Page 58 - SAHCS HIVMed Journal Vol 20 No 1 2019

Page 3 of 10 Review Article

then outcomes by sub-cohort were recorded. LTFU outcomes to 2010. Mean cohort size was 10,711; median was 3737. Only
at 1 and 5 years were collected, when available, for meta- six studies were interventional; the rest were observational.
analysis. If raw numbers were not readily available from the Seven studies (10%) utilised research data; the remainder of
text, the reviewers calculated it from the available text or the studies utilised routinely collected data from ART clinics.
figures and agreed on the numerator and denominator. If a Follow-up time ranged from 9 weeks to 5 years, with a large
study included both pre-ART and ART patients, the statistics variation in how this was calculated. Forty-six cohorts were
were only calculated for ART patients; pre-ART patients solely in primary care clinics, while four were solely in clinics
were excluded. If the study did not provide a definition for located in hospitals and 15 were in both primary care and
LTFU or default, or had another issue needing clarification, hospital clinics. Forty-five cohorts (67%) were in urban
S.K. contacted the corresponding author by email. settings, 7 (10%) were in rural settings and 13 (19%) were in
both urban and rural settings; 2 (3%) studies were missing
Assessment of study quality this information. Twenty-seven cohorts (40%) were in the
Gauteng province, 11 (16%) in the Western Cape, 7 in
Study quality and risk of bias were assessed by evaluating KwaZulu-Natal (10%), 1 (1%) in the Free State, 1 (1%) in
the selection of the cohort, ascertainment of outcomes, Limpopo and 2 (3%) did not include the information;
length of follow-up and the presence of missing data using 18 studies (27%) included data from multiple provinces,
a modified set of criteria based on the Newcastle-Ottawa which included Gauteng, Western Cape, KwaZulu-Natal,
14
domains (Appendix 2). Mpumalanga, Eastern Cape, Limpopo, Free State and North
West provinces.
Meta-analysis
For the 33 adult cohorts that reported age in aggregate, the
For a study to be included in the meta-analysis, it had to median age was 35.8 years, and for the 32 adult cohorts
have raw data available for a total number of patients LTFU reporting CD4 count, the median baseline CD4 was
at 12 months and/or 5 years of ART. Some of these studies 121 cells/µL. Among the paediatric cohorts, the median age
had overlapping data in that the data were collected from was 4.2 years at ART initiation, and the median aggregate
the same clinic population with some overlapping time CD4 percentage was 12.5%. In the four pregnancy cohorts,
periods. If it was not clear if the data were overlapping, the the median age was 28 years (n = 3 cohorts reporting),
reviewers emailed the manuscript authors for verification. If and the median CD4 estimate was 239 cells/µL. In terms of
the data did overlap, the reviewers selected the most recent definitions, 24 adult cohorts defined LTFU as 3 months
cohort with the largest amount of data available. without a clinic visit, 18 adult cohorts defined LTFU as
6 months without a clinic visit and 6 adult cohorts had other
Point estimates and 95% confidence intervals were calculated definitions, such as a different length of time without a clinic
for the proportion of people LTFU and data were visit or no definition of LTFU included in the manuscript text.
pooled following transformation using random-effects Of the paediatric cohorts, 2 cohorts defined LTFU as 3 months
meta-analysis. Differences in the definitions of LTFU without a clinic visit, 6 cohorts defined as 6 months without
(3 months vs. 6 months) and between patient groups (adults a clinic visit and 7 cohorts had other definitions. Among the
vs. children vs. pregnancy) were assessed through pre- pregnancy cohorts, one defined LTFU as 3 months without a
planned subgroup analyses. Point estimates and 95% clinic visit and the other three had other definitions (Online
confidence intervals were displayed visually on a forest plot Appendix 1 and 2 15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,
to visually assess heterogeneity. All data were analysed with 41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,
STATA version 14.0. 77,78,79,80,81 ).

Ethical considerations Of the 96 cohorts reporting mortality, encompassed within
the 67 studies, the median mortality estimate was 7.9%
This article followed all ethical standards for research without (interquartile range [IQR] 4.1% – 11.4%; range 0% – 26%);
direct contact with human or animal subjects.
range of time for reporting was 3 months to 5 years. There was
Results significant variability in how these estimates were calculated;
some were raw data reported at a certain endpoint; some
During the primary database search, 2611 abstract citations were estimated using statistical methods; and some studies
were identified and 2324 were excluded. After removing utilised linkage of patients to the national death registry. Of
duplicates, 163 full-text articles were screened for inclusion those 17 estimates in the lowest quartile (< 4% mortality), all
and six additional articles were included from a bibliography had n < 5000; nine (53%) had n < 1000. Ten of these cohorts
screen of these articles; 67 articles were included in the final (41%) estimated mortality at < 2 years of follow-up, 6 (35%)
review (Figure 1). did not standardise mortality estimates and the remaining
4 (24%) were paediatric studies with longer follow-up.
Of the 67 eligible studies, 48 were adult cohorts, 15 were Of the 16 estimates in the highest quartile (> 11.4% mortality),
paediatric cohorts and four were focussed on pregnant 10 cohorts (63%) had n > 2000, 5 cohorts (31%) had n < 1000, of
women; 57 studies included study or follow-up time prior which 3 were paediatric studies. Only five studies (29%)

http://www.sajhivmed.org.za 51 Open Access