Using vital registration data to update mortality among patients lost to follow-up from ART programmes: evidence from the Themba Lethu Clinic, South Africa

Authors

  • Matthew P. Fox,

    1.  Center for Global Health and Development, Boston University, Boston MA, USA
    2.  Health Economics and Epidemiology Research Office, University of the Witwatersrand, Johannesburg, South Africa
    3.  Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Alana Brennan,

    1.  Center for Global Health and Development, Boston University, Boston MA, USA
    2.  Health Economics and Epidemiology Research Office, University of the Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Mhairi Maskew,

    1.  Right to Care, Johannesburg, South Africa
    2.  Department of Medicine, Clinical HIV Research Unit, University of the Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Patrick MacPhail,

    1.  Right to Care, Johannesburg, South Africa
    2.  Department of Medicine, Clinical HIV Research Unit, University of the Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Ian Sanne

    1.  Right to Care, Johannesburg, South Africa
    2.  Department of Medicine, Clinical HIV Research Unit, University of the Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author

Corresponding Author Matthew Fox, Center for Global Health and Development, Boston University, Crosstown Center, 3rd Floor, 801 Massachusetts Ave, Boston, MA 02118, USA. Tel.: +617 414 1260; E-mail: mfox@bu.edu

Summary

Objective  To estimate the rates of mortality in patients lost to follow-up (LTFU) from a large urban public sector HIV clinic in South Africa.

Methods  We compared vital status using the clinic’s database to vital status verified against the Vital Registration system at the South African Department of Home Affairs. We compared rates of mortality before and after updating mortality data. Predictors of mortality were estimated using Kaplan–Meier curves and proportional hazard regression.

Results  Of the 7097 total patients who initiated highly active antiretroviral therapy at Themba Lethu Clinic by October 1st, 2008 and had an ID number, 6205 were included. 2453 patients (21%) were LTFU, of whom 1037 (42.3%) could be included in the analysis. After matching to the vital registration system, mortality more than doubled from 4.2% (258/6205) to 10.9% (676/6205). Overall 37% of those LTFU died by life-table analysis the probability of survival amongst those LTFU was 69% (95% CI: 66–72%), 64% (95% CI: 61–67%) and 59% (95% CI: 55–62%) by years 1, 2 and 3 since being lost, respectively. Those at highest risk of death after being lost were patients with a history of tuberculosis, CD4 count < 100 cells/μl, BMI < 17.5, haemoglobin < 10 and on <6 months of treatment.

Conclusion  Mortality was substantially underestimated among patients lost from a South African HIV treatment programme despite limited active tracing. Linking to vital registration systems can provide more accurate assessments of programme effectiveness and target lost patients most at risk for mortality.

Abstract

Utilisation de données de l’enregistrement vital pour la mise à jour de la mortalité chez les patients perdus de vue dans les programmes ART: données de la clinique Lethu Themba, Afrique du Sud

Objectif:  Estimer le taux de mortalité chez les patients perdus de vue dans une clinique VIH d’un vaste secteur public urbain en Afrique du Sud.

Méthodes:  Nous avons comparé le statut vital en utilisant la base de données de la clinique le système d’enregistrement vital du ministère sud-africain des affaires intérieures. Nous avons comparé les taux de mortalité avant et après la mise à jour des données de mortalité. Les prédicteurs de mortalité ont été estimés à l’aide des courbes de Kaplan-Meier et de la régression des risques proportionnels.

Résultats:  Sur les 7097 patients au total ayant débuté le HAART à la clinique LT au 1er octobre 2008 et ayant un numéro d’identification, 6205 ont été inclus. 2453 patients (21%) étaient perdus de vue, dont 1037 (42,3%) pourrait être inclus dans l’analyse. Après l’appariement au système de registre vital, la mortalité a plus que doublé, passant de 4,2% (258/6205) à 10,0% (623/6205). Par l’analyse du tableau de vie, la probabilité de survie chez les perdus de vue était de 69,2% (IC95%: 66,3-72), 64,2% (IC95%: 61-67) et 58,7% (IC95%: 55-62) pour les années 1, 2 et 3 respectivement depuis leur disparition. Les patients à plus de risque de décès après avoir été perdus vue étaient ceux ayant des antécédents de tuberculose, un taux de cellules CD4 < 100/μL, Un IMC < 17,5; une hémoglobine <10 et à moins de 6 mois de traitement.

Conclusion:  La mortalitéétait nettement sous-estimée chez les patients perdus de vue d’un programme de traitement du VIH en Afrique du Sud, malgré un retraçage actif limité. Une liaison avec les systèmes de registre vital peut fournir des évaluations plus précises de l’efficacité du programme et cibler les patients perdus les plus à risque de mortalité.

Abstract

Utilizando datos de registros vitales para actualizar la mortalidad en pacientes perdidos durante el seguimiento en programas de TAR: evidencia de la Clínica Themba Lethu, Sudáfrica

Objetivo:  Estimar las tasas de mortalidad en pacientes perdidos durante el seguimiento (PS) en una clínica de VIH de un amplio sector público urbano en Sudáfrica.

Métodos:  Hemos comparado el estatus vital utilizando la base de datos clínica verificada frente al sistema de Registros Vitales del Departamento Sudafricano de Asuntos Internos. Hemos comparado las tasas de mortalidad antes y después de poner al día los datos de mortalidad. Los vaticinadores de mortalidad fueron estimados utilizando curvas de Kaplan-Meier y regresión de riesgos proporcionales.

Resultados:  De un total de 7,097 pacientes que iniciaron una Terapia Antirretroviral Altamente Efectiva (HAART) en la Clínica Themba Lethu (CTL) antes de Octubre 1, 2008 y tenían un número de identificación, 6205 fueron incluidos. 2453 pacientes (21%) fueron PS, de los cuales 1037 (42.3%) podían incluirse en el análisis. Después de realizar la comparación con el sistema de registros vitales, la mortalidad aumentó en más del doble de 4.2% (258/6205) a 10.0% (623/6205). Mediante un análisis de tablas de vida se estimó la probabilidad de supervivencia de estos pacientes PS en un 69.2% (95% IC: 66.3%-72%), 64.2% (95% IC: 61%-67%) y 58.7% (95% CI: 55%-62%) en los años 1, 2 y 3 desde el momento de ser PS, respectivamente. Aquellos en mayor riesgo de muerte después de haber sido PS eran pacientes con una historia de tuberculosis, un conteo de CD4 < 100 cels/μL, IMC <17.5, hemoglobina <10 y con <6 meses de tratamiento.

Conclusión:  La mortalidad estaba sustancialmente subestimada entre pacientes PS de un programa para el tratamiento del VIH en Sudáfrica, a pesar de una búsqueda activa limitada. El conectarse a sistemas de registro vital puede proveer una evaluación más precisa de la efectividad del programa y enfocarse en pacientes perdidos con un mayor riesgo de mortalidad.

Introduction

As access to antiretroviral therapy (ART) has grown exponentially in low and middle income countries since 2004 (WHO 2005a,b, Sepulveda et al. 2007; WHO, UNAIDS, UNICEF 2008), attention has shifted from the immediate need to get patients into care, to the long-term challenges of keeping patients in care and on treatment. While virologic outcomes on ART in low-income countries are positive (Egger et al. 2002; Coetzee et al. 2004; Ivers et al. 2005; Laurent et al. 2005; Lawn et al. 2005) and similar to that of developed areas (Keiser et al. 2008), losses from HIV treatment programmes in low-income countries have been reported to be nearly 40% by 2 years after ART initiation (Rosen et al. 2007). These losses continue to reduce the ability to assess the overall effectiveness of HIV treatment programmes.

Few representative studies exist to inform us on the mortality rates of patients who are lost from HIV care. In some cases, being lost means a patient has found care elsewhere, while in others, particularly in rural areas, patients may have few other options for seeking ART resulting in discontinuation of care and treatment. It has been reported that patients who have been lost and do not seek care elsewhere are likely to die within 1 year of leaving care (Mocroft et al. 1997; Morgan et al. 2002; Badri et al. 2006). According to a recent systematic review and meta-analysis of outcomes of patients lost to HIV care and treatment programmes, 20–60% of patients who could be traced had died (Brinkhof et al. 2009).

Studies of the effectiveness of ART programmes as well as monitoring and evaluation statistics typically separate out death and lost to follow-up (LTFU). Often the majority of reported attrition from ART programmes is in the LTFU category as finding patients who are lost to follow-up is both difficult and requires using limited resources for tracing. However, to be able to determine how effective the global ART scale up is at reducing mortality, it is essential to have accurate data on patient outcomes after leaving care.

One method of reducing the number of missed deaths in ART programmes is through linking with national vital registration systems. While vital registration systems in low- and middle-income countries often have poor sensitivity, in South Africa, where the largest number of HIV-positive patients in the world is living, the National Vital Registration system has been demonstrated to detect nearly 90% of all adult deaths (Dorrington et al. 2001; Timaeus et al. 2002; Statistics SA 2005). Still, few HIV treatment programmes even within South Africa have the resources to be able to devote to investigating outcomes among patients lost. Thus, to better understand vital status outcomes among patients lost from HIV treatment programmes as well as what predicts mortality among those lost, we analysed data from a large urban public sector HIV clinic in South Africa where vital status had been verified against the Vital Registration system at the South African Department of Home Affairs.

Methods

Study site

The study was conducted at the Themba Lethu Clinic (TLC) in Johannesburg, South Africa. TLC is a public sector HIV treatment clinic that began large scale provision of ART in April 2004 as part of the South African government’s Comprehensive Care Management and Treatment of HIV (CCMT) programme. The clinic also receives funding from Right to Care, a South African NGO supporting HIV treatment through PEPFAR. The clinic is one of the largest in South Africa, with more than 22 000 patients ever initiated on ART, more than 14 000 patients actively on ART and thousands more in pre-ART care. Currently, the clinic staff provide upwards of 500 consultations per day with care provided according to the guidelines from the South African National Department of Health (NDH 2004).

Use of Themba Lethu Clinic data and linkage of this data with the Department of Home Affairs Vital Registration system were approved by the Human Research Ethics Committee of the University of the Witwatersrand and by the Ethics Board of the University of Cape Town. Approval for analysis of the data in a de-identified manner was granted by the Institutional Review Board of Boston University.

Data sources

Themba Lethu Clinic.  Themba Lethu Clinic uses a real-time data capturing system called TherapyEdge-HIV™ in which data on all patient encounters (including scheduling information, demographic indicators clinical conditions, laboratory results and patient outcomes) are captured at the time of the encounter by a clinician or member of the clinic staff. At enrolment, South African citizens are asked to give their South African National Identification number, though care is provided to all patients including those who do not have or do not wish to provide one.

We defined LTFU as having missed a clinic appointment (e.g. clinical assessment, ARV pickup, counsellor visit) by at least 3 months after the scheduled visit date. The cumulative incidence of LTFU by Kaplan–Meier analysis was 13.7% during the first year on ART and 21.7% after 3 years on ART. Since 2007, TLC has had an initiative to actively trace those LTFU to improve overall retention in care. At enrolment, patients are asked whether they are willing to be contacted should they leave care and to provide an address, a phone number and a friend’s or relative’s phone number. Close to 95% of patients participate in this programme. If a patient becomes lost, counsellors attempt to contact the patient and return them to care typically by phone and if not reachable via phone then by home visit. Patients found to be receiving care elsewhere are updated as having been transferred and those found to have died have their vital status updated in the database. The objective of the programme is to contact as many lost patients as possible; however, as funding is limited and the clinic is large, not all patients can be contacted. In addition, patients lost who were contacted but could not be reengaged in care would be listed as lost in the database even if they later went on to die.

As of October 2008, a total of 11 694 patients were initiated on ART at TLC, of whom 7097 (60.7%) provided a legitimate South African ID number that could be linked to the registry. Often, in the early years of the clinic’s operation, ID numbers were not collected. In some cases, this was simply the practice of those enrolling patients, while in others, it was because the clinic serves a diverse population of patients, some of whom are not from South Africa and who would not have a national ID number. Patients without an ID number were similar to those included in the analysis in terms of mean baseline CD4 count (104.5 vs. 104.8), haemoglobin (11.3 vs. 11.6) and BMI (21.3 vs. 22.5). There was also little difference in the age sex or race distributions between the two groups.

Of the 7097 patients, we excluded all those who initiated ART prior to the government roll out on April 1st 2004 (N = 184) and all patients who transferred out of care at TLC as their LTFU status could not be determined (N = 320 of whom 8% died). In addition, we excluded all women who had incident pregnancies (N = 316) because we found an inverse association between incident pregnancy and LTFU (HR 0.42; 95% CI 0.29–0.60), partly because the clinic refers pregnant women to another clinic for care during their pregnancy. Finally, we excluded all patients who were LTFU <6 months prior to the linkage (N = 72), as they may not have been included in the registry if they had died because of delays in reporting. This left 6205 subjects who were eligible for the current analysis. At the time of the linkage, 21% (N = 2453) of all highly active antiretroviral therapy (HAART) patients were LTFU, of whom 42.3% (N = 1037) gave a legitimate identification number and were included in the analysis.

National Vital Registration System.  In South Africa, all deaths must be reported to the Department of Home Affairs (DHA) for issuance of a death certificate. While no registry is 100% accurate, in South Africa, the registry is fairly comprehensive. Statistics South Africa estimates that 90% of all adult deaths are captured through this system (Dorrington et al. 2001; Timaeus et al. 2002; Statistics South Africa 2005). In Free State province, over 80% of all deaths identified from their ART programme were found through the registry rather than through patient files (Fairall et al. 2008).

To assist in the identification of patients lost from HIV care, vital status of all patients in the TLC dataset who began ART between April 2004 and September 2008 was verified through the registry by a member of the DHA with ethics approval to conduct the linkage. As South African ID numbers require certain digits to take on specific values, only legitimate ID numbers were validated. When a match was found, the death was recorded along with a date of death. The linked file was returned and used to update the clinic records and the data were then de-identified for analysis.

We note that there is a delay in reporting to and updating the vital registration data; thus, some patients listed as not having died in the current analysis may have died but were not yet reported to the registration system. To account for this, as noted above, all patients who were LTFU <6 months prior to the linkage were excluded from the analysis.

Statistical methods

We calculated the overall mortality in the treatment programme both before and after linking with the vital registration system. We present each as simple proportions and 95% confidence intervals. We then conducted a similar analysis limited to only patients who were lost from the TLC ART programme. We looked for crude differences between groups using proportional hazards regression. We used life table methods to calculate the proportion of patients lost who had died in the years following being lost. To look for predictors of death among those who were LTFU, we calculated crude Kaplan–Meier estimates of mortality and adjusted predictors of mortality were estimated using Cox proportional hazard regression.

Results

Table 1 shows characteristics of the cohort of 6205 patients stratified by their vital status after the linkage. Patients were enrolled fairly evenly over the roughly 4 years of follow-up. The majority of patients were black (94.7%), female (64.1%), between 30–40 years of age (48.7%) and on the standard first-line regimen of stavudine (D4T), lamivudine (3TC) and efavirenz (EFV). Of the 6205 eligible subjects, 258 (4.2%) were known to have died before updating vital status and 1037 (16.7%) were LTFU.

Table 1.   Demographic and clinical characteristics of the 6205 eligible patients on ART at the Themba Lethu Clinic in Johannesburg, South Africa, from April 2004–September 2008 by vital status
Variable Died (%)Alive (%)Hazard ratio of death (95% CI)
  1. CI, confidence interval; TB, tuberculosis; BMI, body mass index; HB, haemoglobin; ART, antiretroviral therapy.

  2. *For patients who died or were lost to follow-up, this was their last value before leaving care. For those still in care, this was their last visit overall.

  3. †Tuberculosis before initiation of ART.

SexFemale398 (10.0)3578 (90.0)Reference
Male278 (12.5)1951 (87.5)1.25 (1.08–1.44)
Race is blackNo41 (12.5)286 (87.5)Reference
Yes635 (10.8)5243 (89.2)0.86 (0.64–1.16)
Year of initiation on first line ART2004132 (13.8)826 (86.2)Reference
2005195 (14.0)1199 (86.0)1.02 (0.83–1.25)
2006170 (11.2)1342 (88.8)0.82 (0.66–1.01)
2007117 (9.6)1103 (90.4)0.70 (0.55–0.88)
200862 (5.5)1058 (94.5)0.40 (0.30–0.54)
Regimen at ART initiationd4T-3TC-EFV593 (11.9)4376 (88.1)Reference
d4T-3TC-NVP29 (6.3)431 (93.7)0.53 (0.37–0.76)
Other54 (7.0)722 (93.0)0.58 (0.45–0.76)
Baseline WHO StageStage I161 (8.4)1747 (91.6)Reference
Stage II67 (9.8)614 (90.2)1.17 (0.89–1.53)
Stage III172 (13.1)1139 (86.9)1.55 (1.27–1.90)
Stage IV82 (18.9)351 (81.1)2.24 (1.76–2.87)
Baseline CD4 count0–50310 (16.9)1519 (83.1)Reference
51–100127 (11.5)975 (88.5)0.68 (0.56–0.82)
101–200129 (7.1)1691 (92.9)0.42 (0.34–0.51)
>20032 (5.5)547 (94.5)0.33 (0.23–0.46)
Last CD4 count*0–50266 (47.1)299 (52.9)Reference
51–100127 (29.8)299 (70.2)0.63 (0.53–0.75)
101–200158 (13.0)1053 (87.0)0.28 (0.23–0.33)
>200117 (2.9)3850 (97.1)0.06 (0.05–0.08)
History of TB†No605 (10.9)4930 (89.1)Reference
Yes71 (10.6)599 (89.4)0.97 (0.77–1.22)
Last viral load*<400166 (3.8)4209 (96.2)Reference
400–10 00039 (9.8)359 (90.2)2.58 (1.85–3.60)
>10 000101 (23.5)329 (76.5)6.19 (4.94–7.77)
Age at initiation18–29.9116 (9.7)1080 (90.3)Reference
30–39.9311 (10.3)2711 (89.7)1.06 (0.87–1.30)
40–49.9180 (12.2)1291 (87.8)1.26 (1.01–1.57)
50+69 (13.4)447 (86.6)1.38 (1.04–1.82)
Last BMI < 17.5*No398 (7.8)4702 (92.2)Reference
Yes155 (47.5)171 (52.5)6.09 (5.25–7.06)
Last HB < 10*No439 (7.9)5119 (92.1)Reference
Yes222 (39.3)343 (60.7)4.97 (4.34–5.70)
Time on ART0–3 months323 (33.5)641 (66.5)Reference
3–6 months110 (20.2)434 (79.8)0.60 (0.50–0.73)
6–12 months107 (14.3)642 (85.7)0.43 (0.35–0.52)
≥12 months132 (3.3)3810 (96.7)0.10 (0.08–0.12)

Vital Status outcomes

Table 2 shows the mortality data within the clinic before and after updating vital status. In total 4.2% of patients were known to have died according to the clinic records (258/6205). After updating the clinic records by matching to the vital registration system, mortality went up to 10.9% (676/6205), more than double. Using product-limit estimates, the cumulative 1- and 3 -year mortality estimates after the linkage were 7.8% and 11.7%, respectively.

Table 2.   Vital status outcomes among ART patients lost to follow-up from the Themba Lethu Clinic, Johannesburg, South Africa*
Themba Lethu Clinic recordsVital registration systemTotal
Registered as diedNot registered as having died
Died23028258
LTFU3866511037
Alive and in care6048504910
Total67655296205
Performance% Mortality (95% CI)N/total
  1. ART, antiretroviral therapy; CI, confidence interval; LTFU, lost to follow-up.

  2. *Data were validated against South Africa’s Department of Home Affairs Vital Registration System.

Mortality before updating4.2% (3.7–4.7%)258/6205
Mortality after updating10.9% (10.1–11.7%)676/6205
Mortality among LTFU after updating37.2% (34.3–40.2%)386/1037

Among those lost from the clinic, after the linkage, 386 patients (37.2%; 95% CI: 34.3–40.2) were known to have died, while when limited to just those with a potential year of follow-up since being lost, 39.0% (95% CI: 35.8?42.3%) were known to have died. As would be expected, the rate of death among those LTFU increased with duration since being lost, with over 65% of those lost in 2004 having died, while less than 30% of those lost in 2007 were deceased (Figure 1). Regardless of the calendar year in which patients became lost, more than 75% of all deaths occurred within 1 year of being lost and nearly 95% occurred within 2 years. By life table analysis (Table 3), we estimate that the cumulative probability of survival amongst those LTFU was 69.2% (95% CI: 66.3–72%) by 1 year since being lost compared to 64.2% (95% CI: 61–67%) and 58.7% (95% CI: 55–62%) by years two and three, respectively.

Figure 1.

 Proportion of subjects who were determined to be lost by the clinic who were confirmed dead in the South African Vital Registration database by year lost from the clinic (Excludes those lost in 2008 (n = 173) as subjects lost in 2008 would not all have completed a potential 1 year since being lost).

Table 3.   Life table survival estimates up to 36 months since being lost among patients on HIV treatment at the Themba Lethu HIV clinic
Interval (months)Number failedNumber censoredSample sizeConditional probability of failureSurvival95% confidence interval
0–2.924801037.00.2411
3–5.9394787.00.050.760.73–0.79
6–8.93099696.50.040.720.70–0.75
9–11.91677578.50.030.690.66–0.72
12–17.922100474.00.050.670.64–0.70
18–23.91998353.00.050.640.61–0.67
24–29.9877246.50.030.610.57–0.64
30–35.9496152.00.030.590.55–0.62
36010050.00.000.570.54–0.61

Predictors of death among those lost

Figure 2 shows crude Kaplan–Meier survival curves in the year after being lost from HIV treatment. Mean follow-up time was 253 days (range 0–365). Of the 1037 subjects included in this analysis, 386 (37.2%) died. Fewer survived among older patients, those with more severe disease progression and those with shorter duration on treatment prior to being lost (logrank P-value < 0.0001 for all). In adjusted analyses (Table 4), the strongest predictors of mortality after being lost were having a low CD4 count before being lost (≤100 vs. >200 HR 3.4; 95% CI 2.3–5.0) and having a BMI < 17.5 before being lost (HR 2.4; 95% CI 1.8–3.1).

Figure 2.

 Crude Kaplan–Meier survival curves for death by 1 year since time of being lost among patients on HIV treatment at the Themba Lethu HIV Clinic Stratified by (a) year lost to follow-up; (b) age at being lost; (c) last CD4 count before being lost (d) last BMI before being lost (e) last haemoglobin before being lost and (f) months on antiretroviral therapy before being lost. (Log rank P-value < 0.0001 for all).

Table 4.   Crude and adjusted hazard ratios of mortality for the 1037 patients lost from an HIV treatment programme at the Themba Lethu Clinic, Johannesburg, South Africa*
VariableCrude model(95% CI)Adjusted model†(95% CI)
  1. CI, Confidence interval; BMI, body mass index; ART, antiretroviral therapy; HB, haemoglobin; LTFU, lost to follow-up.

  2. *Mortality data come from South Africa’s Department of Home Affairs Vital Registration System.

  3. †Adjusted for all other variables in the model using Cox proportional hazards regression.

Age at time of being lost
 ≤401.01.0
 >401.67 (1.34–2.08)1.43 (1.10–1.85)
Last CD4 count before being lost
 >2001.01.0
 100–2000.74 (0.57–0.96)1.80 (1.20–2.70)
 0–1003.83 (3.06–4.79)3.38 (2.27–5.03)
History of tuberculosis
 No1.01.0
 Yes1.31 (0.98–1.74)1.29 (0.93–1.79)
Sex
 Male1.01.0
 Female0.78 (0.63–0.97)1.07 (0.83–1.38)
Last BMI before being lost
 ≥17.51.01.0
 <17.53.69 (2.82–4.83)2.35 (1.76–3.14)
Last HB before being lost
 ≥10.01.01.0
 <10.02.56 (2.03–3.23)1.39 (1.04–1.85)
Year lost from care
 20041.01.0
 20051.52 (1.21–1.91)0.85 (0.48–1.52)
 20060.71 (0.55–0.92)0.62 (0.34–1.13)
 20070.81 (0.63–1.03)0.81 (0.45–1.47)
 20080.94 (0.69–1.27)0.96 (0.52–1.78)
Months on ART before LTFU
 ≥12 months1.01.0
 6–12 months0.82 (0.61–1.11)1.38 (0.90–2.11)
 0–6 months2.34 (1.86–2.94)1.52 (1.05–2.21)

Discussion

Understanding the impact of losses from HIV treatment programmes is critical to accurately evaluating their overall success. In some HIV treatment programmes, LTFU may constitute the largest proportion of overall programme attrition (Rosen et al. 2007). In our study, ascertainment of mortality was strongly underestimated in our clinic even though limited active tracing does exist. After matching clinical data with the vital registration system, the mortality rate more than doubled, from 4.2% to 10.9%. While this increase is large, other programmes without active tracing programmes have estimated that their mortality increased as much as fivefold after adjusting for the high rate of death amongst those lost (Geng et al. 2008).

Our study confirms, in a well defined population, the suspicion that there is high mortality amongst those lost from HIV treatment programmes in South Africa. We found that the probability of death among those lost to follow-up was over 30% at the end of 1 year. It is encouraging, however, to see that the rate of death in the first year after being lost has steadily fallen since 2004. While this overall high mortality rate among those lost means that many who are lost are likely not seeking care from another provider, others may be, particularly as more options for care become available. In a recent systematic review, Brinkhof et al. (2009) estimated that among those lost whose vital status could be determined mortality was as high as 40%, slightly higher than our estimate. Because our study included all patients whose national ID number we had, our study is likely more representative of the clinic population than would be achieved by attempting to contact patients.

Appropriately evaluating treatment programmes means assessing treatment outcomes for all patients. Typically, analyses of ART programme effectiveness report on both mortality and LTFU because the two are closely linked, but many of those who are lost will have died and are therefore misclassified. It has previously been shown that failure to determine the outcomes in patients lost can cause dramatic underestimation of overall mortality. In a cohort in Botswana, missing deaths among those LTFU caused over-estimating of 1 year survival rates by nearly 10% (Bisson et al. 2008). Despite this, as resources targeted for tracing patients who are lost are limited, outcome assessment remains poor.

A 2007 review concluded that in Africa only South Africa, Mauritius and the Seychelles have well-functioning vital registration systems (Setel et al. 2007). Outside of these areas, an alternative proposed solution to the problem of poor outcome assessment is to trace randomly sampled subsets of the population and use estimates from the subset whose vital status can be traced to adjust overall estimates of mortality in the cohort (Geng et al. 2008; Yiannoutsos et al. 2008). Another approach uses a nomogram approach to adjust overall mortality (Egger et al. 2009). When no viable vital registration system exists but resources are available to contact a sample of patients, these approaches are an improvement over crude approaches which ignore LTFU. However, our study demonstrates that even approaches that sample patients to determine their vital status likely suffer from some bias as they miss patients who cannot be contacted and may miss deaths among patients who go on to die after the sampling.

Despite the benefits of using statistical approaches to adjust for deaths among those lost, this approach is not preferable to directly identifying the vital status of those LTFU. Programmes with active tracing have been shown to have substantially lower rates of death than those without (Keiser et al. 2008). This approach has the added advantage of being able to bring patients back into care and potentially reduce the overall mortality rate. Still when not possible, our results show that matching with vital registration systems can be successfully implemented to better estimate the total mortality of patients in ART programmes.

While all patients lost from care are at increased risk for mortality, identifying patients at highest risk for dying after leaving care is of critical importance. In our analysis, those at highest risk of death after being LTFU were predominately those who were sickest when leaving care (low CD4 count, BMI and haemoglobin levels before leaving care) and those on treatment for <6 months prior to being lost. In cases where limited resources exist for getting patients back into care, these sicker patients should be prioritized for tracing and returned to care.

Our findings must be interpreted in light of their limitations. First, because it is not mandatory for patients to provide a National Identification number when registering for treatment at the clinic, we were not able to verify the outcome status of all patients. The clinic in which the study was conducted does not turn away patients regardless of whether or not they have a National ID number. Patients without an ID number could not be included in the analysis and may have been more likely to have died than patients included in the study. If so, our estimates of updated mortality would likely be underestimates. In addition, because stigma still exists around HIV/AIDS, some patients may provide a fake identification number for fear of being identified as being HIV positive or of being turned away. Given this, despite dramatically increasing our overall mortality rate to more accurately reflect overall programmatic impact, we are still likely underestimating the true mortality rate amongst those LTFU at the clinic.

In conclusion, we found that mortality was substantially underestimated among patients lost to follow-up in an HIV treatment programme in Johannesburg, South Africa despite an active tracing programme to return patients lost to care. Linking to the national vital registration system proved to be a practical method to more accurately represent mortality in the cohort. Linking to vital registration systems when feasible can provide a means to more accurately assess the effectiveness of a programme and target LTFU patients most at risk for mortality.

Acknowledgements

We thank Mr. David Bourne for help with linking data for this study and Dr. Andrew Boulle for help in design of the study. We are also extremely grateful to Babatyi Malope-Kgokong for her work in establishing the Themba Lethu Clinical Cohort dataset. We express our gratitude to the directors and staff of the Themba Lethu Clinic (TLC) and to Right to Care, the NGO supporting the study site through a partnership with USAID. We also thank the Gauteng and National Department of Health for providing for the care of the patients at the TLC as part of the Comprehensive Care Management and Treatment plan. Most of all, we thank the patients attending the clinic for their continued trust in the treatment provided at the clinic. Funding was provided by the United States Agency for International Development (USAID) under the terms of agreement 674-A-00-08-00007-00 with Right to Care (RTC). The project described was also supported by Award Number K01AI083097 from the National Institute of Allergy And Infectious Diseases. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Allergy And Infectious Diseases or the National Institutes of Health. The opinions expressed herein are those of the authors and do not necessarily reflect the views of USAID, the Themba Lethu Clinic, or Right to Care. Right to Care provided some of the funding for the current research and also supports the provision of treatment for the patients in the study.

Ancillary