Temporal Trends in Treatment Outcomes for HIV-1 and HIV-2-Infected Adults Enrolled in Côte d'Ivoire's National Antiretroviral Therapy Program

Background In Côte d'Ivoire during 2004–2007, numbers of ART enrollees increased from <5,000 to 36,943. Trends in nationally representative ART program outcomes have not yet been reported. Methodology/Principal Findings We conducted a retrospective chart review to assess trends in patient characteristics and attrition [death or loss to follow-up (LTFU)] over time, among a nationally representative sample of 3,682 adults (≥15 years) initiating ART during 2004–2007 at 34 health facilities. Among ART enrollees during 2004–2007, median age was 36, the proportion female was 67%, the proportion HIV-2-infected or dually HIV-1&2 reactive was 5%, and median baseline CD4+ T-cell (CD4) count was 135 cells/µL. Comparing cohorts initiating ART in 2004 with cohorts initiating ART in 2007, median baseline weight declined from 55 kg to 52 kg (p = 0.008) and the proportion weighing <45 kg increased from 17% to 22% (p = 0.014). During 2004–2007, pharmacy-based estimates of the percentage of new ART enrollees ≥95% adherent to ART declined from 74% to 60% (p = 0.026), and twelve-month retention declined from 86% to 69%, due to increases in 12-month mortality from 2%–4% and LTFU from 12%–28%. In univariate analysis, year of ART initiation was associated with increasing rates of both LTFU and mortality. Controlling for baseline CD4, weight, adherence, and other risk factors, year of ART initiation was still strongly associated with LTFU but not mortality. In multivariate analysis, weight <45 kg and adherence <95% remained strong predictors of LTFU and mortality. Conclusions During 2004–2007, increasing prevalence among ART enrollees of measured mortality risk factors, including weight <45 kg and ART adherence <95%, might explain increases in mortality over time. However, the association between later calendar year and increasing LTFU is not explained by risk factors evaluated in this analysis. Undocumented transfers, political instability, and patient dissatisfaction with crowded facilities might explain increasing LTFU.


Introduction
Similar to other countries in West Africa [1,2], Côte d'Ivoire faces a dual epidemic of HIV-1 and HIV-2 [3]. Current adult HIV-1 prevalence is estimated at 3% [4], while about 5% of HIVinfected adults are HIV-2 or HIV-1&2 dually reactive [5]. Although the burden of the HIV-2 epidemic is limited [1,6], antiretroviral therapy (ART) programs need to provide supplies and training for adequate identification and treatment of HIV-2, which differs from that of HIV-1 [7], complicating the program in an already challenging setting [8], where resources are limited and political instability has culminated in two civil wars in the last decade [9].
Despite these challenges, the Ministry of Health (MOH) and international partners, including the United States (U.S) President's Emergency Plan for Relief (PEPFAR) and the Global Fund to Fight AIDS, Tuberculosis, and Malaria (GFATM), have increased numbers of ART enrollees about 20-fold from less than 5,000 to 104,750 during 2004-2012 [10]. Although sub-national ART programs in Côte d'Ivoire have reported their treatment experience for the period 2004-2008 [5,11], these previous reports cannot be considered nationally representative [12,13]. Investigating and reporting national ART program outcomes is important to provide a representative assessment of program quality and justify continued funding [14][15][16]. Describing program trends over time at a national level, and assessment of factors associated with national outcomes, can help to identify areas for national program improvement activities [17,18].
Therefore, in 2009-2010, we conducted a retrospective, cohort study among a nationally representative sample of adult ART patients starting ART during 2004-2007, to describe trends in patient characteristics at ART initiation over time and trends in mortality and loss to follow-up (LTFU).

Ethics Approval
This study was approved by the Ivorian Ethics Review Committee (Comité National d'Éthique des Sciences de la Vie et de la Santé), the Institutional Review Board (IRB) of the U.S. Centers for Disease Control and Prevention (CDC), and the Harvard School of Public Health IRB. Patient informed consent was not required as only routine, anonymized, monitoring data were collected and analyzed.

Eligibility for ART
During 2004-2007, patients were eligible for ART when diagnosed as having World Health Organization (WHO) stage IV, WHO stage III with CD4 counts #350/mL, or WHO stage I/ II with CD4 counts #200/mL [5]. Prescription of co-trimoxazole (CTX) was indicated for all ART patients with CD4 count #350/ mL.

Patient Monitoring
At ART initiation, monthly for the first 3 months, and quarterly thereafter, standardized MOH-recommended medical records were completed to monitor disease progression or improvement. Patients collected medications monthly from clinic pharmacies where the date of scheduled antiretroviral (ARV) pick-up appointments and actual ARV pick-updates were documented.

Study Design and Population
This was a retrospective cohort study. Patient-level data were abstracted from standardized, MOH-recommended medical records onto study questionnaires by trained abstractors from November 2009 through March 2010. Only medical records of adult patients, $15 years old at ART initiation, who started ART during 2004-2007, were eligible.

Sample Size
Sample size calculations were performed using Epi Info software (CDC, Epi Info 2008, Version 3.5.1, Atlanta, GA). To achieve a 95% confidence interval (CI) of +2.5% around the estimate for 12-month attrition, assuming a design effect of 1.5 [13], and a conservative (i.e., higher than expected) 12-month attrition percentage of 50% [19], a sample size of $2,301 patient records was needed. To meet the needs of a secondary analysis, aimed at assessing site-level predictors of patient outcomes, we aimed to sample 4,000 medical records.

Sampling
Of 124 ART delivery sites in the country by December 31, 2007, 78 had provided ART to $50 adults. Only 833 (2.3%) of all 36,943 adult patients who had received ART by December 31, 2007, were enrolled at sites that had supported ,50 patients on ART by this time. To maintain feasibility, 35 (45%) of the 78 eligible sites were randomly selected, using a two-stage sampling strategy. In stage one, the 78 eligible clinics were divided into three strata based on which organization was largely responsible for implementing the ART program at the site (non-governmental organization, MOH, or GFATM through MOH). Within these three strata, sub-strata were created according to site size (number of ART patients ever enrolled). Site size sub-strata were: small (50-250), medium (251-1,000), and large (.1,000). Within each substratum, SAS 9.2 (SAS Institute Inc., Cary, NC) was used to randomly sample facilities using probability-proportional-to-size sampling. Of the selected 35 clinics, 34 agreed to participate.
In stage two, simple random sampling was used to select 4,000 medical records from the 34 selected, consenting facilities. The total number of medical records selected in each sub-stratum was proportional to the number of eligible records in the corresponding sub-stratum in the general adult ART population by 2007.

Treatment Outcomes
The primary outcomes of interest after ART initiation were documented death and LTFU, and the secondary outcome of interest was the composite outcome of attrition (documented death or LTFU). A patient was considered LTFU if he/she was absent from the health facility in the 90 days preceding data abstraction, and if there was no documentation of death or transferal to another health facility. The date of LTFU was recorded as the date of the most recent visit. Transfers were censored from timeto-event analyses at the date of transfer. Data for time-to-event analysis (i.e. date of ART initiation and date and nature of the final outcome status) were complete.

Exposure Variables
Patient-level characteristics routinely captured on standard MOH medical records (Table 1) were considered a priori risk factors for inclusion in the multivariable models for each of the three outcomes -death, LTFU, and overall attrition. CD4 count and hemoglobin categories [20,21] and weight categories [22] were chosen based on published precedent. Suitability of the ART regimen was assigned according to published international guidelines [7,23]. ART adherence during months 0-6 of ART was estimated by timeliness to drug pick-up appointments (i.e., every day late for a pharmacy drug pick-up appointment during months 0-6 of ART was equivalent to one missed day of ART doses) [24][25][26]. Only site size was included as a site-level variable in this analysis [5,12].  ing the corresponding author, members of the MOH, CDC, and the Elizabeth Glaser Pediatric AIDS Foundation. Missing data, reported for each baseline covariate of interest in Table 1, were assumed to be missing at random (MAR), and were imputed using multiple imputation with chained equations [27]. The ice [28][29][30] procedure in STATA was used to create 20 imputed datasets for each of the following outcomes: (1) documented death, (3) LTFU, and (3) overall attrition [13]. The imputation model included the event indicator, all study variables, and the Nelson-Aalen estimate of cumulative hazard [31]. For all analyses using imputed data, estimates were combined across the imputed datasets according to Rubin's rules [27] using the mim procedure in STATA [32].
To assess the association between baseline characteristics and year of ART initiation, linear, logistic, ordered, or multinomial logistic regression models, accounting for study design, were used for continuous, binary, ordinal, and nominal categorical variables, respectively. To assess the association between baseline characteristics and sex, unadjusted logistic regression, accounting for study design, was used. A competing risks model was used to estimate 6-, 12-, and 24month mortality and LTFU for each annual cohort of adults starting ART during 2004-2007 [33]. Stacked cumulative incidence curves were used to illustrate cumulative probability of death and LTFU over time for each annual cohort of adults starting ART [33].
In time-to-event analysis, Cox proportional hazards regression models that controlled for study design were used to estimate crude and adjusted hazard ratios (AHR) and 95% confidence intervals (CI) for covariates of interest [34]. The proportional hazards assumption was assessed using visual methods and the Grambsch and Therneau test [35]. Kaplan-Meier curves were used to examine cumulative probability of retention (1-attrition) over time stratified by baseline variables.

Trends in Patient Characteristics at ART Initiation
Data from medical records of 3,682 eligible, adult ART patients were abstracted and analyzed. Year of ART enrollment for adult ART patients included in the study was 2004,2005,2006, and 2007, for 6%, 22%, 36% and 36%, respectively. During 2004-2007, 67% of patients were female, median age was 36 years, most patients (95%) were HIV-1-infected, 2% were HIV-2-infected, and 3% were HIV-1&2 dually reactive. These variables did not change significantly over time.
Overall, 59% of patients reported employment, but the proportion reporting employment declined from 65% to 56% during 2004-2007 (p = 0.027). Most patients had WHO stage III (58%) or IV (22%) with no significant changes over time (Table 1). Median ART enrollment weight was 53 kg, but declined from 55 kg to 52 kg during 2004-2007 (p = 0.008). Similarly, the proportion with very low weight (,45 kg) at ART initiation was 20% overall, but increased from 17% to 22% during 2004-2007 (p = 0.014). During 2004-2007, the proportion with hemoglobin ,8 g/dL was 14% and this did not change over time. Median CD4 count overall was 135 cells/mL and did not change significantly over time (p = 0.363).
The proportion of patients prescribed CTX at ART initiation was 59% and this did not change significantly over time. The proportion achieving 95% adherence to pharmacy pick-up appointments decreased from 74% to 60% during 2004-2007 (p = 0.026). The proportion of patients prescribed sub-optimal ART regimens was 8% and did not change significantly over time
Among 182 HIV-2-infected or dually reactive adults, optimal first-line therapy of two NRTIs and a boosted PI were prescribed to 41% of patients. Sub-optimal regimens were prescribed to the remaining 56% of HIV-2-infected or dually reactive adults: two NRTIs with an NNRTI were prescribed to 25%, triple NRTIs to 11%, two NRTIs with an unboosted PI to 10%, and monotherapy to 6% (Table 2).

Predictors of Outcomes
A 10-year increase in age at ART initiation was associated with an 11% reduction in LTFU risk (AHR 0.89; 95% CI, 0.83-0.96) but not mortality risk ( Failure to prescribe CTX was associated with increased risk of LTFU (AHR 1.40; 95% CI, 1.12-1.75), but not documented mortality (Table 4).
In crude analysis, enrollment at smaller sites (,1,000 enrollees) was predictive of mortality, but this was not significant in multivariable analysis.
In crude analysis, year of ART enrollment was associated with both mortality and LTFU (Table 4). However, in adjusted analysis, year of enrollment was only associated with increasing LTFU rates and not mortality (Table 4).
Factors predictive of overall attrition in adjusted analysis included younger age, male sex (Figure 2A), very low weight ( Figure 2B), CD4 count ,50 cells/mL, failure to prescribe CTX at ART initiation, adherence to ART ,95% ( Figure 2C), and later calendar year of ART initiation (Figure 2 D, Table 5).

Discussion
This is the first nationally representative evaluation of Côte d'Ivoire's adult ART program, and the first to evaluate trends in program outcomes over time, and has several important findings.

Declining ART Retention over Time
The most concerning finding of our analysis is the decrease in 12-month retention from 86% for 2004 ART enrollees to 69% for    ,0.001 Abbreviations: LTFU, loss to follow-up; ART, antiretroviral therapy; HR, hazards ratio; CI, 95% confidence interval; AHR, adjusted hazards ratio CTX, co-trimoxazole.

Regimen
*A high proportion of data were missing for adherence (62%). Therefore, similar to other reports [46], we generated multivariate models with and without this variable, to assess effect on hazards ratios for other covariates, and noted no significant differences. doi:10.1371/journal.pone.0098183.t004   The declining 12-month retention estimates are due to nearly three-fold increases in rates of LTFU (from 9.2-28.1/100 PY), and documented mortality (from 1.5-3.9/100 PY). While year of ART initiation was associated with both mortality and LTFU in unadjusted analysis, in multivariable analysis, controlling for other known predictors of death and LTFU, ART initiation year was only associated with LTFU.
The likely explanation for this finding is that other measured mortality risk factors were confounding the crude association between ART initiation year and mortality. Both prevalence of very low weight at ART initiation and non-adherence to ART in months 0-6 of ART, increased among successive annual cohorts of ART enrollees, and were associated with mortality. In contrast, measured risk factors for LTFU in this study do not explain the association between year of ART initiation and LTFU. This analysis has important implications for the program response to declining retention.

Program Response to Increasing Mortality
Addressing increasing prevalence of sub-optimal nutritional status and declining ART adherence may help program managers to reverse trends of increasing mortality rates. Increasing prevalence of nutritional insufficiency may be related to increasing food insecurity [37], which may be related to increasing political instability since the late 1990s that culminated in the second Ivorian civil war in 2011 [9]. Alternately, expansion of the ART program to more rural areas, especially in the north of the country, where food insecurity is more common [37], might explain the worsening baseline nutritional status of ART enrollees. The increasing proportion of ART enrollees who report being unemployed (from 31% to 42% during 2004-2007) supports the theory that food insecurity might underpin increasing prevalence of sub-optimal nutritional status. In Côte d'Ivoire, where 23% of the population live on ,$1.23/day [38], further research to evaluate the health benefits of integrated nutrition programs in adult ART clinics might be warranted [39][40][41].
Addressing food insecurity, for example through clinic-based food assistance [39][40][41], may also help to address the problem of declining ART adherence [39]. Other interventions to improve adherence might include targeting patients who display suboptimal pharmacy-based measures of adherence during months 0-6 of ART, with a package of adherence interventions including viral load testing [42,43]; this targeted approach might improve adherence [43], reduce mortality [44], and reduce LTFU risk [11]. Program Response to Increasing LTFU Identifying the causes of increasing LTFU rates in future studies is important to allow identification of possible interventions. Increasing rates of LTFU have been documented in other countries with rapidly expanding ART programs, including South Africa [45,46] and Mozambique [13]. Two factors may be contributing to increasing LTFU: firstly, with increasing patient load, attention to timely, accurate maintenance of medical records may be compromised, resulting in missing entries for clinic visits or undocumented transfers [46,47]. Developing and implementing effective electronic monitoring systems, with dedicated data management personnel, could improve data quality and accuracy [47].
Secondly, with increasing patient-to-provider ratios, patient waiting times are increased, and waiting rooms become more crowded [45,[48][49][50]. This may be associated with patient and clinician dissatisfaction with clinic conditions, which may be one cause for increasing LTFU [51][52][53][54][55]. Reducing patient-to-provider ratios might be facilitated by several interventions including increasing the workforce, task shifting [56], or decreasing visit frequency for stable patients [57]. One method to decrease visit frequency for stable patients is formation of community adherence support groups (CASG). CASGs comprise groups of 6-10 patients, who take turns to collect the group's ART medications from clinic pharmacies each month. In Mozambique a pilot project significantly reduced 12-month LTFU [58]. Alternately, distribution of ART at locations closer to patient's homes might reduce patient and clinic burden and might improve retention [59].

Co-trimoxazole for ART Enrollees
In our study, failure to prescribe CTX to ART enrollees was associated with increased LTFU and overall attrition. It is unclear whether CTX reduced morbidity, which contributed to reductions in LTFU [60,61], or whether clinician compliance with CTX prescription was a marker of higher quality clinical care. Regardless, there is considerable evidence [13,60,61] supporting the need to prescribe CTX to all ART enrollees.

Male Gender
As has been documented in other African cohorts [5,13,20,22,62], males had a lower baseline median CD4 count than females, a higher risk of LTFU, and marginally increased mortality. Delayed presentation for care might be due to gender norms, which discourage men from admitting ill-health, while higher rates of LTFU might reflect differences in adherence to chronic care [13,62,63]. However, higher background mortality among men in general, regardless of HIV status, might explain gender differences in mortality during ART follow-up [62]. In Côte d'Ivoire's general population, mortality is higher among males than females (472 deaths/1,000 men vs. 385 deaths/1,000 women) [64]. Increased male mortality is attributed to accidents, homicide, suicide [65], and increased opportunistic infections [62,65,66]. In our cohort, higher male LTFU may also be due to underlying increased mortality [22], a proportion of which goes undocumented [67]. However, further research is needed to inform intervention strategies.

Younger Age
In our study, as in others [13,46], younger age was predictive of LTFU risk. Point estimates of LTFU rates were higher in adolescents (aged 15-,20 years at ART initiation) at 19.5/100 PY and young adults (aged 20-,25) at 24.4/100 PY, compared with adults aged 25-,75 at ART initiation (range: 0-18.7/100 PY). This may be because younger people are more mobile. In west Africa, migration for work is particularly common among adults in their twenties and thirties, especially among men [68]. Increased risk for LTFU among adolescents has been documented in other studies [69]. Possible cognitive impairment among perinatally infected children who start ART late as adolescents, lack of youth-friendly services, rigid scheduling, increasing responsibilities, and decreasing involvement of adult caregivers all contribute to the challenge of retaining adolescents and young adults on ART [69]. Youth-specific retention interventions may be needed to keep young adults on ART in Côte d'Ivoire.

HIV-2 and Dual HIV-1&2 Reactivity
As in other countries [8], HIV-2 and dual reactivity were poorly managed, with 56% of affected patients prescribed sub-optimal first-line regimens. Similar to Burkina Faso [8], 25% of HIV-2infected or dually reactive patients were prescribed NNRTIcontaining regimens, to which HIV-2 is resistant [70][71][72]. A further 11% of patients were prescribed two NRTIs with an unboosted PI, which has been associated with poor treatment outcomes in Côte d'Ivoire [73] and Senegal [7,74]. Triple NRTI therapy is also not recommended [7], due to poor outcomes [75,76], and risk of Q151M pan-NRTI resistance [77]. Similarly, mono-and dual-therapy are associated with resistance and poor outcomes [7]. Possible reasons for poor HIV-2 management include insufficient training of clinicians, and low availability of ritonavir-boosted regimens, either due to stock outs [78], or lack of a cold chain prior to availability of heat-stable lopinavir-ritonavir [7]. Clinician training and drug supply and demand issues are being addressed [78].

Limitations
Firstly, these analyses rely on routinely collected and sometimes incomplete data. Missing data on baseline patient characteristics likely introduced non-differential measurement error. Given the proportion of data missing for the adherence variable, prevalence of non-adherence and hazard ratios associated with non-adherence should be viewed with caution, although findings are in line with other publications from Côte d'Ivoire [11,39]. Secondly, our reported LTFU rate is likely an over-estimate due to the probability of undocumented death [67] or undocumented transfer [47] being observed as LTFU. Similarly, our reported mortality rate is likely an underestimate of true mortality [67]. Finally, these data show trends for patients enrolled during 2004-2007 and trends may have changed in more recent years.

Conclusions
Increased prevalence of sub-optimal nutritional status and suboptimal ART adherence, might explain increases in documented mortality over time. Earlier ART initiation before nutritional compromise and targeted adherence interventions might help reverse trends of increasing mortality. Further research to assess the survival benefit of food supplementation for food-insecure ART enrollees, might be warranted. Increasing LTFU rates are not explained by risk factors analyzed in this report. Undocumented transfers, political instability, or patient dissatisfaction with crowded facilities might explain increasing LTFU. Implementing electronic monitoring systems to improve data quality, and innovative LTFU-prevention strategies, possibly targeting men and younger patients, might reverse trends of increasing LTFU.