Conceived and designed the experiments: EL IVB RPW KAF. Performed the experiments: EL IVB SC. Analyzed the data: EL SR. Contributed reagents/materials/analysis tools: IVB SC SR CAS LMU. Wrote the paper: EL. Supervised study enrollment and follow-up at McCord Hospital: JG HH. Coordinated study enrollment and follow-up at both South African sites: SC. Managed the database in Boston: SR. Contributed to the design of the study: RPW JK. Supervised study enrollment and follow-up at St. Mary's Hospital: DR. Contributed to writing the paper: LMU.
The authors have declared that no competing interests exist.
Although loss to follow-up after antiretroviral therapy (ART) initiation is increasingly recognized, little is known about pre-treatment losses to care (PTLC) after an initial positive HIV test. Our objective was to determine PTLC in newly identified HIV-infected individuals in South Africa.
We assembled the South African Test, Identify and Link (STIAL) Cohort of persons presenting for HIV testing at two sites offering HIV and CD4 count testing and HIV care in Durban, South Africa. We defined PTLC as failure to have a CD4 count within 8 weeks of HIV diagnosis. We performed multivariate analysis to identify factors associated with PTLC. From November 2006 to May 2007, of 712 persons who underwent HIV testing and received their test result, 454 (64%) were HIV-positive. Of those, 206 (45%) had PTLC. Infected patients were significantly more likely to have PTLC if they lived ≥10 kilometers from the testing center (RR = 1.37; 95% CI: 1.11–1.71), had a history of tuberculosis treatment (RR = 1.26; 95% CI: 1.00–1.58), or were referred for testing by a health care provider rather than self-referred (RR = 1.61; 95% CI: 1.22–2.13). Patients with one, two or three of these risks for PTLC were 1.88, 2.50 and 3.84 times more likely to have PTLC compared to those with no risk factors.
Nearly half of HIV-infected persons at two high prevalence sites in Durban, South Africa, failed to have CD4 counts following HIV diagnosis. These high rates of pre-treatment loss to care highlight the urgent need to improve rates of linkage to HIV care after an initial positive HIV test.
While the AIDS epidemic continues to devastate sub-Saharan Africa, where it remains the leading cause of death
Maximal effectiveness of ART depends on timely HIV diagnosis, linkage to care, treatment initiation, and retention in care. The success of ART programs in South Africa depends in large part on the ability to identify HIV-infected individuals, to determine ART eligibility and monitor those not yet eligible for ART to facilitate timely transition to treatment, to initiate care among those eligible, and to ensure sustainable access to care over time. High rates of loss to follow-up (LTFU) following ART initiation have been reported from many sites
We obtained written, informed consent from all study participants. The protocol, questionnaires, and consent forms were approved by the McCord Hospital Ethics Committee, the Ethics Committee of St. Mary's Hospital Mariannhill, and the Partners HealthCare Human Research Committee in Boston, MA, USA.
We conducted a prospective cohort study in two sites in KwaZulu-Natal, the province with the highest HIV prevalence in South Africa
St. Mary's Hospital Mariannhill, located 20 kilometers west of Durban, is a Catholic, state-aided, 200-bed general district hospital serving a poorer population from a defined catchment area in the surrounding communities. St. Mary's Hospital serves primarily to the healthcare needs of the “poorest of the poor” living in rural and peri-urban areas of the Durban Metropolitan area. A higher state subsidy enables St. Mary's to charge lower fees for all services than McCord. The iThemba Family Care Centre, based at St. Mary's, began offering ART in February 2003, and is free of charge to patients. Since then, over 3,900 people have had access to ART there. In this report we refer to St. Mary's as the ‘rural hospital.’
Comprehensive care services at both hospitals include outpatient medical care, social work, psychological services, pastoral care, and training and income generation projects for support group members. Both hospitals have been awarded PEPFAR funding to expand the number of patients on ART. Other sources of funding include the KwaZulu-Natal Provincial Department of Health and subsidized patient fees for McCord Hospital and the Department of Health and hospital-raised funds for St. Mary's Hospital. Together these sites have more patients on treatment and more years of experience in treating patients on ART than any other sites in the KwaZulu-Natal, South Africa public sector.
The urban hospital offered HIV testing, counseling, and referral, using rapid HIV tests. Patients eligible for the study were offered testing in the outpatient department. Patients generally received HIV test results within 30 minutes of testing. Those with a positive initial rapid HIV test underwent confirmatory testing with a rapid HIV test from a second manufacturer. Patients diagnosed with HIV were referred to the HIV clinic in an adjacent building, approximately 300 meters away. There, CD4 testing via venipuncture was offered to patients who registered and paid the clinic registration fee of 90 Rand ($12 US, 2007). Venipuncture specimens were sent to the National Health Laboratory Service or Global Laboratories (both located in Durban) for flow cytometry.
The rural hospital's protocol for HIV testing was based on a venipuncture for a standard ELISA (Centaur XP, Siemens Healthcare Diagnostics, Deerfield, IL, USA or Roche E170, Roche, Basel, Switzerland). Patients eligible for the study were offered HIV testing at the primary health care clinic. Patients were asked to return for HIV test results within two weeks of HIV testing. Positive results were confirmed with a Western Blot. CD4 testing for HIV-infected patients was offered at the same site free of charge to the patients immediately after they received an HIV-positive test result. Venipuncture specimens were sent to the National Health Laboratory Service or Lancet Laboratories (both located in Durban) for flow cytometry, and patients had to return for the results after one month.
All English or Zulu speaking adults (≥18 years old) who presented for HIV testing at each of the study hospitals between November 2006 and May 2007 were eligible for enrollment in the South African Test, Identify and Link (STIAL) Cohort, provided they were able to give informed consent, were not pregnant, did not present to care in a stretcher or wheelchair, were not already known to be HIV-infected, and were willing to share HIV test results with research staff. Those enrolled were asked to fill out a voluntary baseline questionnaire, offered in English or Zulu by trained study personnel. For study participants who were newly identified as HIV-infected at the study visit, medical record review was done within 12 weeks of study enrollment to document receipt of HIV test results, presence of follow-up for CD4 count, and results of the CD4 count test.
Patient level data were collected in the following four domains:
The primary outcome of the study—pre-treatment loss to care (PTLC)—was defined by the authors as failure to undergo a CD4 count within 8 weeks of receiving an HIV-positive test result. Thus, PTLC was calculated as the number of newly diagnosed HIV-infected study participants who did not return for a CD4 test within 8 weeks of HIV diagnosis divided by the total number of newly diagnosed HIV-infected study participants.
Bi-variate and multivariate analyses were performed to determine the independent predictors of PTLC. Variables exhibiting association with outcomes at a bi-variate risk ratio (RR) of ≥1.5 or ≤0.60 or at a p-value of <0.10 were advanced into multivariate models to control for confounding and to estimate the independent impact of each factor on PTLC. Log-linear models were utilized for multivariate analysis to estimate risk ratios (RR), since PTLC was not a rare outcome and odds ratios would not accurately estimate risk ratios. We implemented a two-level model building strategy. First, we built models within each of 3 domains: clinical, demographic, and geographic. Then, we developed a series of hierarchical models (demographic, demographic+geographic, and demographic+geographic+clinical) to determine the most parsimonious overall model.
Because patient factors associated with a higher likelihood of PTLC may be inter-correlated, we combined them into a single measure which we termed the ‘index of vulnerability for pre-treatment loss to care.’ This measure was calculated as the sum of the following characteristics, which emerged as predictors of PTLC in multivariate analysis: ≥10 km distance between patient residence and site of care, a history of treatment for TB, and referral for HIV test by a medical care provider as opposed to self-referral. The ‘index of vulnerability’ values could range from 0 (no risk factors) to 3 (all 3 risk factors). We estimated a dose-response relationship between the index of vulnerability and PTLC by calculating a p-value for linear trend. We also estimated the risk ratio of PTLC for persons with various numbers of risk factors compared to persons who had none of the risk factors. All analyses were carried out using Stata statistical software (Stata Statistical Software Release 9, StataCorp, College Station, TX, USA).
Among 1,081 persons screened for the study between November 2006 and May 2007 at the two hospitals, 827 were eligible, and 767 (93%) of those eligible enrolled. The percent eligible among those screened did not differ by site (76% at the rural hospital vs. 77% at the urban hospital), but those eligible were more likely to enroll at the rural hospital than at the urban hospital (97% vs. 90%, p<0.001). The most common reasons for ineligibility were having had a previous positive HIV test, unwillingness to share test results with research staff, and age <18 years (
PTLC: Pre-treatment loss to care.
Factor | Urban | Rural | Overall | |||
N (%) |
HIV-pos. (%) |
N (%) |
HIV-pos. (%) |
N (%) |
HIV-pos. (%) |
|
382 (100) | 211 (55) | 330 (100) | 243 (74) | 712 (100) | 454 (64) | |
Age (years) | ||||||
Under 45 | 299 (78) | 176 (59) | 267 (82) | 205 (77) | 566 (80) | 381 (67) |
45 and over | 83 (22) | 35 (42) | 58 (18) | 34 (59) | 141 (20) | 69 (49) |
Sex |
||||||
Male | 193 (51) | 113 (59) | 195 (59) | 142 (73) | 388 (54) | 255 (66) |
Female | 189 (49) | 98 (52) | 135 (41) | 101 (75) | 324 (46) | 199 (61) |
Education |
||||||
Primary or none | 112 (30) | 61 (54) | 60 (18) | 43 (72) | 172 (25) | 104 (60) |
Some high school | 108 (29) | 70 (65) | 184 (56) | 139 (76) | 292 (42) | 209 (72) |
Matric |
154 (41) | 76 (49) | 83 (25) | 58 (70) | 237 (34) | 134 (57) |
Employment status | ||||||
No full-time job | 195 (52) | 104 (53) | 190 (58) | 141 (74) | 385 (55) | 245 (64) |
Full-time job | 183 (48) | 105 (57) | 137 (42) | 100 (73) | 320 (45) | 205 (64) |
Marital/living arrangement |
||||||
Unmarried/Lives alone | 42 (12) | 26 (62) | 51 (16) | 37 (73) | 93 (14) | 63 (68) |
Unmarried/Lives with others | 204 (60) | 131 (64) | 224 (69) | 170 (76) | 428 (64) | 301 (70) |
Factor | Urban | Rural | Overall | |||
N (%) |
HIV-pos. (%) |
N (%) |
HIV-pos. (%) |
N (%) |
HIV-pos. (%) |
|
Married/With children | 83 (24) | 29 (35) | 30 (9) | 18 (60) | 113 (17) | 47 (42) |
Married/No children | 12 (4) | 5 (42) | 19 (6) | 14 (74) | 31 (5) | 19 (61) |
Distance from home to facility |
||||||
<10 kilometers | 211 (55) | 111 (53) | 90 (28) | 54 (60) | 301 (43) | 165 (55) |
≥10 kilometers | 171 (45) | 100 (58) | 232 (72) | 185 (80) | 403 (57) | 285 (71) |
TB treatment history |
||||||
Yes | 34 (9) | 22 (65) | 63 (19) | 50 (79) | 97 (14) | 72 (74) |
No | 347 (91) | 189 (54) | 267 (81) | 193 (72) | 614 (86) | 382 (62) |
Reason for HIV testing |
||||||
Referred by health care provider | 322 (85) | 181 (56) | 139 (42) | 103 (74) | 461 (65) | 284 (62) |
Self- or other referral | 57 (15) | 28 (49) | 191 (58) | 140 (73) | 248 (35) | 168 (68) |
Self-rating of health |
||||||
Very good | 8 (2) | 3 (38) | 20 (6) | 8 (40) | 28 (4) | 11 (39) |
Good | 237 (64) | 115 (49) | 123 (38) | 80 (65) | 360 (52) | 195 (54) |
Fair | 117 (32) | 82 (70) | 124 (38) | 96 (77) | 241 (35) | 178 (74) |
Poor/bad | 7 (2) | 5 (71) | 60 (18) | 57 (95) | 67 (10) | 62 (93) |
% of total at hospital or overall.
% of total in category.
Equivalent to completing high school in the US school system.
*denotes p-values 0.01<p<0.05 in comparing characteristics across two sites.
**denotes p-values p<0.001 in comparing characteristics across two sites.
Of 712 enrolled patients who were tested and received conclusive results, 454 were HIV-infected, for an overall HIV prevalence of 64% (95% CI: 60%-67%). HIV prevalence was significantly higher among study subjects enrolled at the rural hospital, with 243 of 330 enrollees HIV-infected (74%; 95% CI: 69–78%) compared to subjects enrolled at the urban center, with 211 of 382 enrollees HIV-infected (55%; 95% CI: 50%–60%). HIV prevalence was similar between men and women, and no significant association was found between prevalence of HIV and level of education, employment status, or marital status. Among study participants reporting a history of being treated for TB (9% at the urban hospital vs. 19% at the rural hospital; 14% overall), the prevalence of HIV was 65% in the urban center and 79% in the rural center (74% overall). One hundred thirty-two (62%) of those receiving CD4 counts and their results had CD4 counts <200/µl (68% at the urban hospital and 60% at the rural hospital, p = 0.238;
Nearly half of the 454 newly diagnosed HIV-infected persons (206; 45%) had PTLC (
Factor | N (total) | % PTLC | Crude RR (95% CI) | Adjusted RR (95% CI) |
454 | 45.4 | — | — | |
Age (years) | ||||
45 and over | 69 | 36.2 | 1.00 | 1.00 |
Under 45 | 381 | 46.7 | 1.29 (0.93, 1.80) | 1.30 (0.95, 1.78) |
Sex | ||||
Female | 199 | 43.2 | 1.00 | 1.00 |
Male | 255 | 47.1 | 1.09 (0.89, 1.34) | 1.13 (0.93, 1.38) |
Education | ||||
Primary or none | 104 | 43.3 | 1.00 | |
Some high school | 209 | 45.9 | 1.06 (0.81, 1.38) | |
Matric |
134 | 45.5 | 1.05 (0.79, 1.40) | |
Employment status | ||||
No full-time job | 245 | 43.3 | 1.00 | |
Full-time job | 205 | 47.8 | 1.10 (0.90, 1.35) | |
Marital/living arrangement | ||||
Unmarried/Lives alone | 63 | 46.0 | 1.00 | |
Unmarried/Lives with others | 301 | 45.9 | 1.00 (0.74, 1.34) | |
Married/With children | 47 | 51.1 | 1.11 (0.75,1.63) | |
Married/No children | 19 | 36.8 | 0.80 (0.42, 1.53) |
Factor | N (total) | % PTLC | Crude RR (95% CI) | Adjusted RR (95% CI) |
Distance from home to facility | ||||
<10 kilometers | 165 | 40.6 | 1.00 | 1.00 |
≥10 kilometers | 285 | 48.8 | 1.20 (0.96, 1.50) | 1.37 (1.11, 1.71) |
Site of care | ||||
Rural hospital | 243 | 37.0 | 1.00 | 1.00 |
Urban hospital | 211 | 55.0 | 1.48 (1.21, 1.82) | 1.40 (1.12, 1.77) |
TB treatment history | ||||
No | 382 | 44.0 | 1.00 | 1.00 |
Yes | 72 | 52.8 | 1.20 (0.94, 1.54) | 1.26 (1.00, 1.58) |
Reason for testing | ||||
Self- or other referral | 168 | 31.0 | 1.00 | 1.00 |
Referred by health care provider | 284 | 54.2 | 1.75 (1.36, 2.25) | 1.61 (1.22, 2.13) |
Self-rating of health | ||||
Very good | 11 | 54.6 | 1.00 | |
Good | 195 | 47.2 | 0.86 (0.49, 1.51) | |
Fair | 178 | 43.3 | 0.79 (0.45, 1.40) | |
Poor/bad | 62 | 41.9 | 0.77 (0.42, 1.42) |
RR: risk ratio.
CI: confidence interval.
Equivalent to completing high school in the US school system.
The distribution of the index of vulnerability for PTLC was similar across the two hospitals. Five percent of all HIV-infected persons had all three risk factors (≥10 km from the center, history of TB treatment, and referral by a medical care provider), 36% had two risk factors, 48% had one risk factor, and only 10% had none of these risk factors. HIV-infected study participants with one, two and three of these risk factors for PTLC were 1.88, 2.50 and 3.84 times more likely to have PTLC compared to study participants with no risk factors (p-value for linear trend <0.0001). A greater number of risk factors led to a higher likelihood of PTLC in both the urban and the rural center, as well as the two centers combined (
In a population-based cohort presenting for HIV testing at two South African HIV outpatient sites, we found that nearly half of newly diagnosed HIV-infected persons had pre-treatment loss to care (PTLC) as defined by not following up with a CD4 count. Factors associated with higher rates of PTLC were: living ≥10 km from the health center, a history of TB treatment, and referral for HIV testing by a health care provider as opposed to self-referral. However, even among patients with none of these risk factors, a high proportion (21%) were still lost to care before initiating treatment. Among those who did return for a CD4 count, 62% had advanced HIV disease (CD4 count <200/µl).
We have found a small number of published studies that have examined loss to care prior to starting ART among patients newly diagnosed with HIV infection in resource-scarce settings
Several studies in Africa have tracked patients who failed to keep clinic appointments and ascertained their reasons for nonattendance. Detailed tracking has revealed high rates of mortality among those considered lost, with 27–59% mortality in patients whose status could be determined
Previous research has also shown that cost to the patient is a deterrent to continuing in HIV care
Our finding that patients self-referred for HIV testing have lower rates of PTLC compared to patients referred by health care providers suggest that patients referring themselves for an HIV test are likely more motivated to follow-up with subsequent care. Those referred by physicians may need time to come to terms with the diagnosis and are therefore less prepared to deal with the consequences of knowing their HIV-positive status in the short term. These findings suggest that physicians referring patients for HIV testing should consider complementing that referral with the social support needed to make it effective. In addition to possibly being less motivated, patients referred for HIV testing by a physician may be sicker than self-referred patients, and therefore more likely to have died or been hospitalized at another site. Similarly, patients with a history of TB treatment may have had greater PTLC because they were more immune-compromised than other patients, and may have died or been too sick to return to the care center.
This study has several limitations. We relied on several self-reported measures, including distance from site of care and history of TB treatment. While the findings of this study may apply to settings similar to the two sites in Durban, it may not be possible to generalize the findings to all of South Africa, or to other resource-limited settings. We also used data for patient CD4 follow up at the site of the HIV test for only 8 weeks following receipt of a positive HIV test result. If patients returned after more than 8 weeks, or had a CD4 count done at another site, those results were not reflected in this study.
Despite these limitations, this study also has several methodological strengths. The first lies in the unique nature of the cohort: enrollment was offered at the time of HIV testing, prior to knowledge of HIV status, and was therefore not biased by the HIV test results. We enrolled patients consecutively, minimizing selection bias. Data were collected by study staff not involved in the care of patients. Finally, the study personnel recruiting patients were bilingual in English and Zulu, ensuring clear communication between study participants and staff.
The major finding of this study is that nearly half of newly diagnosed HIV-infected patients at two sites in South Africa had pre-treatment loss to care; they failed to obtain a CD4 count within 8 weeks after initial HIV diagnosis. Since a CD4 count is the first step toward linkage to effective HIV care and toward determining eligibility for ART, these findings—and appropriate remedies—are critical to ensure the success of HIV treatment scale-up efforts in South Africa. Of those who did obtain a CD4 count and result, nearly two-thirds had advanced HIV disease (CD4<200/µl) and thus met guidelines for beginning antiretroviral therapy. This high proportion of patients presenting with advanced disease highlights the urgent need to identify HIV-infected patients earlier in the course of their illness. One recent study from South Africa reported 34% mortality in patients lost to follow up prior to ART initiation
Understanding reasons for pre-treatment loss to care is critical for the development of targeted interventions to increase the number of people who access and remain in HIV care. This significant rate of loss to care warrants the expansion of adherence and retention efforts focused on those starting ART to all persons from the time they receive an initial HIV-positive test result. Strategies to increase pre-treatment linkage to care might include providing reimbursement for transportation to the clinic, providing CD4 testing free of charge to the patient, and contacting patients with reminders about appointments and test results. The substantial loss of patients after their initial HIV-positive test result, and prior to initiating HIV care, highlights the urgent need for strategies to be developed to improve successful linkage to care and retention in South Africa.
The authors thank the study's research assistants: Nompumelelo Badumuti, Success Mabunda, Aletta Maphasa, Yolisa Mgobhozi, and Lindeni Sangweni. Presented in part at the XVII International AIDS Conference, Mexico City, August 3–8, 2008.