Survey: Killing fields: Mortality in South Africa

Uncounted graves

South Africa’s indifferent quality of statistics has been disguising an alarming reality of death in our society. Although the registration of deaths is improving, neither annual births nor deaths are as yet accurately reported and recorded in South Africa. Demographers have to resort to complex estimation and modelling.

In September 2001 the President of the Medical Research Council, Dr Malegapuru Makgoba, created discomfort in official circles when he reported on the Council’s finding that in the year 2000 Aids-related deaths had risen to 40 per cent of adult mortality, becoming the biggest single cause of death and significantly increasing overall mortality¹.

The results of a special study conducted by the Helen Suzman Foundation, in association with MarkData, (see Field Study text at end) more than confirm the MRC findings, suggesting that mortality rates are very much higher than the best estimates hitherto made.

The full study will be published as a report in due course but the first findings show that South Africa is experiencing a monumental tragedy of early death within its families and communities.

Aids is a major contributor but, because of early deaths through violence, it is not the only nemesis.

Measuring, modelling and guessing — mortality rates in South Africa

Notwithstanding inadequate death registration over past decades, demographers had estimated a steady decline in mortality in the South African population up to the early nineties. The incidence of death is usually expressed as a simple mortality rate, namely the number of deaths per annum per 1000 of the population. For example, the demographers Mostert, Oosthuizen and Hofmeyr² tracked the estimated decline in mortality from nearly 20 per 1000 in 1945 down to just over 8 per 1000 in 1990.

Hence by the late eighties, advances in medicine and improvements in public health and living conditions had driven mortality in South Africa down to below the average for Africa (around 12 per 1000) and to a level close to the world average. At that stage mortality was not expected to decline further because it was reaching its biological limits and because the steady ageing of the population would tend to increase the incidence of mortality correspondingly. All else being equal, mortality rates were expected to rise very gradually from around 8 per 1000 towards the average for Europe of roughly 10 per 1000.

All else was not equal, however, and in the nineties HIV/Aids started to take its toll. Demographers started to revise their estimates. By 1998/99 Van Aardt et al, The Institute for Futures Research, ING-Barings (WEFA) and others estimated that mortality was already rising significantly³. The estimates of the crude mortality rate among these authors varied from around 10 to over 13 per 1000 in the late nineties. The Institute for Futures Research expected it to increase from 11,7 per 1000 for 1996 — 2001 to just below 14 per 1000 until 2016, after which it would fall gradually.

In 2001 the Medical Research Council (MRC) undertook an extensive review of available mortality records and related these to both additional administrative data in the Department of Home Affairs and to the ASSA 600 demographic model developed by the Actuarial Society for population forecasting purposes.

The MRC was able to adjust the mortality records for incompleteness. On this basis it estimated that by 1999/2000 there were 412 000 adult deaths per annum (15 years and over). It concluded that, while the population had grown by 37 per cent over the preceding decade, mortality had risen by some 73 per cent largely because of the rise in Aids-related deaths⁴. Taking the estimate of 412 000 adult deaths and adding an estimate of deaths up to 14 years⁵, suggests that there were some roughly 492 000 deaths among all ages in 1999/2000. Relating this figure to the midyear estimates of the population of Stats SA (an average population of 43.37 million for 1999 and 2000) suggests that the mortality rate at that time was roughly 11.3 per 1000, just under the estimate of the Institute for Futures Research.

Against this background, and allowing for rising Aids-related deaths, our sample survey in March 2002 was expected to reflect current mortality running at between 12 and 14 deaths per 1000 people. As it turns out, this expectation was way too low.    

Spiralling mortality today

In response to the survey questions on deaths over the preceding 12 months, 8.9 per cent of households reported one or more deaths in the period. Some 1.1 per cent of households reported two deaths and there was a single case in the sample of three deaths. Altogether 214 deaths were reported. The 2164 households in the sample encompassed 9457 people of all ages, to which one must add the people who died, yielding a sample population of 9671. The result therefore was 214 deaths among 9671 people, or a mortality rate of no less than 22,1 per 1000 — almost double the expected rate. 

The MarkData interviewers reported that there was discomfort among some respondents about questions concerning deaths in the household and they felt that some deaths were concealed. Consequently we assumed that the real number of deaths in the sampled households might be slightly under-represented.

The interviewers asked about deaths over the past 12 months and specified the period as being from the appropriate date in March 2001 up to the date of the interview. The interviewers were careful to communicate the time period very thoroughly. It is possible, however, that a few people might have included deaths occurring earlier than 12 months prior to the interviewing. It is also possible, however, that a few others would have omitted to mention deaths occurring as far back as 12 months. We have assumed that any errors in the dating of deaths would be roughly self-cancelling. 

The estimated mortality rate of 22,1 per 1000, being based on a random sample, is subject to a margin of possible statistical error, which can be estimated. This calculation⁶ suggests that the “true” mortality rate could lie somewhere between 19,2, and 25,0, per 1000 people. Even the lower limit is very high indeed.

We can also estimate the total number of deaths in the period March 2001 to March 2002 from the sample, allowing for sampling error as above. The estimated total number of deaths in the South Africa population from March 2001 to March 2002 was between some 1 127 000 and 866 000, with a midpoint estimate of 996 500 or around one million.  

In effect these rates mean that mortality in South Africa has been dragged back to levels not seen since WW2 when medical science and living conditions were far inferior to what they are today. In Table One historical trends estimated by Mostert, van Tonder and Hofmeyr are compared with the range suggested by the current survey.

What kills South Africans?

We asked respondents what had caused the death or deaths that they reported. Obviously their perceptions are not medically precise but most of them knew what broad condition or event had caused the death. The causes they gave are noted in Table Two.

Some of these causes are what one would expect. Aids as such is seldom the immediate cause of death because it strikes through lowering immunity to other diseases. Particularly the categories above labelled tuberculosis (TB), pneumonia, bronchial and lung ailments as well as “allergies” are the typical (but not the only) conditions from which Aids sufferers eventually die.

With Aids and other STDs in the table they account for some 24 per cent of cases and some 38 per cent of deaths among people of 15 years and older. The substantial contribution that these conditions make to mortality is suggestive of high Aids mortality at this stage. Some of the deaths from the conditions in the categories mentioned would not be due to Aids, but on the other hand some other diverse illnesses could be Aids related. The indications from these figures are compatible with the trends in the MRC report quoted earlier, which concluded that some 40 per cent of deaths in 2000 were due to Aids.

The survey results also reflect a high level of deaths due to external causes (accidents, poisoning and other violent deaths). These unnatural causes amount to no less than 14 per cent of the deaths reported, and as such are indicative of very turbulent conditions in our society. Causes of death were last officially reported in 1996, and at that stage external causes of death amounted to 18 per cent of mortality. The drop in the relative incidence of unnatural deaths is no doubt due to the increase in deaths from a variety of Aids-related causes.

Who is dying?

The survey estimates broken down by province appear in Table Three.

The differences between provinces are due to various factors and not only differences in HIV infection. It is interesting to note that the patterns in the table show a rough resemblance to the pattern of HIV infection as captured in the Department of Health annual surveys among women attending public antenatal clinics, with KwaZulu-Natal leading and relatively low rates in the Northern Cape, Western Cape and Limpopo⁷.

The survey results produce a lower than expected level of mortality in Mpumulanga which is not supported by the trends in respect of HIV infection, and may therefore be due to statistical error in the small sub- sample in that province.

The pattern of deaths by age of the deceased is presented in Table Four. The survey estimates are compared with the pattern established by the Medical Research Council for 2000 based on the registered deaths adjusted for missing identity documents⁸.

The comparison between the figures for 2000 and 2002 has to be approached cautiously because the latter is based on a sample and is subject to sampling error. Nevertheless, the differences are interesting and suggestive of the mounting effect of Aids-related deaths.

In the 2002 estimates deaths of people between 60 and 79 — when non-Aids related illness is most common — tend to be lower than those recorded in the 2000 figures. The exception here is the high percentage of deaths among people who are 80 years of age or older.

But the expected trend in most societies where modern medicine is practised is for the age of death to rise over time. Part of the difference could also be due to sampling errors and part of it due to the fact that the ages of very old people in traditional, semi-literate communities may be over-estimated by their relatives because their dates of birth may be unknown.

One notes in the 2002 figures that deaths in the 0 to 14 year range are also lower. This is another cohort in which Aids-related deaths are uncommon. The 2002 figures suggest that mortality in the ages of highest Aids risk, 20-29, is significantly higher than in the 2000 MRC estimates.

Broadly, then, the results according to age would seem to indicate an increasingly serious impact of HIV/Aids in the economically active age range of 20 to 29 years. Young adults are clearly at very great risk. The genders of the deceased were:

     Male: 50,9 per cent

     Female: 49,1 per cent

The level of male deaths in the survey is slightly lower than the MRC adjusted records for 2000. This is probably due to the declining relative impact of deaths due to external causes (violence, accidents) as the impact of Aids-related deaths rises among both sexes.

The grim message

It would be tempting for readers to discount these results because they are based on a sample survey rather than on complete records of mortality. As has been noted, however, South Africa’s official records are problematic and incomplete. Furthermore, surveys provide reliable insights into broad trends in society and they are often essential in augmenting incomplete official statistics.

The validity of the survey methodology may also be questioned in respect of mortality data. Readers may wonder whether people interviewed tell the truth about deaths in their households.

But why should people want to lie? In the interviewing they not only reported on deaths but also provided detailed information on the ages, occupations, health conditions and other circumstances of the deceased family members. The deaths reported were not fabricated, although some deaths may have been omitted. The greatest danger of inaccuracy lies in the recall of the dates of death, but here the biases are probably self-cancelling.

What cannot be denied, however, is that the results show a very dramatic current increase in mortality, and that it is largely due to the mounting effects of Aids. Mortality may be up to twice as high as everyone thinks it is. The survey results may not be completely precise. But they are at the very least a warning that the silent biological bomb of HIV/Aids is already exploding in our midst. There can be no more vacillation. The crisis demands concerted action by government, the private sector and all concerned South Africans. 

The Field Study

As part of a national survey conducted in March 2002 to assess the impact of HIV/Aids on the family, a fully representative sample of 2164 households was probed on deaths that had occurred among members living in the households in the 12 months preceding the interviews. Questions were asked about the number of members of the specific household who had died in the period, their ages, gender, occupation and what the cause of the death was perceived to have been.

The sample was that of the regular MarkData Omnibus, a syndicated survey that is conducted three times a year. It was a two-stage stratified probability sample covering all types of housing in metropolitan, other urban, commercial farming and traditional rural areas, including shack areas and hostels.

Census enumerator areas stratified by province and nine categories of settlement type are selected in the first stage. In the second stage individual sites are selected randomly. Face to face interviews are conducted with a randomly selected adult respondent within households by experienced fieldworkers in the languages of choice of the respondents.

All interviewing occurs under senior field supervision and a minimum 15 per cent back check is carried out to ensure validity of the interviews. These samples produce results that correspond to the basic characteristics of the population reflected in official statistics and other major independent surveys.

Notes

1.  See Business Day, 17 September, 2001, and Rob Dorrington et al, The impact of Hiv/Aids on adult mortality in South Africa, Medical Research Council, September 2001.

2.  WP Mostert, JS Oosthuizen and BE Hofmeyr, (1998) Demograph: textbook for the South African student, Pretoria: HSRC.

3. Van Aardt, van Tonder and Sadie in Bureau for Market Research, UNISA, Pretoria, report 270, 1999; Institute for Futures Research, High, Medium and Low Projections of the South African Population, Stellenbosch: 1999; and Kristina Quattek/ WEFA/ ING-Barings, The Demographic Impact of Aids on the SA Economy, Johannesburg, 1999.

4. Dorrington et al, The impact of Hiv/Aids on adult mortality in South Africa, Medical Research Council, September 2001.

5.  Allowing for the same ratio of child to adult deaths as that in the ASSA 600 demographic model.

6. The calculation of sampling error has been based on the 214 reported deaths as a proportion of the sampled household population of 9671 (deaths included) at the 95 per cent level of confidence. The estimate of sampling error is approximate because the sample is not a simple random design. On the one hand the error will be larger because the sample is a two-stage and not a simple random design, but on the other hand the precision is improved by the fact that the sample was stratified (controlled) according to 9 settlement types in each of the 9 provinces.

7.   See South African Institute of Race Relations, South Africa Survey, 2001/2002, Johannesburg, 2001, page 293.

8.  See Appendix Table A4 in Rob Dorrington et al, The impact of Hiv/Aids on mortality in South Africa, Medical Research Council, September 2001.

 
Table 1

Crude mortality rates in South Africa: earlier estimates of Mostert et al compared with the current survey findings (rounded to nearest whole number) 

Western Cape      7,4 per 1000
Northern Cape     15,3
Eastern Cape       25,1
Free State            32,1
KwaZulu-Natal     36,8
Mpumulanga         13,5
Limpopo                12,8
Gauteng                24,7
North West            22,1