QAD/21s301; Total nos of Pages: 7;
The epidemiology of HIV in South African
Mark Colvina, Cathy Connollyband Lorna Maduraic
Objective: To determine the prevalence and distribution of HIV in South African
conducted in 22 public and private sector organizations in all nine provinces of South
Africa on full-time, formally employed personnel who provided consent to participate.
Outcome measures: The primary outcome was HIV prevalence.
Results: The crude HIV prevalence among the 32015 participants was 10.9%. HIV
prevalence was higher among men (11.3%) than among women (9.8%) and among
black Africans (16.6%) than among other race groups (2.7%). Although managers and
employees with post-school education had a lower HIV prevalence than lower skilled
employees, this only partly accounted for the race differences.
race, age and sex structure of the workforce. This indicates that there is some other
factor that is associated with the organization and has an impact on HIV prevalence.
? 2007 Lippincott Williams & Wilkins
AIDS 2007, 21 (suppl 0):S1–S7
Over the past decade, employers in the private and public
sectors of South Africa have been increasingly concerned
about the impact that the HIV epidemic may have on
their operations. The potential impacts of HIV-related
mortality and morbidity include decreased productivity,
rising production costs and a higher employee turnover.
At the same time, employers have been under increasing
pressure to respond to the epidemic by providing
prevention and treatment services.
In order to determine the extent of the impacts and to
plan for the future, many organizations have undertaken
HIV prevalence studies on their workforces. Obtaining
data on the epidemiology of HIVallows an organization
to conduct human resource and cost-impact planning,
enables it to anticipate treatment and support require-
ments, facilitates the implementation of prevention
measures, and allows the impact of workplace HIV
interventions to be monitored over time.
As a result of concerns regarding the confidentiality of
company information and potentially negative publicity,
however, most research in this field is not published.
of HIV in South African workplaces by combining
datasets from studies conducted in 22 workplaces, and to
compare and contrast workplace data with data obtained
from other surveys.
Since 1997 the authors and colleagues have conducted
HIV prevalence surveys in a variety of private sector
companies and government institutions across all nine
provinces of South Africa. Each survey is individually
commissioned by the management of the organization
and therefore the researchers have no control over the
selection of organizations.
From theaCentre for AIDS Development, Research and Evaluation (CADRE), Durban, South Africa, thebSouth African Medical
Research Council (MRC), Durban, South Africa, and thecGlobal Laboratories, Durban, South Africa.
Correspondence to Mark Colvin, CADRE, Private Bag X07, Dalbridge 4014, South Africa. No reprints available from the authors.
Received: ????; revised: ?? ??; accepted: ????.
ISSN 0269-9370 Q 2007 Lippincott Williams & Wilkins
QAD/21s301; Total nos of Pages: 7;
Although participationin the surveyswas voluntary, these
were not ‘volunteer’ studies, in that we did not simply
invite any employee to participate until the required
sample size was reached. Instead, everyone who was
sampled was encouraged to participate, and participation
levels were monitored and reported on.
In organizations with less than 500 employees, there was
no sampling and all employeeswere invited to participate.
In organizations with more than 500 employees, shifts or
departments were randomly selected and then the whole
shift or department was invited to participate. Random
sampling at the individual level is not usually feasible or
acceptable in a workplace setting.
It is only after thorough consultation with management,
trade unions and employees has been undertaken that a
survey is implemented. Groups of sampled employees are
brought together in a suitable venue such as a training
room or board room, and are again informed about the
study and their right to refuse to participate. Those
choosing to participate complete brief questionnaires that
obtain demographic information in addition to questions
on their knowledge, attitudes and sexual practices.
Consenting employees also provide an oral fluid or
laboratory-based enzyme-linked immunosorbent assay
testing system or a government-approved HIV rapid test.
test and this is in line with World Health Organization
guidelines in settings where the prevalence of HIV
exceeds 10% .
HIV test results are linked to the questionnaire by means
of a bar code but no HIV result can be linked back to an
individual. Questionnaires are double entered onto a
computer database and merged with the HIV results. The
dataset is kept by the researchers and only a summary
report is given to the company.
The demographic characteristics of the combined
workforce are presented as well as the maximum and
minimum values per company to illustrate the variation
Previous research has shown that age, sex, race, time of
survey and geographical region are all associated with the
prevalence of HIV . Differences in the composition of
these factors among company employee populations will
thus influence the overall HIV prevalence in a particular
institution. Direct standardization is used to present a
single summary statistic that takes account of age, sex and
racial differences. The combined workforce population
was used as the ‘standard population’, so the standardized
rates should be used for comparison purposes only. The
standardized rates were then adjusted for temporal and
geographical trends using weights generated from a
generalized linear model for binomial distributions based
on data from the Department of Health’s annual antenatal
HIV prevalence survey. The weights used were: 1.196,
2005, respectively, and 2.302, 1.2673, 1.0 for low, middle
and high prevalence provinces. Where provinces were
spread over several provincial areas a weighted average
In order to explore both individual and company
characteristics affecting HIV prevalence, multilevel mod-
elling techniqueswere used. The individual characteristics
available were: age, race, sex and province. Not all com-
panies collected data on job band and education, so there
variables were not included in the model. Company
characteristics were size of company, sector and yearof the
survey. Sector was divided into three categories: public/
utilities/parastatals; service and manufacturing.
Because the employees are clustered by company, the
standard assumption of independent observations is likely
to be violated because of the correlation among
employees within the same company. Company is
therefore added to the model as a random effect to take
account of the clustering. This random effect also enables
us to estimate the intraclass correlation coefficient that is a
measure of the degree of similarity among employees of
the same company, which is unobserved. In other words,
it is that ‘proxy’ measure of some other, as yet
undetermined, factor unique to a particular company.
Approval for workplace HIV prevalence studies has been
given by the Labour Court of South Africa provided that
the conditions laid down by the court are followed . In
addition, ethical approval is obtained from universities or
parastatal research organizations, depending on where the
study is being conducted. All trade unions within a
particular organization have to give approval and support
for the research before the study is undertaken.
Participation in a study is voluntary and anonymous in
that no names, work numbers or other identifying
information are obtained from employees. Because the
study is anonymous, verbal consent is obtained rather
than written consent, as the latter would render the
process no longer anonymous. Participant information
sheets are made available to employees and the contents
explained to them. For ethical reasons and to maximize
participation levels, substantial effort goes into informing
the workforce about the study. Briefing sessions, e-mail
messages, pamphlets and other methods are used in this
avail themselves of this service.
S2AIDS2007, Vol 21 (suppl 0)
QAD/21s301; Total nos of Pages: 7;
One company was surveyed twice at different timepoints
the sites had significantly different operations. In the
analysis, each survey is treated as if it was an independent
company. The surveys were performed over a period of
6 years from 1999 to 2005. Some companies had multiple
sites in various provinces of the country, whereas others
were located in a single province.
Participation levels could not be measured in four surveys
because of difficulties in reconciling actual participants
with those who were sampled. Participation levels in the
remaining 18 surveys ranged from 49.5 to 98.8%, with
21374 employees out of 27909 (75.6%) participating
overall. The mean number of employees surveyed per
company was 1455 (median 841) with a range of 245
The median age of employees was 39 years (range 14–85)
and varied by company from 33 to 48 years. The median
age of employees was 40 years for men (range 14–85) and
35 years for women (range 18–85). These medians also
varied by company: men 33–47 years and for women
29–49 years. Overall, 70.5% of employees were men;
however, one company had as few as 10.3% and another
Africans, but this varied from 37.3 to 91.8%. The
demographics of participants are given in Table 1.
A total of 3500 employees were HIV positive out of the
32015, giving an unadjusted HIV prevalence of 10.9%.
This prevalence by organization ranged from 2.1 to
30.2%. HIV prevalence levels by year, region and selected
demographic variables are given in Table 1.
HIV prevalence was higher in men than women: 11.3%
compared with 9.8%, but this is not statistically
significant. As in the general population, HIV prevalence
peaked earlier in women (age group 20–29 years) than
men (age group 30–39 years). The age and sex
distribution of HIV infection is given in Figure 1.
HIV prevalence also varied by province, from a low of
5.5% in the Western Cape and 6.1% in the Northern
Cape to 16.7% in KwaZulu Natal and 18.5% in the
HIV prevalence was significantly higher in black Africans
(16.6%) when compared with other race groups (2.8%).
Table 2 shows HIV prevalence by job band and education
HIV prevalence is shown for all the races and then
separately for Africans and other race groups. HIV
prevalence was significantly lower among managers when
compared with skilled and unskilled labour [odds ratio
(OR) 0.8;95% confidence interval (CI)0.6–0.9;and OR
0.4; 95% CI 0.3–0.5, respectively). A siignificantly lower
HIV prevalence was also associated with increasing
educational levels: secondary (OR 0.7; 95% CI 0.6–0.8)
and postsecondary (OR 0.3; 95% CI 0.3–0.4) when
The lower rates among managers and employees with
postschool education remain statistically significant when
racial groups were analysed separately.
Table 3 gives the crude HIV prevalence, HIV prevalence
standardized for age, sex and race, the adjustment factor,
and the final prevalence adjusted for geographical and
HIV in South African workplaces Colvin et al.S3
Table 1. Distribution of participants and HIV prevalence by year,
region and various demographic variables.
Other race groups
Low prevalence provinces
Mid prevalence provinces
High prevalence provinces
Fig. 1. HIV prevalence by sex and age group.
QAD/21s301; Total nos of Pages: 7;
overall prevalence in public sector companies was the
lowest (11.8%) followed by utilities/parastatals (14.8%),
and private sectors companies (also 14.8%).
Figure 2 showsthatwhentheresultsarestandardizedusing
a variety of adjustment approaches, the variability of the
data decreases. The crude prevalence ranged from 4.3 to
31.3%. The range of the standardized prevalence narrows
considerably, 3.3–17.0% and then widens again when
adjusted for trends, 4.9–24.3%. In some cases the
adjustment resulted in a big change (31.3–10.9%), but
in most cases the change was less than 5%. Adjusting for
demographic, regional and time-based variables does not,
however, fully account for the variability between
Table 4 shows the odds ratios for factors associated with
HIV infection adjusted for individual and company-level
variables. Age, race and province of work are all
significantly associated with HIV prevalence in the entire
workforce. Interestingly, the difference between men and
women is not significant. The intraclass correlation
among companies, r¼0.04, P<0.001, is significantly
different from zero, indicating that there are still
differences among companies not explained by the
demographic factors in the model.
Company-level factors, as described in the Methods
section, are then added to the model. The association
between individual factors and HIV prevalence remains
unaffected by the addition of the second-level factors.
Company size was correlated with sector and dropped
from the model. The other two factors are both
significantly associated with HIV prevalence. Surveys
performed in 2002/2003 and 2004/2005 both had a
S4AIDS 2007, Vol 21 (suppl 0)
Table 2. HIV prevalence by job band and education.
LevelEntire workforce Africans Other race groups
% HIVORPN % HIV ORPN % HIVORP
OR, Odds ratio.
aNumbers less as a result of missing data: 28% missing job band and 55% educational level.
Table 3. Crude and adjusted company-level HIV prevalence.
PublicEastern Cape municipality
Gauteng local government department
Northern Province municipality
National transport parastatal
Food distribution warehouse
Contract cleaning companies
National agricultural company
aHIV prevalence adjust to combined population using age (three groups) sex and race (two groups).
bAdjustment factor for temporal and geographical trend.
cFinal adjusted prevalence.
QAD/21s301; Total nos of Pages: 7;
in 1999/2001 (OR 1.4; 95% CI 1.1–1.9; and OR 1.5;
95% CI 1.2–1.9). HIV prevalence was significantly
greater in service sector companies when compared
with public sector or manufacturing sector companies
(OR 1.8; 95% CI 1.1–2.8).
Company variables show similar patterns when races are
analysed separately, but the associations are not as great
and do not reach statistical significance.
The intraclass correlation is unchanged when company
suggests that, although company factors explain an
additional component of thevariability inHIVprevalence
among companies, significant variability among compa-
nies still remains unexplained.
This is the first comprehensive epidemiological descrip-
tion of the epidemiology of HIV in South African
workplaces. The only other previously published reports
on the prevalence of HIV in South African workplaces
[4,5] did not obtain information on the race group of
participants and thus there was likely to be substantial
confounding by race.
The crude HIV prevalence levels reported here are
substantially lower than those reported from surveys
among antenatal women, but are broadly similar to those
reported among other workplace studies and among
employed individuals in general population studies.
Table 5 shows some of these comparisons. It must be
acknowledged, however, that these are crude levels and
further analysis needs to be performed to disaggregate the
of a future paper.
Pre-employment and intra-employment linked testing
for HIV has been illegal in South Africa since 1997, and
there is no evidence that it was widely used before that or
HIV in South African workplaces Colvin et al. S5
Fig. 2. Comparison of the variability of HIV prevalence by
using different methods of adjustment.
Table 4. Odds ratios for factors associated with HIV infection adjusted for individual and company-level variables.
Entire work force AfricansOther races
OR95% CIP OR95% CIP OR95% CIP
0.9 0.9–1.1 0.90.9–1.2 0.50.7–1.1 0.3
CI, Confidence interval; OR, odds ratio.
QAD/21s301; Total nos of Pages: 7;
since. Therefore, we do not believe that such ‘screening’
is the reason for a lower HIV prevalence among
Consistent with other large surveillance studies in South
Africa [6,7], race remains the variable most strongly
associated with HIV prevalence. This association is only
partly explained by race differences in socioeconomic
status as measured by job band and level of education.
There is no evidence that race per se increases the risk of
infection, and so the challenge for researchers is to
determine for what factors race is a proxy measure.
The association between education and HIV and socio-
economic status and HIV reported here is remarkably
similar to the findings from the 2002 Human Sciences
Research Council (HSRC) national survey. In both
occurred in those with a tertiary education. It is possible
that individuals with a higher qualification have better
access to information and live in a social context wherein
they are better able to protect themselves from acquiring
HIV. It is probably for the same reasons that in both the
2002 HSRC study and our workplace studies, it was
reported that managers have thelowest prevalence of HIV.
Asregardsthegrowth rateof the HIVepidemic within the
working population, it appears to be stabilizing. Multi-
variate analysis (in contrast to the crude data) showed that
between 1999/2000 and 2001/2002, there was not a
significant growth rate over the 2002/2004 period.
HIV prevalence studies in the general population [7–9]
typically find that the prevalence of HIV is significantly
higher among women than men. The reverse was found
in this study, and in Evian’s study the prevalence was
significantly lower among women. The possibility exists
that being employed may be protective for women, and
this hypothesis warrants further investigation.
reduces some of the variation in HIV prevalence between
companies, substantial differences between companies
remain. One implication of this finding is that HIV
cannot be accurately modelled for a particular company
purelyon the basis of knowing the demographicstructure
of the workforce.
Of interest is whether the variation in HIV prevalence by
conditions), or whether company is a proxy measure for
some as yet unknown factor. Future workplace research
should evaluate workplace prevention programmes and
risky sexual behaviours and a reduction in the incidence
Organizations self-selected the commissioning of HIV
prevalence studies and so the representivity of sectors
cannot be assumed. Although participation levels were
relatively high, non-participation bias may have occurred.
It is speculated that those who are HIV positive or suspect
that they may be, are less likely to participate in a
seroprevalence study. Employees who were off sick at the
workforce may be biased downwards. Finally, as more and
more people come to know their HIV status, the
probability of such bias occurring increases. This bias
may also be differential by race because, probably, more
showed that 47% of whites and 35% of Indians compared
with 26% of Africans have had an HIV test .
The factor used to adjust the standardized prevalence for
geographical and temporal change assumes that the HIV
shape to that of largely unemployed pregnant African
women. The increased variability introduced by this
factor suggests the possibility of differing epidemic curves
for different populations, which needs to be examined
Sponsorship: All survey costs were paid for by the
individual organizations concerned.
Conflicts of interest: None.
AIDS 2007, Vol 21 (suppl 0)
Table 5. Comparisons of crude HIV prevalence levels as reported by
this and other studies.
DatasetNHIV prevalence (%)
This workplace data
Evian workplace studya
Educators study 2004/2005c
Antenatal survey 2003e
HSRC, Human Sciences Research Council.
aThe HIV prevalence study by Evian et al.  was conducted in
workplaces in three southern African countries during 2000–2001. In
this table, only the South African companies were included .
bThis study was conducted in 2002 at Anglo platinum mines and sites
in three provinces in South Africa .
cThe Educators Study was an HIV prevalence study conducted
among public sector teachers  employment were included [4,5].
dThe two studies conducted in 2002 and 2004 were national
community-based HIV prevalence surveys [4,5]. In this table only
respondents who claimed to be in full time, formal employment were
is conducted by the Department of Health .
QAD/21s301; Total nos of Pages: 7;
1. UNAIDS/WHO. Guidelines for using HIV testing technologies
in surveillance. ISBN 0173-92-063-7. Geneva, Switzerland;
in South Africa – results of a national, community-based survey.
S Afr Med J 2004; 94:776–781.
Irving and Johnson Limited. The Labour Court of South Africa.
C1126/2002. In: The Labour Court of South Africa, Cape Town,
South Africa; 2002.
Evian C, Fox M, Macleod W, Slotow SJ, Rosen S. Prevalence of
HIV in workforces in southern Africa 2000–2001. S Afr Med J
prevalence testing – merits, methodology and outcomes of a
survey conducted at a large mining organisation in South Africa.
S Afr Med J 2006; 96:134–139.
ShisanaO, SimbayiL.Nelson Mandela/HSRCStudyofHIV/AIDS:
Media Household Survey 2002. Cape Town: HSRC Press; 2002.
South African National HIV Prevalence, HIV Incidence, Behaviour
Abdool-Karim Q, Abdool-Karim SS. The evolving HIV epidemic
in South Africa. Int J Epidemiol 2002; 31:37–40.
ShisanaO, PeltzerK, Zugu-Dirwayi N, LouwJS.The healthofour
educators: a focus on HIV/AIDS in South African public schools.
Cape Town: HSRC Press; 2005.
HIV in South African workplaces Colvin et al. S7
Manuscript No. 21s301
During the preparation of your manuscript for typesetting, some queries have arisen. These
are listed below. Please check your typeset proof carefully and mark any corrections in the
margin as neatly as possible or compile them as a separate list. This form should then be
returned with your marked proof/list of corrections to the Production Editor.
QUERIES: to be answered by AUTHOR/EDITOR
AUTHOR: The following queries have arisen during the editing of your manuscript. Please
answer the queries by marking the requisite corrections at the appropriate positions in the
Typeset by Thomson Digital
for Lippincott Williams & Wilkins
Please provide history date.
Please provide some keywords.
Table 5 – please clarify meaning of this
phrase – it doesn't make sense as it is – is
Table 5 – please check the references
cited in the table footnote – I don't think
they were originally correct.
Fig. 2 – please explain ANC here.