Bulletin of the World Health Organization

The population-based burden of influenza-associated hospitalization in rural western Kenya, 2007–2009

Daniel R Feikin a, Maurice O Ope b, Barrack Aura a, James A Fuller c, Stella Gikunju a, John Vulule d, Zipporah Ng’ang’a e, M Kariuki Njenga a, Robert F Breiman a & Mark Katz a

a. Centers for Disease Control and Prevention, PO Box 1578, Kisumu, Kenya.
b. Ministry of Public Health and Sanitation, Nairobi, Kenya.
c. Johns Hopkins Bloomberg School of Public Health, Baltimore, United States of America.
d. Center for Global Health Research, Kenya Medical Research Institute, Kisumu, Kenya.
e. Jomo Kenyatta University of Agriculture and Technology, Nairobi, Kenya.

Correspondence to Daniel R Feikin (e-mail: dfeikin@jhsph.edu).

(Submitted: 04 August 2011 – Revised version received: 08 December 2011 – Accepted: 09 December 2011.)

Bulletin of the World Health Organization 2012;90:256-263A. doi: 10.2471/BLT.11.094326


Although influenza has been widely studied in developed countries, little is known about its epidemiology, seasonality and burden in developing countries.16 The epidemiology of influenza and the hospitalization rates associated with the illness may be different in sub-Saharan Africa and more developed areas for several reasons. In temperate regions influenza is highly seasonal, with transmission peaking during winter, whereas in the tropics influenza viruses are thought to circulate year round.7 In Africa, the risk factors for severe outcomes from influenza, such as chronic heart disease, chronic lung disease, infection with human immunodeficiency virus (HIV) and smoking, differ from those found in Europe and North America.4,810 Regional differences in the age distribution of human populations and their use of health services may also affect the rate of hospitalization for influenza.

In Kenya, people hospitalized with respiratory illnesses are rarely tested for viral pathogens. Influenza vaccine and drugs active against influenza viruses are not readily available, and little is known about the relative contribution of influenza to severe respiratory disease. In any country, data on influenza can help guide clinical management and allow the health ministry to estimate the influenza burden so that the potential impact of targeted interventions, such as immunization, can be assessed.

Following the 2006 outbreak of avian influenza in Nigerian poultry, Kenya established prospective surveillance for human influenza in several hospitals, with particular emphasis on the detection of the H5N1 subtype of the influenza A virus.11 We describe here the results of population-based, district-wide surveillance in western Kenya for hospitalization with influenza virus in the two years before A(H1N1)pdm09 reached Kenya.12


Study site

The study was based in the Bondo district of western Kenya, which had a population of 238 780 at the time of the 1999 national census. The district’s population is culturally homogenous and almost entirely rural and poor, as it consists mainly of individuals of Luo ethnicity who live by subsistence farming and fishing.13 Most roads in the district are unpaved. The area is holoendemic for malaria and has a high child mortality rate (212 deaths per 1000 live births in 2008).13 It also has a high prevalence of HIV infection in young adults. In 2003–2004, for example, HIV prevalence was 15.4%, or twice the corresponding national value, among those aged 13–34 years who lived in Bondo district.8

Bondo district has seven inpatient facilities, including one district hospital, as well as 45 outpatient health facilities. The residents of Bondo district seek care in these facilities and also from private local providers, traditional healers, drug sellers and community health workers.14 The hospital in the neighbouring district of Siaya and the provincial hospital in Kisumu are 27 and 57 km from Bondo town, respectively, and rarely admit patients from Bondo district.15 In one part of Bondo district, 25 000 people are offered free, high-quality care at Lwak Mission Hospital, including admission, as part of population-based infectious disease surveillance.16,17 Influenza vaccination is not available in the district in either the public or the private sector.

Data collection

In May 2007, the United States Centers for Disease Control and Prevention (CDC) and the Kenya Medical Research Institute (KEMRI) established prospective population-based surveillance for influenza in all seven inpatient facilities in Bondo district. Trained surveillance officers, mostly nurses, enrolled all consenting patients who were hospitalized for respiratory illness, defined as cough, difficulty breathing or (for patients aged ≥ 5 years) pleuritic chest pain. Patients who had had a cough for more than 2 weeks when they presented were excluded because the study’s focus was on acute respiratory illness. Also excluded were those presenting within 3 days of a previous hospitalization for the same illness. A structured questionnaire and scannable forms (Cardiff Teleform, Vista, United States of America) were used to collect the data, which included demographic characteristics, clinical symptoms and signs, treatment history, avian influenza risk factors and medications. Data were stored in Access (Microsoft, Redmond, USA) databases and regularly checked for errors using cleaning programs; incorrectly completed forms were returned to the field for correction.

Specimens and laboratory methods

Dacron-tipped swabs were used to collect a nasopharyngeal and an oropharyngeal sample from each enrolled subject, and then the pair of swabs was placed, together, in a vial containing 1 ml of virus transport medium (VTM) without antibiotics. In four of the health facilities these specimens were transported on the day of their collection, at 2−8 °C, to the KEMRI/CDC laboratories in Kisumu, approximately 60 km from the collection sites. In the other three health facilities, which were relatively far, the specimens were rapidly placed in tanks of liquid nitrogen and these tanks were taken to the KEMRI/CDC laboratories in Kisumu every 2 weeks. Once at Kisumu, the VTM in each specimen (after thawing, if frozen) was divided into four aliquots and stored at −70 °C. About once a month, the frozen aliquots were transported, on dry ice, to the KEMRI/CDC laboratory in Nairobi (a distance of about 350 km), where one aliquot per enrolled patient was tested for influenza virus ribonucleic acid (RNA) by means of real-time reverse transcriptase polymerase chain reaction (RT–PCR).

The total RNA from a 100-µl subsample of an aliquot was extracted using a QIAamp viral RNA minikit (Qiagen Inc., Valencia, USA), in accordance with the manufacturer’s instructions. One-step RT–PCR was then carried out using the AgPath kit (Applied Biosystems, Foster City, USA) and a protocol developed by the CDC’s Influenza Division in Atlanta (S Lindstrom, unpublished data, 2010). The protocol is available upon request, under a material transfer agreement, from the CDC in Atlanta. Specimens yielding crossover threshold (CT) values of ≤ 39.9 were considered positive, whereas those giving higher CT values or no CT values were considered negative. Specimens found positive for influenza A virus RNA were subtyped for seasonal H1 and H3 using another real-time RT–PCR, with primers, probes and positive control samples of influenza viruses provided by the CDC in Atlanta.

Blood smears from all febrile patients (≥ 38.0 °C) admitted to Lwak Mission Hospital and from all children admitted to Bondo District Hospital were prepared, Giemsa-stained and checked for malarial parasites by light microscopy. At the five other study facilities, such blood smears were prepared at the clinician’s discretion. Although there was no routine testing of inpatients for HIV, some enrolled patients were checked for HIV as part of a case–control study of influenza risk factors.10

Data analysis

Rates were calculated using age-specific population projections for Bondo district from the 1999 census, assuming an annual growth rate of 1.8%.18 Adjustments were made for the proportion of eligible patients who did not have a swab taken (because the patient refused, or was admitted when surveillance staff were not available, or left the hospital or died before specimens could be collected) and for the proportion of patients from whom swabs were taken but not tested for influenza virus. These adjustments were made separately for each health facility and included in the study, each age group considered and by season (high in May–October, low in November–April), each time assuming that the proportion of the enrolled but untested patients who were infected with influenza virus was the same as the corresponding proportion of tested patients who were found positive for influenza virus RNA. The binomial distribution method was used to calculate 95% confidence intervals (CIs) around each rate.19 Case-fatality rates (CFRs) were estimated, with similar adjustments for the proportion of eligible patients who had swabs tested, age group, season and health facility. The categorical variables were compared by means of χ2 tests. Stata version 11 (StataCorp. LP, College Station, USA) was used for all data analyses.

Ethical review

This surveillance project was approved by the KEMRI National Ethical Review Committee and exempted, as non-research, by the CDC Institutional Review Board. Written informed consent was obtained from all adult participants and from the parents or legally authorized representatives of the enrolled children.


Between June 2007 and May 2009, 3924 patients were hospitalized with respiratory illness and enrolled in the present study (Table 1). Most (58.6%) were children aged < 5 years. the="" median="" length="" of="" stay="" was="" 2 days.="" although="" nasopharyngeal="" and="" oropharyngeal="" swabs="" were="" collected="" from="" 2389="" (60.9%)="" of="" the="" patients="" with="" respiratory="" illness,="" only="" the="" swabs="" from="" 2079="" (87.0%)="" of="" these="" patients="" were="" tested="" (table 1);="" the="" swabs="" from="" the="" other="" 310="" patients="" were="" not="" of="" adequate="" quality="" to="" be="" tested="" or="" were="" mislaid.="" when="" the="" enrolled="" patients="" from="" whom="" swabs="" were="" collected="" were="" compared="" with="" the="" other="" enrolled="" patients,="" no="" marked="" differences="" were="" seen="" in="" terms="" of="" demographic="" characteristics="" or="" median="" length="" of="" hospital="" stay.="" although="" those="" who="" had="" swabs="" collected="" had="" a="" slightly="" higher="" cfr,="" the="" difference="" was="" not="" statistically="" significant="" (3.6%="" versus="" 2.7%;="">P = 0.12). The proportion of eligible patients who were swabbed was much smaller in the Lwak Mission Hospital than in the other six study facilities (20.3% versus 89.2–98.8%; P < 0.001 for="" each="" pair-wise="" comparison;="">

Influenza virus RNA was detected in 204 (9.8%) of the 2079 swabs tested; 177 (8.5%) and 28 (1.3%) of the swabs tested were PCR-positive for influenza A virus and influenza B virus RNA, respectively. One patient’s swab was positive for both influenza A and B virus RNA (Table 1). Influenza virus RNA was detected in 6.8% and 14.0% of the swabs from patients aged < 5 and="" ≥ 5 years,="" respectively.="" four="" patients="" with="" pcr-confirmed="" influenza="" –="" three="" children="" aged="">< 5 years and="" one="" older="" patient="" (giving="" a="" cfr="" of="" 3.6%="" and="" 0.8%="" for="" the="" two="" age="" groups,="" respectively)="" –="" died.="" the="" overall="" cfr="" was="" similar="" for="" patients="" who="" tested="" positive="" and="" negative="" for="" influenza="" virus="" (2.0%="" versus="" 3.7%;="">P = 0.24). The subtypes of influenza virus were successfully identified for 25 patients infected with influenza A virus: 18 harboured H3N2 and 7 carried seasonal H1N1 (although the first case of A(H1N1)pdm09 in Kenya was not identified until late June 2009, after surveillance for the present study had ended)12.

Females accounted for slightly more than half of the patients with respiratory illness (54.3%), influenza A virus (59.3%) or influenza B virus (67.9%). The median age of patients admitted with respiratory illness was 3 years (range: 2 days to 99 years; (Table 3). The patients found infected with influenza A virus were generally younger than those found infected with influenza B virus, with median ages of 6 years (range: 1 month to 77 years) and 26.5 years (range: 6 months–80 years), respectively. Overall, 53 patients aged > 12 years were found PCR-positive for influenza virus and were tested for HIV. Of these, 13 patients (24.5%) – all from the 22 patients aged ≥ 18 years – were found to be HIV-infected. The cases found infected with influenza virus had been ill for a median of 4 days when their swabs were collected. Such cases were less likely to have presented with fever (i.e. a body temperature of at least 38.0 °C) than the patients found negative for influenza virus (17.2% versus 27.6%; P <>

Blood samples for the preparation of smears were collected from 2623 (67%) of the inpatients with respiratory illness. The influenza-virus-positive cases investigated were just as likely to be found smear-positive for malaria as the influenza-virus-negative cases (36.2% versus 33.8%; P = 0.62). Only one patient found infected with influenza virus had a positive blood culture, with Group B Salmonella isolated. Although more swabs were tested in the first year of surveillance than in the second, the proportion of swabs found positive for influenza virus was lower in the first year (7.7% versus 12.3%; P < 0.01; table 4).="" in="" both="" years,="" most="" influenza-virus-positive="" cases="" were="" detected="" between="" may="" and="" october,="" with="" the="" monthly="" numbers="" of="" cases="" peaking="" in="" july="" and="" august="" (fig. 1)="" –="" a="" time="" when="" western="" kenya="" is="" usually="" dry="" and="" relatively="" cool="" (21–23="" °c).="" in="" the="" second="" year,="" but="" not="" the="" first,="" there="" was="" a="" second="" smaller="" peak="" of="" cases="" in="" january="" and="" february="" –="" a="" period="" that="" is="" also="" usually="" dry="" in="" western="" kenya="" but="" warmer="" (24–25="">

Fig. 1. Laboratory-confirmed influenza among patients hospitalized with respiratory illness, Bondo district, Kenya, June 2007–May 2009
Fig. 1. Laboratory-confirmed influenza among patients hospitalized with respiratory illness, Bondo district, Kenya, June 2007–May 2009

Over the two-year study period, the mean annual rate (per 100 000 population) of hospitalization for respiratory illness was 699.8, whereas the corresponding rates of influenza-associated hospitalizations were 56.2 for influenza virus infection of any type, 50.3 for influenza A virus and 6.1 for influenza B virus (Table 5, available at: http://www.who.int/bulletin/volumes/90/4/11-094326). Compared with older children and adults, children aged < 5 years had="" a="" much="" higher="" rate="" of="" influenza-associated="" hospitalization:="" 143.7="" cases="" (95%="" ci:="" 119.6–167.8)="" per="" 100 000="" versus="" 36.7="" cases="" (95%="" ci:="" 33.0–44.4)="" per="" 100 000.="" overall,="" the="" rate="" of="" influenza-associated="" hospitalization="" was="" also="" higher="" among="" females="" than="" males="" (fig. 2).="" there="" was="" a="" significantly="" higher="" rate="" of="" influenza-associated="" hospitalization="" among="" females="" aged="" 20–34="" or="" ≥ 50 years="" than="" among="" males="" of="" the="" same="" age="">

Fig. 2. Age- and gender-specific annual influenza-associated hospital admissions (per 100 000 people), Bondo district, Kenya, June 2007–May 2009
Fig. 2. Age- and gender-specific annual influenza-associated hospital admissions (per 100 000 people), Bondo district, Kenya, June 2007–May 2009
Note: The vertical lines indicate 95% confidence intervals.

The mean annual rate of influenza-associated inpatient mortality was 1.5 deaths (95% CI: 0.5–2.5) per 100 000 population. There were 7.6 influenza-associated deaths (95% CI: 2.1–13.2) per 100 000 children aged < 5 years but="" only="" 0.2="" (95%="" ci:="" 0.0–0.6)="" per="" 100 000="" individuals="" aged="">

No avian influenza virus (H5N1) was detected, but 19.1% of the enrolled participants reported having heard rumours of sick or dead birds in their neighbourhood in the past two weeks; 9.3% said that they had been within 1 m of a sick or dead bird, and 4.8% that they had touched a dead bird.20


Our prospective surveillance in a district in Kenya is one of the first to report directly-measured rates of influenza-associated hospitalization in African adults and children.4 In developing countries, sentinel hospital surveillance can provide data for describing influenza epidemiology and seasonality, characterizing the circulating strains of influenza virus (to guide vaccinations) and monitoring influenza pandemics. Such surveillance can seldom be used to define the burden of influenza, however, because it often occurs in referral hospitals, where the denominator population is difficult to define and the health-seeking patterns may not be representative of the norm.21,22

The rate of influenza-associated hospitalization that we report in children of Bondo district who were aged < 5 years (i.e.="" 143.7="" cases="" per="" 100 000="" child-years)="" is="" lower="" than="" the="" corresponding="" values="" reported="" in="" south="" africa,="" thailand="" and="" the="" united="">2325 In urban Soweto, South Africa, for example, 309 annual cases of hospitalized influenza were recorded per 100 000 HIV-uninfected children and 1444 such cases were recorded per 100 000 HIV-infected children.25 The lower rate of influenza-associated hospitalization seen in Bondo district might only reflect the local epidemiology of influenza, as related to the patient population and/or climate. However, it can probably be attributed, at least in part, to the relatively low levels of health-seeking for severe illness in this region of Kenya, probably attributable to cultural beliefs, the costs of health care and the large distances to the nearest health-care facilities.15,2628 In a survey of health utilization in Bondo district in 2005, only 26% of children aged < 5 years and="" 16%="" of="" older="" children="" and="" adults="" were="" hospitalized="" for="" a="" reported="" episode="" of="" pneumonia="" in="" the="" previous="">26 The observation that health-care utilization for severe illness is low in this part of Kenya is reinforced by the finding that most deaths in the area occur at home.13 If hospital utilization for severe influenza-associated illness in Bondo district is as low as for pneumonia, the incidence of severe influenza-related respiratory illness in this area could be up to fourfold higher for children and sixfold higher for adults than those reported in this article.

Low levels of hospital utilization despite severe illness may also explain why, in the present study, the elderly had some of the lowest rates of hospitalization for respiratory illness, whereas in developed countries the elderly have some of the highest rates.15,29,30 In rural Kenya, it is generally the elderly who find it hardest to travel the often long distances to their nearest hospital and it is generally also they who are least comfortable with the non-traditional medicine practised in hospitals. The elderly of Bondo district are probably also less likely to suffer from an underlying cardiopulmonary medical condition than their counterparts in developed countries,31 partly because the prevalence of smoking in Kenya is still very low. Most influenza-associated hospitalizations of adults in the United States result from exacerbation of underlying medical conditions.29

Unlike developed areas, Bondo district had an influenza-associated hospitalization rate in young adults that was similar to – not lower than – the rate among the elderly.30,32 Although this could, again, result from the relatively low levels of care-seeking among the elderly, we believe it may also reflect the high prevalence of HIV infection among young adults in this area of Kenya.8 HIV infection may also explain why influenza-associated hospitalizations were more common among the young women of Bondo district than among the young men, since the young women have the higher prevalence of HIV infection.8 In an earlier study in the same district, HIV infection was associated with a 3.5-fold risk of influenza-associated hospitalization among adults.10 Among South African children infected with influenza virus, HIV infection was associated with an estimated eightfold higher risk of hospitalization.33

Although influenza virus activity in temperate climates peaks during winter seasons, influenza virus in Bondo district circulates throughout the year – as seen in other tropical and subtropical climates, such as those of Indonesia, Singapore, Thailand and Viet Nam46 – but tends to have a broad peak of activity coincident with winter in the southern hemisphere. Bondo town is about 11 km south of the Equator.

The proportion of swabs found positive for influenza virus RNA and the CFRs for influenza-associated hospitalizations varied among the seven health facilities included in the present study, although the total number of deaths was small. Lwak Mission Hospital admitted many more patients with respiratory illness than any other of the health facilities, and almost three times more than the district hospital, probably because the hospital has a programme for the population-based surveillance of infectious diseases. This programme offers free, high-quality health care to participants, which promotes greater health care utilization.16,17 The variability in findings between health facilities highlights the limitations of using one sentinel hospital to draw conclusions about the epidemiology and burden of influenza in a given district.

Our study had several limitations. First, we used only molecular methods to detect influenza virus, and some infected cases may not have been detected. In Thailand, serological testing with haemagglutinin inhibition tests identified influenza cases not detected by RT–PCR.34 Second, some of the patients found swab-positive for influenza virus may not have been hospitalized for influenza but for another cause, such as bacterial pneumonia or malaria. Influenza virus might have been present from a recent or concurrent upper respiratory tract infection. Influenza virus has been detected in healthy children used as controls.35 Third, only about 60% of the enrolled patients with respiratory illness had a swab collected. If patients who were not swabbed had a different risk of developing influenza than patients who were swabbed, our adjusted rates could be inaccurate. There were, however, no significant differences in age, length of hospitalization or CFR between those who had and those who did not have swabs taken, and the extrapolation to include untested patients therefore seems appropriate. Excluding untested hospitalized patients with respiratory illness from the hospitalization rate calculation would have led to a clear underestimate of the true rate. Fourth, to define seasonality and annual fluctuations in the rates of influenza-associated hospitalization clearly, surveillance for a period longer than 2 years is required. Fifth, by excluding those who had had a cough for more than 2 weeks at presentation, we may have underestimated the burden of influenza by excluding patients hospitalized with tuberculosis or chronic lung diseases, both of which have been identified as risk factors for influenza-associated hospitalization.10

In conclusion, our population-based study highlights the disproportionately high burden of influenza among children aged < 5 years and="" young="" adults="" in="" western="" kenya.="" the="" findings="" suggest="" that="" influenza="" prevention="" activities="" could="" substantially="" reduce="" severe="" respiratory="" disease="" in="" other="" developing="" countries,="" especially="" in="" young="" children="" and="" other="" people="" at="" particular="" risk="" of="" severe="" respiratory="" illness,="" such="" as="" those="" living="" with="">10,25


This study was conducted as a collaborative activity between the Kenya Medical Research Institute and the United States Centers for Disease Control and Prevention. We thank the District Medical Office of Health of Bondo district. This paper is published with the permission of the director of the Kenya Medical Research Institute.

The findings and conclusions are those of the authors and do not necessarily represent the views of the United States Centers for Disease Control and Prevention.

Competing interests:

None declared.