ca Fujieda 2006

Methods	Prospective cohort study carried out in 54 clinics around Japan during the 02 to 03 season. The study assessed the effectiveness of TIV against ILI. Baseline questionnaires were filled in at enrolment and then an "attack" questionnaire in which every week for 17 weeks parents recorded children's body temperature in 3 steps of 1°C There authors report ILI surveillance Japan‐wide with peak isolates of A and B viruses in Jan to Feb. The authors describe an analysis stratified by age and other potential confounders (which are reported in Table 1). Systematic differences in age, birth and current body weight, number of siblings, family members, number and space in rooms etc are significantly different between hemicohorts
Participants	2913 children (1512 vaccinees and 1401 non‐vaccinees) under 6 years of age (52% males). Allocation was on an alternation basis according to the provision of parental informed consent and the following child whose parents did not give consent was allocated to the control arm. Attrition is not mentioned. Data by age group and location are reported but not extracted
Interventions	TIV (A/New Caledonia/20/99(H1N1), A/Panama/2007/99(H3N2) and B/Shandong/7/97) or no vaccination in 1 or 2 shots according to age. Producer not described. Matching not reported
Outcomes	Serological N/A Effectiveness ILI: acute febrile illness occurring during the highest epidemic period in each study area (but it is ILI, not influenza as claimed by the authors). Fever reported as below 38°C, between 38°C and 39°C, and 39°C or more (but no description of how temperature was taken by parents or whether follow‐up was complete) Safety N/A
Funding Source	Government
Notes	The authors conclude that the adjusted OR and its 95% CI were calculated by the proportional odds model using logistic regression with 3‐level outcome variables (< 38.0/38.0 to 38.9/> or = 39.0 degrees C). A significantly decreased OR of vaccination was observed (OR: 0.76; 95% CI 0.66 to 0.88), corresponding to a vaccine effectiveness (1‐OR) of 24% (95% CI 12% to 34%). When the analysis was confined to those aged > or = 2 years, a more pronounced OR (0.67; CI 0.56 to 0.79) was obtained with a vaccine effectiveness of 33% (21% to 44%). On the other hand, no significant vaccine effectiveness was detected among very young children; the ORs were 1.84 (CI 0.81 to 4.19) for those < 1 year of age and 0.99 (CI 0.72 to 1.36) for those 1.0 to 1.9 years of age and 1.07 (CI 0.80 to 1.44) when these 2 age groups were combined. Thus, among very young children vaccine effectiveness could not be demonstrated Lack of description of matching, unacceptable ILI definition (fever only), recall bias, measurement bias, unknown attrition, systematic differences between hemicohorts etc. make the study at high risk of bias. Of note in the Results is the reporting of the range of percentage of A and B isolates in each study area as a proportion of samples submitted during the height of the epidemic by sentinel physicians from symptomatic cases: 3% to 61%. In other words if data from this non‐random sampling is generalisable, up to 97% of ILIs were not due to influenza
Risk of bias
Bias	Authors' judgement	Support for judgement
PCS/RCS‐Selection Exposed cohort	Unclear risk	Volunteer
PCS/RCS‐Selection Non Exposed cohort	Unclear risk	Volunteer
PCS/RCS‐Comparability	High risk	Several difference between groups at baseline
PCS/RCS‐Assessment of Oucome	Low risk	Secure record
Summary assessments	High risk	Plausible bias that seriously weakens confidence in the results