Written informed consent was obtained for data and specimen collection at the clinics and households. For children <13 years of age, written informed consent was obtained from parents or guardians for specimen collection. For minors aged 13–17 years of age, written informed consent was obtained from parents or guardians and written assent from the minor him/herself for specimen collection. The protocol and consent forms were reviewed and approved by the Institutional Review Boards of KEMRI (#932) and CDC (#4566).

CDC’s International Emerging Infections Program and the Kenya Medical Research Institute (KEMRI) have conducted population-based, infectious disease surveillance since late 2005 in Asembo, Nyanza Province, in rural western Kenya [2], [3]. All households in 33 villages within 5 kilometers of the referral facility, St. Elizabeth Lwak Mission Hospital (Lwak Hospital), were offered enrollment –78% were enrolled. The surveillance population on July 1, 2007, included 21,420 persons ≥5 years old, who had resided permanently in the area for at least 4 calendar months [4], [5]. Enrollment was continuous since the project’s beginning. Malaria transmission was holoendemic [5], [6]. HIV prevalence was high (17% in adults ≥18 years in 2008) [7].

From March 1, 2007, to February 28, 2010, all enrolled participants received free medical care by KEMRI/CDC-trained nurses and clinical officers at Lwak Hospital for acute illnesses [2], [3]. Lwak hospital has 40 inpatient beds and manages most hospitalizations, only referring complicated patients. No chest radiographs were taken during the study period.

ARI was defined, using a variation on the definition of severe acute respiratory illness suggested for influenza surveillance, as cough or difficulty breathing or chest pain and documented axillary temperature ≥38.0^oC or oxygen saturation <90% or hospitalization [2], [8]. For ARI patients, blood, nasopharyngeal and oropharyngeal specimens using polyester-tipped swabs, and urine were collected. Giemsa-stained malaria blood smears were performed on patients with history of fever or documented temperature ≥38.0°C.

Community interviewers visited enrolled households every two weeks and questioned participants using a standardized questionnaire, in the local language, about all illness episodes in the past two weeks, including symptoms and health-seeking [3]. For this analysis, we defined ARI from the household visits as cough, difficulty breathing or chest pain and reported fever. The ARI cases from the household visit were only used to assess health-seeking patterns in the area and not directly included in the incidence calculations. No specimens were collected at household visits.

From January 1, 2009, asymptomatic controls were enrolled from Lwak Hospital. Eligible controls were those who presented with non-severe illness (i.e., not requiring hospitalization), for immunizations, or for medicine refills. Eligible controls could not have had fever, any respiratory symptoms or diarrhea in the past two weeks. Each month we attempted to enroll a targeted number of controls, frequency-matched to cases by age and HIV status. The monthly target was 23 controls aged ≥5 years old (six HIV-infected), yielding power to detect a significant difference in detection of a pathogen between 12% of cases and 5% of controls. Nasopharyngeal and oropharyngeal swabs were collected on controls.

Clinic nurses attempted to collect five to ten ml of blood for culture, which was inoculated into commercially-produced blood culture bottles and bacterial growth identified using standard methodology, which we have described previously (BACTEC™ Aerobic PLUS™, Becton Dickinson, Belgium) [9], [10].

Naso- and oropharyngeal swabs were placed together in 1 ml viral transport media without antibiotics and transported the same day at 2°C–8°C to KEMRI/CDC laboratories near Kisumu, approximately 60 km from Lwak Hospital, where each specimen was divided into four aliquots and stored at −70°C. In monthly batches, a frozen aliquot was transported on dry ice to the KEMRI/CDC laboratory in Nairobi, where lab technicians, blinded to case-control status, tested it after one freeze-thaw cycle, using quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR). Nucleic acid was extracted from 100 µL aliquots of each sample using Qiagen’s QIAamp viral RNA minikit (Qiagen Inc, California), according to manufacturer's instructions. Prior to August 2008, TaqMan® Universal PCR Master Mix (Applied Biosystems, California) was used for qRT-PCR; later qRT-PCR was carried out using AgPath-ID™ One-Step RT-PCR Reagents (Applied Biosystems). Samples were tested for the presence of adenovirus, respiratory syncytial virus (RSV), human metapneumovirus (hMPV), influenza A and B viruses, and parainfluenza virus (PIV) types 1–3 using qRT-PCR assays using previously published assays [11], [12]. Each clinical specimen was also tested for the human ribonuclease P gene to measure nucleic acid integrity and to confirm sample adequacy. A qRT-PCR test result was considered positive if an exponential fluorescence curve was produced that crossed the assigned threshold at C_T <40.0 [12].

Between January 1, 2009, and February 28, 2010, enhanced testing was done, which included testing respiratory swabs for three additional viruses (rhinovirus, enterovirus, and parechovirus) and atypical bacteria, and testing of asymptomatic controls. For testing of swabs for these additional pathogens, total nucleic acid extracts were prepared from 100µl of specimen per aliquot using MagMAX Viral RNA Isolation Kit (Applied Biosystems) in Nairobi and transported to the Viral Respiratory Diseases laboratory at CDC Atlanta on dry ice. Singleplex qRT-PCR for rhinovirus, enterovirus and parechovirus were performed using previously published methodologies [13], [14], [15]. The rhino- and enterovirus assays targeted the 5′ noncoding region of an area of high sequence similarity between human rhinovirus and enterovirus and exhibit some cross-reactivity (S. Oberste, personal communication). Therefore, positive rhinovirus and/or enterovirus qRT-PCR were reported together as rhino/enterovirus positive. For atypical bacteria, multiplex qRT-PCR was performed at KEMRI/CDC laboratories in Nairobi for Mycoplasma pneumoniae, Chlamydia pneumoniae and pan-Legionella species using published assays [12], [16].

Urine specimens were tested for pneumococcal urine antigen using the BinaxNOW® kit (Binax Inc., Maine) starting in May 2007; kits were not consistently available throughout the entire study period. Sputum specimens were not collected by protocol, but only based on the clinician’s suspicion of pulmonary tuberculosis, and microscopy was done at Lwak Hospital using Ziehl-Neelsen staining.

HIV testing was performed as part of a home-based testing initiative during 2008 when all persons ≥13 years in the surveillance area were offered HIV testing (two parallel rapid HIV tests), as described previously [7]. Seventy-eight percent of eligible adults agreed to be tested. We assumed that a person’s HIV status during home-based testing was the same throughout the study period. HIV-testing was not performed routinely in the clinic on most patients during this period.

Structured questionnaires on scannable paper forms detailing the current illness were completed for all sick visits at Lwak Hospital. (TeleForm® software, Cardiff™, Vista, CA). At the household visits, data were collected using personal digital assistants (PDAs) [3].

Analysis was performed using SAS (version 9.2, Cary, NC). Proportions were compared using chi-square test or Fisher’s exact test, and medians by Wilcoxon Rank Sum test. Rate ratios and 95% confidence intervals were calculated using Fisher’s method (Computer Programs for Epidemiologists, PEPI, version 4.0x) for crude rates and the delta method for adjusted rates [17]. Correlation between monthly rates of ARI and percent positive for each pathogen were tested by Pearson’s or Spearman’s correlation coefficient, depending on normality of the data.

We compared detection of each virus by qRT-PCR between cases and asymptomatic controls. Odds ratios (OR) and 95% confidence intervals were calculated using unconditional logistic regression, adjusting for age group, season when the swab was taken (December-February, hot and dry; March-May, long rains; June-August, cooler and dry; September-November, short rains) and HIV status (positive, negative, unknown). We used the OR to calculate pathogen-attributable fractions, which estimated the proportion of cases positive for each virus in which the virus was the likely cause [18], [19]. The pathogen-attributable fractions were calculated as (OR-1)/OR; pathogen-attributable fractions were only calculated for viruses with ORs that were statistically significant (p<0.05). For purposes of this analysis, we assumed the pathogen-attributable fraction was 1.0 for S. pneumoniae based on the negligible probability of detection of S. pneumoniae in blood or urine of asymptomatic controls [20], [21].

ARI incidence was calculated as the number of ARI clinic visits per 100 person-years of observation. Revisits for the same illness were not counted as separate episodes. Permanent residence status in the surveillance area was used to determine person-time contribution. Adjusted rates of clinic visitation were calculated accounting for the percentage of all clinic visits made for ARI that went to Lwak Hospital as opposed to other area clinics, as determined from the household visits [3]. Etiology-specific incidence was calculated by applying the proportions of each etiology to the adjusted incidence of ARI (Figure 1). Inpatients were over-represented among ARI patients tested at Lwak Hospital (35% inpatients), compared to ARI patients who visited a health facility in the community (5% inpatients), as determined by the household surveillance. Therefore, the inpatient and outpatient etiologic results from Lwak Hospital for each pathogen were weighted using a 19:1 outpatient:inpatient ratio. After adjusting for the outpatient/inpatient distribution, the pathogen-attributable fraction for each pathogen was applied to adjust the etiology-specific proportions. For those pathogens found to have an OR that included 1.0, no incidences were calculated because the role of the pathogen as a cause of ARI was not supported by our data.

Among 3,406 ARI patients, 8.6% reported receiving one or more antibiotics before presentation (4.7% cotrimoxazole, 2.3% penicillin or amoxicillin, 2.0% other). Epidemiologic characteristics were similar between patients who did (n = 1,677, 49%) and did not (n = 1,729, 51%) have blood cultures taken (Table 1). The median blood volume for culture was 4.0 ml. Four percent of blood cultures grew contaminants (Table 2). Among uncontaminated cultures, the most common bacteria isolated were Streptococcus pneumoniae (3%) and non-Typhi Salmonella (3%). The percentage of patients with S. pneumoniae identified increased to 16% when adding urine antigen results to blood cultures (23% inpatients versus 14% outpatients, p = 0.003). Among the 131 patients who had S. pneumoniae detected, 10 (8%) were positive by blood culture only, 114 (87%) by urine antigen only, and 7 (5%) by both tests. The case-fatality ratio was higher among those with positive blood cultures for S. pneumoniae (7%) compared with those with only urine antigen detected (0.8%, p = 0.049). S. pneumoniae were detected more frequently among HIV-positive patients (28%) than HIV-negative patients (13%, p = 0.002).

Mycoplasma pneumonia was detected in 0.7% of patients; no Legionella or Chlamydia pneumoniae were detected (Table 2). Among 47 sputum samples tested by smear for Mycobacterium tuberculosis, 7 (15%) were read as positive.

Epidemiologic characteristics were similar between patients who did (n = 1,216, 36%) and did not (n = 2,190, 64%) have naso/oropharyngeal swabs taken (Table 1). Overall during the time period of full viral testing, 68% of swabs were positive for at least 1 virus, detection being higher among outpatients (71%) than inpatients (58%, p = 0.008, Table 2). The detection of at least one virus decreased with increasing age (p<0.001). The most commonly detected virus was rhino/enterovirus (33%) followed by influenza A virus (22%). Detection of other viruses ranged from <1% to 9%. Influenza A virus was more often detected in outpatients (27%) than inpatients (11%, p<0.001) and among those <35 years old (23%) than those ≥35 years old (11%, p<0.001). Influenza B virus was also more common among outpatients (8%) than inpatients (2%, p<0.001). All other viruses were similarly detected among inpatients and outpatients. There were no significant differences in viral detection between HIV-positive and HIV-negative patients.

Of 569 patients with full pathogen testing, 29% had >1 pathogen identified (Table 3). Detecting >1 pathogen was more common among HIV-positive patients (40%) than HIV-negative patients (25%, p = 0.04), but in similar proportions among inpatients (24%) and outpatients (31%, p = 0.21). The coinfection prevalence among those who had influenza (41%) was lower than that among those with other pathogens (66%, p<0.001). When limited to only those pathogens significantly more common in cases than controls, 18% of ARI patients had >1 pathogen identified; still no differences between inpatients (15%) and outpatients (19%) existed for coinfection (p = 0.51).

Cases (n = 766) had a younger age distribution than controls (n = 273), with 68% and 42% aged 5–17 years old, respectively (p<0.01, Table 4). More cases (44%) than controls (33%) were enrolled in the hot, dry season (p<0.01). Among those with HIV-testing results available, 33% of cases and 39% of controls were HIV-positive (p = 0.28).

After adjustment for age, season, and HIV status, detection of influenza A virus, influenza B virus, RSV and hMPV in the naso/oropharynx was associated with being a case (Table 4). Influenza A virus and hMPV were more strongly associated with outpatients than inpatients, although the confidence intervals overlapped. In contrast, RSV was associated with being an inpatient, but not an outpatient. PIV2 and PIV3 were also more common among cases, but did not reach statistical significance. Rhino/enteroviruses and adenovirus were equally common among cases (33% and 11%, respectively) and controls (24% and 12%, respectively).

The number of ARI cases tended to peak in June and July each year, with a smaller peak in October-December (Figure 2). There was a trend towards increasing ARI cases during the three year period (p = 0.002, linear regression). The number of ARI cases was highest in January-February 2010 when pandemic H1N1 influenza A virus circulated widely in Asembo [22]. Of the pathogens associated with case status in the case-control study, the monthly rate of ARI was correlated with the percentage positive for influenza A virus (p = 0.018), influenza B virus (p = 0.026), and hMPV (p = 0.012), but not for S. pneumoniae and RSV. Of note, when we limited the analysis to the 32 months prior to November 2009, when pandemic H1N1 virus first appeared, the monthly correlation between ARI case rates and viral respiratory pathogens was no longer significant. The percentage of malaria blood smears positive among ARI patients ranged from 8 to 69% by month, and tended to be higher in the rainy months (Figure 2). The monthly ARI rate was correlated with the malaria prevalence throughout the 36 months (p = 0.015), as well as in the 32 months prior to pandemic H1N1 (p = 0.007).

The adjusted overall incidence of ARI resulting in a clinic visit was 12.0 per 100 person-years for persons ≥5 years of age and 8.4 per 100 person-years for persons ≥18 years of age (Table 5). Among HIV-positive persons ≥18 years old, the rate of ARI was 21.4 per 100 person-years, compared to 5.6 for HIV-negative persons (RR 3.8, 95% CI 3.5–4.2). The highest pathogen-specific incidence among persons ≥5 years old was influenza A virus at 2.6 per 100 person-years, followed by pneumococcus at 1.7 per 100 person-years (Table 6). Rates of all pathogens were higher among HIV-positive than HIV-negative persons.