SEARCH

SEARCH BY CITATION

Keywords:

  • non-communicable diseases;
  • diabetes;
  • hypertension;
  • asthma;
  • epilepsy;
  • developing countries;
  • epidemiology;
  • capture–recapture

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Non-communicable diseases (NCDs) are becoming increasingly common and important in developing countries, yet their enumeration is problematic. We have attempted to enumerate NCD patients in a rural district of KwazuluNatal, South Africa, using the techniques of electronic data linkage and capture–recapture (CR). We examined four major NCDs (hypertension, diabetes, asthma and epilepsy). Basic patient details were recorded onto EpiInfo software over a 6-week period, from the main hospital clinic at Hlabisa, as well as the 10 outlying peripheral health clinics. Using electronic data linkage of lists from the main hospital, the peripheral clinics, and repeat prescription cards, a district NCD register was produced of 2455 patients. The mean age was 51 ± 16 years (1 SD) and 76% were female. Of the total NCD patients, 62% had hypertension (age 57 ± 12 years, 82% female), 16% epilepsy (age 35 ± 17 years, 49% female), 13% asthma (age 45 ± 19 years, 60% female) and 12% diabetes (age 54 ± 13 years, 61% female). Estimated population crude prevalence rates for known NCD cases on the register were 7.4% for hypertension, epilepsy 0.2%, asthma 0.2% and diabetes 0.2%. We also attempted a CR analysis to assess completeness of data, by comparing overlap between patients attending peripheral clinics, and the central Hlabisa Hospital clinic. Matching by name, age, and diagnosis proved feasible, but there was little overlap, and CR calculations were invalid because of the relative independence of sources. We conclude that NCDs are common in rural Africa, and that a simple NCD district register is a potentially feasible and inexpensive option. Capture–recapture analysis is feasible, but requires suitable lists with acceptable overlap of patients.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Non-communicable diseases (NCDs) are growing in importance in developing countries (Zimmet & Lefebre 1996). They are rising in prevalence for a variety of reasons, including population expansion as well as prolonged life expectancy. Increasing urbanization and `westernization' of populations have also increased the risk of certain NCDs, notably type 2 diabetes (Papoz et al. 1988; Cohen et al. 1998), asthma (Keeley et al. 1991; Ng'ang'a et al. 1998) and hypertension (Poulter et al. 1990). The major NCDs are also chronic and incurable, and thus represent an enormous personal, health care and economic burden (Chale et al. 1992). For example, insulin supply in many developing countries is erratic, mainly because of simple economic factors (Lester 1985; Alberti 1994).

For these reasons, up to date and accurate estimates of the prevalence of major NCDs in developing countries are enormously important, if the provision of even basic health care is to be planned. Accurate epidemiological surveys in low-resource countries are problematic – they are often time-consuming and expensive, frequently involve areas where the overall population is not censused and often very mobile. Hence most high-quality prevalence surveys of NCDs have involved specific studies of subpopulations, using gold standard diagnostic tests, such as glucose tolerance tests for diabetes in rural Tanzania (McLarty et al. 1989), and exercise-provocation testing for asthma in South Africa (Keeley et al. 1991).

We report our experience of using two simpler and less expensive epidemiological systems to enumerate four major NCDs (hypertension, diabetes, asthma and epilepsy) in an area of rural South Africa. First we attempted to construct a district register for NCDs, using as many data sources as possible – a system known as electronic data linkage (Morris et al. 1997). Secondly, we aimed to assess the potential undercount of the constructed register using the capture–recapture (CR) technique (International Working Group 1995). This is a statistical method which analyses multiple patient lists for overlap, and allows estimates of the total population (both counted and uncounted) to be made (Laporte et al. 1993; Laporte 1994; Ismail & Gill 1999).

Capture–recapture assumes that existing lists are incomplete, and provides a tool to estimate – with confidence intervals (CI) – the total population of persons with the disease in question, including diagnosed cases not appearing on current lists. The lists used need to be from sources as independent as possible, and there needs to be a reasonable degree of mixing between the lists, as the method depends on calculations based on the proportion of overlapping patients, compared with the total list sizes (Ismail & Gill 1999).

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Hlabisa district is an area of northern Kwazulu Natal in South Africa, population 205 000 (1991 census). It is almost entirely rural, consisting of scattered villages and dwellings connected by non-tarmac dirt roads. The district is served by one hospital in the small town of Hlabisa as well as by 10 satellite health centres, which are scattered around the district. They are staffed by nurses, led by an experienced sister, and visited approximately weekly by a doctor from Hlabisa Hospital. Patients with NCDs are seen either at a regular medically staffed weekly clinic at Hlabisa Hospital, or at the peripheral clinics, where they receive treatment from nurses using agreed protocols of management (Coleman et al. 1998), with medical referral (either to the doctor next visiting the clinic, or directly to Hlabisa) if necessary. All patients with NCDs attend a health care facility monthly, for review and to receive appropriate drug prescriptions for the next month. Some patients who are deemed particularly stable, and who attend the Hlabisa NCD Clinic, may be given repeat prescription cards which enable them to pick up prescriptions at the hospital pharmacy on a monthly basis until their next medical review (usually in 3 months). There are no other systems of obtaining drugs for the treatment of NCDs in Hlabisa District. In view of the need for monthly attendance, together with centrally held repeat prescription cards, we collected data intensively over a 6-week period. Unless patients absconded without treatment, they would have to attend at least once, and in many cases twice, during this period.

We recorded patients with named diagnoses of hypertension, diabetes, asthma and epilepsy. Diagnostic criteria for hypertension were based on national guidelines at the time, recommending drug therapy for sustained blood pressure (BP) levels of 170/100 (Heart Foundation Hypertension Consensus Symposium 1992). Diabetes was diagnosed on WHO criteria at the time of the study (WHO 1985), i.e. a fasting blood glucose level > 7.0 mmol/l (laboratory measured), or random blood glucose > 10.0 mmol/l; together with typical hyperglycaemic symptoms. Diagnosis of asthma was based on a significant history of breathlessness and wheeze, supported by a low peak flow measurement improved with salbutamol inhalation, or a clinical response to a trial of bronchodilator therapy. Epilepsy was diagnosed on the basis of a witnessed history of typical grand-mal seizures, or other atypical attacks responsive to a trial of anticonvulsant therapy.

Electronic data linkage

Information was recorded from three sources: the Hlabisa weekly NCD Clinic, satellite health clinics and repeat prescription cards. The NCD Clinic at Hlabisa enters basic patient details with diagnosis into an attendance ledger, and at the 10 satellite clinics similar information is kept in exercise books (although patients with NCDs attend at any time, rather than in focused clinics). Repeat prescription cards are held at the Hlabisa Hospital dispensary. Maximum available information was full name, sex, age and diagnosis. Information on sex and diagnosis was firm. Age was available for most patients, but if in doubt not recorded. The name was carefully recorded because of the difficult and sometimes variable spelling of Zulu names, and all names were independently checked by a Zulu nurse.

Data was entered into a laptop computerized database using EpiInfo 6.04 software (Dean et al. 1994). The database was then carefully re-examined manually and electronically to check for duplication. Duplicates were removed, and all patients were then summated to produce a single NCD register.

Capture–recapture

The technique of CR, as applied to medical epidemiology, depends on at least two separate lists of patients with a specific characteristic (here an NCD diagnosis) being available in a population. The lists should not be overly dependent or independent (i.e. adequate, but not excessive, overlap between the lists), and patients should have a random chance of appearing on the different lists. The population studied should also be stable in terms of death and migration (Laporte et al. 1993). We postulated that the NCD lists we prepared would fulfil these criteria. Hlabisa is the central town for the area and many people travel from the surrounding areas to the town for work, school, market, visiting friends and relatives, etc. It seemed reasonable to suppose therefore that NCD patients may randomly attend either a peripheral health clinic or the central hospital NCD Clinic (which they are allowed to do). Our short and intense study period also removed the potential problem of any significant death or migration. For the CR calculations, we used the same database as prepared above, but did not exclude duplicate attenders (unless it was to the same health facility). We also collapsed the Hlabisa NCD Clinic list and the repeat prescription card list, as they were heavily dependent (those with repeat prescription cards came mainly from the Hlabisa Clinic). Combining lists to reduce dependence is an accepted technique in CR methods used to maintain independence of lists (Bruno et al. 1994; Ismail et al. 1999, 2000). In this case, the Hlabisa Clinic list would be heavily dependent on the repeat prescription card list, as such cards are only given to those patients attending Hlabisa Clinic.

The final two lists formed were thus the Hlabisa list which was compared with the clinic list (from the peripheral clinics) by two-list CR (Laporte et al. 1993; Laporte 1994; International Working Group 1995; Ismail & Gill 1999). Statistical analysis was undertaken using the Statistical Package for Social Sciences (SPSS) (Norusis 1993). We used all available data to match patients between the two lists (name, age, sex and diagnosis). This was because Zulu names often vary, as people may have two first names – one from their parents, and one from their grandparents. Additionally, many older patients do not know their age accurately.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Characteristics of patients

A total of 2455 patients with NCDs were recorded. Hypertension was by far the commonest NCD (62% of all diagnoses), followed by epilepsy (16%), asthma (13%) and diabetes (12%) (Table 1). The overall mean age (± 1 SD) was 51 ± 16, and there was a large female predominance (76%). Women were also significantly older than men (53 ± 15 vs. 49 ± 19, P < 0.01). For individual NCDs, age structure was similar for hypertension and diabetes, but asthmatics and epileptics had a lower mean age (45 and 35, respectively; Table 1). Compared with the overall sex distribution, females were over-represented in the hypertensive group (82 vs. 76%, P < 0.001); but under-represented (compared with the whole group) in the asthmatic group (60 vs. 76%, P < 0.02), the epileptic group (49 vs. 76%, P < 0.001), and inpatients with diabetes (62 vs. 86%, P < 0.001). Children (defined as below 15 years of age) rarely had NCD diagnoses – 12% of epileptic and 7% of asthmatic patients were in this age group, but only 1% of those with diabetes, and none with hypertension (see Table 1).

Table 1.  Details of the four major non-communicable diseases (NCDs), amongst 2455 Hlabisa District residents Thumbnail image of

Clinic attendance patterns

We examined characteristics of patients attending the main NCD Clinic at Hlabisa Hospital, compared with those attending peripheral health centre clinics. There were 18% of the total NCD population attending the Hlabisa Clinic, and these patients were younger than those attending peripherally (45 ± 19 vs. 53 ± 15, P < 0.001). Females were less likely to attend at Hlabisa (P < 0.001), mainly because of reduced numbers of female hypertensives, and, to a lesser extent, diabetic patients. There was no gender difference with epilepsy, and there were slightly more asthmatic female attenders at Hlabisa. For individual diseases, asthma (P < 0.03) and hypertension (P < 0.001) were relatively under-represented at Hlabisa; and epilepsy (P < 0.001) and diabetes (P < 0.001) relatively over-represented.

NCD prevalence estimates

Using the information from electronic data linkage, the estimated crude population prevalence rates of individual NCDs are shown in Table 1– 7.4% for hypertension, and approximately 0.2% each for diabetes, asthma and epilepsy. These figures of course assume complete case ascertainment, which is unlikely, for reasons discussed below.

For the capture–recapture analysis, there was sufficient matching information available (identical name, age, sex and diagnosis) for 2268 (92%) patients. The degree of overlap between the Hlabisa list, and all the other peripheral clinics, is shown in Figure 1. As can be seen, there was no overlap at all for the asthmatic and epileptic groups, and very little for those with hypertension and diabetes. Capture–recapture calculations could therefore only be made in the latter two groups, and because of the low level of overlap the prevalence estimates had very wide CI – 12.8% (95% CI 1.2–24.4%) for hypertension, and 2.3% (95% CI 0.1–4.5%) for diabetes.

image

Figure 1. Distribution of non-communicable diseases (NCDs) between Hlabisa Hospital and peripheral clinics, with capture–recapture analysis.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Hlabisa District is not dissimilar from other rural areas of Africa, with a central hospital and outlying clinics. The clinics are reached by dirt roads and communication is difficult. Record-keeping is rudimentary, and laboratory support very basic. Drug supplies, however, are generally better than in other sub-Saharan African countries. Also typical is the system of monthly drug supply at clinic for patients with chronic diseases. Our 6-week surveillance period was intended to identify all NCD patients in the district, as they would need drug supplies within this period. The basic data recording system using EpiInfo software proved reliable and effective. EpiInfo is simple to use and widely available at no charge. Because of the complex spelling of several Zulu names, as well as some duplicate attendances at the same clinic during the study period, some visual cleaning of the dataset was needed. This was however, not a major problem, and our experience shows that a simple NCD register in rural Africa is a feasible option.

The patient characteristics show that as expected, hypertension was the main problem numerically (62% of all NCD diagnoses), as has been recorded previously (Sever et al. 1980). Amongst our NCD patients females strongly predominated (76%), especially in the group with hypertension (82%). The mean age of the group was relatively high at 52 years; particularly in those with hypertension (57) and diabetes (54). This illustrates NCD prevalence increase with age as life expectancy rises in tropical populations.

Comparing characteristics of the patients attending the Hlabisa NCD clinic with those attending peripheral health clinics showed that 82% attended peripheral clinics – the result of an agreed policy of referral of uncomplicated patients to their local clinic. Similarly, the significant excess of asthmatic and epileptic patients at the Hlabisa NCD clinic was a result of treatment protocols – these required doctor-based intervention for drug treatment of epilepsy, and for insulin treatment of diabetes. The younger mean age at Hlabisa may reflect greater concern over such patients by peripheral clinic nursing staff resulting in central referral. The male predominance at Hlabisa may reflect their greater mobility.

Although our methodology of simple electronic data recording (with easily accessible and user-friendly software) was able to produce a district NCD register, there are of course concerns about its completeness. We are not aware of any major cross-boundary movement of patients to adjacent health care facilities, but some may have occurred. Migration is always a potential problem in studies such as this, and there may have been a loss of male patients by this mechanism – perhaps working away in the nearby industrial centres of Richard's Bay or Durban (Tanser et al. 2000). This could partially explain the female excess of NCD patients we noted. Some patients would have been lost because of missed appointments and interrupted treatment.

In an attempt to correct the potential undercount discussed above, we used the accepted system of CR. For valid CR calculations, strict criteria are required, concerning mixing of the lists, lack of dependence and stability of the population. We believed at the planning stage of our project that these criteria would be valid, but then unexpectedly encountered almost complete independence of lists. Thus in the asthma and epilepsy groups, no patients attended both central and peripheral health utilities in the study period, and in the diabetes (2) and hypertension groups (8) the numbers were very small. Dependency problems such as this are well recognized as potential problems with CR analyses (Brenner 1995). Interestingly, the prevalence rates calculated from the 2 lists for hypertension and diabetes were 13 and 2%, respectively, – more realistic than the crude rates calculated from the NCD register. However, the lack of list dependence resulted, as expected, in very wide CI of 1.2–24.4% for hypertension, and 0.1–4.5% for diabetes. The calculations cannot thus be regarded as valid. There are two possible reasons for the lack of overlap between peripheral and central clinics. Firstly, it may be that patients prefer to attend one clinic facility only, where they are presumably well-known. Alternatively (or in addition), it may be that our study period was too short. With a 1-month appointment system for drug supplies, 6 weeks was designed as a period to capture all NCD patients at least once. However, in this period, the minority of patients would need to attend twice, thus reducing the likely degree of overlap between central and peripheral clinics.

The crude prevalence estimates from our NCD register figures (by data linkage) are compared with other surveys in rural Africa in Table 2. None of these studies are of course exactly comparable, being from other geographical locations and sometimes with different diagnostic criteria. The hypertension surveys from Tanzania (Swai et al. 1993) and Nigeria (Kaufman et al. 1996), however, used similar diagnostic criteria (BP > 140/90), but were population-based surveys, uncovering known as well as unknown disease. Our hypertension prevalence (7.4%) of known disease only is thus comparable, and may represent a potentially greater overall figure. For diabetes, rates of 0.9% (McLarty et al. 1989) and 1.3% (Aspray et al. 2000) from Tanzania and 0.8% (Mbanya et al. 1997) in Cameroon have been reported in rural areas, calculated from population glucose tolerance test (GTT) surveys. Interestingly, these studies also recorded known diabetic cases, and the rates of these were 0.1% in Tanzania (McLarty et al. 1989; Aspray et al. 2000) and 0.3% in Cameroon (Mbanya et al. 1997) – similar to our figure of 0.2%. The rates quoted in Table 2 for epilepsy (Tekle-Haimanot et al. 1990) and asthma (Yemaneberhan et al. 1997) were based on household surveys of known disease only and are somewhat higher than our Hlabisa figures.

Table 2.  Non-communicable disease prevalence (diagnosed disease) in Hlabisa District, from the NCD. Register figures, compared with other rural African surveys Thumbnail image of

In conclusion, we believe that construction of a district register for NCD patients in rural Africa is eminently feasible, using the software and system we have developed. Such a database allows enumeration of known patients with simple demographic characterization. However, prolonged recording and updating is likely to be necessary, to ascertain those absconding from routine appointments. Our attempts to undertake a CR analysis was unsuccessful because of lack of overlap between the two lists we compared (Laporte et al. 1993; International Working Group 1995). More prolonged recording may have allowed such analysis, and in other areas with greater movement between clinics, the system may be more suitable. Importantly however, we believe that we have shown its feasibility in rural Africa. We used full name, age and diagnosis as markers. The data needed visual cleaning for repeat entry and spelling mistakes, but otherwise worked satisfactorily – we had sufficient data for cross-matching on 92% patients. Capture–recapture analysis has been attempted in Tanzania, where it failed because of problems with Arab-based names involving interchangeability between first and second names (Black et al. 1994). Despite these problems, and the overlap difficulties we encountered, CR techniques in Africa are potentially feasible if adequate lists with good mixing are available.

We do not doubt that our numbers of NCD patients are underestimates, but we have shown that the techniques of electronic data linkage and CR can be applied to difficult areas of the rural tropics. Also, we emphasize that we were enumerating known and diagnosed disease, under current health care treatment. It is of interest that with all our methodological limitations, our prevalence figures for hypertension and diabetes accord well with house-to-house surveys elsewhere in rural Africa.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

We are grateful to Dr Sean Drysdale, and all the doctors and nurses of Hlabisa Hospital, who helped and co-operated with this project. In particular, Matilda Khumalo helped greatly with accurate data recording.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • 1
    Alberti KGMM (1994) Insulin dependent diabetes mellitus: a lethal disease in the developing world. British Medical Journal 309, 754 755.
  • 2
    Aspray TJ, Mugusi F, Rashid S et al. (2000) Rural and urban differences in diabetes prevalence in Tanzania: the role of obesity, physical inactivity and urban living. Transactions of the Royal Society of Tropical Medicine and Hygiene 2000 94, 637 644.
  • 3
    Black JFP, McLarty DG, Mtasiwa D (1994) Capture–recapture techniques – difficult to use in developing countries (letter). British Medical Journal 308, 531531.
  • 4
    Brenner H (1995) Use and limitations of the capture-recapture method in disease monitoring with two dependent sources. Epidemiology 6, 42 48.
  • 5
    Bruno G, Biggeri A, Laporte RE et al. (1994) National diabetes programs. Application of capture–recapture to count diabetes? Diabetes Care 17, 548 554.
  • 6
    Chale SS, Swai ABM, Munjinja PGM, McLarty DG (1992) Must diabetes be a fatal disease in Africa? Study of costs of treatment. British Medical Journal 304, 1215 1218.
  • 7
    Cohen MP, Stern E, Rusecki Y, Ziedler A (1998) High prevalence of diabetes in young adult Ethiopian immigrants to Israel. Diabetes 37, 824 828.
  • 8
    Coleman R, Gill G, Wilkinson D (1998) Non-communicable disease management in resource-poor settings: a primary care model from rural South Africa. Bulletin of the World Health Organization 76, 633 640.
  • 9
    Dean AG, Dean JA, Coulombier D et al. (1994) Epiinfo, Version 6: a Word Processing Database and Statistics Program for Epidemiology on Microcomputers. Centers for Disease Control and Prevention, Atlanta, USA.
  • 10
    Heart Foundation Hypertension Consensus Symposium (endorsed by the Medical Research Council and Hypertension Society of Southern Africa) (1992) Management guidelines for hypertension in Southern Africa. South African Medical Journal S1–S4.
  • 11
    International Working Group for Disease Monitoring and Forecasting (1995) Capture-recapture and multiple-record systems estimation I: history and theoretical development. American Journal of Epidemiology 142, 1047 1058.
  • 12
    Ismail AA & Gill GV (1999) The epidemiology of type 2 diabetes and its current measurement. In: Frontiers of Clinical Diabetology (eds GV Gill & IA MacFarlane) Bailliere-Tindal, London, UK, pp. 197–220.
  • 13
    Ismail AA, Beeching NJ, Gill GV, Bellis MA (1999) Capture–recapture adjusted prevalence rates of type 2 diabetes are related to social deprivation. Quarterly Journal of Medicine 92, 707 710.
  • 14
    Ismail AA, Beeching NJ, Gill GV (2000) How many data sources are needed to determine diabetes prevalence by capture–recapture? International Journal of Epidemiology 29, 536 541.DOI: 10.1093/ije/29.3.536
  • 15
    Kaufman JS, Rotimi CN, Brieger WR et al. (1996) The mortality risk associated with hypertension: preliminary results of a prospective study in rural Africa. Journal of Human Hypertension 10, 461 464.
  • 16
    Keeley D, Neill P, Gallivan S (1991) Comparison of the prevalence of reversible airways obstruction in rural and urban Zimbabwean children. Thorax 46, 549 553.
  • 17
    Laporte RE (1994) Assessing the human condition: capture–recapture techniques. British Medical Journal 308, 5 6.
  • 18
    Laporte RE, McCarty D, Bruno G, Tajima N, Baba S (1993) Counting diabetes in the next millennium, application of capture-recapture technology. Diabetes Care 16, 528 534.
  • 19
    Lester FT (1985) Insulin therapy: problems in an African country. Diabetic Medicine 2, 405 407.
  • 20
    Mbanya JCN, Ngogang J, Salah JN, Minkoulou E, Balkau B (1997) Prevalence of NIDDM and impaired glucose tolerance in a rural and an urban population in Cameroon. Diabetologia 40, 824 829.DOI: 10.1007/s001250050755
  • 21
    McLarty DG, Swai ABM, Kitange HM et al. (1989) Prevalence of diabetes and impaired glucose tolerance in rural Tanzania. Lancet i, 871 875.
  • 22
    Morris AD, Boyle DIR, MacAlpine R et al. (1997) The diabetes audit and research in Tayside, Scotland (DARTS) study: electronic data linkage to create a diabetes register. British Medical Journal 315, 524 528.
  • 23
    Ng'ang'a LW, Odhiambo JA, Mungai MW et al. (1998) Prevalence of exercise-induced bronchospasm in Kenyan school children: an urban-rural comparison. Thorax 53, 919 926.
  • 24
    Norusis MJ (1993) Statistical Package for Social Science (SPSS) – Windows base System Users Guide (Release 6.0). SPSS, Chicago.
  • 25
    Papoz L, Ben Khalifa F, Eschwege E, Ben Ayed H (1988) Diabetes mellitus in Tunisia: description in urban and rural populations. International Journal of Epidemiology 2, 419 422.
  • 26
    Poulter NR, Khaw KT, Hopwood BEC et al. (1990) The Kenyan Luo migration study: observations on the initiation of a rise in blood pressure. British Medical Journal 300, 967 972.
  • 27
    Sever P, Gordon D, Peart WS, Beighton P (1980) Blood pressure and its correlates in urban and rural Africa. Lancet ii, 60 64.
  • 28
    Swai ABM, McLarty DG, Kitange HM et al. (1993) Low prevalence of risk factors for coronary heart disease in rural Tanzania. International Journal of Epidemiology 22, 651 659.
  • 29
    Tanser F, LeSeur D, Solarsh G, Wilkinson D (2000) HIV heterogeneity and proximity of homestead to roads in rural South Africa; an exploration using a geographical information system. Tropical Medicine & International Health 5, 40 46.
  • 30
    Tekle-Haimanot R, Abebe M, Gebre-Mariam A et al. (1990) Community-based study of neurological disorders in rural central Ethiopia. In: Neurological Disorders in Ethiopia (ed R Tekle-Haimanot) Umea University, Umea. pp. 83--141.
  • 31
    World Health Organization (1985) Diabetes Mellitus – Report of a WHO Study Group. WHO Technical Report Series 727. WHO, Geneva.
  • 32
    Yemaneberhan H, Bekele Z, Venn A, Lewis S, Parry E, Britton J (1997) Prevalence of wheeze and asthma and relation to atopy in urban and rural Ethiopia. Lancet 350, 85 90.DOI: 10.1016/s0140-6736(97)01151-3
  • 33
    Zimmet P & Lefebre P (1996) The global NIDDM epidemic: treating the disease and ignoring the symptom. Diabetologia 39, 1247 1248.DOI: 10.1007/s001250050565