To the Editors:
South Africa has the largest antiretroviral therapy (ART) program worldwide with 970,000 people receiving ART in 2009.1 As patient numbers have expanded, rising levels of loss to follow-up (LTFU) have significantly reduced the effectiveness of the national ART program, and increasing virological failure has become apparent.2,3 A causal association between increased facility patient caseload and worsening ART outcomes in South Africa is, however, not established.3
A recent study from Mozambique indicated increased attrition from clinics with high pharmacy staff burden.4 However, there is little other sub-Saharan data comparing ART outcomes between facilities with varying patient caseloads. This knowledge is valuable, as the most efficient approaches to further upscale treatment access while maintaining good ART outcomes are critical to the success of ART programs. To determine the effect of increasing facility patient burden on ART program effectiveness, this study compared program outcomes among 54 government primary health care ART facilities in 4 South African provinces.
A multicohort study of adults enrolled at ART facilities supported by a nongovernmental organization (NGO) was conducted. The NGO provides clinical staff, infrastructure, clinical mentoring, community-based adherence support, and electronic data collection systems. Facilities are located in 4 provinces (Western Cape, KwaZulu-Natal, Eastern Cape, and Mpumalanga) in urban and rural areas.
All ART-naive adults (16 years and older) enrolled on triple ART between January 2004 and September 2009, with documented date of birth, sex, date of starting ART, at least 1 day of follow-up, and who were enrolled at least 6 months before site database closure were included in the analyses. Patients were followed up until March 2010 or until the NGO exited from a site. Routine patient data were collected prospectively by facility-based data capturers at each clinic visit. Continual data cleaning and quality control routines were implemented to enhance data validity.
Outcomes measures were death, LTFU, and virological suppression after starting ART. Attrition was defined as patient losses due to LTFU or death. LTFU was defined as no patient visit for 3 months or more beyond the last missed appointment. Viral load was measured 6 monthly on treatment, and virological suppression was defined as a viral load <400 copies per milliliter.
The cumulative total of ART-naive adults enrolled at a facility at the end of the enrollment period was the measure of clinic patient load used. To guide the choice of how to categorize the sample according to facility load to optimize data fit, the Akaike information criterion from Cox models of facility load associated with outcomes was employed. A single cutoff of 950 patients was thus chosen, with facilities enrolling more than 950 patients classified as high caseload clinics and facilities enrolling fewer as low caseload clinics. Subgroup analyses were performed within high caseload clinic patients by comparing outcomes of patients enrolled prior to 6 months before enrollment of the 950th patient per site (the “early period” when the clinics had a low caseload) with patients enrolled thereafter (later period).
Sensitivity analyses were conducted by modelling patient load alternatively as a 3-level ordinal variable, split at the tertiles of the sample distribution of cumulative facility enrollment (low tertile < 1067; mid tertile 1067-2031; high tertile > 2031 patients/facility, respectively).
Kaplan-Meier estimates of outcomes were calculated. Time till initial virological suppression analyses were limited to patients having 1 or more viral load measurement within the first 12-month measurement period after starting ART. Multivariable Cox regression was used to estimate associations of patient and site-level factors with attrition and LTFU adjusting for baseline patient variables (age, sex, World Health Organization clinical stage, CD4 cell count, year of starting ART, tuberculosis treatment, pregnancy, and initial regimen) and site-related variables (province and rural/urban nature of site). Missing baseline values were classified as separate categories within variables,5 to retain observations in multivariable models. All adjusted models included all available baseline variables. All confidence intervals (CIs) quoted are 95% CIs. Statistical analyses were performed using Stata 11.1 (Stata Corporation, College Station, Texas). The study was approved by the University of Cape Town Research Ethics Committee.
Overall, 40,861 adults were included, of whom 11,830 (29.0%) were treated at 38 low caseload clinics and 29,031 (71.1%) were treated at 16 high caseload clinics. The median number enrolled per site was 488 [interquartile range (IQR): 291-736) and 2031 (IQR: 1275-3599) at low and high caseload clinics, respectively. The median age was 34.4 years (IQR: 29.3-41.1 years), being equivalent between groups (P = 0.26). Patients at high caseload facilities had a lower proportion of males (30.8% vs. 33.3%; P < 0.0005) and a lower median baseline CD4 cell count [117 cells/μL (IQR: 61-167) vs. 130 cells/μL (IQR: 71-177)]. Availability of baseline CD4 cell count results was 82.5%, with no difference between groups (P = 0.73). A higher proportion of patients were enrolled in earlier calendar years at high caseload facilities (44.5% enrolled pre-2008 vs. 23.7% at low caseload facilities, P < 0.0005).
The duration of follow-up was 52,602 person-years. Cumulative LTFU was reduced at low caseload clinics, being 6.4% (CI: 5.9% to 7.0%) and 11.2% (CI: 10.4% to 12.2%) after 12 and 24 months of ART, respectively, vs. 8.6% (CI: 8.3% to 9.0%) and 14.9% (CI: 14.3% to 15.4%) at high caseload facilities, P < 0.0005 (Fig. 1). Mortality was equivalent between the groups, being 5.1% (CI: 4.9% to 5.4%) and 6.8% (CI: 6.5% to 7.1%), after 12 and 24 months overall, respectively (P = 0.40). After 36 months of ART, retention in care (1-attrition) remained superior at low caseload clinics, being 78.8% (CI: 76.7% to 80.8%) vs. 73.9% (CI: 73.1% to 74.7%), P < 0.0005.
In adjusted multivariable analyses, high caseload clinics had independently increased probabilities of attrition and LTFU, adjusted hazard ratio (AHR) 1.26 (CI: 1.17 to 1.35; n = 40,861) and AHR 1.67 (CI: 1.52 to 1.83), respectively. Patients enrolled in later calendar years had independently increased risks of LTFU [AHR 2.25 (CI: 1.95 to 2.59), 2008-2009 vs. 2004-2005].
The subgroup of patients enrolled at high caseload clinics during the early period (during low caseload) totalled 9200 (31.7%) patients. This subgroup had independently better outcomes compared with patients enrolled during the later (high caseload) period; AHR of attrition and LTFU being 0.81 (CI: 0.75 to 0.88) and 0.74 (CI: 0.67 to 0.82), respectively. Mortality was equivalent, AHR 0.94 (CI: 0.82 to 1.09).
In sensitivity analyses using the alternate exposure categorization, patients in the mid and high tertiles of facility caseload had independently increased attrition, AHR 1.20 (CI: 1.12 to 1.29) and AHR 1.37 (CI: 1.23 to 1.52), respectively. LTFU was increased, AHR 1.42 (CI: 1.29 to 1.55) and AHR 2.03 (CI: 1.77 to 2.33) in the mid and high tertiles, respectively. In both cases, a dose-response relationship was evident, with hazards of attrition and LTFU in the high tertile compared with the mid tertile being AHR 1.17 (CI: 1.05 to 1.31) and AHR 1.48 (CI: 1.29 to 1.70), respectively.
Viral load result availability after 6 months of ART was better at low caseload clinics, being 61.1% vs. 58.4% (P < 0.0005). The proportion of patients achieving virological suppression at all time points on treatment was 87.2% (CI: 86.8% to 87.5%; n = 42,340), with no difference between groups (P = 0.30). The probability of achieving initial virologic suppression within 12 months of starting ART was, however, increased at low caseload clinics, being 88.3% (CI: 87.5% to 89.2%) vs. 85.0% (CI: 84.4% to 85.5%); P<0.0005, n = 22,251. [Low tertile 89.0% (CI: 88.2% to 89.7%), mid tertile 85.1% (CI: 84.3% to 86.0%), and high tertile 83.4% (CI: 82.5% to 84.3%); log rank trend P < 0.0005].
Patients at facilities with larger patient loads had increased attrition due to LTFU and reduced initial virologic suppression in this multicohort study. Attrition increased, and initial virological suppression decreased in dose-dependent manners as facility patient loads expanded. LTFU increased in more recent years, similar to other South African cohorts.2,3,6 Low caseload clinics, however, had a decreased probability of LTFU despite having enrolled the majority of their patients in more recent years.
Low ART health care provider to patient ratios have been associated with a reduced probability of starting ART after enrollment at HIV care facilities,7 and increased attrition with low pharmacy staff to patient ratios.4 However, no association between nonpharmacy health care provider to patient ratios and ART program attrition was evident in Uganda and Mozambique.4,8 Facility-level factors are nevertheless important determinants of ART outcomes.7 Large patient loads at facilities may overwhelm clinic administration systems and infrastructure, be associated with longer waiting times, and provide a higher caseload for community adherence support counsellors. Dissatisfaction with waiting times is an important predictor of discontent among ART patients.9
Strengths of this study include that a large number of patients from a broad range of sites and settings were analyzed, and individual-level data were collected prospectively, enabling adjustment of patient-level factors associated with outcomes. Limitations include that health care provider to patient ratios and facility infrastructure variables were not included in the analyses as data for all sites were unavailable. Attrition was raised at high caseload sites, although increased misclassification of patients as LTFU instead of deceased at high caseload sites might account for a degree of the raised LTFU observed. Facility patient burden may be assessed using alternate measures such as monthly visit count; cumulative patient enrollment is, however, a useful measure that is currently tracked in South African routine settings.
In conclusion, these results further establish an association between increasing facility patient load and poorer ART outcomes. Further research should be conducted to elucidate underlying mechanisms explaining this and to identify potentially remedial interventions.
The authors acknowledge Emmanuel Okoli, Kheth'Impilo colleagues, PEPFAR, Absolute Return for Kids, and the SA Health Department.
Geoffrey Fatti, MBChB, MPH*
Ashraf Grimwood, MBChB, MPH*
Eula Mothibi, MBChB, FCP*
Jawaya Shea, MHPE†
*Kheth'Impilo, Cape Town, South Africa †Child Health Unit, School of Child & Adolescent Health, University of Cape Town, Cape Town, South Africa