The authors have read the journal's policy and declare the following competing interests: MO does not have any competing interests that would influence this paper, but does work with the team developing the software used to collect the source data whose guidance and mentorship to the national government are for the benefit of all public health facilities using the software, not just those facilities that are contributing to this paper. This is a non-commercial entity. The other authors have declared that no competing interests exist.
‡ AB and GM are joint senior authors on this work.
Retention in care is an essential component of meeting the UNAIDS “90-90-90” HIV treatment targets. In Khayelitsha township (population ~500,000) in Cape Town, South Africa, more than 50,000 patients have received antiretroviral therapy (ART) since the inception of this public-sector program in 2001. Disengagement from care remains an important challenge. We sought to determine the incidence of and risk factors associated with disengagement from care during 2013–2014 and outcomes for those who disengaged.
We conducted a retrospective cohort study of all patients ≥10 years of age who visited 1 of the 13 Khayelitsha ART clinics from 2013–2014 regardless of the date they initiated ART. We described the cumulative incidence of first disengagement (>180 days not attending clinic) between 1 January 2013 and 31 December 2014 using competing risks methods, enabling us to estimate disengagement incidence up to 10 years after ART initiation. We also described risk factors for disengagement based on a Cox proportional hazards model, using multiple imputation for missing data. We ascertained outcomes (death, return to care, hospital admission, other hospital contact, alive but not in care, no information) after disengagement until 30 June 2015 using province-wide health databases and the National Death Registry. Of 39,884 patients meeting our eligibility criteria, the median time on ART to 31 December 2014 was 33.6 months (IQR 12.4–63.2). Of the total study cohort, 592 (1.5%) died in the study period, 1,231 (3.1%) formally transferred out, 987 (2.5%) were silent transfers and visited another Western Cape province clinic within 180 days, 9,005 (22.6%) disengaged, and 28,069 (70.4%) remained in care. Cumulative incidence of disengagement from care was estimated to be 25.1% by 2 years and 50.3% by 5 years on ART. Key factors associated with disengagement (age, male sex, pregnancy at ART start [HR 1.58, 95% CI 1.47–1.69], most recent CD4 count) and retention (ART club membership, baseline CD4) after adjustment were similar to those found in previous studies; however, notably, the higher hazard of disengagement soon after starting ART was no longer present after adjusting for these risk factors. Of the 9,005 who disengaged, the 2 most common initial outcomes were return to ART care after 180 days (33%;
Twenty-three percent of ART patients in the large cohort of Khayelitsha, one of the oldest public-sector ART programs in South Africa, disengaged from care at least once in a contemporary 2-year period. Fifty-eight percent of these patients either subsequently returned to care (some “silently”) or remained alive without hospitalization, suggesting that many who are considered “lost” actually return to care, and that misclassification of “lost” patients is likely common in similar urban populations.
A challenge to meeting ART retention targets is developing, testing, and implementing program designs to target mobile populations and retain them in lifelong care. This should be guided by risk factors for disengagement and improving interlinkage of routine information systems to better support patient care across complex care platforms.
In a retrospective study, Samantha Kaplan and colleagues document the continuity of antiretroviral treatment and care in Khayelitsha, South Africa.
As of 2015, over 11 million people in sub-Saharan Africa were receiving antiretroviral therapy (ART) for treatment of HIV/AIDS, indicating the scale of patients who need lifelong treatment.
An important component in controlling the HIV/AIDS epidemic is retaining patients in lifelong care. In addition to keeping individuals healthy, adherence to ART also helps curb the epidemic via viral suppression and prevention of onward transmission of the virus.
The Khayelitsha ART program in Cape Town, South Africa is one of the oldest and largest public-sector ART programs in South Africa, and retention in care has been an ongoing issue needing attention.
Previous studies have estimated rates of patients who are lost to follow-up (LTFU), but the program has grown substantially, and South African ART guidelines have changed repeatedly in recent years, supporting the need to reassess rates of disengagement in a larger and changing patient population.
We conducted a cohort study using clinic database extracts of 39,884 patients ≥10 years of age who visited one of the 13 Khayelitsha ART clinics during 2013–2014.
9,005 (22.6%) patients disengaged from care (>180 days not attending clinic) at least once during 2013–2014, and an additional 987 (2.5%) silently transferred to another clinic in the same province. Using a statistical model, we estimated a cumulative incidence of disengagement of 25.1% at 2 years of study time.
Factors associated with disengagement were age <30 years, male sex, and pregnancy at first ART visit. Factors associated with retention were baseline CD4 count <350 cells/μl and membership of an ART adherence club.
In an analysis of outcomes for those who disengaged using electronic tracing methods, about a third of patients returned to care after disengagement, and an additional quarter were alive but not in care. During short-term follow-up, 3% of all patients who disengaged died.
Rates of disengagement in this contemporary urban Khayelitsha cohort continue to be high; however, many patients return to care, sometimes at different clinics than the one in which they were originally initiated on ART. This suggests that many patients are potentially misclassified as lost to care, making program estimates for ART retention inaccurate, which may be more broadly applicable to similar urban sub-Saharan African ART cohorts.
Additionally, our data suggest that systems need to account for and track mobile populations to retain them in care and prevent morbidity, mortality, and spread of HIV infection.
With the 2015 World Health Organization (WHO) guidelines recommending treatment for all HIV-infected individuals regardless of CD4 status and the continued high HIV incidence rates in endemic areas, there are increasing numbers of patients eligible for and starting lifelong antiretroviral therapy (ART). In order for health systems to meet the UNAIDS 90-90-90 treatment targets of patients receiving sustained ART and maintaining viral suppression, retention in care is an essential focus [
In Khayelitsha township (population approximately 500,000) in Cape Town, South Africa, an HIV treatment program was established in 2001 as a partnership between Médecins sans Frontières and the provincial government at 3 public-sector primary care clinics. This represented the first initiative to provide ART in the South African public sector [
Over the 15 years since the Khayelitsha ART program’s inception, clinics have grown in size considerably [
This study was approved by the Yale University Human Investigation Committee (Protocol # 1504015732) and the University of Cape Town Human Research Ethics Committee (HREC REF: 568/2015) per the protocol “Sub-study of protocol ‘Enhanced routine surveillance of patients in HIV care in Khayelitsha’ (HREC 395/2005)” (
A cohort study was conducted using data from all provincial and municipal public-sector ART clinics (
The study included all patients who had at least 1 visit at a Khayelitsha ART clinic between 1 January 2013 and 31 December 2014 regardless of the date they initiated ART, provided it was prior to 31 December 2014 (
ART, antiretroviral therapy.
ART eligibility criteria, patient monitoring, and treatment regimens have progressively changed since program initiation in 2001 (
In all Khayelitsha clinics, doctors and nurses who see patients enter visit data onto structured paper clinical records, which are subsequently captured on site into an electronic patient information system by data capturers. The municipal and provincial sites use different electronic patient information systems, but data are formatted and then exported by clinics based on an internationally implemented data exchange standard for HIV treatment data [
Civil identification numbers, when available, were used to ascertain or confirm dates of death up to 30 June 2015. Death dates were also ascertained from clinic records, if available. Western Cape province unique health identifiers were linked with the province-wide laboratory, pharmacy, and health facility visit databases to determine outcomes for those who disengaged, if available, up to 30 June 2015, and to supplement laboratory data where these were missing until 31 December 2014.
Key study terms are defined below. Additional definitions can be found in
Data were analyzed using STATA/SE version 14.0 (StataCorp, College Station, TX, USA).
The cohort analysis was conducted in two parts: 1) analysis of time to disengagement from care in the cohort and analysis of risk factors for disengagement, using cumulative incidence curves and Cox proportional hazards models; and 2) for patients who disengaged from care, a description of outcomes after disengagement and times to these outcomes.
Entry into the study was at the beginning of 2013 for those who started ART prior to the analysis window, or at ART initiation if after 1 Jan 2013. Time to disengagement (failure) was defined as time from date of study entry to date of first disengagement within the window of 1 January 2013–31 December 2014 (
For patients who officially transferred out within the study window, if they returned to care in Khayelitsha ≤180 days later, we ignored the transfer outcome and followed them until their next outcome (disengagement, transfer, death, or alive and in care). If the gap was >180 days (or they returned to care outside of Khayelitsha at any point), we classified them as transfer out and censored them at the transfer date. Silent transfers were censored on their last visit date in Khayelitsha. They were then reclassified as not disengaged. If their gap in care was >180 days, the patient remained classified as disengaged.
A patient was censored if they a) died (censored on death date); b) transferred (censored on transfer date); or c) were alive without disengagement at study end (administrative censor, on 31 December 2014). If a patient appeared in the National Death Registry within 90 days of a presumed date of disengagement, then they were reclassified to a death and not a patient who had disengaged. If a patient was seen in the cohort window for only 1 visit, we added 1 day to their outcome date so that they were included in the survival analysis [
This analysis included only those who disengaged, and analyzed their subsequent outcomes after the first date of disengagement. For purposes of determining outcomes, the date of administrative censoring was extended to 30 June 2015. All outcomes after disengagement were ascertained from Western Cape province electronic data systems and the National Death Registry. Possible primary outcomes for those who disengaged were: 1) death; 2) return to care at a different facility within 180 days (silent transfer); 3) return to care after 180 days; 4) hospital admission; 5) other hospital contact (outpatient or emergency visit); 6) alive on 30 June 2015 if they had a national identification number but no death date was found in the National Death Registry; or 7) no information, still disengaged. Primary outcomes were identified as the first of these outcomes after the date of disengagement. Secondary outcomes were time to return to care and time to death after disengagement. Overall deaths and hospitalizations at any point post disengagement until 30 June 2015 were also calculated. We conducted a sensitivity analysis of disengagement outcomes for those with national identification numbers to restrict analyses to only those with reliable vital status ascertainment.
A total of 53,461 patients initiated ART at any Khayelitsha site since program inception through 31 December 2014. For the cohort study, we excluded 11,839 patients who did not have a visit in the time period between 1 January 2013 and 31 December 2014, and 1,607 who were less than 10 years old at 1 January 2013. An additional 131 were excluded due to incomplete data (
Variable | Patients who did not disengage | Patients who disengaged (>180 days) | Whole cohort | ||||||
---|---|---|---|---|---|---|---|---|---|
# of participants ( |
30,879 | # of patients with complete data | % complete data | 9,005 | # of patients with complete data | % complete data | 39,884 | # of patients with complete data | % complete data |
Age at 1 January 2013, years (median, IQR) | 35.2 (29.2–41.6) | 30,879 | 100.0% | 32.0 (26.6–39.0) | 9,005 | 100.0% | 34.4 (28.5–41.0) | 39,884 | 100.0% |
Male sex ( |
8,862 (28.7%) | 30,879 | 100.0% | 2,825 (31.4%) | 9,005 | 100.0% | 11,687 (29.3%) | 39,884 | 100.0% |
Months on ART at 31 December 2014 (median, IQR) | 38.3 (16.8–68.4) | 30,879 | 100.0% | 15.8 (4.6–41) | 9,005 | 100.0% | 33.6 (12.4–63.2) | 39,884 | 100.0% |
Median year of starting ART (median, IQR) | 2011 (2009–2013) | 30,879 | 100.0% | 2012 (2010–2013) | 9,005 | 100.0% | 2012 (2009–2013) | 39,884 | 100.0% |
Baseline CD4 count, cells/μL (median, IQR) | 185 (101–274) | 27,398 | 88.7% | 200 (114–299) | 7,815 | 86.8% | 188 (104–280) | 35,213 | 88.3% |
Baseline CD4 by category (cells/μL; |
>350: 2,703 (8.8%) | >350: 1,164 (12.9%) | 9,005 | 100.0% | >350: 3,867 (9.7%) | 39,884 | 100.0% | ||
200–350: 9,659 (31.3%) | 200–350: 2,770 (30.8%) | 200–350: 12,429 (31.2%) | |||||||
50–200: 11,835 (38.3%) | 50–200: 3,076 (34.2%) | 50–200: 14,911 (37.4%) | |||||||
<50: 3,201 (10.4%) | <50: 805 (8.9%) | <50: 4,006 (10.0%) | |||||||
missing: 3,481 (11.3%) | missing: 1,190 (13.2%) | missing: 4,671 (11.7%) | |||||||
Most recent CD4 count as of 31 Dec 2014, cells/ |
443 (284–616) | 26,559 | 86.0% | 325 (195–495) | 7,936 | 88.1% | 415 (259–593) | 34,495 | 86.5% |
Most recent viral load >1,000 on ART as of 31 Dec 2014 ( |
1,812 (10.5%) | 17,191 | 55.7% | 1,183 (23.4%) | 5,060 | 56.2% | 2,995 (13.5%) | 22,251 | 55.8% |
Achieved viral suppression on ART (<400) (n, %) | 22,669 (95.6%) | 23,734 | 76.9% | 4,513 (85.9%) | 5,252 | 58.3% | 27,212 (93.9%) | 28,986 | 72.7% |
Initiated ART during pregnancy (women only) ( |
2,554 (11.7%) | 21,911 | 99.5% | 1,231 (20.0%) | 6,150 | 99.5% | 3,785 (13.5%) | 28,061 | 99.5% |
TB treatment at ART initiation ( |
6,612 (21.5%) | 30,721 | 99.5% | 1,981 (22.1%) | 8,962 | 99.5% | 8,593 (21.7%) | 39,683 | 99.5% |
Ever transferred within Khayelitsha ( |
1,315 (4.3%) | 30,879 | 100.0% | 371 (4.1%) | 9,005 | 100.0% | 1,686 (4.2%) | 39,884 | 100.0% |
Transferred into ART care ( |
3,401 (11.0%) | 30,879 | 100.0% | 1042 (11.6%) | 9,005 | 100.0% | 4,443 (11.1%) | 39,884 | 100.0% |
ART club membership, ever ( |
6,000 (30.1%) | 19,957 | 64.6% | 409 (7.2%) | 5,703 | 63.3% | 6,409 (25.0%) | 25,660 | 64.3% |
Weight at ART baseline (median, IQR) |
64 (55.6–75) | 18,589 | 60.2% | 63 (55–74) | 5,197 | 57.7% | 64 (55.5–74.9) | 23,786 | 59.6% |
Baseline ART regimen drug 1 ( |
TDF 18,189 (58.9%) | 30,879 | 100.0% | TDF 6,268 (69.6%) | 9,005 | 100.0% | TDF 24,457 (61.3%) | 39,884 | 100.0% |
d4T 7,322 (23.7%) | d4T 1,424 (15.8%) | d4T 8,746 (21.9%) | |||||||
AZT 3,116 (10.1%) | AZT 661 (7.4%) | AZT 3,777 (9.5%) | |||||||
ABC 95 (0.3%) | ABC 24 (0.3%) | ABC 119 (0.3%) | |||||||
missing 2,157 (7.0%) | missing 628 (7.0%) | missing 2,785 (7.0%) | |||||||
Baseline ART regimen drug 3 ( |
EFV 22,815 (73.9%) | 30,879 | 100.0% | EFV 7,157 (79.5%)) | 9,005 | 100.0% | EFV 29,972 (75.2%) | 39,884 | 100.0% |
NVP 5,590 (18.1%) | NVP 1,108 (12.3%) | NVP 6,698 (16.8%) | |||||||
LPV/r 200 (0.7%) | LPV/r 56 (0.6%) | LPV/r 256 (0.6%) | |||||||
Other 15 (0.05%) | Other 5 (0.06%) | other 20 (0.05%) | |||||||
missing 2,259 (7.3%) | missing 679 (7.5%) | missing 2,938 (7.4%) | |||||||
Most recent ART regimen drug 1 as of 31 Dec 2014 ( |
TDF 24,646 (79.8%) | 30,879 | 100.0% | TDF 7,406 (82.2%) | 9,005 | 100.0% | TDF 32,052 (80.4%) | 39,884 | 100.0% |
AZT 4,258 (13.8%) | AZT 916 (10.2%) | AZT 5,174 (13.0%) | |||||||
d4T 1,386 (4.5%) | d4T 538 (6.0%) | d4T 1,924 (4.8%) | |||||||
ABC 74 (0.2%) | ABC 18 (0.2%) | ABC 92 (0.2%) | |||||||
missing 515 (1.7%) | missing 127 (1.4%) | missing 642 (1.6%) | |||||||
Most recent ART regimen drug 3 as of 31 Dec 2014 ( |
EFV 24,706 (80.0%) | 30,879 | 100.0% | EFV 7,542 (83.8%) | 9,005 | 100.0% | EFV 32,248 (80.9%) | 39,884 | 100.0% |
NVP 2,933 (9.5%) | NVP 667 (7.4%) | NVP 3,600 (9.0%) | |||||||
LPV/r 2,783 (9.0%) | LPV/r 687 (7.7%) | LPV/r 3,470 (8.7%) | |||||||
Other 116 (0.4%) | Other 10 (0.1%) | other 126 (0.3%) | |||||||
missing 341 (1.1%) | missing 99 (1.1%) | missing 440 (1.1%) | |||||||
Previous gap in care of >180 days ( |
3,677 (11.9%) | 30,879 | 100.0% | 1,737 (19.3%) | 9,005 | 100.0% | 5,414 (13.6%) | 39,884 | 100.0% |
*Variables/data available at provincial clinics only
ABC, abacavir; ART, antiretroviral therapy; AZT, zidovudine; d4T, stavudine; EFV, efavirenz; LPV/r, lopinavir/ritonavir; NVP, nevirapine; TB, tuberculosis; TDF, tenofovir.
From the perspective of the Khayelitsha clinic system, a total of 9,992 (25.1%) patients disengaged from care. However, after linkage of these patients to Western Cape data systems, we found that patients who disengaged (excluding those who silently transferred) numbered 9,005 (22.6%). From this point on in the manuscript, “patients who disengaged” refers to those who disengaged, excluding silent transfers.
As of 31 December 2014, of the total cohort, 592 (1.5%) died, 1,231 (3.1%) transferred out, 987 (2.5%) were silent transfers and visited another ART or primary care clinic in the same province (Western Cape) within 180 days of their last visit date, 9,005 (22.6%) disengaged, and 28,069 (70.4%) were in care. Of those in our study cohort, 1,179 (3.0%) patients disengaged prior to 2013 but returned to care prior to 1 January 2013. 4,156 (10.4%) disengaged prior to 2013 but returned within the study window.
Of the patients who disengaged, 5,463 (60.7%) had South African national identification numbers and could be linked to the National Death Registry. Total mortality for the entire cohort as of 30 June 2015 was 2.4% (
The cumulative incidence of disengagement from care, before any other event could occur, was estimated to be 25.1% at 2 years, analyzed by time in the study (
The strongest adjusted associations with disengagement were most recent CD4 count <350 cells/μl (CD4 200–350 hazard ratio (HR) 2.03; 95% CI 1.91–2.15; CD4 50–200 HR 3.07; 95% CI 2.84–3.31; CD4 <50 HR 3.34; 95% CI 2.92–3.83, all relative to CD4 > 350), use of d4T (stavudine) at last visit (HR 1.72; 95% CI 1.57–1.89), and pregnancy at ART start (HR 1.58; 95% CI 1.47–1.69) (
Univariate: IMPUTED ( |
Multivariable: IMPUTED |
|||
---|---|---|---|---|
Variable | HR | 95% CI | HR | 95% CI |
1.58 | 1.42–1.75 | 1.38 | 1.24–1.54 | |
1.50 | 1.42–1.57 | 1.46 | 1.38–1.54 | |
ref | ref | ref | ref | |
0.87 | 0.82–0.93 | 0.90 | 0.85–0.96 | |
0.90 | 0.82–1.00 | 0.91 | 0.82–1.01 | |
1.24 | 1.03–1.50 | 1.08 | 0.89–1.31 | |
ref | ref | ref | ref | |
1.73 | 1.62–1.84 | 1.58 | 1.47–1.69 | |
1.22 | 1.17–1.28 | 1.14 | 1.08–1.20 | |
1.17 | 1.13–1.21 | - | - | |
1.06 | 1.01–1.19 | strata | strata | |
1.10 | 1.04–1.18 | - | - | |
1.10 | 0.99–1.22 | - | - | |
1.13 | 1.07–1.20 | strata | strata | |
2.80 | 2.65–2.97 | strata | strata | |
1.06 | 1.02–1.11 | 1.06 | 1.01–1.12 | |
ref | ref | ref | ref | |
0.71 | 0.67–0.76 | 0.60 | 0.56–0.65 | |
0.77 | 0.71–0.82 | 0.46 | 0.43–0.50 | |
0.78 | 0.72–0.86 | 0.39 | 0.35–0.44 | |
ref | ref | ref | ref | |
1.91 | 1.81–2.02 | 2.03 | 1.91–2.15 | |
2.89 | 2.72–3.07 | 3.07 | 2.84–3.31 | |
3.14 | 2.79–3.52 | 3.34 | 2.92–3.83 | |
ref | ref | - | - | |
1.48 | 1.35–1.61 | - | - | |
1.68 | 1.56–1.80 | - | - | |
1.71 | 1.59–1.83 | - | - | |
ref | ref | - | - | |
1.98 | 1.76–2.23 | - | - | |
2.54 | 2.34–2.77 | - | - | |
2.87 | 2.60–3.18 | - | - | |
0.38 | 0.36–0.41 | 0.58 | 0.53–0.64 | |
0.23 | 0.21–0.25 | 0.29 | 0.26–0.32 | |
0.99 | 0.99–1.00 | - | - | |
ref | ref | - | - | |
0.95 | 0.86–1.06 | - | - | |
0.81 | 0.86–1.06 | - | - | |
0.96 | 0.87–1.06 | - | - | |
ref | ref | - | - | |
0.92 | 0.85–0.99 | - | - | |
1.03 | 0.80–1.33 | - | - | |
1.33 | 0.53–3.33 | - | - | |
ref | ref | ref | ref | |
1.81 | 1.66–1.98 | 1.72 | 1.57–1.89 | |
ref | ref | ref | ref | |
1.04 | 0.96–1.14 | 1.17 | 1.08–1.28 | |
1.03 | 0.95–1.12 | 0.65 | 0.60–0.71 | |
0.39 | 0.21–0.72 | 0.21 | 0.11–0.39 |
ABC, abacavir; ART, antiretroviral therapy; AZT, zidovudine; d4T, stavudine; EFV, efavirenz; HR, hazard ratio; LPV/r, lopinavir/ritonavir; NVP, nevirapine; ref, reference; TB, tuberculosis; TDF, tenofovir; VL, viral load; yrs, years
*All variables with estimates listed were included in the multivariable model.
Other associations included younger age group (<30 years) and male sex. The factors associated with retention were ART adherence club membership (HR 0.29; 95% CI 0.26–0.32), a suppressed HIV viral load at any point during ART (HR 0.58; 95% CI 0.53–0.64), and a baseline CD4 count <350 cells/μl (CD4 200–350 HR 0.60; 95% CI 0.56–0.65; CD4 50–200 HR 0.46; 95% CI 0.43–0.50; CD4<50 HR 0.39; 95% CI 0.35–0.44, all relative to CD4 > 350) (
Of those who disengaged (
Patients who disengaged + silent transfers ( |
All patients who disengaged ( |
Patients who disengaged with identification numbers, allowing accurate mortality ascertainment ( |
|
---|---|---|---|
2,976 (30%) | 2,976 (33%) | 1,877 (34%) | |
2,255 (23%) | 2,255 (25%) | 2,248 (41%) | |
1,944 (19%) | 1,944 (22%) | - | |
1,218 (12%) | 1,218 (13%) | 896 (17%) | |
987 (10%) | - | - | |
545 (5%) | 545 (6%) | 384 (7%) | |
67 (1%) | 67 (1%) | 58 (1%) |
*Alive as of 30 June 2015 refers to patients who had valid national identification numbers but were not found in care anywhere in the Western Cape nor were they found to be dead. Therefore, “alive” is the only outcome we could ascertain.
A Kaplan-Meier analysis of time to returning to care for patients who disengaged estimated that approximately 50% of patients who disengage will return to care by 2.5 years (
DOH, Department of Health.
In this study, we examined disengagement from ART care during 2013–2014 among patients of the large, peri-urban cohort in Khayelitsha—one of the oldest public-sector ART cohorts in South Africa. Roughly 1 in 5 patients disengaged from care, demonstrating a high rate of disengagement and a key challenge to reaching the UNAIDS 90-90-90 treatment targets. Factors associated with disengagement were age <30 years, male sex, pregnancy at ART initiation, and last CD4 count <350 cells/μl. Factors associated with retention were ART adherence club membership and baseline CD4 <350 cells/μl. However, despite the high incidence of disengagement, many of those who disengaged did not do so permanently. While 48% (
Previous studies of the Khayelitsha cohort have reported on disengagement (termed “lost to follow-up, LTFU”, in those studies) and mortality up until 2009 [
Our assessment of risk factors associated with disengagement was limited to those in the Khayelitsha database. Male sex and pregnancy at ART initiation were strongly associated with disengagement, consistent with previous studies [
Our results indicate that a baseline CD4 count <350 cells/μl was associated with retention, and that a most recent CD4 count <350 cells/μl was predictive of disengagement. While these findings could suggest that those who are less sick early in treatment are more likely to disengage in line with previous studies [
We also found an association with retention for adherence club membership in Khayelitsha—which is confounded by indication—as more stable, engaged patients (on ART for >12 months, virally suppressed) are referred to adherence clubs. Twenty-six percent of patients in the entire cohort were in clubs by June 2014, and recent data have shown that adherence club membership is associated with a 67% reduction in LTFU [
We found that the higher hazard of disengagement soon after starting ART seemed to be accounted for by other risk factors in that this was no longer present in adjusted analyses (
The combination of high proportions of patients being silent transfers or returning to care after disengagement in this setting casts the challenge as cycling in and out of care and between facilities, rather than as definitive losses to care. This represents a major shift in the way ART care is now being delivered. Many of these patients spent substantial time out of care and were likely viremic. This cyclical engagement undermines the ultimate goals of the UNAIDS 90-90-90 agenda because during these interruptions in care, HIV may progress in individuals, and transmission of HIV to others is more likely.
In the current analysis, the 10% of disengagements that were in fact silent transfers underscore the relative importance of undocumented transfer being a reason for apparent disengagement. While the estimate may sound lower than previous findings of this figure, which is closer to 20% in the Western Cape [
In terms of return to care, one study in the Cape Town area found that 33% of those who disengaged returned to care with the probability of resuming within 3 years of 42%, a median of 228 days after disengagement [
However, half of patients who disengaged in this study did not return to care in the Western Cape province. Over 15% of the patients in this study were admitted to a hospital after disengagement, and 3% died. These adverse events could illustrate the potential clinical consequences of disengagement with associated healthcare costs. Other studies have illustrated such problems that arise when patients disengage from ART care: one study indicated that HIV contributed to over 60% of medical admissions to a South African district hospital in Cape Town from 2012–2013, and 19.3% of these HIV-infected patients had interrupted ART therapy [
The Khayelitsha ART program is one of the largest and oldest public-sector ART programs in South Africa, lending credibility to our findings and conclusions. Additionally, we focused on a contemporary cohort, which allows us to draw conclusions that are applicable to current ART programs across the country, as well as in the region. The ability to track patients using a unique identifier across different health services and laboratory data, and link patients to the National Death Registry for mortality ascertainment [
Potential interventions to improve patient tracking and retention, as well as provide a greater degree of flexibility in the system may include: rational use of patient held cards/records (which exist in Khayelitsha but are not always utilized), nationally accessible electronic health records, increasing the period for ART prescription to allow patient travel, editing clinical management protocols to include sections on transferring and receiving patients, improving the ability of clinics to communicate with each other and, perhaps most importantly, accepting that patients cycle through care and promoting healthcare worker understanding of the need to adapt care around patients’ lifestyles and mobility through health worker training [
We likely underestimate true silent transfer and return to care, as we only tracked patients in the Western Cape province. Because of the mobility of this population, it is likely that additional patients returned to care in other provinces, particularly the Eastern Cape. If person-level data were cascaded up to the national level in line with recommendations [
We had a short period for follow-up. We were also unable to ascertain causes of death. As we included only patient follow-up from 2013–2014 by design in order to maximize the generalizability of our findings to current programs, this limited our ability to more fully describe temporal trends in first disengagement. Additionally, we recognize the nonuniform collection of data, which necessitated dropping particular variables from our analyses and performing multiple imputation for missing data. It is also possible that some clinic and/or lab visits were not captured in the database. This is somewhat mitigated by the large size of the dataset and power of the analysis.
Finally, the legacy of South African apartheid has contributed to specific mobility issues and social problems in communities such as Khayelitsha that may not be generalizable to other urban settings in sub-Saharan Africa.
The Khayelitsha ART program is one of the oldest in South Africa and has grown greatly in size. The merging of all provincial and municipal clinic data in this study is unique and lends completeness to the cohort data as representative of an entire community. The ability to electronically trace patients throughout the Western Cape province and link to the National Death Registry allowed us to correct misclassified patient outcomes. For these reasons, the findings from this study are of value in directing appropriate patient and clinic-based interventions to improve long-term retention and meet the 90-90-90 HIV treatment targets in high prevalence urban settings such as this one throughout sub-Saharan Africa.
Our finding that many patients who disengaged return to care suggests that many patients in ART programs in Africa, particularly in urban settings, cycle in and out of care. This suggests a shift in the provision of ART care: that the linear UNAIDS model of HIV diagnosis, ART initiation, and viral suppression may need to be reconceptualized to account for this cycling in and out of care and the mobile populations served by ART programs in many sub-Saharan African countries. As “Universal Test and Treat” is implemented in both South Africa and the rest of sub-Saharan Africa, more patients will enroll in ART care with higher CD4 counts and need to be retained. It will be important to find ways to adapt services to accommodate mobile populations to retain patients in care and prevent morbidity, mortality, and HIV transmission. Even if these systems are slow to improve, clinics across sub-Saharan Africa that are familiar with and accommodating of the mobility of their patients will be able to better care and advocate for them.
(DOC)
(DOCX)
(DOCX)
Bracketed dates indicate origin dates (first antiretroviral therapy [ART] visit); grey rectangles indicate entry date into the cohort.
(PNG)
(PNG)
(PDF)
*For those who disengaged returning to care, the 180-day lag before patients return to care is a function of our definition that those who returned to care <180 days later were designated silent transfers and were censored on their date of return to care elsewhere in the Western Cape province.
*For death post-disengagement, the 3-month lag before patients die is a function of our definition to reclassify those who died ≤90 days after disengagement as deaths and were no longer classified as those who disengaged.
(TIF)
AZT, zidovudine; ddI, didanosine; d4T, stavudine; DVR/r, dolutegravir/ritonavir; EPTB, extrapulmonary tuberculosis; EFV, efavirenz; FTC, emtricitabine; HBV, hepatitis B; m, months; LPV/r, lopinavir/ritonavir; MDR/XDR TB, multi-drug resistant / extensively drug resistant tuberculosis; NVP, nevirapine; RAL, raltegravir; TB, tuberculosis; TDF, tenofovir; 3TC, lamivudine; VL, viral load; WHO, World Health Organization.
(DOCX)
(DOCX)
* Variables/data available at provincial clinics only.
ABC, abacavir; AZT, zidovudine; d4T, stavudine; EFV, efavirenz; LPV/r, lopinavir/ritonavir; NVP, nevirapine; TB, tuberculosis; TDF, tenofovir.
(DOCX)
**This model contained imputed data but was restricted to those patients with national identification numbers to allow for accurate mortality ascertainment.
ABC, abacavir; ART, antiretroviral therapy; AZT, zidovudine; CI, confidence interval; d4T, stavudine EFV, efavirenz; LPV/r, lopinavir/ritonavir; NVP, nevirapine; ref, reference; TB, tuberculosis; TDF, tenofovir; yrs, years.
(DOCX)
*Different from initial outcomes for those who disengaged as presented in
(DOCX)
* Variables/data available at provincial clinics only.
ABC, abacavir; AZT, zidovudine; d4T, stavudine; EFV, efavirenz; LPV/r, lopinavir/ritonavir; NVP, nevirapine; TB, tuberculosis; TDF, tenofovir.
(DOCX)
*Those who returned to care included all patients who returned to a clinic in the Western Cape ≥180 days after their last visit. Those who did not return to care included patients who did not have any follow-up information found in the Western Cape databases. We excluded silent transfers, and patients who had disengaged but whose first data point after disengagement was either a) death or b) hospital contact.
**Variables selected for this model were the same variables selected for the multivariable Cox model in
ART, antiretroviral therapy; CI, confidence interval; d4T, stavudine; EFV, efavirenz; LPV/r, lopinavir/ritonavir; NVP, nevirapine; ref, reference; TB, tuberculosis; yrs, years.
(DOCX)
The authors would like to acknowledge the staff of the Centre for Infectious Disease Epidemiology and Research (CIDER) at the University of Cape Town, the clinical, administrative, and Provincial Health Data Centre staff of the Western Cape Government Department of Health, and the City of Cape Town Health Department. SK would also like to acknowledge Dr. Gerald Friedland for mentorship at Yale School of Medicine, and Franz Simon, PhD student at Yale University, for assistance in creating
antiretroviral therapy
stavudine
geographic information system
hazard ratio
lost to follow-up
National Health Laboratory Service
the United States President’s Emergency Plan for AIDS Relief
proportional hazards assumption
World Health Organization