Vaccination of Health Care Workers to Protect Patients at Increased Risk for Acute Respiratory Disease

Evidence is limited but sufficient to sustain current vaccination recommendations.

Health care workers (HCWs) may transmit respiratory infection to patients. We assessed evidence for the effectiveness of vaccinating HCWs to provide indirect protection for patients at risk for severe or complicated disease after acute respiratory infection. We searched electronic health care databases and sources of gray literature by using a predefi ned strategy. Risk for bias was assessed by using validated tools, and results were synthesized by using a narrative approach. Seventeen of the 12,352 identifi ed citations met the full inclusion criteria, and 3 additional articles were identifi ed from reference or citation tracking. All considered infl uenza vaccination of HCWs, and most were conducted in long-term residential care settings. Consistency in the direction of effect was observed across several different outcome measures, suggesting a likely protective effect for patients in residential care settings. However, evidence was insuffi cient for us to confi dently extrapolate this to other at-risk patient groups.
R espiratory disease is a leading cause of deaths worldwide, and infl uenza and pneumococcal infections are major contributors. Certain groups, such as persons >65 years of age or with chronic underlying health problems (1) are particularly vulnerable to severe respiratory disease and have poorer outcomes after infection than does the general population. These persons are likely to be frequent users of health care facilities, and outbreaks have been described in a range of high-risk environments, including acute care (2,3), pulmonary (4), and infectious diseases wards (5); organ transplant departments (6); children's wards (7,8); neonatal intensive care units (9); and nursing homes (10,11). Severe respiratory infections often occur despite high vaccine coverage rates among patients, suggesting that seroconversion is suboptimal (10). Although the origin of infection often is diffi cult to establish, evidence from some outbreaks (5,7,(10)(11)(12)(13)(14) suggests that transmission from HCWs to patients is likely.
It is estimated from previous infl uenza seasons that ≈20% of HCWs have evidence of infection (15), although not necessarily acquired in the workplace. Young healthy adults often have asymptomatic infection, and ≈28%-59% might experience subclinical infection (15). Many persons with mild or subclinical illness continue to work while infectious, and even when illness is recognized, virus might be shed before symptom onset. In a randomized controlled trial among health care professionals, Wilde et al. demonstrated that infl uenza vaccine was 88% effi cacious for reducing serologically confi rmed infl uenza A infection and 89% effi cacious for reducing serologically confi rmed infl uenza B infection (16). Therefore, vaccination of HCWs has been widely recommended to provide direct protection for themselves and indirect protection for their patients (1,17).
Despite efforts to encourage infl uenza vaccination of HCWs, coverage has been historically poor. Recently, ethical arguments for mandatory infl uenza vaccination have been raised that focus not only on the direct and indirect benefi ts to staff and patient health but also on the economic consequences. Burls et al. (18) suggested that at a cost of £51-£405 (US$85-$675) per life-year saved, mandatory vaccination is likely to be cost-effective. However, evidence for the effectiveness of vaccinating HCWs for protecting vulnerable patients is limited.
Two recent systematic reviews considered the evidence for indirect protection of vulnerable patient groups after staff infl uenza vaccination (18,19). They suggest that vaccination of HCWs might be effective for reducing death and infl uenza-like illness (ILI) among elderly residents, but we are unaware of comparable data related to other at-risk groups. We aimed to identify and assess further evidence for the effect of vaccinating HCWs on patient groups most vulnerable to severe or complicated respiratory illness.

Methods
The full study protocol is registered with the UK National Institute for Health Research International Prospective Register of Systematic Reviews (www.crd.york. ac.uk/PROSPERO [registration no. CRD420111092]). We searched several electronic health care databases, sources of evidence-based reviews, guidelines, and gray literature in accordance with the specifi cations of each database ( Figure). In addition, we contacted domain experts and vaccine manufacturers to identify unpublished data and undertook citation and reference tracking for all included papers. Thesaurus-indexed and free text terms were defi ned for the population, intervention, and outcome parameters; peer reviewed; and adapted as necessary for each search engine.
Eligibility criteria were defi ned a priori as follows: • Types of study: any experiment, observational study, or systematic review reporting on the effectiveness of vaccination (including infl uenza or pneumococcal vaccines) of HCWs for protecting patients at higher risk for severe or complicated respiratory infection.
• Types of participants: persons at higher risk for severe or complicated illness as a result of acute respiratory infection (as defi ned in World Health Organization [1] and Advisory Committee on Immunization Practices guidance [17]), who have received or are receiving care from an HCW.
• Types of intervention: infl uenza or pneumococcal vaccination of any worker providing medical, nursing, social, or personal health care (because no uniformly accepted defi nition of an HCW exists, it was defi ned by the peer-reviewed terms specifi ed in the search strategy).
• Types of outcome measure: cases or consultations, death or hospitalization for acute respiratory disease, infl uenza, ILI, or pneumococcal disease.
Published and unpublished reports from any year that were written in Chinese, English, French, Japanese, Portuguese, Russian, or Spanish were considered. A 3-stage process was used to assess eligibility for inclusion screening fi rst by title, then abstract, and then full text. Two reviewers undertook this in parallel for stages 1 and 2 and independently for stage 3. Consensus was reached by discussion; when reviewers disagreed, a third reviewer was consulted for a fi nal decision. Where multiple reports were identifi ed for the same piece of original research, the most recent peer-reviewed source was selected.
Two reviewers independently extracted data from each included, by using a predefi ned, piloted template. The risk for bias was assessed by using the Cochrane Collaboration tool (20) for experimental and prospective cohort studies, the Downs and Black tool (21) for other observational studies, and the US Agency for Healthcare Research and Quality (22) domain and element-based evaluation instrument for systematic reviews. Again, consensus was reached by discussion, with engagement of a third reviewer as necessary. No additional information was sought from corresponding authors. Data were synthesized qualitatively by using a narrative approach in accordance with the framework described by the Economic and Social Research Council and recommended by the University of York Centre for Reviews and Dissemination (23).

Study Selection
We identifi ed 12,352 citations ( Figure): 10,713 from health care databases and the remainder from additional sources. Seventeen studies met the inclusion criteria at the full text stage; 3 others were identifi ed from citation or reference tracking. Of these, 14 were primary research articles; 4 were cluster randomized controlled trials (RCTs), and 10 were observational studies. Four of the remaining 6 articles were different versions of a report relating to 1 systematic review, and the other 2 were different versions of a report relating to a second systematic review. One of these systematic reviews (18) provided a qualitative analysis of 2 of the earliest cluster RCTs (24,25), and the other (19) provided a quantitative meta-analysis of all 4 cluster RCTs (24-27) and 1 additional observational study (28). We used the most recent and detailed version of each review published in a peer-reviewed source in this study.
All of the primary studies considered infl uenza vaccination of HCWs (online Appendix Table 1, wwwnc. cdc.gov/EID/article/18/8/11-1355-TA1.htm); therefore, we discarded our planned subanalysis relating to pneumococcal vaccination. Only 4 studies (24-26,29) defi ned HCW, even though this defi nition is likely to affect the probability of transmission and therefore the magnitude of observed effects. Where reported, vaccination among staff ranged from ≈35% to 70% in the intervention arm and from none to 32% in the control arm of experimental studies and from 12% to 90% in observational studies. Eleven of the primary research studies were conducted in long-term care facilities; the remainder were conducted in renal dialysis facilities (30), a pediatric hospital (31), and an adult oncology hospital (32) (1 study each). Where reported, vaccination coverage among patient populations ranged from 0% to ≈90%, and few studies considered additional infection control practices, such as hand washing, duration of contact, or use of face masks, which vary and again infl uence the propensity for transmission.

Cochrane Collaboration Tool
Concerns arose largely from the lack of blinding of participants or study personnel (Table 1). Although the effect was likely to be minimal with regard to the primary outcome for all 4 RCTs (all-cause mortality), it might have resulted in underestimation or overestimation of additional, more subjective, outcome measures, such as incidence of ILI.

Downs and Black Tool
The Downs and Black tool ( Table 2) considers 5 assessment domains, but because most observational studies identifi ed were primarily descriptive, we excluded the power domain in this review. Scores ranged from 3/27 (34) to 10/27 (29,30,35), with higher scores representing lower risk for bias. None of the studies provided suffi cient detail about the patient population, and only 1 (29) described principal confounders. Other concerns about reporting related to lack of detail of study objectives (29,32,34), a priori defi nition of outcome measures (32,(34)(35)(36)(37) or those lost to follow up (35), failure to provide suffi cient detail of statistical analysis (29,30,(34)(35)(36)(37), lack of randomization or blinding, and failure to adjust outcome measures.

Agency for Healthcare Research and Quality Tool
We assessed the 2 identifi ed systematic reviews (18,19) by using the Agency for Healthcare Research and Quality tool (22). Both appeared to be at a comparatively low risk for bias, providing a clearly defi ned research question, search strategy, inclusion and exclusion criteria, and description of outcomes. However, details were lacking about blinding of reviewers to authorship and measurement of agreement in extracting data, which might have resulted in measurement bias.

Cases or Consultations for Acute Respiratory Disease
One RCT reported data (25) for 2 measures of consultation for respiratory disease; episodes of lower respiratory tract infection and suspected viral illness (Table 3). In addition, the estimate for lower respiratory tract infection was adjusted for clustering by Thomas et al. (19). Both measures demonstrated reduced odds, and results were signifi cant for suspected viral illness when vaccinated and nonvaccinated patients were considered together.
The study by Potter et al. (25) was considered to be at a higher risk for bias than the other RCTs identifi ed; thus, the strength of evidence for these outcomes is questionable. In addition, the measures considered are nonspecifi c, and the observed effects cannot necessarily be attributed to reduced infl uenza infection. Nasopharyngeal samples were taken from a subset of patients within 48 hours after symptoms developed; no samples were positive for infl uenza on immunofl uorescence assay.
Three RCTs (25-27) measured cases of ILI, and these data were pooled by Thomas et al. (19) to demonstrate a statistically signifi cant reduction in odds. Two observational studies (28,33) also measured cases of clinically defi ned ILI, demonstrating statistically signifi cant reductions in risk, although the threshold of staff vaccination coverage used to categorize facilities in these studies varied (Oshitani [28] considering facilities where more or fewer than 10 staff were vaccinated, and Saito [33] comparing facilities with <40%, 40%-59%, and >60% coverage among staff). A third observational study (29) reported no correlation between  staff vaccination coverage and cases of infl uenza in patients, although the relative change in vaccination coverage (79%-91%) was small and thus any difference in the number of cases was probably diffi cult to detect. The magnitude of reported effects varied, most notably by infl uenza season in the study of Hayward et al. (27), and with patient vaccination status in the study of Potter et al. (25). One study measured general practitioners consultations for ILI (27). An inconsistent effect was demonstrated across different periods of infl uenza activity, but pooled data suggested an overall statistically signifi cant reduction in the odds of consultation after vaccination of HCWs.
Three observational studies (28,35,37) demonstrated a statistically signifi cant protective effect of staff vaccination against clinically defi ned outbreaks of ILI in patients ( Table  4). The thresholds used to categorize facilities on the basis of staff vaccination coverage again varied among studies, and these data were considered to be at relatively high risk for bias.
Measures of laboratory-confi rmed infection (online Appendix Table 3, wwwnc.cdc.gov/EID/article/18/8/11-1355-TA3.htm) were less frequently reported and generally based on small samples of data at high risk for bias. Five studies measured laboratory-diagnosed infl uenza (24,25,31,32,36), although 1 reported no statistical analysis (25). Different methods of defi ning laboratory confi rmation were used (online Appendix Table 3). Thomas et al. (19) pooled data from the 2 RCTs (24,25) to demonstrate a small nonsignifi cant protective effect. This result is supported by evidence from 2 additional observational studies (31,32), which indicated a statistically signifi cant reduction in the proportion of laboratory-confi rmed cases of nosocomial infl uenza among inpatient pediatric and oncology patients after implementation of vaccination campaigns. In addition, Monto et al. (36) measured outbreaks of laboratory-diagnosed infl uenza, and this was the only study not to demonstrate a protective effect of vaccinating HCWs. The authors reported a higher, but nonsignifi cant, median vaccination coverage among staff in homes experiencing outbreaks.

Deaths from Respiratory Infection, ILI, or Acute or Respiratory Disease or Its Complications
Evidence for 5 measures of death was identifi ed ( Table 5). All 4 RCTs (24-27) considered all-cause death as their primary objective, providing the strongest evidence on the basis of study design. Although not defi ned a priori as an outcome of interest for this review, data were therefore extracted. These were pooled by Thomas et al. (19) to demonstrate a statistically signifi cant protective effect.
Although at higher risk for bias, supporting data were provided for 4 more-specifi c measures. Thomas et al. (19) pooled data from 2 RCTs, 1 measuring deaths after pneumonia (25), the other measuring respiratory deaths (26), and demonstrated a small nonsignifi cant protective effect. However, the validity of this pooled analysis was questionable because how these outcomes were defi ned was not clear. Nonsignifi cant reductions in risk also were observed for laboratory-diagnosed infl uenza at death (24) and death after ILI (27). Again, the direction of the observed effects was largely consistent with other measures, providing further support for a hypothesis of indirect protection.

Admission to a Health Care Facility or Any Other Suggestion of Impact
Hospitalization was measured in 2 RCTs (26,27), pooled data suggesting a small, nonsignifi cant effect (Table 6). One RCT also measured hospitalization for respiratory causes (26) and 1 admission to hospital with ILI (27), although neither demonstrated any apparent effect. This result is particularly noteworthy given the observed decrease in deaths and might refl ect health-seeking behaviors.

Discussion
Evidence is limited for the effectiveness of vaccination of HCWs for protecting patients at higher risk for severe or complicated respiratory illness. Despite the broad question posed, extensive searching, and large number of resultant hits, our search resulted in a low yield of studies, all of which focused on infl uenza with no consideration for pneumococcal infection. This fi nding is perhaps not surprising because pneumococcal vaccination is not routinely recommended for HCWs and little, if any, evidence exists of nosocomial spread. A consistent direction of effect was observed across multiple outcome measures, with virtually all studies noting a trend toward a protective effect of vaccinating HCWs. This consistency adds to the degree of confi dence in interpreting our overall fi ndings. Given that most studies were carried out in long-term care facilities, we conclude that vaccination of HCWs against infl uenza is likely to offer protection for this patient group. However, future reviews that specifi cally examine the effect of vaccinating other outpatient providers, such as home HCWs and hospital staff in acute care, short-stay settings, would clearly be of value. These fi ndings are more diffi cult to extrapolate to other at-risk groups, although some, albeit limited, evidence was identifi ed from other settings to suggest a similar effect.
The results of all 4 RCTs (24-27) and 1 of the observational studies identifi ed (28) previously had been pooled in a quantitative meta-analysis (19). The authors of this analysis concluded that evidence is lacking that vaccinating HCWs prevents infl uenza infection in elderly patients because the apparent benefi ts were confi ned to nonspecifi c outcome measures. We considered additional observational data that demonstrate consistency in the direction of the observed effects across specifi c and nonspecifi c outcome measures. Although the strength of evidence for more-specifi c measures is generally much weaker, these fi ndings add greater weight to the hypothesis of a potential protective effect.
The recent position statement by the Society for Healthcare Epidemiology of America (38) suggests that further studies are not needed because the biological rationale for vaccination does not vary by practice setting. However, effect size might vary considerably because of patient characteristics and care patterns (staff deployment and duration of inpatient stay), and further evidence is needed among the most at-risk groups where benefi ts are probably greatest, to enable prioritization of resources, particularly where vaccine shortages or resource limitations might exist.
Previous authors have suggested that vaccination of HCWs might enable development of herd immunity. Realistically, herd immunity is diffi cult to achieve in health care settings, especially acute care short-stay settings, because of patient admissions and discharges, visitors, and staff turnover. That said, herd immunity might not be necessary to benefi t patients; modeling studies (39) suggest a direct association between coverage and attack rates. Such studies (39) also suggest variation in the potential for transmission of infection by different staff groups, which should be explored in further detail. This fi eld of research has some inherent problems. These diffi culties result in part from the diffi culty of isolating the effect of HCW vaccination, disentangling it from other factors that might infl uence patient outcomes, such as patient vaccination (as demonstrated by Potter et al. [25]) and background infl uenza activity (as demonstrated by Hayward et al. [27]). Staff vaccination itself might be linked to additional confounding variables, such as organizational culture and professional beliefs. In fact, such confounding might explain the difference in fi ndings between the work of Monto (36) and the other authors. Prospective collection of information relating to relevant transmission factors and infection control measures that were largely overlooked by the studies in this review should be used to enable appropriate adjustment in future studies. Furthermore, the most appropriate outcome measures are diffi cult to defi ne because not all persons with laboratoryconfi rmed infection have symptoms of illness and vice versa. Future studies thus need to demonstrate consistent effects for a range of clearly defi ned outcomes by using valid measures across several different infl uenza seasons, with suffi cient power to detect true underlying effects.
The fi ndings of our review are subject to several limitations. Because 11 of the 14 primary research articles considered outcomes in long-term care facilities, generalizability to other at-risk groups is limited. In addition, we did not attempt to contact authors of original studies, and the conclusions drawn are limited by the reported detail. Although the number of reviewers was limited as far as possible, some inconsistency might have occurred in the selection, extraction, and assessment of data introducing potential bias, particularly where the opportunity for subjective judgment existed. We attempted to minimize inconsistency by using several standard assessment tools, but their use was limited by lack of information where components were not conducted because of the nature of the study design. Meta-analysis of the 4 RCTs identifi ed had already been conducted, and although we identifi ed additional observational data, the observed heterogeneity limited any further quantitative analysis.
Some wider possible effects of HCW vaccination, such as reduction in absenteeism because of illness, are beyond the scope of this review. Ethically, autonomy needs to be balanced with nonmalefi cence, and this need must be addressed when policy decisions about vaccination are considered. Anikeeva et al. (40) reported that in a review of 15 studies focusing on the reasons staff accept infl uenza vaccine, self-protection was the most important. However, patient protection also was perceived as important, particularly among HCWs in settings with higher risk patients (40). Nevertheless, HCWs would be justifi ed in claiming that the current evidence base is not especially strong and heavily weighted toward the benefi ts to patients receiving care in long-term care facilities, although limited evidence would not necessarily legitimize nonacceptance.
The existing evidence base is suffi cient to sustain current recommendations for vaccinating HCWs on the grounds that some protection of high-risk patients against infl uenza seems likely. However, vaccination should be considered 1 element of a broad package of infection prevention and control measures, such as good hand and respiratory hygiene, environmental cleaning, protection against respiratory droplets, and cohorted care during outbreaks. Well-designed studies that strengthen the evidence base might increase compliance with guidelines, resulting in improved coverage.