RSV has been associated with seasonal mortality of young children in the US 
, Brazil 
and Indonesia 
and pneumonia deaths have been noted to correspond with the peak season of bronchiolitis in Bangladesh in 2003 
. Using four years of data, we found that although both RSV detections and respiratory deaths were relatively high in the winter months, the increase in RSV detections was not accompanied by an increase in death during individual years. Of the eight study months with four or more respiratory deaths, only one matched a peak in RSV detections. This could indicate that RSV does not impact respiratory deaths in the young child in Bangladesh. There are, however, other explanations for not finding a significant association between RSV and other respiratory virus outbreaks and respiratory deaths. The timing of RSV outbreaks in the community where specimens were tested may not have matched the timing of RSV outbreaks in the community where verbal autopsies were performed, there may have been a problem with the accuracy in classification of respiratory deaths, or the number of observed deaths was too few to accurately assess correlations with RSV or the other respiratory virus detections. The potential problems identified in this study suggest ways to improve future efforts to assess community mortality from respiratory virus infections.
First, it is essential that surveillance for respiratory virus circulation represent activity in the community where deaths are assessed. Surveillance data from the United States suggests that year-to-year variation in timing of RSV outbreaks in the same community as well as community-to-community variation in the same year 
and indicates the importance of using detection data from the same community where the mortality is assessed. Previous studies of RSV detections Bangladesh found overall peak activity during winter months in Dhaka 
while a rural community had peak activity during the monsoon season and not during winter months 
. In our study, RSV detections showed well-defined increases in activity but the timing of these increases was too varied to infer longer-term patterns. There are differences in both population density and living conditions between urban Dhaka and Matlab, which may affect RSV transmission. Population density is an important contributor patterns of RSV season in the United States and has been found to affect duration and timing of seasons 
. Of note, the incidence of clinical pneumonia in Matlab is less than half that in Dhaka 
. Although these two sites are separated by less than 50 kilometers, variation in timing of local RSV outbreaks has been previously noted within short distances 
and the degree of variation may be greater in this setting. To be confident about linking RSV detections to health outcomes in a single community, data on RSV circulation from the same community are needed.
The small number of respiratory deaths identified by the verbal autopsies made it difficult to establish correlations in RSV activity with these deaths. Verbal autopsy has been validated, shown to be sensitive for respiratory death surveillance 
and used previously in Bangladesh to provide cause of death data to policies and programs 
. However, in this study, one or two more or fewer deaths would change the timing of peaks in deaths. This would make correlations of RSV detections with these peaks unstable and highly susceptible to random events that would affect a peak in the number of deaths. It is possible that respiratory deaths would lag behind an increase in viral detection, however we see no consistent increase in deaths 1 or 2 months after RSV detection.
Etiology studies have shown that RSV is predominantly associated with pneumonia and severe respiratory illness in infants 
and for this analysis, we assumed that deaths followed by respiratory illness in this age group could be attributed to RSV. However, overlap in respiratory virus seasons and co-infections complicate efforts to link detections to a single disease in the community. In this study, the timing of influenza and human metapneumovirus detections were distinct from that for RSV detections, suggesting that circulation of these viruses would not have affected determination of RSV deaths. The timing of adenovirus and parainfluenza virus detections did overlap with RSV detections for some months. Studies of circulation of respiratory viruses in some tropical climates have shown both overlap among these viruses and less consistent seasonal patterns of circulation than those in more temperate locations 
. Thus, we cannot be sure if circulation patterns of any of the respiratory viruses observed in Dhaka were indicative of that in Matlab where deaths were assessed. It is important, however, to understand potential overlaps in circulation and studies looking at links between health outcomes and virus detections should include detection of the spectrum of potentially etiologic pathogens.
To accrue sufficient numbers of deaths for a more robust analysis of RSV-associated mortality in developing countries, studies will likely require several sites with each site using detections and death data come from the same location. Another approach that has been considered for determining RSV-associated mortality is a probe study that measures the reduction in mortality with use of an effective RSV-specific preventive measure. The only presently available preventive measure is immune prophylaxis. When available, an effective vaccine could also serve this purpose. Although our findings do not reliably inform RSV-associated mortality, they do identify factors that should be considered in the design of future studies of RSV disease burden in developing countries.