The intention is to introduce a rotavirus vaccine into the vaccination programme of Kenya in the near future. Based on efficacy data in sub-Saharan Africa this introduction is expected to result in a significant reduction of the disease burden caused by this virus. Two vaccines are available with similar efficacy for protection against severe rotavirus in the first year of life. Applying vaccine prices per dose of US$3.5 for RotaTeq or US$2.5 for Rotarix in the analysis both vaccines can be considered cost effective using the WHO threshold. Furthermore, the conclusion of ‘cost effective’ for both vaccines is robust. Firstly, in our analysis herd protection was not included although this has been reported 
, inclusion of this will improve the cost effectiveness. Secondly, even if only the death and disease burden as registered in the health care system are considered (rather than that assumed to arise in the community not recorded directly by the health system) the verdict remains ‘cost effective’. Notwithstanding the above, in all scenarios Rotarix will be more cost-effective, and create less strain on capacity than RotaTeq due to the fewer vaccine doses (lower cost; less storage volume).
The major factor influencing the cost effectiveness estimation is rotavirus-associated mortality, most of which is assumed to arise outside the health care system. The uncertainties originate in the estimates of (i) the mortality rate in children aged less than five years, (ii) the proportion of mortality due to diarrhoea and (iii) the proportion of diarrhoea caused by rotavirus, where the last in the list is the most uncertain. The under five mortality is based on a household survey of Kenyan mothers who gave birth in the preceding five years with a sample size of only 3973 and thus of dubious precision 
. However, since mortality tends to be under-reported, the uncertainty will in all likelihood lead to under-estimated disease burden. The estimated value of 74 per 1000 live births is lower in comparison to similar surveys in the recent history 
. In Ethiopia, a neighbouring country, 23% of the mortality was found to be caused by diarrhoea in a recent study 
; the estimated percentage based on a meta-analysis was 19% in Africa 
, suggesting our estimate (14.5%) is on the conservative side. The percentage diarrhoea mortality caused by rotavirus was set at the value observed from the district and provincial hospital data (29%). The proportion of disease caused by rotavirus was higher in Kenyatta National Hospital (42%), where concomitantly the average Vesikari score was substantially higher compared to lower level hospitals. If, as this suggests, the percentage of rotavirus is higher among the more severe diarrhoea cases, then the mortality due to rotavirus may be under-estimated in our study using the lower prevalence of rotavirus in diarrhoea cases. Nonetheless, the overall estimated incidence of rotavirus mortality (65 per 100,000) is comparable with the 68 per 100,000 that was estimated by Tate et.al 
(which uses the same method, but different data).
In addition, the analysis is strongly influenced by estimates of incidence of rotavirus associated disease burden in the hospital and outpatient facilities. The average incidence of hospitalisation of 279 per 100,000 is high relative to the 132 per 100,000 estimated by Tate et al. 
for Siaya District of Western Kenya. The true incidence of hospitalisation will differ by region, this due to true differences in disease incidence and access to health care 
. Accordingly, the incidence estimate for diarrhoeal hospital admissions in our study was scaled to the national estimates of diarrhoeal admission incidence measured in the KDHS. Uncertainty is also associated with the incidence of rotavirus related outpatient visits. The estimated incidence of outpatient visits of 3016 per 100,000 is low compared to the 21,800 clinic visits per 100,000 children under five estimated by Tate et al 
, but high compared to the 932 per 100,000 estimated in India 
. We based our incidence on the recorded outpatient visits which are nationally counted. The alternative, to use disease incidence 8-fold higher based on KDHS questions to mothers about diarrhoea in their infants (154,834 cases per 100,000), was investigated in the sensitivity analysis leading to more favourable cost effectiveness.
There are several issues relating to uncertainty surrounding the estimates of vaccine efficacy for rotavirus vaccines that might have a bearing the reported results. Firstly oral vaccines are less efficacious in developing countries 
, for ill-defined reasons. Secondly the vaccine does not protect against all serotypes; there is an ongoing antigenic drift and emergence of new variants due to reassortment and animal introductions 
. The cross protection for Rotarix and RotaTeq might be different for those genotypes causing differences between the efficacy of the two vaccines. Thirdly the follow-up of vaccinated children has been short. The maximal follow-up has been 2 years after receiving the vaccine, applying the vaccine efficacy beyond this point is therefore speculative, in general the efficacy declined steeply after the first year of age. Fourthly the timing and the number of doses applied in the clinical trials differ from the schedule considered for introduction in Kenya (2 dose; at 6 and 10 weeks).
The verdict of “cost effective” depends on the applied threshold for the cost per DALY. We used the threshold of US$1,600 based on the Kenyan GDP per capita corrected for purchasing power parity, which is in line with WHO Choice definition and regional thresholds 
. Adoption of the value used by Tate et.al 
, that is, the absolute GDP per capita (US$580; 2006), would not alter the verdict of cost effective for any of the scenarios explored in this study () excepting where community mortality was excluded. In this case neither vaccine would be considered cost effective. There are previous estimates for the cost effectiveness of rotavirus vaccine in Kenya 
. Applying the full coarse costs we are able to compare with Tate et al. 
who estimated a slightly higher cost per DALY for both vaccines; $168 for Rotarix (full course $7.4) and $343 for RotaTeq (full course $12.9). Comparison with Atherly et al. 
is more difficult, as their assumption of the inclusion of administration costs and the number of doses is unclear, however all tested possibilities resulted in cost per DALY below $100 per DALY for Rotarix, and around $100 per DALY for RotaTeq. Our estimates are therefore in between previous estimates.
Finally there are other important arguments to consider in the decision between RotaTeq and Rotarix apart from cost effectiveness. These include the impact on the cold chain and the reliability of the vaccine at administration. RotaTeq has more doses and a larger volume per dose, leading to a more severe impact on the cold chain compared to Rotarix. To achieve the predicted vaccine impact it is essential that the vaccine is in good condition when it is administered. In regions with unreliable power supply the cold chain is harder to maintain. Rotarix has proven thermal stability 
for vaccines stored at 37C for 7 days if correctly handled. Rotarix comes with a vaccine vial monitor, making it possible to see at administration if the vaccine has suffered from storage. This strongly reduces the probability that faulty vaccine is administered. In contrast, at the time of writing, RotaTeq did not have proven thermal stability and it was not available with vaccine vial monitoring technology.