The study team obtained written informed consent from the parents of the SARI cases. Icddr,b Institutional Review Board (Research Review Committee and Ethical Review Committee) approved the research protocol. CDC Institutional Review Board reviewed and approved reliance on icddr,b IRB approval.

We conducted our study in Matlab, a rural sub-district in eastern Bangladesh located about 55 km southeast of Dhaka. Since 1966, icddr,b has maintained a health and demographic surveillance system (HDSS) in this community where earlier ARI studies were conducted [6], [7]. The icddr,b surveillance site was divided into two areas; an icddr,b service area in blocks A–D and an area with usual government health services. Each block was comprised of 10–36 villages. The Maternal and Child Health and Family Planning programme provided a range of free services to pregnant women and children aged <5 years in the icddr,b service area through home visits by community health research workers (CHRWs) and through one icddr,b sub-center outpatient medical facility in each block [13]. The icddr,b sub-centers served as the first level of referral. The Matlab hospital run by icddr,b and located in the A block served as the second level of referral. Fee-based government health care facilities were also present in our study areas, which provided health services to all age groups. We conducted this study in all 67 villages of the icddr,b service area and all 12,850 children aged <5 years residing in the area were eligible for inclusion [14]. We assumed that following children in the community and at the icddr,b hospital would allow us to identify the majority of those who developed SARI because during 2009–2010, only 2% of children aged <5 years in the icddr,b service area sought pneumonia care from government health care providers [15].

We conducted this study during June through October 2010 rainy season. We hired and trained 34 female village health workers to identify potential SARI cases using a checklist for our case-definition (symptoms, age-specific tachypnea and the danger signs). Our case definition for SARI was cough or difficulty breathing with age-specific tachypnea (>60/minute for infant aged 0–1 months, ≥50/minute for infant aged 2–11 months and ≥40/minute for child aged 12–59 months) and one or more danger signs (i.e. unable to drink/breastfeed, lethargic or unconscious, intractable vomiting, convulsions, chest in-drawing or stridor while calm). Through active case detection in the community, we also sought children with cough or difficulty breathing with age-specific tachypnea, without danger signs but still requiring hospitalization. We did not require that children be admitted to the hospital in order to meet the SARI case definition because in the icddr,b service area approximately one in four children aged <5 years were cared for at home rather than being brought to a health care provider for pneumonia care [4].

Health workers made weekly home visit and asked parents about their children’s symptoms and danger signs during the preceding seven days. The field research assistants also made daily visits to the icddr,b Matlab hospital and reviewed the hospital admission record books to identify new SARI cases among hospitalized children. At the end of each day the village health workers and research assistants informed the study physician about potential SARI cases in the community or in hospital via mobile phone. After receiving information about potential SARI cases, the study physician visited the patients at their homes within 48 hours or at the Matlab hospital within 24 hours of notification to confirm that they fulfilled the SARI case definition, enroll cases, and administer a structured questionnaire using a handheld computer. The study physician did not refer community cases to the hospital but advised guardians of all enrolled SARI cases to seek care from an icddr,b subcenter or the Matlab hospital for treatment if they had not done so prior to the physicians’ visit.

The questionnaire contained participants’ demographic and socioeconomic status information (i.e., age, sex, family size, number of rooms in the household, housing materials, ownership of durable goods and monthly household income), where families had sought care for the current SARI event, medical and treatment history (pre-existing medical conditions, symptoms, supplemental oxygen therapy, antibiotics and antivirals), study physician’s physical examination findings (temperature, cyanosis, respiratory rate, breath sounds, rhonchi, crepitation, chest indrawing, stridor and convulsion), peripheral oxygen saturation measured with a portable pulse oximeter and health status at discharge. The study physician collected nasopharyngeal swabs for viral testing from all the enrolled cases and 0.5–3 mls of blood samples using aseptic techniques for bacterial culture from hospitalized cases within 24 hours of admission. We considered it to be a new episode of SARI if the child was symptom free in the preceding 7 days.

The collected swabs were placed in viral transport media, stored and transported at 2–8°C to the icddr,b virology laboratory on the same day of collection. We tested swabs for RSV, influenza A and B viruses, human metapneumoviruses (HMPV), adenoviruses and human parainfluenza viruses (HPIV) 1–3 by real time reverse transcription polymerase chain reaction (RT-PCR) [primer/probe sequences and assay protocols for non-influenza respiratory viruses available from CDC upon request] [16]. Influenza A positive samples were further characterized for seasonal influenza A(H1N1), A(H3N2), A(H1N1)pdm09 and A(H5N1) using real time RT-PCR [16].

The collected blood samples were inoculated into BacTAlert 3D bottle and then transported at ambient temperature to the icddr,b clinical microbiology laboratory within 24 hours. BacTAlert (Organon Teknika Corp., Durham, N.C) machine positive samples were then sub cultured into Blood agar, MacConkey agar and Chocolate agar. After 24 hours of incubation, the agar plates were observed for the growth of Streptococcus species, Staphylococcus species, Hemophilus species, Klebsiella pneumoniae, Enterococcus faecalis, Acinetobacter species, Pseudomonas species and Candida species. Acinetobacter species was identified using an analytical profile index (bioMérieux, France). We considered growth of coagulase negative Staphylococcus as a likely contaminant. Antibiotic susceptibility of the isolated pathogens were performed and interpreted following Clinical and Laboratory Standards Institute guideline [17].

The study team called the parents of all hospitalized cases 21 days after hospital discharge to determine the children’s survival status as influenza-associated deaths caused by secondary bacterial infections could occur during this time [18]. We then contacted all the hospitalized cases again at the end of November 2010 (30 days after last SARI case enrollment) to determine if they had recovered completely from their SARI.

We performed nonparametric equality-of-median tests and two-sample test of proportions (Z-test) to assess demographic characteristics among hospitalized versus non-hospitalized cases (i.e. children who were cared by their families or by outpatient health providers throughout their illness). To compare the age distribution of hospitalized and non-hospitalized cases with RSV, we performed two-sample Wilcoxon rank-sum (Mann-Whitney) tests. We calculated virus-specific incidences of SARI among hospitalized and non-hospitalized cases by dividing the number of SARI events by the person-time each child contributed to the study. We report the annual incidence (person-years, py) of influenza-associated SARI from five months of data because we assumed that the majority of influenza-associated SARIs occurred during the historical influenza epidemic period [12]. We report the incidence of SARI for other viruses in person-weeks (pw) because we lacked a full year of data. We calculated incidences and 95% confidence intervals (CIs) for incidences using Poisson distribution. We calculated 95% CI for the proportion of blood specimens with a bacterial pathogen using binomial distributions, however, because we collected blood specimens only from hospitalized SARI cases.

We followed 12,850 children aged <5 years for 279,029 pw (5,337 py), a mean 22 pw per child observed. Of these, 2,528 (20%) children were aged ≤12 months (Table 1). Fifty-one percent of children were male. The study physician identified 141 SARI cases during the study period; 76 (54%) at home (non-hospitalized cases), 58 (41%) in the hospital and 7 (5%) initially identified at home that were later hospitalized (Figure 1). Two children developed two separate episodes of SARI. According to WHO/IMCI criteria [19], 126 (89%) of these 141 SARI cases met the case definition for severe pneumonia or very severe pneumonia.

The median age of the hospitalized cases was 9 months (interquartile range, IQR 3–15 months) and the median age of the non-hospitalized cases was 23 months (IQR 11–36 months) (p<0.001) (Table 2). Sixty-eight percent of hospitalized cases were male compared to 59% of non-hospitalized cases (p = 0.3).

The median time interval between the first symptom associated with the illness and nasopharyngeal swab collection for hospitalized cases was 4 (IQR 3–5) days and for non-hospitalized cases was 2 (IQR 2–4) days (p = 0.005). At least one virus was detected in specimens from 34 of 65 (52%) hospitalized cases and 42 of 76 (55%) non-hospitalized cases (p = 0.7). The highest proportion of virus detected throughout the study period was RSV (34% in hospitalized and 39% in non-hospitalized cases, p = 0.5) (Table 3) (Figure 2). The median age of hospitalized cases with RSV was lower than non-hospitalized cases (age 7 versus 13 months, p = 0.04). The age-adjusted proportion of hospitalized SARI cases testing positive for individual viruses was similar to that of non-hospitalized cases. We did not identify any influenza A H5 viruses from the SARI cases.

A pathogenic bacteria was isolated from blood cultures of 4 (6% [95% CI 2%–15%]) of 64 hospitalized cases. The pathogens included Streptococcus non pneumococcal species, Enterococcus faecalis, Acinetobacter hemolyticus and Candida species.

The overall incidence of SARI was 50.5/100,000 pw (95% CI 42.5–59.6). The highest virus-specific hospitalized SARI incidence was among children aged 1–2 months with RSV (68.1/100,000 pw) followed by HPIV3 (13.6/100,000 pw) (Table 4). The highest virus-specific non-hospitalized SARI incidence was among children aged 3–6 months with RSV (37.9/100,000 pw). The incidence of influenza-associated hospitalization was 1.1/100,000 pw (0.6/1,000 py).

Overall 51% (72/141) of all identified SARI cases sought care at a hospital or clinic. Of these 72 case-patients, none sought care at the government hospital and 65 sought hospital care at icddr,b, and were hospitalized. Sixty-three percent (41/65) of hospitalized SARI cases only visited icddr,b hospital for treatment while the remaining 37% (24/65) of cases visited two or more health care providers before they were admitted to the icddr,b hospital (Table 5). One-fifth (15/76) of non-hospitalized SARI cases had not visited any health care provider when they were visited by the study physician at a median of 2 days after symptom onset (IQR 1–3 days). The remaining non-hospitalized cases (61/76) visited one or two health care providers for treatment.

Hospitalized cases were admitted a median of 2 days (IQR 1–3) after development of first symptom associated with the illness and were hospitalized for a median of 4 days (IQR 3–6). The predominant signs among hospitalized cases were tachypnoea (88%), rhonchi (83%), crepitation (83%) and chest in-drawing (72%) (Table 6). The predominant signs among non-hospitalized cases were tachypnoea (100%), crepitation (93%) and rhonchi (80%). No danger sign was present in 12% (8/65) of the hospitalized cases when the study physician visited the cases at the hospital. Although non-pneumococcal streptococcus was isolated from a blood culture from one of these eight cases, no viruses were detected from them. Ninety-four percent (61/65) of hospitalized cases received one or more antibiotics and 38% (25/65) of cases received supplemental oxygen. Fifty-nine percent (45/76) of non-hospitalized cases received one or more antibiotics. Fifty-one (78%) cases were discharged from the hospital after improvement of their clinical condition, but were advised to continue treatment at home. During the study period, we identified no cases who died with respiratory symptoms in the community before they could be enrolled and all enrolled cases recovered.

We conducted this study during the June to October 2010 influenza season which limits our capacity to estimate annual incidences of hospitalization for other viruses. Data from this children’s cohort and the national hospital-based influenza surveillance suggests that our study period may have included the majority of influenza infections. Influenza epidemics have been observed in hospital-based influenza surveillance during May-September in 2007–2008 and June–September in 2010 [12], [50]. From the national hospital-based influenza surveillance in Bangladesh, influenza was identified from January through November in 2010 and 63% (357/563) of the influenza illnesses were detected during our study period [50]. Yet, our study included only a proportion of the annual RSV illnesses. In 2010, four surveillance platforms in different locations in Bangladesh suggests that 52% (48/92) of RSV hospitalizations occurred outside our study period (icddr,b internal report). Three to five whole years of data collection could enable us to better calculate virus-specific annual hospitalization rates for all viruses. It is possible that a proportion of children with respiratory infections may have developed complications as a results of their illness, but did not because of their early identification and prompt treatment by icddr,b. The population may also have had increased health awareness and improved health seeking behavior because of periodic visits with our health workers. Therefore, the clinical outcomes of this cohort probably do not reflect those of other areas in Bangladesh. Most of the cases received medications including antibiotics either at a healthcare provider visited prior to hospitalization or immediately after hospitalization before collection of blood sample by the study physician. It was not feasible to collect blood samples from the admitted SARI cases before initiation of medication following hospitalization because field staff sought and identified SARI cases by visiting the hospital only once daily. Thus the study physician collected blood sample within 24 hours of hospitalization.