Rates and causes of child mortality in an area of high HIV prevalence in rural South Africa
Corresponding Author, Anupam Garrib, Department of Public Health Medicine, Nelson R. Mandela School of Medicine, Private Bag 7, Congella 4013, South Africa. Tel.: +27 31 2604287; Fax: +27 31 2604211; E-mail: email@example.com
Objective To determine child mortality rates in a rural area of South Africa with high HIV prevalence.
Methods A community-based survey was conducted between 1 January 2000 and 31 December 2002 on deaths in children under the age of 15 years. Children were followed up through four monthly home visits. Cause of death was ascertained by verbal autopsy. Rates were calculated using Poisson regression.
Results Mortality ratios were 59.6 deaths per 1000 live births for infants and 97.1 for children under 5 years of age. Infant and under-5 mortality rates were, respectively, 67.5 and 21.1 deaths per 1000 person-years. HIV/AIDS was attributed to 41% of deaths in the under-5 age group, with a mortality rate of 8.6 per 1000 person-years. Lower respiratory infections caused an estimated 24.9 deaths per 1000 person-years in children under 1 year of age.
Conclusions In rural South Africa, infant and child mortality levels are high, with HIV/AIDS estimated as the single largest cause of death. Interventions to reduce child mortality are required urgently.
Objectif Déterminer les taux de mortalité infantile dans une zone rurale d'Afrique du sud à haute prévalence du VIH.
Méthode Etude de communauté menée entre le 1er janvier 2000 et le 31 décembre 2002 sur les décès d'enfants de moins de 15 ans. Les enfants ont été suivis par des visites dans les familles effectuées quatre fois par mois. La cause de décès a été confirmée par une autopsie verbale. Les taux ont été calculés en utilisant la régression de Poisson.
Résultats Les proportions de mortalitéétaient de 59,6 pour 1000 naissances viables pour les jeunes enfants et de 97,1 pour ceux de moins de 5 ans. Les taux de mortalité pour les jeunes enfants et ceux de moins de 5 ans étaient respectivement de 67,5 et de 21,1 pour 1000 personnes années. 41% des décès étaient attribués au VIH/SIDA chez les moins de 5 ans, équivalent à un taux de mortalité de 8,6 par 1000 personnes années. Les infections respiratoires basses contribuaient à environ 29,4 de décès pour 1000 personnes années chez les enfants de moins de 1 an.
Conclusions En milieu rural d'Afrique du sud, les taux de mortalité infantile et ceux des jeunes enfants sont élevés, le VIH/SIDA estimé comme étant la cause unique la plus importante. Des interventions visant à réduire la mortalité infantile sont urgemment nécessaires.
Objetivo Determinar las tasas de mortalidad infantil en un área rural de Sudáfrica con una alta prevalencia de VIH.
Método Investigación basada en la comunidad sobre las muertes de niños menores de 15 años, conducida entre el 1 de enero del 2000 y el 31 de Diciembre del 2002. Los niños fueron seguidos mediante visitas a sus casas cada 4 meses y las causas de muerte se determinaron mediante autopsias verbales. Las tasas se calcularon con una regresión de Poisson.
Resultados Las razones de mortalidad fueron de 59.6 muertes por 1000 nacidos vivos para lactantes y de 97.1 para los menores de 5 años. Las tasas de mortalidad para lactantes y menores de 5 años fueron, respectivamente, 67.5 y 21.1 muertes por 1000 personas año. Se atribuyó un 41% de las muertes de menores de 5 años al VIH/SIDA, con una tasa de mortalidad de 8.6 por 1000 personas año. Se estimó que las infecciones del tracto respiratorio bajo causaron 24.9 muertes por 1000 personas año en niños menores de 1 año.
Conclusiones En Sudáfrica rural, los niveles de mortalidad en lactantes y niños son altos, estimándose que el VIH/SIDA es por sí solo la principal causa de muerte. Se requieren, con urgencia, intervenciones para reducir la mortalidad infantil.
Dramatic reductions in child mortality rate occurred in the twentieth century in both developed and developing countries. These gains are now being eroded in many developing countries by the progression of the HIV/AIDS epidemic, such that the decline in child mortality rates has slowed considerably in some countries and begun to reverse in others (Ahmad et al. 2000).
Information on rates and causes of child mortality is essential to developing appropriate policy and intervention strategies to improve child health. However, in most of Africa, this is constrained by a lack of reliable data, with a particular gap in child health data from the least developed countries with the highest levels of child mortality (Rudan et al. 2005). Less than 10% of Africa is covered by death registration systems, and the deaths of infants and younger children are less likely to be reported than the deaths of adults (Jewkes & Wood 1998; Mathers et al. 2005). The information on cause-specific mortality patterns in sub-Saharan Africa is also generally of questionable quality (Rudan et al. 2005). For those who die at home, the death is unlikely to be attended by medical personnel, making the information on death certificates unreliable. At health facilities, diagnostic facilities are limited and record keeping is poor.
South Africa is a middle-income country (per capita GNI of $3650 per year) with extreme levels of inequality, which is experiencing a rapid spread of a generalized HIV/AIDS epidemic. Significant improvement in death reporting in the country has occurred, and in 2000, it was estimated that approximately 90% of adult deaths were registered, although the proportion of child deaths reported is uncertain (Dorrington et al. 2001). In Africa, population-based surveys are essential for estimating child mortality rates and cause-specific mortality patterns. In this study, we used an established demographic surveillance system for determining mortality rates, and a verbal autopsy (VA) method to estimate cause of death, in a rural area of South Africa with high HIV prevalence.
The Africa Centre Demographic Information System (ACDIS) is an established demographic surveillance site in northern KwaZulu-Natal, South Africa, as described previously (INDEPTH Network 2002). Approximately, 11 000 inhabited homesteads in a 435-km2 area have been mapped, comprising around 90 000 people, 65 000 of whom are residents and 25 000 non-residents. A household within ACDIS was a self-defined social unit, with individuals being resident or non-resident members. A resident member was someone who had spent most nights in the surveillance area since the last fieldworker visit. Visits to households were conducted three times a year, tracking both resident and non-resident members. Births, deaths, family composition, economic circumstances and health are monitored in the longitudinal surveillance of this population.
The surveillance site is predominantly rural with a small peri-urban pocket in the southern part. In 2000, 38% of households had access to piped water, 39% reported having no toilet facilities and unemployment rates were 39% (Chimbwete & Herbst 2003). Twelve per cent of the population was under the age of 5 years.
The surveillance area has six government primary health care clinics run by nurses and a district hospital approximately 40 km away. Health care in the public sector has been free for pregnant women and children under the age of 6 years since 1994. Utilization of health services in this community is high (Case et al. 2005). In 1997, it was estimated that 91% of women in the area attended an antenatal clinic at least once during their pregnancy, 83% delivered in a health facility, and 76% of children had received all recommended vaccine doses for a child under the age of 1 year (Wilkinson et al. 1997). A programme providing nevirapine for the prevention of mother-to-child transmission (PMTCT) of HIV began in December 2001.
In the study area, HIV prevalence among women attending public antenatal clinics was estimated to have increased from 6.9% among women aged 20–24 in 1992 to 50.8% in the same age group by 2001 (Abdool Karrim & Abdool Karim 2005). In 2003, KwaZulu-Natal province had the highest HIV prevalence rate (38%, 95% CI: 35–40) among attendees at public antenatal clinics in South Africa (Department of Health 2003). Published data on the causes of death in adults from this site showed that HIV/AIDS was the leading cause of death among adults, responsible for 48% of deaths in adults over the age of 15 years and 74% of deaths among women aged between 15 and 44 (Hosegood et al. 2004).
Determining cause of death using verbal autopsies
Deaths were reported through routine home visits. A professional nurse then visited the household and conducted a structured VA interview with a caregiver to the deceased or someone who witnessed at close hand the progression of the terminal illness, preferably the mother. Verbal informed consent was taken for the interview. Interviews were conducted in the local language, Zulu, and transcribed onto the questionnaire by the VA nurse. If any health records were available, information on these was copied onto the questionnaire.
There were three sections to the VA questionnaire: an open narrative section in which the informants were asked to relate in their own words the course of the illness, a checklist of signs and symptoms and structured age-specific questionnaires enquiring in detail about specific signs and symptoms. The questionnaires were analysed independently by two physicians who attributed at least an underlying and, if possible, also contributory causes of death to each case. Criteria for common causes of death were provided as a guide to the physicians reviewing the records, but this was not prescriptive and the physicians were asked to use their clinical judgment in attributing causes of death. An ‘undetermined’ cause of death was assigned, if the doctors could not come to agreement on the cause of death or if there was inadequate information on which to make a diagnosis. Only a single underlying cause of death per case was used in this analysis.
Persons included in this analysis were all children under the age of 15 years, who were ACDIS members, resident or non-resident, between 1 January 2000 and 31 December 2002. All deaths occurring in ACDIS during the same period, to children who were under the age of 15 years at the time of death, were included. Stillbirths were excluded from the analysis.
The observation period began at the start of membership of a household in the study area, which would have occurred at the start of the study, birth, or in-migration into the area. Follow-up time ended at the end of the study period, on reaching 15 years of age, or at the end of membership of a household in the study area, which would have occurred at death or out-migration from the area. The numbers of live births per year were calculated from data collected on the outcomes of pregnancies identified in ACDIS.
Categorical data were compared between groups using chi-square tests in univariate analyses. Overall and cause-specific mortality rates, stratified by age at death, were calculated using Poisson regression methods. The following standard age categories were used in the analysis and were defined a priori: neonatal period from birth to 28 days, post-neonatal period from 29 days to 1 year, 1 to 4 years, 5 to 9 years and 10 to 14 years. In addition, where appropriate, age groups were combined for analysis into an under-1 age group and an under-5 age group. Infant and under-5 mortality ratios were also calculated using live births as denominator, reflecting the probability of death in these age groups, and were included for comparison with other published data. These are referred to as mortality ratios. Data were analysed using stata 7 (Stata Corporation, College Station, TX, USA).
Both the demographic surveillance and the VA component were approved by the Research Ethics Committee of the University of KwaZulu-Natal.
In the cohort these were 46 192 individuals under the age of 15 years and a total of 102 120 person-years of follow-up. There were 792 deaths in children, of which 52% (414) were in boys. Of the 87 neonatal deaths, most (82%) occurred in the first week of life. The neonatal deaths accounted for 13% of the 685 deaths in children under the age of 5 years. Infant deaths accounted for 61% of deaths in children under the age of 5 years and deaths in children under the age of 5 years accounted for 87% of all deaths.
Over the 3-year period, neonatal, infant and under-5 mortality rates (95% CI) per 1000 person-years were, respectively, 183.2 (148.5–226.0), 67.5 (61.3–74.2) and 21.1 (19.5–22.7) (Table 1). Mortality rates were lower in girls than in boys in all age groups. In 2002, this difference was significant in the under-1 and under-5 age groups with rate ratios (95% CI) of 1.4 (1.0–2.1) and 1.3 (1.0–1.7), respectively. There was no significant increase in mortality rates over the 3-year period for both the under-1 and under-5 age groups. Using the number of live births per year as the denominator, the overall neonatal mortality ratio was 12.3 deaths per 1000 live births; infant mortality ratio was 59.3 deaths per 1000 live births per year, while the under-5 mortality ratio was 97.0 deaths per 1000 live births per year (Table 2).
Table 1. Mortality rates (95% CI) per 1000 person-years by age group and sex in ACDIS, 2000–2002
|Neonatal||39||232||168.1 (122.8–230.0)||48||243||197.6 (148.9–262.2)||87||475||183.2 (148.5–226.0)|
|Post-neonatal||159||2835||56.4 (48.3–65.9)||173||2916||59.3 (51.1–68.8)||332||5752||57.9 (52.0–64.5)|
|1–4 years||126||13 218||9.5 (7.9–11.3)||140||13 086||10.7 (9.1–12.6)||266||26 303||10.1 (8.9–11.4)|
|5–9 years||41||17 244||2.4 (1.8–3.2)||37||17 290||2.1 (1.6–3.0)||78||34 535||2.3 (1.8–2.8)|
|10–14 years||13||17 571||0.7 (0.4–1.3)||16||17 483||0.9 (0.6–1.5)||29||35 055||0.8 (0.6–1.2)|
|All ages combined||378||51 100||7.4 (6.7–8.2)||414||51 019||8.1 (7.4–8.9)||792||102 120||7.8 (7.2–8.3)|
|Under-1||198||3067||64.9 (56.5–74.6)||221||3160||69.9 (61.3–79.8)||419||6227||67.5 (61.3–74.2)|
|Under-5||324||16 284||19.9 (17.8–22.1)||361||16 246||22.2 (20.0–24.6)||685||32 530||21.1 (19.5–22.7)|
Table 2. Neonatal, infant and under-5 mortality ratios (number of deaths) per 1000 live births in ACDIS, 2000–2002
|Number of live births||2532||2358||2173||7063|
|Neonatal mortality rate† (deaths)||11.1 (28)||8.9 (21)||17.5 (38)||12.3 (87)|
|Post-neonatal mortality rate‡ (deaths)||48.6 (123)||50.0 (118)||41.9 (91)||47.0 (332)|
|Infant mortality rate§ (deaths)||59.6 (151)||58.9 (139)||59.4 (129)||59.3 (419)|
|Under-5 mortality rate¶ (deaths)||86.9 (220)||98.0 (231)||107.7 (234)||97.0 (685)|
Attributed causes of death
Birth asphyxia (35.6%) and low birthweight and pre-maturity (26.4%) were recorded as the leading causes of death in the neonatal period. Most of these deaths occurred in the early neonatal period, with 44% occurring on the day of birth (Table 3).
Table 3. Causes of death by age group, ACDIS 2000–2002
|Birth asphyxia and trauma||31 (36%)||LRTI||143 (43%)||HIV/AIDS||162 (61%)||HIV/AIDS||26 (33%)||HIV/AIDS||5 (17%)|
|Low birthweight||23 (26%)||HIV/AIDS||113 (34%)||Diarrhoeal diseases||17 (6%)||Road traffic accidents||11 (14%)||Malignancies||4 (14%)|
|LRTI||12 (14%)||Diarrhoeal diseases||12 (4%)||LRTI||16 (6%)||Meningitis||6 (8%)||Other infectious and parasitic||4 (14%)|
|Other perinatal conditions||6 (7%)||Septicaemia||12 (4%)||Protein energy malnutrition||16 (6%)||Drownings||5 (6%)||Road traffic accidents||4 (14%)|
|HIV/AIDS||4 (5%)||Undetermined||12 (4%)||Undetermined||14 (5%)||Epilepsy||4 (5%)||Meningitis||3 (10%)|
|Undetermined||4 (5%)||Protein energy malnutrition||9 (3%)||Congenital anomalies||8 (3%)||Undetermined||4 (5%)||Drownings||2 (7%)|
|Congenital anomalies||3 (3%)||Congenital anomalies||6 (2%)||Other unintentional injuries||7 (3%)||Other infectious and parasitic||4 (5%)||Tuberculosis||2 (7%)|
|Other infectious and parasitic||2 (2%)||Meningitis||6 (2%)||Meningitis||5 (2%)||Other unintentional injuries||4 (5%)||Cardiovascular diseases||1 (3%)|
|Protein energy malnutrition||1 (1%)||Other unintentional injuries||6 (2%)||Road traffic accidents||5 (2%)||Protein energy malnutrition||4 (5%)||Diarrhoeal diseases||1 (3%)|
|Septicaemia||1 (1%)||Tuberculosis||3 (1%)||Malignancies||3 (1%)||LRTI||3 (4%)||Epilepsy||1 (3%)|
| || ||Birth asphyxia and trauma||2 (1%)||Other infectious and parasitic||3 (1%)||Digestive diseases||2 (3%)||Other neoplasms||1 (3%)|
| || ||Low birthweight||2 (1%)||Homicide and violence||2 (0.8%)||Congenital anomalies||1 (1%)||Respiratory diseases||1 (3%)|
| || ||Other infectious and parasitic||2 (1%)||Septicaemia||2 (0.8%)||Genitourinary diseases||1 (1%)|| || |
| || ||Cot death||1 (0.3%)||Tuberculosis||2 (0.8%)||Malignancies||1 (1%)|| || |
| || ||Drownings||1 (0.3%)||Birth asphyxia and trauma||1 (0.4%)||Respiratory diseases||1 (1%)|| || |
| || ||Genitourinary diseases||1 (0.3%)||Digestive diseases||1 (0.4%)||Septicaemia||1 (1%)|| || |
| || ||Road traffic accidents||1 (0.3%)||Drownings||1 (0.4%)|| || || || |
| || || || ||Respiratory diseases||1 (0.4%)|| || || || |
|Total||87 (100%)|| ||332 (100%)|| ||266 (100%)|| ||78 (100%)|| ||29 (100%)|
In the post-neonatal period (29 days to 1 year), lower respiratory tract infection (LRTI) was attributed to 43% of deaths and HIV/AIDS to 34% of cases. HIV/AIDS remained the commonest assigned cause of death in the 1–4 years age group (61%). Injury-related deaths became more common in the 5–9 and 10–14 years age groups, although HIV/AIDS remained the most common attributed cause of death. Road traffic accidents and drowning were the most commonly attributed causes of accidental deaths. Seventy-six per cent (16/21) of the deceased in road traffic accidents were pedestrians, most of these were in the 5–9 years age group. The proportions of deaths due to non-communicable diseases were the highest in the 5–9 and 10–14 years age groups.
Overall, for all children under the age of 15 years, HIV/AIDS was the most common attributed cause of death (39%, 310/792). Most of the remaining deaths (45%, 357/792) were attributed to communicable, nutritional and perinatal conditions, with LRTI attributed to 174. Five per cent (42/792) of deaths were attributed to non-communicable diseases, 6% (49/792) to injuries and 4% (34/792) of cases were assigned an undetermined cause of death.
Mortality rates because of HIV/AIDS varied significantly between age groups, ranging from 19.8 (95% CI: 16.5–23.8) deaths per 1000 person-years of observation in the post-neonatal age group to 0.1 (95% CI: 0.06–0.3) deaths per 1000 person-years of observation in the 10–14 years age group. For all children under the age of 5 years, HIV/AIDS was attributed to 8.6 (95% CI: 7.6–9.6) deaths per 1000 person-years of observation.
Mortality rates (95% CI) attributed to LRTI were 24.9 (21.3–29.1) deaths per 1000 person-years of observation in the under-1 age group, 4.9 (4.0–6.1) deaths per 1000 person-years for children under 5 years of age and 1.7 (1.5–2.0) deaths per 1000 person-years of observation for children under the age of 15 years. In each year, the highest number of LRTI-attributed deaths occurred in the winter months of July or August. For the 3 years combined, the largest proportion of LRTI deaths occurred in the month of July. Overall, diarrhoea was only attributed to 4% of deaths, 6% in the 1–4 years age group. The highest mortality rate because of diarrhoea was in the post-neonatal age group with a rate of 2.1 (1.2–3.7) deaths per 1000 person-years.
This study shows high levels of child mortality, with infant and under-5 mortality ratios of 60 and 97 deaths per 1000 live births, highlighting HIV/AIDS as a priority condition. Levels of child mortality in South Africa are increasing and much of this has been attributed to the impact of the HIV epidemic. The lack of readily available health information from developing countries, critical for the management of child survival programmes, remains a serious problem. This study contributes empirical data, and addresses a gap in available data on child health in rural South Africa.
The neonatal mortality ratio in this population is low at 12.6 deaths per 1000 live births. This could be a reflection of the high rate of utilization of services for maternal health in the population; however, it could also be that some neonatal deaths have been missed. Contrary to most other sub-Saharan countries, the majority of women in the area deliver within health facilities. Despite this, birth asphyxia and birth trauma were the most commonly attributed causes of death in neonates.
In South Africa, few studies have estimated national and provincial child mortality rates (Table 4). Estimates that are available show that the mortality rates in the KwaZulu-Natal province are higher than the national estimates. In this study, infant and under-5 mortality ratios are higher than those calculated by the Demographic and Health Survey, but lower than those calculated by the South African National Burden of Disease Study for the KwaZulu-Natal Province. These studies have used very different methods and results are difficult to compare. While it is possible that mortality rates have been increasing rapidly because of the effect of the HIV/AIDS epidemic, continued monitoring is necessary to determine their trend.
In 39% of deaths in children under the age of 15 years, HIV/AIDS was assigned as the cause of death, and a similar proportion in the under-5 age group. This is similar to the results of the South African National Burden of Disease Study (Bradshaw et al. 2003). As discussed, KwaZulu-Natal province and the study site have very high HIV prevalence among pregnant women. The proportion of deaths because of HIV/AIDS is, therefore, not surprising, as this is a major determinant of the impact of the epidemic on child mortality. With the commencement of the PMTCT programme, the number of deaths because of HIV/AIDS in children should reduce in the coming years, although this study did not have the power to detect this. Continued monitoring of the infant and under-5 mortality rates at the population level will provide an indication of the effectiveness of the programme, as these rates should decrease with increased coverage of the programme.
A similar demographic site in the Limpopo Province of South Africa, estimated an under-5 mortality ratio of 56 deaths per 1000 live births between 1998 and 2000. They found, using VA, that HIV/AIDS was attributed to 22% of deaths in children under the age of 5 years; however, HIV prevalence among women attending public antenatal clinic in that province was much lower than that in KwaZulu-Natal, 13.2% (11.7–14.8) during 2000 (Agincourt Health and Population Unit 2002; Department of Health 2003).
The contribution of HIV/AIDS to overall under-5 mortality in sub-Saharan Africa is increasing, with the greatest rise in the countries of southern Africa. Mortality of HIV-infected children is high, and there is evidence to suggest that the mortality of children of infected mothers is higher than the mortality of children whose mothers are not infected (Adetunji 2000; Newell et al. 2004). In a high HIV prevalence area such as the KwaZulu-Natal Province, strengthening the programme for the PMTCT of HIV is key to preventing a large proportion of child deaths. The provision of antiretroviral therapy is also important, both to infected children and to infected women.
Diarrhoea and pneumonia remain among the most common causes of deaths in children in sub-Saharan Africa, while the proportions of deaths because of malaria, AIDS and neonatal disorders varies significantly between countries in the region (Black et al. 2003). This is demonstrated in a recent comparative analysis of mortality from several demographic surveillance sites in Sub-Saharan Africa (Adjuik et al. 2006). Lower respiratory infections were found to be a common cause or contributory factor in childhood deaths in this study and were attributed as a cause of death in 22% of cases. This suggests that there is a need for targeted interventions to improve the management of lower respiratory infections in the area. Diarrhoea was also a commonly attributed cause of death and was attributed to 4% of deaths in children under the age of 5 years, much lower than the 10% estimated in the South Africa National Burden of Disease Study. However, 4% is probably an underestimate of the true burden of diarrhoea, as it is a common contributory factor in AIDS deaths. As multiple causes of death were not included in this analysis, the true burden of diarrhoea may not be reflected in the results. The same is probably also true for deaths because of malnutrition.
As much of sub-Saharan Africa is without routine systems to monitor cause-specific mortality patterns, it is important to further develop tools that can be used to determine cause of death where such information is not available. VAs have been used within surveillance systems to generate cause-specific mortality data in many countries where routine systems are incomplete. The limitations of this methodology have been discussed extensively in the literature, particularly for malaria and LRTI, where their sensitivity and specificity against hospital diagnosis has been shown to be very low in some settings (Snow et al. 1992; Anker et al. 1999; Kahn et al. 2000). There is little information on the validity of VAs to distinguish HIV deaths in children or the use of VAs in high HIV prevalence areas (Dowell et al. 1993). Deaths because of HIV/AIDS may have similar signs and symptoms as deaths because of malnutrition or tuberculosis, and could be misclassified. Without a validation study of the VA, it is difficult to estimate the true burden of HIV-related deaths in this population. However, the consistency with the results of the South African National Burden of Disease Study in terms of the proportion of deaths because of HIV is supportive of the findings of this study.
At the best, these data give an indication of broad trends and provide a baseline against which changes in cause-specific mortality can be assessed in the future. Monitoring of overall mortality rates as the HIV/AIDS epidemic matures will allow the assessment of the impact of the epidemic, as well as the impact of interventions to address the epidemic. In conclusion, our study suggests that AIDS is now the leading cause of death in children under the age of 15 years in rural South Africa. Deaths from LRTI are also high in this setting.
Analysis was based on data collected by the Africa Centre for Health and Population Studies, University of KwaZulu-Natal through the ACDIS. We thank the ACDIS field and data centre staff for their contributions and support. We thank the community for generously sharing information with us during a difficult time for their families. The VA study was designed and implemented by Dr A Vanneste. Members of the VA team, Thembeka Mngomezulu, Mendiswa Thusi, Faith Madela, Themba Dumisa, Wandile Manqele, Sabelo Msweli, are also acknowledged. We thank Drs Johannes Viljoen, Peter van Gelderen Stort, Victoria Mubaiwa, Kalavani Moodley and Tselane Thebe for the consistent high quality of their work. We also thank A Stein and V Hosegood for helpful discussions.