A systematic review of the effectiveness of antimicrobial rinse-free hand sanitizers for prevention of illness-related absenteeism in elementary school children

A comprehensive search was conducted to identify all relevant studies regardless of publication status. Six electronic databases were searched for studies published in any language. The databases included: Biological Abstracts (1990-May 2003), CINAHL (1982–2003), the Cochrane Controlled Trials Register (1981–2003), EMBASE (1980-May 2003), HealthSTAR (1975-May 2003), and MEDLINE (1966-May 2003). A detailed search strategy was developed for use in MEDLINE and an iterative process was completed to refine the MEDLINE search for each database. Descriptions of the database search strategies are presented in Appendix 1 (see Additional File 1). OVID served as the primary search interface, and the SDI feature was used to monitor newly posted citations, the most recent date September 30, 2004. Due to the low occurrence of studies in this subject area, no filters were used to identify specific study types or reviews.

The reference lists of all relevant articles were reviewed for additional studies. A letter was sent to all corresponding authors of the articles identified by hand-search, excluding two newly eligible citations posted between May 2003 and September 2004 [15, 16], or by searching bibliographic databases. Additionally, contact experts and the industrial companies that manufactured the antimicrobial hand gels used in the included trials (GOJO Industries and Woodward Laboratories, Inc.) were contacted in attempts to identify other eligible trials. A detailed list of contacts is provided in Appendix 2 (see Additional File 2). Finally, conference proceedings for the American Journal of Infection Control (2000–2004) and recently published issues of the American Journal of Infection Control (February 2003 to August 2004) were searched by hand.

Studies were evaluated for inclusion on the basis of four criteria: target population, intervention, outcome, and study design. The target population of interest consisted of elementary school children between 4 and 12 years of age (including senior kindergarten and grades 1 through 8). The interventions of interest were those that administered antimicrobial rinse-free hand hygiene programs compared with no intervention or placebo treatment arm in a school setting. The outcome of interest was the comparison of the number of absences due to communicable illnesses in children who received the antimicrobial rinse-free hand hygiene intervention with the number of such absences in those who received a placebo or no intervention. We included cluster randomized controlled trials (RCTs) and cluster non-randomized controlled trials regardless of publication status.

All relevant citations, titles and abstracts, were imported into a reference database where duplicates were manually removed. Priority in downloading was given to MEDLINE. Reviews were excluded, but the bibliographies from such articles were examined for relevant studies. The screening was completed in an unblinded manner; there is inconclusive evidence that blinding introduced bias into the process [17]. One individual (EM) independently screened the titles and abstracts of each citation and identified all citations for full review. One reviewer was deemed appropriate as it was thought that the reviewer would error on the side of caution. Hard copies of all potentially relevant citations were retrieved, and two reviewers (EM, NLS) independently assessed each article using the aforementioned eligibility criteria, excluding the newly published citations in which eligibility was assessed by EM [15, 16]. Disagreements were discussed and a final decision was made by means of open consensus. A pilot test assessing the eligibility criteria on a sample of articles was not performed. Recent studies by both Juni et al. 2002 and Moher et al. 2003 indicate that the exclusion of trials in languages other than English (LOE) does not bias measures of effectiveness- however, both are cautionary, advocating language inclusive search strategies [18, 19]. Due to limited fiscal resources, an English language restriction was applied at this level, however the number of citations in LOE that met eligibility criteria will be noted.

Two reviewers (EM, NLS) independently abstracted data from all studies meeting the eligibility criteria, excluding one abstract where EM independently abstracted pertinent information [15], using pre-printed data collection forms presented in Appendix 3 (see Additional File 3). Information pertaining to the descriptive details of the study (e.g., year published, language of publication, publication status), design (e.g., randomized controlled trial), population (e.g., age, grade level), intervention (e.g., type of antimicrobial rinse-free hand sanitizer, inclusion of an educational component), and primary outcome(s) (e.g., absences due to illness) were collected. Adverse advents were not considered due to the relatively benign nature of the intervention. Reviewers resolved differences by means of open consensus. For the case of crossover study designs, data from the both arms of the study was abstracted. A pilot test assessing the data collection form on a sample of articles was not performed.

Two reviewers (EM, NLS) independently assessed the quality of each of the included studies, excluding the abstract by Thompson (2004) previously mentioned [15], using the previously validated 3-item Jadad scale, which assesses the quality of the report in terms of randomization generation, double blinding, and withdrawals and drop-outs by intervention group [20]. Studies were not given a quantitative score; rather this was used as a qualitative tool. (Due to the nature of the intervention, not all items apply). Additionally, if trials were randomized, allocation concealment was assessed and qualitatively evaluated as adequate, inadequate or unclear [21]. Disagreements were resolved through open consensus. A pilot study applying the quality assessment criteria on a subset of studies was not completed.

Data synthesis and analysis was performed in accordance with the Cochrane Reviewers' Handbook [22]. Firstly, data was qualitatively synthesized to examine the overall pattern of studies with respect to study design, population, intervention, and outcome characteristics. Sources of clinical and statistical heterogeneity were identified and results were examined. All data was abstracted as reported. The primary outcome, frequency of absences due to communicable illness was analyzed. Percent relative differences were presented along with 95 percent confidence intervals as the estimate of intervention effectiveness in the four studies which calculated rate and risk ratios as the measure of association [10, 27, 31, 41]. Percent relative differences and 95 percent confidence intervals were calculated enabling the results to be compared between studies, without altering the measure of association reported in the studies (rate and risk ratios). In the studies where data could not be abstracted, measures of association were reported [15, 16]. The validity of performing a quantitative synthesis was considered, however based on a qualitative inspection of heterogeneity and estimates of intervention effectiveness this was not deemed appropriate. Thus, sensitivity and subgroup analyses were not performed, and publication bias was not assessed quantitatively.

A flow diagram of the search results is illustrated in Figure 1. From the searches of the electronic databases, a total of 211 citations were identified, of which 70 were duplicates, resulting in the identification of 141 unique citations. In all, 18 potentially relevant trials were retrieved from the searches of the relevant databases. Hand-searching of the reference lists of relevant articles and conference proceedings resulted in a further 7 trials, which were also reviewed for consideration. Thus, 25 studies were determined potentially relevant [10, 23‐46]. Using both titles and abstracts, no trials meeting the inclusion criteria in LOE were found during the study selection phase.

After review of the full text of these studies, 21 were excluded for the following reasons: no outcomes of interest (n = 1), inappropriate population (n = 1), inappropriate interventions (n = 9), inappropriate study design (n = 3), irrelevant subject matter (n = 5), and review (n = 2). Thus, a total of 4 trials fulfilled the inclusion criteria [10, 27, 31, 41]. However, during the time between manuscript submission and revision, 2 additional citations were deemed to be eligible [15, 16], bringing the total to n = 6 eligible studies, one of which was a published manuscript [16], and the other a published abstract [15].

Of the 6 remaining studies, 2 were crossover studies [16, 27], 1 was a placebo-controlled cluster randomized controlled trial (RCT) [41], 2 were cluster non-randomized controlled trials (NRCT) [10, 31], and the published abstract was a cluster trial, however randomization was unclear [15]. McNemars's test was used to assess observer agreement, chi-square = 2.00; df = 1, p = 0.1573; and there was 92 percent agreement between the two reviewers with respect to study relevance for the initial four trials included [10, 27, 31, 41]. All of the relevant trials are described in Table 1, and an assessment of their quality of reporting is presented in Table 2 (the study by Thompson 2004 was excluded as only the abstract was available). Our overall agreement was 89 percent with respect to data collection for the five included studies outlined in Table 1. For quality abstraction, percent agreement was 80 percent. When examining items relating to blinding and assessment of withdrawals and dropouts, observer agreement was 100 and 80 percent, respectively. Kappa statistics were not calculated as sample size was insufficient.

All six trials were conducted in the United States, 4 with reported industrial sponsorship, and were published between 2000 to 2004. The trials varied in size (range = 138 to 6080 students; range = 1 to 18 schools), and geographic locations (Pennsylvania [10], California [27, 37], Ohio/Tennessee/Delaware/California [31], New England [16]). One of the studies assessed only private schools [28]; another study assessed both private and public schools [41], three assessed only public schools [10, 16, 31], while one's type of school assessed was not available [15]. Additionally, there was considerable variation in the type of school included both within and between trials: Christian private school, public elementary schools, same-sex schools and co-ed schools. The duration of the studies ranged from 5 weeks to 10 months, the longest trial being that of the RCT [41].

The trials also varied with respect to the intervention administered. White et al. (2001) and Dyer et al. (2000), provided each student with alcohol-free instant hand sanitizer [27, 41], whilst Hammond et al. (2000), Guinan et al. (2002), Morton et al. (2004), and Thompson et al. (2004) provided each class with alcohol-based instant hand-sanitizer dispensers [10, 15, 16, 31]. Education was concurrently provided for both the control and intervention arms in two studies [16, 27, 41], education on germs and hygiene was provided only to the intervention arm in one study [10], one study did not provide any education however study reinforcement was provided for teaching staff [31], and one study provided education to the intervention arm but as only the abstract was available for this study it was unclear if the control arm received any education [15].

The quality of reporting of the 5 trials that were examined in detail was low. Only one study was described as being randomized and double-blinded, however, it failed to describe a detailed and appropriate method of randomization and allocation concealment was unclear [41]. Four of the five studies, as previously mentioned, discussed withdrawals and dropouts however the description was quite basic and detailed flow-diagrams outlining the passage of participants through the trial were not supplied [16, 27, 31, 41]. White et al. (2001) reported a significant number of dropouts (857 of 1626 students did not complete the study) however no explanations were offered [41]. Four studies received industrial sponsorship either by GOJO industries or Woodward Laboratories. In addition, two studies received financial support from another external source [16, 41]. Other characteristics of poor quality reporting included: no sample size calculation defined for all five studies, and the statistical methods were vague. No studies took into consideration clustering when analyzing their results. Our overall agreement for all items of quality was greater than 80 percent; again observer agreement statistics were not calculated as the sample size was insufficient.

All six studies varied in their definition of communicable illness-related absenteeism, refer to Table 1. Out of the five studies with published manuscripts, four of the studies reported the estimate of intervention effectiveness in terms of risk/rate ratios, subsequently calculating percent relative effect, and one reported and odds ratio for a pair-matched study; results are reported in Table 3, Table 4, and Table 5. The percent relative effect measures the decreased rate of the occurrence of absenteeism when the rate ratio is the measure of association or the decreased risk of absenteeism when the relative risk is the measure of association. Tests of significance were completed in all five of the studies using chi-squared tests or t tests however no confidence intervals were calculated. Two of the studies used rate ratios as the measure of association [10, 31], two studies used relative risks as the measure of association [27, 41], and the other used the odds ratio [16]. Table 3, 4 and 5 distinguish between the measures of association used. All studies found a statistically significant effect of the antimicrobial rinse-free hand gel interventions. In the study by Hammond et al. (2000), the experimental group had a 20% (95% CI = 19–21%) reduction in absences due to communicable illness, the experimental group in the trial completed by Guinan et al. (2002) had a 49% (95% CI = 42–56%) reduction, White et al. (2001) demonstrated a 33% (95% CI = 17–45%) reduction in the experimental group, Dyer et al. (2000) had a 34 % (95% CI = 10–50%) reduction in absences due to communicable illness in the experimental group in the first phase and a 56 % (95% CI = 31–72%) in the second phase, and Morton et al. (2004) reported a significant odds ratio for McNemar's test (chi-square = 7.787; p = 0.0053).