Carman 2000.

Methods	Purpose: to assess the effects of staff vaccination against influenza on resident mortality in long‐term care hospitals Design: cluster‐randomised study (C‐RCT) conducted in Scotland during the 1996 to 1997 influenza season. The study identified 10 long‐term care geriatric hospitals in West and Central Scotland with a policy of vaccinating all patients against influenza if they had no contraindications and then only on the request of the patients or their relatives. Pairs of hospitals in each of these clusters were matched on patient enrolment and then in a Latin square design were randomised by a table of random numbers for the HCWs to be offered influenza vaccination or not Anonymous questionnaires were sent to ward nurses on 31 March 1997 to ask if they had received influenza vaccination and these data were used to estimate vaccine acceptance for all HCWs in hospitals where influenza vaccine had not been offered to HCWs. In each hospital a random sample chosen by computer of 50% patients was selected for virological monitoring Data from the Scottish Centre for Infection and Epidemiological Health and from GPs were used to define the start of the influenza season. Combined nasal and throat swabs were taken from patients every 2 weeks from 14 December 1996 to 14 February 1997. Opportunistic samples were also taken from patients whom the ward nurses thought had influenza. Samples were taken within 12 hours of death of any patient who died. Samples were analysed by RT‐PCR analysis Results were summarised for the 2 groups of LTCIs. Hospitals were not well‐matched for patient vaccination rates and Barthel scores (Wikipedia 2009) and post‐hoc statistical adjustments could not be made because of missing data. The outcome was the empirical logic of mortality for each cluster (= natural logarithm of the odds on death) Statistics: the power calculation was based on the previous study by Potter 1997 and the authors computed that with 1600 patients in 20 hospitals they would have ≥ 80% power to detect a decrease in mortality from 15% to 10% with alpha = 0.05 (2‐tailed), allowing for the clustered design. The power calculation for virological sampling showed that 500 patients would be required to give 80% power at 5% significance (2‐tailed) to detect a decrease in influenza infection from 25% to 15% Mortality rates were compared in the 2 groups with the Mann‐Whitney test. "Incomplete data for patient‐level covariates meant that a full multilevel approach to the analysis was not possible without making strong, implausible and untestable assumptions about the mechanisms that led to the incomplete data. Instead, we calculated summary statistics to describe the mix of patients in each hospital and these values were included in a multiple linear‐regression analysis. The response variable in these analyses was the empirical logit of each hospital's mortality rate that is, the natural logarithm of the odds on death."
Participants	Country: Scotland Setting: 20 long‐term care hospitals in Glasgow Eligible participants: 749 participants were residents of facilities in the arm in which 1217 HCWs were offered vaccination (620 accepted) and 688 in the arm in which HCWs were not offered vaccination. Day and night nurses, doctors, therapists, porters and ancillary staff (including domestic staff and ward cleaners) were offered influenza vaccination Age: 82 Gender: 70% F
Interventions	Intervention: influenza vaccination. The type, dosage and route are not described. A good match in the study year between the prevailing strain and the vaccine strains was reported Control: no influenza vaccination
Outcomes	RT‐PCR and tissue culture for influenza A or B. A random sample of 50% of patients in each hospital was selected for virological monitoring of influenza infections by nose and throat swabs every 2 weeks, which were sent for RT‐PCR analysis and tissue culture. "At the times when study nurses took routine samples, they took additional opportunistic nose and throat swabs from non‐randomised patients who the ward nurses thought had an influenza‐like illness. The ward staff were asked to take routine nasal swabs within 12 hours of death for any patient who died." Mortality (all causes) (N.B. clinical outcomes were not reported but were used to investigate the viral circulation in the facility)
Notes	The situation that 10 long‐term care hospitals had a policy of routinely vaccinating residents for influenza vaccination and 10 did not, permitted a Latin square design RCT of offering influenza vaccination or not to HCWs within each of these clusters Analysis was not according to intention‐to‐treat Mean cluster size: 72 (based on mortality data) Intra‐cluster correlation coefficient: 0.023 as reported in Hayward 2006 Despite no difference in isolation of influenza viruses between clusters, the authors conclude that vaccines are protective. In addition, they fail to comment on the implausibility of the vaccines' effect on aspecific outcomes (ILI) and lack of effect on influenza
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Hospitals were randomly allocated ... by random‐numbers table."
Allocation concealment (selection bias)	Unclear risk	Not stated
Blinding (performance bias and detection bias) All outcomes	Unclear risk	Not stated
Incomplete outcome data (attrition bias) All outcomes	High risk	In the 10 hospitals where HCWs were offered vaccination 749 patients were included and "a random sample of 375 patients was offered virological screening by nose/throat swab"; 258 accepted. In the 10 hospitals where HCWs were not offered vaccination 688 patients were included and a random sample of 344 were offered virological screening by nose/throat swab; 269 accepted. Note comments by authors in the Description section above on incomplete data. Polymerase chain reaction (PCR) samples were obtained from only 17% of deaths. 4 samples from each patient surveyed were planned from protocol: 1798 samples were obtained from 719 patients (2.5 samples/patient)
Selective reporting (reporting bias)	Low risk	No selective reporting
Other bias	High risk	Selection bias: the total number of long‐term care hospitals in West and Central Scotland is not stated. In the long‐term care hospitals in which HCWs were offered vaccination, residents had higher Barthel scores Performance bias: only 51% of HCWs in the arm received vaccine in the long‐term care hospitals where vaccine was offered and 4.8% where it was not; 48% of patients received vaccine in the arm where HCWs were offered vaccination and 33% in the arm where HCWs were not Statistical bias: the analysis was not corrected for clustering, unlike the Potter 1997 pilot; in the long‐term care hospitals where HCWs were offered vaccination, the patients had significantly higher Barthel scores and were more likely to receive influenza vaccine (no significance level stated) and due to missing data these differences could not be adjusted for other than by estimation. Statistical power may also have been a problem as the detection rate of 6.7% was lower than the estimated rate of 25% used in the power calculation