There is a dearth of information about influenza spread in tropical environments, and how transmission in semi-closed environments such as schools or military camps differs from the general population as high attack rates have been reported 
. In addition, there is little information about the effectiveness of the pandemic vaccine in tropical regions where a large proportion of the world's population resides, and where early availability and access to such vaccines may pose a challenge.
The pandemic H1N1 outbreaks in the military Camp A in December 2009, January and May 2010, and the influenza B outbreak in March 2010 demonstrate the unique vulnerability of recruit populations. The physical and psychological challenges associated with starting military life, including living in close proximity, can result in increased influenza transmission 
. In addition, as new recruits enter from the general population, they may introduce infections prevalent in the community, which are then amplified in the closed environment as recruits are initially confined to camp. This probably also applies to other closed and semi-closed communities such as schools and boarding facilities.
Once the vulnerability of the new recruits was recognised and inactivated monovalent pandemic H1N1-2009 vaccine became available, a mass vaccination campaign with the vaccine was introduced from December 2009 for new recruits in Camp A. Our simple Bayesian model showed that the risk for pandemic H1N1 was 86% lower in Camp A than in other camps after vaccine introduction, and similarly the GLM accounting for possible confounders showed a significant decrease of 54% over what was expected in the absence of vaccination. Rates of pandemic H1N1 remained low despite national outbreaks.
Our study provides additional evidence of the effectiveness of monovalent pandemic influenza vaccination. Puig et al showed a 90% vaccine effectiveness in preventing H1N1-associated hospitalisations 
; Simpson et al, using vaccination status among ILI cases, showed a vaccine effectiveness of 95% 
. Another multi-centre case-control study in Europe found a lower 66.0% to 71.9% vaccine effectiveness using different statistical computations 
. Ours is the first study, to our knowledge, conducted in the tropics among a semi-closed population that takes into account the burden of disease both internally and externally to determine the actual effectiveness of the vaccine based on laboratory confirmed infections.
After the introduction of pandemic H1N1 vaccine, we observed a large rise in the risk of influenza B in Camp A compared to the other camps, with a high ratio of RRs of 18 (though with a broad 95% BCI that spanned 1); when accounting for potential confounders and autocorrelation via the GLM, the risk rose to 66 times that in the unvaccinated group, with enough precision to exclude the possibility of no change. Potential reasons could be the dominance of influenza B in the ecological niche of the vulnerable recruits given the effective suppression of pandemic H1N1 by vaccination. Considering that the strain of influenza B responsible for the outbreaks in 2010 was matched to the prevailing trivalent seasonal influenza vaccine, a universal seasonal influenza vaccination program would probably have prevented this.
In addition to providing vaccines to vulnerable populations, the timing of vaccination is important. The two pandemic H1N1 outbreaks in Camp A in January and May 2010 showed that early outbreaks could still occur due to the lag time between vaccination and establishment of immunity. This is substantiated by pandemic H1N1 cases who had been vaccinated within two weeks prior to disease onset which constituted the majority of the “vaccine failures”. One possible solution would be to vaccinate at-risk groups before they congregate: for example, military recruits before they enlist or school children before starting each school year. This may also reduce community outbreaks, since transmission in schools may catalyse community transmission 
There are some limitations to this study. As influenza transmission patterns change over time, it is inherently difficult to compare between periods before and after vaccination from time series data, and we relied on statistical tools to provide an inductive estimate of the effect of vaccination. Given the non-independent nature of infectious diseases spread, it is difficult to assess vaccine effectiveness directly with a limited number of cohorts, substantial levels of contact within cohorts, and the distorting signal of the consequent herd immunity. To quantify the uncertainty in our estimates, we used indirect methods that rely on intrinsic replication within the time series data prospectively collected. In addition, the populations of the camps may differ in demographics and on prior exposure to the 2009 H1N1 virus. We used two statistical methods to compare between the camps, and between periods in Camp A to minimize confounding. Studies whose purposes do not include surveillance or assessing group-level vaccine effects might use different designs; with extrinsic replication, either in multiple cohorts or using non-interacting participants, allowing more direct measurements of vaccine effectiveness at the individual level. Finally, we did not survey milder cases without fever and asymptomatic cases, as we had to balance completeness with a focused surveillance program. We have however shown that vaccination reduces febrile cases and therefore the potential impact of the program.
Pandemic 2009-H1N1 vaccine has been effective in reducing the burden of 2009-H1N1 among recruits in the Singapore military, and should be made available early for future pandemics. Seasonal influenza vaccination programs for new entrants to such communities may significantly reduce the impact of influenza further, and should be considered routinely.