as salary for the ascertainer, US $500 for fuel, US $500 per month for other consumables and administrative costs. We observed a strong relationship between death and loss to follow-up, which suggests that analyses of mortality in the pre-ART population based on passively ascertained vital status information is likely biased and, possibly, severely so.31,32 We found the rate and total uptake of ART was neither optimally fast nor complete. Factors associated with uptake identify groups of patients who may benefit from targeted interventions, namely, men and patients with little or no education. Enhanced program features, such as point-of-care, distributed care sites, and text messages to notify patients already in care when they become eligible for ART, may improve ART uptake, delivery, and potentially survival. ART is highly efficacious; improving the utilization of ART under program conditions through implementation strategies can make this intervention more effective in routine care.
The authors are grateful to Hassan Baryahikwa, Megan Lazzar, and Michael Kanyesigye.
1. Nam SL, Fielding K, Avalos A, et al.. The relationship of acceptance or denial of HIV-status to antiretroviral adherence among adult HIV patients in urban Botswana. Soc Sci Med. 2008;67:301–310.
2. Medley A, Garcia-Moreno C, McGill S, et al.. Rates, barriers and outcomes of HIV serostatus disclosure among women in developing countries: implications for prevention of mother-to-child transmission programmes. Bull World Health Organ. 2004;82:299–307.
3. Siedner MJ, Lankowski A, Haberer JE, et al.. Rethinking the “pre” in pre-therapy counseling: no benefit of additional visits prior to therapy on adherence or viremia in Ugandans initiating ARVs. PLoS One. 2012;7:e39894.
4. Tuller DM, Bangsberg DR, Senkungu J, et al.. Transportation costs impede sustained adherence and access to HAART in a clinic population in southwestern Uganda: a qualitative study. AIDS Behav. 2009;14:778–784.
5. Geng EH, Bangsberg DR, Musinguzi N, et al.. Understanding reasons for and outcomes of patients lost to follow-up in antiretroviral therapy
programs in Africa
through a sampling-based approach. J Acquir Immune Defic Syndr. 2010;53:405–411.
6. Rosen S, Fox MP. Retention in HIV care between testing and treatment in sub-Saharan Africa
: a systematic review. PLoS Med. 2011;8:e1001056.
7. Karcher H, Omondi A, Odera J, et al.. Risk factors for treatment denial and loss to follow-up
in an antiretroviral treatment cohort in Kenya. Trop Med Int Health. 2007;12:687–694.
8. Tayler-Smith K, Zachariah R, Massaquoi M, et al.. Unacceptable attrition among WHO stages 1 and 2 patients in a hospital-based setting in rural Malawi: can we retain such patients within the general health system? Trans R Soc Trop Med Hyg. 2010;104:313–319.
9. Zachariah R, Harries AD, Manzi M, et al.. Acceptance of anti-retroviral therapy among patients infected with HIV and tuberculosis in rural Malawi is low and associated with cost of transport. PLoS One. 2006;1:e121.
10. Amuron B, Namara G, Birungi J, et al.. Mortality
and loss-to-follow-up during the pre-treatment period in an antiretroviral therapy
programme under normal health service conditions in Uganda. BMC Public Health. 2009;9:290.
11. Bassett IV, Wang B, Chetty S, et al.. Loss to care and death before antiretroviral therapy
in Durban, South Africa
. J Acquir Immune Defic Syndr. 2009;51:135–139.
12. Micek MA, Gimbel-Sherr K, Baptista AJ, et al.. Loss to follow-up
of adults in public HIV care systems in central Mozambique: identifying obstacles to treatment. J Acquir Immune Defic Syndr. 2009;52:397–405.
13. Geng EH, Emenyonu N, Bwana MB, et al.. Sampling-based approach to determining outcomes of patients lost to follow-up in antiretroviral therapy
scale-up programs in Africa
. JAMA. 2008;300:506–507.
14. Prentice RL, Kalbfleisch JD, Peterson AV Jr, et al.. The analysis of failure times in the presence of competing risks. Biometrics. 1978;34:541–554.
15. Satagopan JM, Ben-Porat L, Berwick M, et al.. A note on competing risks in survival data analysis. Br J Cancer. 2004;91:1229–1235.
16. Marubini E, Valsecchi MG. Analysing Survival Data from Clinical Trials and Observational Studies. Chichester, United Kingdom: John Wiley & Sons, Ltd; 1995.
17. Frangakis CE, Rubin DB. Addressing an idiosyncrasy in estimating survival curves using double sampling in the presence of self-selected right censoring. Biometrics. 2001;57:333–342.
18. Schafer JL. Multiple imputation: a primer. Stat Methods Med Res. 1999;8:3–15.
19. Schafer JL, Graham JW. Missing data: our view of the state of the art. Psychol Methods. 2002;7:147–177.
20. Zolopa A, Andersen J, Komarow L, et al.. Early antiretroviral therapy
reduces AIDS progression/death in individuals with acute opportunistic infections: a multicenter randomized strategy trial. PLoS One. 2009;4:e5575.
21. Geng EH, Kreiswirth BN, Burzynski J, et al.. Transmission trends for human immunodeficiency virus associated tuberculosis in New York City. Int J Tuberc Lung Dis. 2005;9:661–666.
22. Decroo T, Telfer B, Biot M, et al.. Distribution of antiretroviral treatment through self-forming groups of patients in Tete province, Mozambique. J Acquir Immune Defic Syndr. 2011;56:e39–e44.
23. Decroo T, Van Damme W, Kegels G, et al.. Are expert patients an untapped resource for ART provision in sub-Saharan Africa
? AIDS Res Treat. 2012;2012:749718.
24. Boehme CC, Nabeta P, Hillemann D, et al.. Rapid molecular detection of tuberculosis and rifampin resistance. New Engl J Med. 2010;363:1005–1015.
25. Van Rie A, Page-Shipp L, Scott L, et al.. Xpert® MTB/RIF for point-of-care diagnosis of TB in high-HIV burden, resource-limited countries: hype or hope? Expert Rev Mol Diagn. 2010;10:937–946.
26. Tsai AC, Bangsberg DR, Emenyonu N, et al.. The social context of food insecurity among persons living with HIV/AIDS in rural Uganda. Soc Sci Med. 2011;73:1717–1724.
27. Miller CL, Bangsberg DR, Tuller DM, et al.. Food insecurity and sexual risk in an HIV endemic community in Uganda. AIDS Behav. 2011;15:1512–1519.
28. Weiser SD, Young SL, Cohen CR, et al.. Conceptual framework for understanding the bidirectional links between food insecurity and HIV/AIDS. Am J Clin Nutr. 2011;94:1729S–1739S.
29. Tanser F, Barnighausen T, Cooke GS, et al.. Localized spatial clustering of HIV infections in a widely disseminated rural South African epidemic. Int J Epidemiol. 2009;38:1008–1016.
30. Tanser F, Lesueur D, Solarsh G, et al.. HIV heterogeneity and proximity of homestead to roads in rural South Africa
: an exploration using a geographical information system. Trop Med Int Health. 2000;5:40–46.
31. Geng EH, Glidden DV, Bangsberg DR, et al.. A causal framework for understanding the effect of losses to follow-up on epidemiologic analyses in clinic-based cohorts: the case of HIV-infected patients on antiretroviral therapy
. Am J Epidemiol. 2012;175:1080–1087.
32. Yiannoutsos CT, An MW, Frangakis CE, et al.. Sampling-based approaches to improve estimation of mortality
among patient dropouts: experience from a large PEPFAR-funded program in Western Kenya. PLoS One. 2008;3:e3843.