ca Allison 2006.

Methods	5‐practice retrospective cohort study taking place in Colorado during the 2003 to 2004 season. The study assessed the effectiveness of an undescribed vaccine in preventing ILI in healthy children aged 6 to 21 months. Participant data and immunisation status were identified from reimbursement registers and a web‐based immunisation register. Analysed data come from the period 1 November to 31 December 2003, when influenza A circulated in a prevalent fashion according to hospital isolates. Respiratory syncytial virus started circulating at the end of December, so the authors attempted to restrict analysis to the period of maximum influenza circulation. This, of course, does not mean that other pathogens may not have been co‐circulating. The results are presented for 2 peaks of ILI attendances, 1 corresponding with influenza A circulation and the other with RSV circulation ("influenza and RSV seasons").
Participants	5193 healthy children aged 6 to 21 months. The 21‐month limit was chosen because of billing constraints. Participants were mostly white and privately insured. The authors classified participants as FV, PV, or UV, but as some UV became PVs and FVs as the season progressed, denominators are unstable. In addition, FV includes those who had 1 dose from the previous season, further increasing the confusion. At 1 March 2004 when the study ended there were 2087 FV, 1040 PV, and 2066 UV.
Interventions	1‐ and 2‐dose vaccinations versus do nothing. The vaccine must have been TIV, which is the only one registered in this age group in the USA. No mention is made of content or matching.
Outcomes	Serological N/A Effectiveness Physician's office attendance for: ILI or P&I as defined in ICD‐9. These were assessed only for first visits to the family physician. Safety N/A
Funding Source	Government
Notes	The authors conclude that "a total of 28% of the patients had an ILI office visit and 5% had a pneumonia/influenza visit. HRs for FV versus UV were 0.31 (95% CI 0.3 to 0.4) for ILI and 0.13 (95% CI 0.1 to 0.2) for pneumonia/influenza, corresponding to a vaccine effectiveness (1 ‐ HR 100) of 69% for ILI and 87% for pneumonia/influenza. The corresponding HRs for PV versus UV were 1.0 (95% CI 0.9 to 1.2) and 1.1 (95% CI 0.8 to 1.5) Conclusions Although 2 doses of vaccine were 69% effective against ILI office visits and 87% effective against pneumonia/influenza office visits, 1 dose did not prevent office visits during the 2003 to 2004 influenza season." Summary estimates are presented as HR, and the authors used a Cox proportional hazards model, so no disaggregated numerators are available. As denominators are also moving, the study results are difficult to interpret. Data are reported by influenza (ILI and P&I) and RSV (ILI) seasons. Asymmetrical reporting? It is difficult to assign a design to this study, as the text is unclear on timings and buried in the text is the phrase "This study was conducted as part of a randomised controlled trial of registry‐based reminder recall in 5 private paediatric practices in Denver, Colorado from September 1, 2003 through February 29, 2004 (Kempe A, Daley MF, Barrow J, Allred N, Hester N, Beaty BL, et al). Implementation of universal influenza immunisation recommendations for healthy young children: results of a randomised, controlled trial with registry based recall. Pediatrics 2005;115:146‐54). There is also an implausible sharp division between influenza and RSV around New Year's Eve. High risk of bias
Risk of bias
Bias	Authors' judgement	Support for judgement
PCS/RCS‐Selection Exposed cohort	Low risk	Selected group of users, secure records
PCS/RCS‐Selection Non Exposed cohort	Low risk	From the same community as the exposed cohort
PCS/RCS‐Comparability	Unclear risk	It is unclear whether the study took into account all possible confounders.
PCS/RCS‐Assessment of Oucome	Unclear risk	Physician's office attendance for: ILI or P&I as defined in ICD‐9. These were assessed only for first visits to the family physician.
Summary assessments	Unclear risk	No description of vaccine, content, or matching, no disaggregate numerators by event and arm, unstable denominators, low generalisability of results