Causes of deaths using verbal autopsy among adolescents and adults in rural western Kenya

Authors


Corresponding Author A.M. van Eijk, 172 Herbert Chitepo Road, Harare, Zimbabwe. Tel.: +263 4 796040/302971; Fax: +263 4 2022981; E-mail: vaneijka@zimcdc.co.zw; AmvanEijk@yahoo.com

Summary

Objective  To establish causes and patterns of deaths among adolescents and adults (age >11 years) using verbal autopsy (VA) in a rural area of western Kenya where malaria and HIV are common.

Methods  Village reporters reported all deaths in Asembo and Gem (population 135 000), an area under routine demographic surveillance. After an interval of ≥1 month, a trained interviewer used a structured questionnaire to ask the caretaker about signs and symptoms that preceded death. Three clinical officers independently reviewed the interviews and assigned two unranked causes of death; a common cause was designated as the cause of death.

Results  In 2003, 1816 deaths were reported from residents; 48% were male and 72% were between 20 and 64 years of age. Most residents (97%) were ill before death, with 60% of illnesses lasting more than 2 months; 87% died at home. Care was sought by 96%; a health facility was the most common source, visited by 73%. For 1759 persons (97%), a common cause of death was designated. Overall, 74% of deaths were attributed to infectious causes. HIV (32%) and tuberculosis (TB) (16%) were the most frequent, followed by malaria, respiratory infections, anaemia and diarrhoeal disease (approximately 6% each). Death in a health facility was associated with young age, higher education, higher SES, a non-infectious disease cause and a shorter duration of illness.

Conclusion  In this area, the majority of adult and adolescent deaths were attributed to potentially preventable infectious diseases. Deaths in health facilities were not representative of deaths in the community. Programmes to prevent HIV and TB infection and to decrease mortality have started. Their impact can be evaluated against this baseline information.

Abstract

Objectif:  Etablir les causes et modes de décès chez les adolescents et adultes (âge > 11 ans) en utilisant les autopsies verbales dans une zone rurale de l’ouest du Kenya où la malaria et le VIH sont fréquents.

Méthodes:  Les rapporteurs ont signalé tous les décès dans les villages d’Asembo et Gem (population 135000), une zone sous surveillance démographique en routine. Après un intervalle de ≥ 1 mois, un enquêteur formé a utilisé un questionnaire structuré pour obtenir des soignants, les signes et les symptômes qui ont précédé les décès. Trois officiers cliniques ont indépendamment examiné les entretiens et ont assigné deux types de causes de décès, une cause commune a été désignée comme la cause du décès.

Résultats:  En 2003, 1816 décès ont été signalés par les résidents, 48%étaient des hommes et 72% avaient entre 20 et 64 ans. La plupart des résidents (97%) étaient malades avant de mourir, avec 60% des maladies durant plus de 2 mois, 87% des décès sont survenus à domicile. Des soins ont été recherchés chez 96% des cas, un service de santé a été la source la plus courante, visité par 73% des cas. Pour 1759 personnes (97%) une cause courante de décès a été désignée. Au total, 74% des décès étaient attribués à des causes infectieuses. Le VIH (32%) et la tuberculose (16%) étaient les plus fréquents, suivie par la malaria, les infections respiratoires, l’anémie et les maladies diarrhéiques (environ 6% chacun). Le décès dans un service de santéétait associé au jeune âge, le niveau d’éducation supérieur, un statut socioéconomique plus élevé, une cause non infectieuse et une plus courte durée de la maladie.

Conclusion:  Dans cette région, la majorité des décès d’adultes et d’adolescents était attribuée à des maladies infectieuses potentiellement évitables. Les décès dans les services de santé n’étaient pas représentatifs des décès dans la communauté. Des programmes de prévention de l’infection par le VIH et la tuberculose et pour réduire la mortalité ont commencé. Leur impact pourra être évalué par rapport à cette information de base.

Abstract

Objetivo:  Establecer las causas y patrones de muerte entre adolescentes y adultos (edad > 11 años) utilizando la autopsia verbal (AV), en un área rural de Kenia occidental en donde la malaria y el VIH son comunes.

Métodos:  Los informantes de los poblados reportaron todas las muertes ocurridas en Asembo y Gem (población 135,000), un área bajo seguimiento demográfico rutinario. Después de un intervalo de ≥1 mes, un entrevistador entrenado, utilizó un cuestionario estructurado para preguntar al cuidador sobre los signos y síntomas que precedieron a la muerte. Tres oficiales clínicos independientes revisaron las entrevistas y asignaron dos causas de muerte sin darles un ranking; una causa común fue designada como la causa de muerte.

Resultados:  En el 2003, se reportaron 1816 muertes de residentes; 48% eran hombres y 72% tenían entre 20 y 64 años. La mayoría de los residentes (97%) estaban enfermos antes de morir; un 60% de las enfermedades tenían una duración de más de 2 meses; el 87% murió en su casa. Un 96% buscó cuidados, siendo los centro sanitario la fuente más común, visitados por un 73%. Para 1759 personas (97%) se designó una causa común de muerte. En total, un 74% de las muertes estaban atribuidas a causas infecciosas. El VIH (32%) y la tuberculosis (16%) fueron las más frecuentes, seguidas por la malaria, infecciones respiratorias, anemia y enfermedad diarreica (aproximadamente 6% cada una). La muerte en un centro sanitario estaba asociada a una edad joven, a un mayor nivel educativo, un mayor estatus socioeconómico, a una causa no relacionada con infecciones y a una duración más corta de la enfermedad.

Conclusión:  En esta área, la mayoría de las muertes de adultos y adolescentes estaban atribuidas a enfermedades infecciosas potencialmente prevenibles. Las muertes en centros sanitarios no eran representativas de las muertes en la comunidad. Los programas para prevenir las infecciones por VIH y TB y para reducir la mortalidad ya han comenzado. Su impacto podrá evaluarse comparativamente con esta información de base.

Introduction

The disease burden and mortality because of infectious diseases are high in developing countries, but non-communicable diseases also become increasingly important as the socio-economic conditions in a country improve (World Health Organization 2002). Unfortunately, accurate statistics on causes of death are lacking in many developing countries because of inadequate capture in hospitals and government reporting systems (Arudo et al. 2003). Knowledge about cause-specific mortality is important because it can help to prioritise health information strategies and investments.

Western Kenya has a high burden of infectious diseases: HIV infection is a major cause of morbidity and mortality among adults and children, and malaria an important cause among children. The regional HIV prevalence reported in the 2003 Demographic Health Survey for Nyanza Province in western Kenya was 15% (18% among women and 12% among men), considerably higher than the national prevalence of 7% (Central Bureau of Statistics Kenya, Ministry of Health Kenya & ORC Macro 2004). In a 2003 cross-sectional survey, the age-adjusted prevalence rates in men and women 13–34 years of age were 11% and 21%, respectively, in Asembo, an area in western Kenya (P. Amornkul personal communication). The rate of tuberculosis (TB) case notifications (364/100 000) in Nyanza Province in 2003 was the highest of all provinces outside Nairobi (Kenya Ministry of Health 2003). However, accurate data on cause-specific mortality in this area are limited because most deaths occur outside healthcare facilities, and causes are not routinely captured by vital events monitoring systems. Moreover, stigma associated with HIV likely results in under attribution to this cause.

In 2001, we established the KEMRI/CDC demographic surveillance system (DSS) in rural western Kenya (Adazu et al. 2005). As part of this system, we introduced verbal autopsy (VA) among adolescents and adults in 2003. Verbal autopsy uses systematic interviews of the bereaved on signs and symptoms of the last illness to assign a cause of death. Although VA has limited sensitivity and specificity for certain death causes, it offers a better approximation than the use of routine data in many developing countries (Murray & Lopez 1997; Chandramohan et al. 1998; Gajalakshmi et al. 2002; Gajalakshmi & Peto 2004; Setel et al. 2006). In this paper, we describe VA findings for 2003 among the adolescent and adult population in western Kenya. We were also interested to explore potential differences in household socio-economic status (SES) by cause of death, and in differences among persons by place of death (health facility vs. home). Finally, we assessed the reported prevalence of some non-communicable diseases, and use of alcohol and lifestyle drugs among the deceased, and their association with the main causes of death.

Methods

Study area

The demographic surveillance area (DSA) is located in a rural part of Nyanza province in western Kenya in Asembo (Rarieda Division, Bondo district) and Gem (Yala and Wagai Divisions, Siaya District). The DSA covers 217 villages (75 in Asembo and 142 in Gem) spread over approximately 500 km2 along the shores of Lake Victoria; the population is approximately 135 000. Before launching the health and DSS, the DSA was the site of a randomised controlled trial of insecticide-treated bednets (ITNs) from 1997 to 2002 (Phillips-Howard et al. 2003; Lindblade et al. 2004).

Study population

The DSA population is culturally homogeneous (more than 95% are of the Luo ethnic group) and lives in dispersed settlements. Malaria transmission is intense and perennial, although it was reduced by the use of ITNs (Gimnig et al. 2003). In 2003, HIV-prevention and treatment programmes were developed for the area, but were not yet fully established; voluntary testing and counselling were available in some local centres.

Verbal autopsies

We modified for use a standardised VA questionnaire developed by DSS sites around the world to accommodate the western Kenya situation (http://www.indepth-network.org/core_documents/indepthtools.htm). Information on the deceased person’s terminal illness was requested from a caregiver and included queries on healthcare sought during the illness. One or two village reporters were selected in each village in the study area to report village births and deaths as they occurred. Beginning in 2003, VAs were conducted using trained interviewers for all deaths among adults and adolescents (>11 years) at least 1 month after the date of death to respect the mourning period. Information was obtained on residents only, defined as persons who were living at least 4 months in the DSA prior to death. The permanent identification numbers of the persons were used to permit linking of VA and clinical and demographic data; we were able to retrieve these for 93% of the deaths in 2003. Using information from the VA questionnaires, a panel of three clinical officers (2 years of post-secondary school medical training) independently assigned up to two causes of death; concurrence by at least two of the clinical officers was required to assign that diagnosis as a cause of death. If there was disagreement among the three original clinical officers, the information was discussed in a separate meeting by another three clinical officers, and a final diagnosis was assigned. A local system for coding was used; data were later regrouped into the classification used for the global burden of disease study and a study into cause-specific mortality rates in sub-Saharan Africa (Murray & Lopez 1997; Adjuik et al. 2006). We added anaemia as a separate category (including final diagnoses of anaemia or sickle cell anaemia) because we did not feel we could attribute anaemia to nutritional deficiencies; the aetiology of anaemia in this area is often multifactorial.

Definitions and analysis

Socio-economic information was obtained from the KEMRI/CDC DSS database. Part of this information could not be retrieved because the death and subsequent closure of the household occurred before the information on socio-economic status (SES) was collected. Information on household assets was used to derive a wealth index using the method developed by Filmer & Pritchet (2000). All households were categorised by quintile: a medium/low SES was defined as an SES in the bottom three quintiles of the wealth index. A low educational status was defined as less than 8 years of education, the number of years generally required to finish primary school. Rainy season included the months of April, May, June and July (long rains), and October and November (short rains). Maternal deaths were classified in two ways: (1) a diagnosis of maternal death by the review panel, or (2) any death which occurred during pregnancy, childbirth or within 42 days after birth. Injuries included accidents, suicides and poisoning. Proportions were compared using the chi-square test or Fisher’s exact test. We used logistic regression to assess factors associated with a death in a health facility; factors with a P-value of 0.05 or more were removed from the model. Factors examined included age, gender, season of death, level of education, SES, marital status, type of work and duration of the last illness. The statistical program SAS (SAS system for Windows version 8, SAS, Cary, NC, USA) was used for analyses. All tests were two-sided; P < 0.05 was considered significant.

Ethical review and informed consent

Verbal autopsy was conducted as part of the KEMRI/CDC DSS. The DSS protocol has been reviewed and approved by the institutional review boards of both CDC (Atlanta, GA) and KEMRI (Nairobi, Kenya). Informed written consent was obtained from compound heads for participation of their families in all aspects of the KEMRI/CDC DSS.

Results

Characteristics of the persons who died

The mid-year population of the study area in 2003 was 136 497 (64 009 males and 72 488 females). Of this number, 61.8% (84 401; 39 287 males and 48 114 females) were aged 12 years and above. A total of 3267 deaths and 5182 births were recorded in 2003. Of the deceased, 60.1% (n = 1965) were 12 years or older. Verbal autopsies were successfully conducted for 92.4% (n = 1816) of the deaths for the age group 12 years and older and 7.6% (149 cases) were lost to follow-up. The households of lost cases relocated shortly after the deaths and could not be traced. The distribution of the mid-year population, the deaths, and death rates by age and sex are shown in Table 1.

Table 1.   Deaths by age group and gender, Asembo/Gem, 2003
Age (years)PopulationDeathsDeath rates (per 1000)
MaleFemaleMaleFemaleMaleFemale
12–1915 52814 68130471.93.2
20–3912 48816 18231541225.225.5
40–64786412 31333730942.925.1
65+3407493826325277.251.0
Total39 28748 114945102024.121.2

For 3.1% (57 persons), no diagnosis could be determined using the VA panel review. The median age at death among all 1816 persons was 45.7 years (range 12–102); 50% were married (Table 2). Seventy-six per cent of the people whose education level was known (1260/1651) had not completed primary school. Eighty-six per cent died at home (1567/1816). Overall, care prior to death was sought during the terminal illness by 94% (1705/1816); a health facility was the most common source, visited by 73% (1240/1705). For 1762 persons (97%) who had a history of illness, characteristics of the last illness are presented in Table 2: 60% of these illnesses lasted >2 months.

Table 2.   Characteristics of adolescent and adult residents who died in Asembo/Gem in 2003
  All (%) n = 1816SES available (%) n = 1321*SES not available (%) n = 495
  1. *Chi-square test: P < 0.05 comparing “SES available”vs.“SES not available” for age group, gender, marital status, education level and type of labour.

  2. †Marital status unknown for four persons; type of work unknown for 11 persons.

Age group at time of death
 <20 years64 (3.5)58 (4.4)6 (1.2)
 20–39 years658 (36.2)487 (36.9)171 (34.6)
 40–64 years645 (35.5)476 (36.0)169 (34.1)
 ≥65 years449 (24.7)300 (22.7)149 (30.1)
Male866 (47.7)691 (52.3)175 (35.4)
Marital status†
 Married915 (50.4)746 (56.6)169 (34.2)
 Separated184 (10.1)139 (10.6)45 (9.1)
 Widowed543 (29.9)297 (22.5)246 (49.8)
 Single170 (9.4)136 (10.3)34 (6.9)
Education level
 No school attended388 (21.3)240 (18.2)148 (29.9)
 <8 years872 (47.9)675 (51.1)195 (39.4)
 ≥8 years391 (21.5)300 (22.7)91 (18.4)
 Unknown167 (9.2)106 (8.0)61 (12.3)
Low/medium SES787 (59.6)  
Unskilled labour†1419 (78.6)1012 (77.1)407 (82.7)
Died in rainy season986 (54.3)717 (54.3)269 (54.3)

Causes of deaths

Overall, 71% of the 1816 deaths were attributed to infectious causes, 15% to non-communicable diseases, 8% to nutritional or maternal death causes, 3% to injuries and 3% were indeterminate.

HIV or TB was assigned to 48% of deaths and were the most frequent causes, followed by four other categories: malaria, respiratory infections, anaemia and diarrhoeal disease, each accounting for approximately 6% of deaths. Cardio-vascular disease was the most common non-communicable disease, but was only assigned in 5% of the deaths.

The diagnosis “Injuries” was assigned to 3% (n = 54). Leading causes of these injury-related deaths were assaults (n = 10), poisoning (including alcohol, n = 9), accidental trauma (fall, tooth extraction, n = 9), suicide (n = 9) and road accidents (n = 7). Forty-six persons were reported to have died from cancer (2.5%). Cancers of the upper digestive tract (throat, oesophagus, gastric cancer) were the most frequently mentioned (n = 19), followed by cancer of the lower digestive tract (n = 9).

The number of women who died in the fertile age range of 13–45 years was 519; 16 (3%) were reported by the review panel as related to pregnancy, delivery or miscarriage. However, if the temporal relation to delivery was considered as a criterion for maternal death, 39 women (8%) died either pregnant (18), or within 42 days of delivery or miscarriage (21). There were 5182 live births in 2003, generating a maternal mortality of 753 per 100 000. However, for nine women, the time period between miscarriage and death was unknown; HIV/AIDS was the assigned cause of their deaths.

Causes of death for selected characteristics

HIV was reported as a cause of death significantly more often among women, whereas injuries, other infectious diseases and diabetes were more common among men (Table 4 and Figure 1). Among the 64 (4%) deaths occurring among those aged 12–19 years, HIV/AIDS was common among women (12/38) followed by TB (5), and maternal conditions (4). Among men in the same age group, malaria (7/26) was more frequently reported, followed by other diseases (4) and injuries (3) (Table 3).

Table 4.   Causes of death overall, and by gender and age group among adolescent and adult residents, Asembo/Gem, 2003
DiagnosisAll, % (n = 1816)By gender, %Age group 12–19 years, %Age group 20–39 years, %Age group 40–64 years, %Age group ≥65 years, %
MaleFemaleMaleFemaleMaleFemaleMaleFemaleMaleFemale
n = 866n = 950n = 26n = 38n = 281n = 377n = 333n = 312n = 226n = 223
  1. NA, not applicable.

  2. *P < 0.05 comparing male vs. females.

  3. †Includes: anaemia and sickle cell disease.

  4. ‡Included: assaults (n = 10), trauma (sharp or blunt trauma, e.g. fall or tooth extraction) (9), suicide (9), poisoning (including alcohol) (9), road accidents (7), snake bite (3), burns (2), drowning (2), bee sting (1), dog bite (1) and allergic reaction (1).

  5. §Included: upper digestive tract (19), lower digestive tract and liver (10), site not clear (5), cervix (4), breast (3), lung (2), eye (1), leukaemia (1), osteosarcoma (1).

  6. ¶Included: puerperal sepsis (6), miscarriage (5), other complications (3), obstructed labour (2).

  7. **Included: epilepsy, sudden death, senility.

  8. ††Included four boys with epilepsy.

  9. ‡‡Among 1759 residents with a cause assigned.

HIV/AIDS31.727.5*35.57.7*31.637.4*49.633.037.29.39.9
TB16.216.615.87.713.220.618.316.518.312.88.5
Malaria6.56.56.526.97.96.44.55.46.45.89.9
Respiratory infections6.36.46.33.95.35.76.14.85.19.78.5
Anaemia†6.15.76.405.35.05.05.77.17.18.1
Diarrhoeal diseases6.16.55.702.67.85.05.43.57.110.3
Cardio-vascular Diseases5.35.45.23.900.40.35.45.512.013.9
Injuries‡3.04.9*1.311.52.66.4*0.33.31.34.42.7
Meningitis2.72.72.77.77.93.62.43.02.90.42.2
Malignant neoplasms§2.51.93.23.901.11.11.5*4.53.15.4
Other infectious diseases2.03.1*1.1000.71.33.9*0.35.31.8
Genito-urinary diseases1.42.01.0000.402.10.64.03.1
Digestive diseases1.31.71.07.700.40.52.11.32.21.4
Diabetes1.12.1*0.200002.10.34.9*0.5
Respiratory diseases0.90.61.302.60.71.30.61.00.41.4
Maternal conditions¶0.9NA1.7NA10.5NA2.1NA1.3NANA
Hepatitis0.20.10.20000.5000.40
Other**2.73.22.215.4††2.62.1*0.30.30.37.58.1
Indeterminate3.13.43.03.97.91.41.34.83.23.54.5
  n = 837n = 922n = 25n = 35n = 277n = 373n = 317n = 302n = 218n = 213
Infectious disease cause‡‡74.371.7*76.756.0*80.083.4*90.175.776.252.853.5
Figure 1.

 Proportion of deaths by HIV and TB by age and gender, Asembo/Gem, Kenya 2003. *P < 0.05 comparing proportion of HIV/AIDS diagnoses among women vs. men.

Table 3.   Characteristics of last illnesses among 1762 persons with a history of an illness before death, Asembo/Gem 2003
 n (%)
  1. †Multiple sources of care could have been sought.

Duration of illness (n = 1762)
 ≤1 week225 (12.8)
 >1 week–2 months465 (26.4)
 >2 months–11 months529 (30.0)
 ≥1 year529 (30.0)
 Unknown duration14 (0.8)
Care sought for last illness1693 (96.1)
Type of care sought† (n = 1693)
 Health facility1228 (72.5)
 Pharmacy/drug seller1139 (67.3)
 Religious leader628 (37.1)
 Traditional healer553 (32.7)
 Private practitioner351 (20.7)
 Bush doctor296 (17.5)
 Traditional birth attendant199 (11.8)
Place of death (n = 1762)
 Home1536 (87.2)
 Facility201 (11.4)
 Elsewhere25 (1.4)
Type of death cause (n = 1762)
 Infectious disease1305 (74.1)
 Non-infectious disease391 (22.2)
 Accidents/violence20 (1.1)
 Undetermined46 (2.6)

In the age group 20–39 years (36% of the deaths), HIV and TB were assigned to the majority of deaths among both men and women (58% of 281, and 68% of 377, respectively). In the older age group (40–64 years, 36% of the deaths), HIV and TB were associated with half of the deaths (50% among 333 men, and 55% among 312 women). In the oldest age group (≥65 years, 25% of all deaths), HIV and TB were associated with approximately one-fifth of the deaths (22% among 226 men and 18% among 223 women), followed among men by cardio-vascular diseases (12%) and respiratory infections (10%); among women in this age group, other frequently reported causes included cardio-vascular (14%) and diarrhoeal (10%) diseases. Overall, infectious disease causes were more common in the younger age groups and among women (Table 4).

There was no seasonal association between infectious or other reported causes of death (Table 5). Persons with a higher level of education were more likely to be reported as having died from HIV or an infectious disease (P < 0.01) and less likely to die of cardio-vascular diseases or malignant neoplasms (P = 0.03 for both); however, these associations disappeared when adjusting for age (Mantel Haenszel statistic: P > 0.05).

Table 5.   Causes of death by season, educational status, SES, and place of death among adolescent and adult residents, Asembo/Gem, 2003
DiagnosisAll, %By season, %By educational status†, % By SES†, %By place of death†, %
RainyDry<8 years≥8 yearsLowerHigherHFHome
n = 1816n = 986n = 830n = 1258n = 391n = 787n = 534n = 211n = 1567
  1. SES, socio-economic status; HF, health facility.

  2. *P < 0.05.

  3. †Educational status missing for 167 persons (9.2%); SES missing for 495 persons (27.3%); for place of death 38 persons excluded who died elsewhere (2.1%).

  4. ‡Includes: anaemia, sickle cell disease.

  5. §Included: epilepsy, sudden death, senility.

  6. ¶Among residents with a cause assigned.

HIV/AIDS31.730.333.329.3*39.632.231.722.8*33.4
TB16.215.916.515.316.916.015.513.716.6
Malaria6.57.05.96.56.47.45.87.66.3
Respiratory infections6.37.25.36.06.96.06.17.16.3
Anaemia‡6.16.55.56.54.65.83.92.8*6.6
Diarrhoeal diseases6.16.65.46.65.16.55.22.8*6.6
Cardio-vascular diseases5.34.46.45.9*3.15.75.27.65.0
Injuries3.02.53.52.53.83.32.87.1*1.8
Meningitis2.72.43.02.62.62.23.25.7*2.2
Malignant neoplasms2.52.52.53.0*1.02.32.64.72.2
Other infectious diseases2.02.02.12.31.81.82.81.92.0
Genito-urinary diseases1.41.51.31.9*0.31.31.91.01.5
Digestive diseases1.31.51.11.6*0.31.12.12.8*1.1
Diabetes1.11.40.70.91.80.8*2.42.40.9
Respiratory diseases0.90.81.11.00.50.80.40.51.0
Maternal conditions0.90.81.00.81.30.3*2.14.3*0.5
Hepatitis0.20.10.20.20.30.30.200.2
Other§2.73.12.23.4*1.32.81.51.02.9
Indeterminate3.13.33.03.62.12.5*4.74.32.8
  n = 954n = 805n = 1213n = 383n = 767n = 509n = 202n = 1523
Infectious disease cause¶74.374.474.271.7*82.375.274.764.4*76.1

Seventy-three per cent of the deaths could be matched with the KEMRI/CDC DSS database and had an indicator of SES. Persons without an indicator of SES were more likely to be older (>65 years), female, widowed, and to have no formal education (Table 2). Overall, there was no difference in disease causes by SES (Table 5).

Place of death

The highest proportions of deaths in health facilities by disease were reported for maternal conditions (56%), injuries (28%), diabetes (25%), digestive diseases (25%), meningitis (24%) and cancers (22%). Overall, deaths by an infectious disease were less likely to occur in a health facility than deaths by a non-infectious cause (Table 5). Persons younger than 65 years whose illness had lasted less than 1 month were more likely to die in a health facility, whereas persons with a lower level of education and SES, or an infectious disease cause were more likely to die at home (Table 6). By age group, significant factors associated with a death in a health facility among persons <40 years were a lower SES (adjusted OR [AOR] 0.50, 95% CI 0.29–0.86) and a shorter duration of illness (≤7 days: AOR 5.61, 2.55–12.35, 8–14 days AOR 4.78, 95% CI 1.77–12.87). In the age group 40–64 years, these were a lower SES (AOR 0.52, 95% CI 0.29–0.90), unskilled labour (AOR 0.29, 0.17–0.50), an infectious disease (AOR 0.42, 0.25–0.71) and a shorter duration of illness (≤7 days: AOR 7.92, 3.52–17.83; 8–14 days AOR 6.19, 95% CI 2.04–18.76; 15–28 days: AOR 2.89, 1.18–7.12). Among the elder persons aged ≥65 years, only SES remained important (lower SES: AOR 0.26, 0.12–0.60).

Table 6.   Factors associated with death in a health facility among adolescent and adult residents with a history of illness and death at home or in a health facility, Asembo/Gem, Kenya 2003
CharacteristicUnivariate analysis OR (95% CI)P-value†Multivariate analysis OR (95% CI)P-value‡
  1. OR, odds ratio; CI, confidence interval; SES, socio-economic status; NS, not significant.

  2. †Likelihood ratio test age: P = 0.0002, SES: P < 0.0001, duration of illness: P < 0.0001.

  3. ‡Likelihood ratio test age: P = 0.0001, SES: P = 0.0003, duration of illness: P < 0.0001.

  4. §Married vs. widowed, single or divorced. Infectious disease cause only for residents with an assigned disease cause (57 missing). Rainy season not significant in univariate analysis (data not shown). Significant odds ratios printed in bold.

Age group at death
 <20 years3.66 (1.75–7.67)0.00064.41 (1.88–10.34)0.0006
 20–39 years1.66 (1.06–2.59)0.02731.97 (1.11–3.47)0.0198
 40–64 years2.29 (1.48–3.52)0.00022.85 (1.68–4.83)<0.0001
 ≥65 yearsReference Reference 
Married§1.69 (1.25–2.28)0.0007NS 
Male1.40 (1.04–1.88)0.0254NS 
<8 years of school0.53 (0.38–0.74)0.00020.57 (0.39–0.84)0.0044
SES
 Low/medium0.39 (0.28–0.55)<0.00010.51 (0.35–0.74)0.0005
 HighReference Reference 
 SES unknown0.35 (0.24–0.52)<0.00010.47 (0.30–0.74)0.0012
Unskilled labour0.40 (0.29–0.54)<0.00010.43 (0.30–0.62)<0.0001
Infectious disease cause0.60 (0.44–0.83)0.00190.46 (0.31–0.69)0.0001
Duration of illness
 ≤1 week4.54 (2.83–7.28)<0.00014.73 (2.76–8.11)<0.0001
 8–14 days3.64 (1.93–6.89)<0.00013.70 (1.83–7.49)0.0003
 15–28 days2.64 (1.55–4.50)0.00042.44 (1.35–4.43)0.0033
 1–2 months1.56 (0.85–2.84)0.14841.68 (0.87–3.27)0.1251
 3–12 months1.63 (1.03–2.56)0.03601.80 (1.11–2.93)0.0182
 >1 yearReference Reference 

Non-communicable diseases, use of alcohol and lifestyle drugs

We asked the bereaved if the deceased was known with a chronic disease and if he or she was known to use alcohol or lifestyle drugs. This information was included in the form for the panel reviewers. Both high blood pressure (346 out of 1816 or 19%) and TB (330 or 18%) were common chronic diseases in this population; however, their distribution across age groups differed, with the highest prevalence of TB among the 20–39-year-olds (26%) and the highest prevalence of hypertension among persons 40 years and older (23%). Diabetes and epilepsy were mentioned for 20 (1%) and 23 (1%) persons, respectively. Considering the stigma on the diagnosis of HIV, it is encouraging that for 6.7% of the deceased (122 persons), the respondent reported that the deceased suffered from HIV; 77% (94) of these deceased were assigned HIV as death cause. Among the persons reported with TB as chronic disease, 43% (143) were assigned TB and 38% (125) were assigned HIV as death cause by the reviewers.

Thirty-two per cent (586) of the residents who died were reported to ever drink alcohol; 260 of them (44%) drank daily and 215 (37%) were reported to be drunk daily. Beer (30%) and traditional brews (96%) were the most common types of alcoholic consumptions reported. Alcohol use was more common among men (50% compared with women 17%, P < 0.01), and was highest in the age group of 40–64 years (44%). There was no association between alcohol use and SES or a death reported caused by HIV or TB or an infectious disease. However, persons using alcohol were more likely to die from an injury (26 or 4%vs. 28 or 2% among non-alcohol users, P = 0.02).

Lifestyle drugs were only reported to be used by 94 of the deceased persons (5%). Marijuana was most commonly reported (91%). The use of lifestyle drugs was higher among men (10% or 83 out of 866 vs. 1%, 11 out of 950, among women, P < 0.01), and highest in the age group of 20–39 years (47 out of 658 or 7%), and more common among persons with lower SES (6% or 50 out of 787 vs. 4%, 19 out of 534, P = 0.03). Persons using lifestyle drugs were more likely to have HIV as a death cause (44% or 41/94 vs. 31% or 531/1709 among non-users, P = 0.02).

Discussion

According to VA, infectious diseases were associated with three-quarters of deaths among adolescent and adult residents in this rural area of Kenya. HIV and TB were the main causes assigned, and together accounted for almost half of the reported causes of death. The burden of HIV/AIDS is tremendous in this area, and comparable to DSS sites in South Africa (Adjuik et al. 2006). Although the HIV/AIDS burden is initially higher in women younger than 25 years, men suffer a similar burden by their late twenties. Similar gender patterns for HIV have been described before in an HIV survey in Kisumu, western Kenya, and for HIV-related deaths in Zimbabwe (Buve et al. 2001; Lopman et al. 2006). Contrary to South African DSS sites, the burden because of injuries was low (>10% in South Africa vs. 3% in our study), whereas the pattern of other infectious diseases was more comparable to other (non-south) African sites (Adjuik et al. 2006). However, malaria as a death cause was relatively less common (6.5%) than in all other African sites with malaria (9–16.7%). Of the non-communicable diseases, cardio-vascular disease was less frequent than most other African sites (5.3%vs. 7.8–13.7%).

There is a high co-morbidity of HIV and TB (Raviglione et al. 1997); in Nyanza Province, approximately 72% of TB patients are HIV-infected (J. Odhiambo, personal communication). However, when examining the age pattern of the separate diseases, the gender difference by age group was mainly seen for HIV and not for TB. Tuberculosis also showed less fluctuation by socio-economic or educational status. The reason for this is not clear.

The Demographic and Health survey in 2003 reported an HIV prevalence in Nyanza province of 15% (12% among men and 18% among women aged 15–49 years) (Central Bureau of Statistics Kenya, Ministry of Health Kenya & ORC Macro 2004); the age-adjusted prevalence rates of HIV in men and women 13–34 years old in a survey in Asembo in 2003–2004 were 11% and 21%, respectively (P. Amornkul, personal communication). When HIV and TB were considered together (and assuming that 72% of the TB deaths were related to HIV), an estimated 43% of the deaths among these residents in Asembo/Gem were related to HIV. When we restricted the age to 15–49 years, this number increased to 57% (52% among males and 61% among females). Our estimate is lower than has been reported for other areas with a high HIV prevalence using different methods for the assessment of HIV-related mortality. In Malawi, among women aged 15–49 years and their husbands, an algorithm method estimated HIV-related deaths at 76% (76% for males and 75% for females) at a time when the HIV prevalence was estimated at 16% (Doctor & Weinreb 2003). In Zimbabwe, with an HIV prevalence of 15–21%, 61% of deaths among males aged 17–54 and 70% among females aged 15–44 were attributed to HIV (Lopman et al. 2006). In a similar study in Tanzania, with a population with HIV prevalence of 4%, 35% of deaths among persons aged 15–54 years were attributed to HIV (30% among men and 41% among women) (Todd et al. 1997). Verbal autopsy alone may lead to an underestimate of the magnitude of HIV-related deaths. Many deaths are multifactorial, and involve several medical conditions; it is probable that not all HIV-related deaths were captured as such using VA.

The maternal mortality rate obtained (753 per 100 000) was higher than recorded for the whole of Kenya in 2003 (414 per 100 000) (Central Bureau of Statistics Kenya & ORC Macro 2004). Our finding may still be an underestimate; if the deaths of nine women with unknown period between miscarriage and death occurred within 42 days of the miscarriage, our estimate would be 926/100 000. A rate of 753–926/100 000 would fall into a similar range as other high HIV-prevalence countries such as Lesotho (national HIV estimate: 23%, maternal mortality 762/100 000) and Zambia (national HIV estimate: 16%, maternal mortality 729/100 000) (Central Statistical Office Zambia, Central Board of Health Zambia & ORC Macro 2003; Ministry of Health and Social Welfare Lesotho et al. 2004). Maternal HIV infection may affect the maternal mortality estimate both ways: because of HIV-related severe illness, more pregnancies may end in miscarriage or result in (infectious) complications in pregnancy and an increase in maternal mortality. As a contrast, HIV-related severe illness may also mask pregnancy as we saw in our study. In addition, HIV infection has been associated with reduced fertility (Lewis et al. 2004). But overall, HIV infection has been associated with an increase in maternal mortality (Bicego et al. 2002).

The use of alcohol was common among the male adult population. The use of lifestyle drugs is likely to be an underestimate, given the nature of reporting and the sensitivity of the issue. Traditional brews have been associated with intoxication and death in Kenya because of poisonous ingredients, and this may explain the higher number of injuries/poisoning reported among alcohol users (Willis 2000). Alcohol use has been associated with an increased risk of injuries (Borges et al. 2006). However, a sudden death in the African setting is often interpreted by relatives and possibly also medical reviewers as poisoning, or alcohol-related poisoning if the deceased was known to use alcohol, although other causes could be possible; the classification of sudden death causes should therefore be viewed with caution, as reported previously (Kahn et al. 2000).

Non-communicable disease is the leading cause of death in the world, except in Africa (World Health Organization 2005). Developing countries in Africa have a triple burden: the highest mortality in the world from infectious diseases, increasing non-communicable diseases, and a considerable burden of injuries (World Health Organization 2005). Of the non-communicable diseases reported in our study population, the prevalence of high blood pressure is of concern; cardio-vascular disease was the most common non-communicable disease cause of death. Cancers of the gastro-intestinal tract were the most frequently reported among deaths of malignancies. Notably, gastric and oesophagus cancer have been associated with Helicobacter pylori infections and as such may have a communicable component as well (Perez-Perez et al. 2004). Recently, cost-effective interventions have been summarised, which governments could consider to reduce the national burden of non-communicable diseases, including the promotion of reduction of salt intake and potentially the reduced intake of saturated and trans fats (Gaziano et al. 2007). Nevertheless, the burden of non-communicable diseases in this rural area is of a small scale compared with the infectious causes; even in the oldest age group, only half of all deaths were assigned a non-communicable disease cause, contrasting with developed countries where chronic diseases account for over 80% (Murray & Lopez 1997). It is clear that in this rural area, immediate gain in delaying mortality would be achieved from effective programmes providing care for HIV-infected persons and programmes directed towards the prevention of HIV infection.

Verbal autopsy has several well-known limitations including recall bias. Although these might be minimised somewhat if the interview follows closely after the event, such timing may not be culturally appropriate. In our site, a minimum delay of 1 month was recommended and implemented after consultation with community members. Other factors that may affect the outcome include the questionnaire design, choice of interviewers and respondents, cause of death ascertainment mechanisms, and procedures for coding and tabulation of data (Chandramohan et al. 1994;Garenne & Fauveau 2006). Ideally, the method of VA should be validated; however, a limitation of such studies is that they generally compare death causes obtained by VA with the death causes obtained from clinical records for persons who died in health facilities (Kahn et al. 2000; Setel et al. 2006; Yang et al. 2006). Similar to births in this region, the fast majority of deaths among residents (86%) occurred at home. Deaths which occurred in health facilities were clearly not representative of deaths in the community, and were associated with a higher level of education, SES, a skilled job, the duration of the last illness and age of the deceased. Verbal autopsy can only be considered as a limited tool to assess the broad picture of cause-specific mortality in areas with insufficient systems to register and certify causes of death.

However, results of validation studies of VA for adults in areas affected by the HIV epidemic are encouraging. A small early VA study in Uganda reported a sensitivity and specificity of more than 80% for HIV (Kamali et al. 1996). A larger study using data from hospitals in Tanzania, Ethiopia and Ghana found a sensitivity of >75% for direct maternal causes, injuries and HIV or TB, and of 82% for communicable diseases combined (Chandramohan et al. 1998). In Tanzania, where a shorter form was used with 80% of overlap in questions compared with the Indepth form, VA showed a sensitivity of over 50% for TB, HIV and malaria (Setel et al. 2006). A sensitivity of 89% for communicable diseases was detected in a validation study in South Africa using a form similar to ours (Kahn et al. 2000). Lower values for sensitivity and positive predictive value have generally been reported for non-communicable diseases (Chandramohan et al. 1998; Kahn et al. 2000).

Although not many people died in a health facility, care was sought in a health facility by 70%, and care from any source was sought by over 90%. This could imply that there is potential to treat diseases and prevent deaths using the health facility system, and studies could examine what amendable patient-factors and health facility-related factors can be identified which can improve the outcome of diseases among adults in this region.

In conclusion, in this rural area in western Kenya, among adults and adolescents, the majority of deaths were due to potentially preventable infectious diseases. HIV and TB were the most important causes, despite the weakness of the VA instrument in capturing HIV-related deaths in settings where stigma is an issue. Programmes to prevent HIV infection and decrease HIV- and TB-associated mortality are urgently needed and have been implemented in the area. Their impact can be evaluated against this baseline information.

Acknowledgements

We thank the respondents for their help in collecting this information, and all the staff for helping to collect and process it. We acknowledge the help of John Arudo, Amek Nyaguara, Kim Lindblade, Dan Rosen and Kayla Laserson. We thank the director of KEMRI for support.

Disclaimer

The views or assertions contained in this paper are the private opinions of the authors and are not to be construed as official reflecting the views of the U. S. Public Health Service or Department of Health and Human Services. Use of trade names is for identification only and does not imply endorsement by U. S. Public Health Service or Department of Health and Human Services.

Ancillary