Impact of a multicomponent hand hygiene–related intervention on the infectious risk

in nursing homes: A cluster randomized trial

Background: The aim of this study was to assess the impact of a multifaceted hand
hygiene (HH) program on the infectious risk in nursing homes (NHs).
Methods: This was a 2-arm cluster randomized trial; French NHs were allocated
randomly to the intervention (13 NHs) or control (13 NHs) groups. The intervention
consisted of implementing a bundle of HH-related measures over 1 year, including
increased availability of alcohol-based handrub, HH promotion, staff education, and
localwork groups. The primary end pointwas the incidence rate of acute respiratory
infections and gastroenteritis reported in the context of clustered cases episodes.
Secondary end points were mortality, hospitalization, and antibiotic prescription
rates.
Results: Baseline characteristics did not differ between groups. The overall handrub
consumption was higher in the intervention group over the 1-year intervention period.
Because of underreporting, data on the primary end points were of insufficient quality
for analysis. Hospitalizations did not differ between the 2 groups. However, the
intervention group showed significantly lower mortality (2.10 vs 2.65 per 100
residents per month, respectively; P = .003) and antibiotic prescriptions (5.0 vs 5.8
defined daily doses per 100 resident days, respectively; P <.001). These results
were confirmed by the longitudinal multivariate analysis adjusted for NH and resident
characteristics and for seasonality (mortality rate ratio, 0.76).
Conclusions: A multifaceted HH intervention may have a short-term impact on
mortality in NHs. Nevertheless, other strategies may remain necessary to reduce
morbidity.

BACKGROUND
Nursing homes (NHs) are both care facilities and home to many older adults
in developed countries worldwide. This combination leads to a unique situation in
which the infection risk is difficult to control.1 Indeed, in addition to the effect of
aging, NH residents often have multiple comorbidities, physical disabilities, and
cognitive impairment, which all contribute to make them highly vulnerable to
infection.2 Additionally, because NHs are living places, they offer a wide range of
opportunities for socialization, includingshared meals, recreational activities, and
rehabilitation exercises, therefore increasing the probability of person-to-person
transmission of pathogens.1
This translates to a high prevalence of infections in NHs, estimated in France
at 11.2% (95% CI, 4.4%-16.2%).3 In the United States,the total number of infections
occurring in NHs was estimated at 1.6-3 million per year, figures similar to those
observed for nosocomial infections in acute care settings.1 Respiratory infections
and gastroenteritis are the most frequent infections leading to outbreaks in NH
settings, with respiratory viruses and norovirus as the most common etiologic
agents.2,4
These infections account for a large share of NH mortality, with the rate of
deathassociated with infections in NHs estimated at 0.6 per 1,000 resident days
(range, 0.04-0.71).5 They are also associated with frequent hospital transfers and
increased antimicrobial use and with high costs, that may exceed $1 billion per year
in the United States.6
 For all these reasons, infection prevention and control is a majorfocus of NH
attention. In particular, hand hygiene (HH) is recognizedas one of the simplest and
most cost-effective tools for infectioncontrol in health care settings.7 However, there
are several barriersto the implementation of efficient HH policies in NH settings.First,
these policies may seem too costly for NHs, for instance becauseof increased
handrub solution expenditures.8 Second, staff educationis complicated by a high
turnover, making frequent trainingsessions mandatory.9 Third, most available HH
recommendationswere originally designed for acute care settings. Hence, a
reluctanceto apply these recommendations in NHs is often reportedbecause staff
members feel that they are not adapted to long-termcare and tend to overmedicalize
resident-staff interactions.10,11
Consequently, compliance with international HH recommendationsis low in
NH staff: it was reported as 14.7% in Canada,12 17%in Italy,13 and 11.3% in
Taiwan.14 A recent review of the evidence onHH in NHs underlined the importance
of adapting HH recommendationsand programs to NH settings to improve this
compliancerate.15 This review also showed that the current evidence on HHimpact
in long-term care settings is lacking and mostly based onsmall-scale observational
studies.
In this context, the objective of this study was to assess, througha randomized
controlled trial, the effectiveness of a participativemulticomponent HH intervention
specifically designed for NHs. Thisintervention aimed at overcoming most of the
aforementioned barriersto HH in NHs, notably by associating the staff to the design
oflocal guidelines, and by increasing their awareness of the infectiousrisk.

METHODS

Study design
This study was a 2-arm cluster randomized controlled trial, complyingwith the
Consolidated Standards of Reporting Trials guidelines.16 Each participating NH
constituted 1 cluster. The intervention was implemented in intervention NHs for a
period of 12 months, from April 1, 2014-April 1, 2015, whereas usual practice
continued in control NHs. Directors of the included NHs agreed to participate in the
study and be randomly allocated to either the intervention group or control group
before the beginning of the data collection process. Individual written consent was
not required from the residents or their relatives. However, in all intervention NHs,
information letters about the study were posted near the entry and sent to doctors,
and meetings were organized to inform residents and families. The trial was
registered in the ISRCTN Registry (study no. ISRCTN16474757).No ethics approval
was required for this study according to French legislation because no individual-
level data were collected and the intervention did not involve any additional therapy
or procedure for residents.

Participants
The study was performed in French NHs belonging to the Koriangroup, a
privately held chain of NHs. After a call for participation, a total of 27 NHs distributed
over France volunteered to take part in the study. We selected a NH in Paris for the
pilot study, and included the remaining 26 NHs, assigning 13 NHs to the intervention
group using simple randomization, whereas the other 13 NHs were assigned to the
control group.
Usual and control practices
Usual HH practices continued in NHs allocated to the controlgroup. This
included general training on infection control offered to volunteer staff, access to
handrub solution dispensers nearresident rooms, and HH promotion on the annual
World Hand Hygiene Day (May 5).

Intervention
The intervention was based on a bundle of HH-related measuresaimed at NH
staff, residents, visitors, and outside care providers. These measures included
facilitated access to handrub solution using pocket-sized containers and new
dispensers, a campaign to promote HH with posters and event organization, the
formation of local work groups in each NH to work on HH guidelines, and staff
education using e-learning on infection control and HH training performed by the
same nurse for all NHs. Online quizzes were proposed at the end of the training
program, and staff who did not achieve a high enough score were invited to go
through the e-learning again at a later date. A detailed description of the intervention
is available from the authors on request.
All aspects of the intervention started being implemented simultaneously in
April 2014 in all intervention NHs. Based on success rates for online quizzes and
personal communication with the nurse in charge of the training, 3-6 monthswere
necessary for the program to take.

Study end point measures

The primary end point was the incidence rate of acute respiratoryinfections
(ARI) and acute gastroenteritis (AGE) reported in the context of episodes of
clustered cases, defined as at least 5 cases within 4 days among NH residents or
staff. ARIs were defined as the combination of at least 1 respiratory symptom with 1
symptom of systemic infection. AGE was defined as the sudden onset of diarrhea or
vomiting in the absence of a noninfectious etiology.
Secondary end pointswere the mortality rate, hospitalization rate, and
antibiotic prescription rate (measured in defined daily doses [DDDs] per 100 resident
days17).

Data collection
Data were collected monthly at the NH level in all participatingNHs over the 3
months preceding the intervention (baseline period), during the 1-year intervention,
and over the 3 months after the end of the intervention (follow-up period). Data
collectionwas centralized by theNHgroup headquarters in Paris, and made use of
existing computerized databases. No individual data were collected.
The collected data on NH characteristics and organization included location,
number of beds (capacity and monthly occupancy rate), and staffing levels (monthly
planned worked hours for each staff category and absenteeism data). Resident
characteristics included age distribution, proportion of men, pathology, and
dependency rates. The resident-level dependency rate was measured using the
French GIR (iso-resource group) group, based on a scale corresponding to the
length of assistance time required to help the individual accomplish daily living
activities, with a score ranging from 1 for highly dependent individuals to 6 for
independent individuals.18,19 The mean weighted GIR (GMP) was collected at the
NH level, as was the mean weighted pathology score (PMP) of residents. GMP and
PMP are commonly used in French long-term care institutions.
 In NHs from the intervention group, the average success rate for online
quizzes on HH performed at the end of the education intervention was also collected.
ARI and AGE data came from a national surveillance system that was
implemented in France in 2010 as part of a national plan toreduce the infection risk
in long-term health care settings.20 Thesurveillance is based on voluntary and
standardized notifications to health authorities of any AGE or ARI clustered case
episodes. The sensitivity of this surveillance system was assessed at 81% for both
AGE and ARI during a pilot study in 2011-2012.21
Data on resident mortality, hospitalization, and antibiotic prescriptions were
collected monthly at the NH level using the national computerized database of the
Korian NH group.
Finally, the mean amount of handrub solution used per resident day was
assessed in each NH over the entire study period as a proxy for HH frequency. This
amount was estimated based on a follow-up of the quantity of handrub solution
bought by the NH, which is routinely monitored in all Korian NHs.

Statistical analyses
Sample size computation was based on recently published estimates of the
incidence rate of lower respiratory infections in NHs before and after a HH-based
intervention,22 from 0.97 to 0.53 per 1,000 resident days (a 45% reduction). Using
these data, and assuming an intracluster correlation coefficient of 0.04 (leading to an
inflation factor of 4.16 to adjust for the clustered design23), the calculation indicated
that 12 NHs had to be included in each arm to achieve a 90% power at a 2-tailed
significance level of 0.05. We overrecruited by 1 NH in each arm.
Outcomes were compared between the intervention and control groups using
χ2 tests or Fisher exact tests for categorical variables and t tests for continuous
variables. Generalized estimating equations (GEEs) were fitted to analyze monthly
longitudinal data for mortality, hospitalizations, and antibiotic prescriptions, while
taking into account the cluster structure and adjusting for resident and NH
characteristics, and for seasonality. R version 3.2.2 (R foundation for statistical
computing, Vienna, Austria) was used for all analyses.

RESULTS

Baseline characteristics of participating NHs

The geographic distribution of participating NHs among administrative regions
was not different between the intervention and control groups (Fisher exact test, P =
.27).
Table 1 provides the characteristics of NHs from the control and intervention
groups observed at baseline, during the January-March 2014 period (before
beginning the intervention in April 2014).
There was no significant difference at baseline in size, bed occupancy, double
room frequency, staff-to-resident ratio, or staff absenteeism of NHs and in mean age,
sex ratio, dependency level (GMP) or pathology level (PMP) of residents. Influenza
vaccination rates of residents and staff for the 2014-2015 season did not differ either.

HH practices
Based on the 2 quarters before implementation of the intervention,the
baseline quantity of consumed handrub solution in participating NHs was 4.5 mL per
resident per day and did not differ between the control and intervention groups. Over
the 1-year intervention period, the mean quantity of consumed handrub solution was
significantly higher in the intervention group (7.9 mL per resident per day) than in the
control group (5.7 per resident per day)(Welch t test, P = .009).
Data on AGE and ARI clustered cases episodes
Only 3 AGE clustered cases episodes were reported over the1-year study
period in the context of clustered cases episodes: 2 occurred in the intervention
group and 1 in the control group. Six ARI clustered cases episodeswere reported in
the intervention group, all taking place in January-March 2015; a single episode was
reported in the control group. Because a large influenza epidemic occurred in France
in January 2015, with a documented severe impact on older adults,24 we suspected
that underreporting occurred. The data were verified qualitatively after the end of the
intervention through individual phone interviews with each participating NH. Based
on these interviews, ARI clustered cases episodes had actually occurred in 12 out of
13 control NHs; however, only 1 had been notified to health authorities. No
unreported clustered cases episodes were identified in the intervention NHs.
Because of the lack of reliable data and suspected severe underreporting in
the control group, we decided not to analyze the available reported AGE and ARI
data.

Secondary outcomes
There was no difference at baseline (January-March 2014)between
intervention and control NHs in the mortality rate (2.3% vs 2.1% per month),
hospitalization rate (6.5% vs 6.0% per month), or antibiotic prescription rate (5.4 vs
5.2 DDDs per 100 resident days), respectively.
Over the 1-year intervention period, mortality and antibiotic prescriptions were
significantly lower in intervention than in control NHs (2.10% vs 2.65% per month,
respectively; P = .003; 5.0 vs 5.8 DDDs per 100 resident days, respectively; P
<.001). In particular, there was a 30% lower mortality in the intervention group in
January-March 2015 (2.45 vs 3.64 deaths per 100 residents per month, respectively;
P = .004). There was no significant difference in hospitalization rates between the 2
groups at any time during the study.
Figure 1 depicts changes in the quarterly mortality rate, hospitalization rate, and
antibiotic prescription rate in both groups from January 2014-March 2015. There was
a significant increase in mortality from baseline over the January-March 2015 quarter
in the control group (P <.001), but not in the intervention group. There were no
significant temporal changes in hospitalization rates. In both groups, antibiotic
prescriptions decreased during spring and summer, and increased in winter, with
significantly more prescriptions in January-March 2015 compared with baseline (P
<.001). In the controlgroup, this increase was also significant over the October
December 2014 quarter.

Multivariate longitudinal data analysis

To account for seasonality, we performed multivariateanalyses of 1-year
longitudinal mortality, hospitalization, and antibiotic prescription data, using a GEE
approach with a Poisson distribution.
Four factors were significantly associated with mortality over the whole
intervention duration (Table 2). The first was the study period: the peak in mortality in
January-March 2015 reflected the large influenza epidemic that occurred in France
during this time period, with a major impact on older adults.23 The second was the
intervention: a mortality rate ratio of 0.76 (exp[−0.27]) was observed between
intervention and control NHs over the entire study period. The third was the
paramedical staff-toresident ratio: mortality was higher when less paramedical staff
(nurses and auxiliary nurses) were available per resident. Finally, the fourth was the
influenza vaccination of the NH staff: paradoxically,resident mortality increased with
vaccination coverage among the staff.
On the other hand, the modeling of the hospitalization data showed a
significantly protective effect of the vaccination of residents against influenza.
Hospitalizations were also significantly higher in NHs with high proportions of double
rooms.
Finally, the study period was significantly associated with antibiotic
prescriptions, with more prescriptions between October 2014 and March 2015,
compared with the April-June 2014 period, reflecting higher antibiotic prescription
rates in winter. In addition, there were significantly more antibiotic prescriptions in
NHs hosting older residents.

DISCUSSION

In this randomized controlled trial, a multifaceted HH interventionwas

associated with a short-term decreased mortality in NHs, but not with morbidity.
Based on a recent review, there are very few randomized trials on HH in NHs.
However, another trial was performed in France in 2010. Although the implemented
multicomponent intervention was similar to ours, no statistically significant difference
was found in the investigated end point, which was the total infection rate, leading
the authors to conclude that their intervention had no impact on the infectious risk.
The difference in conclusions between our 2 studies may be ascribed to 2 main
reasons. First,the context was different because the 2014-2015 French seasonal
influenza epidemic led to many cases among older adults, whichwas not the case in
2010 when pandemic influenza mostly affected young people. Second, the
significant impact we found was for the mortality rate, an outcome which was not
provided in the 2010 study.

Discussion of study results

Overall, the mortality rate in the NHs participating to our study(26% per year
over the year 2014) was consistent with that observed in a large French survey
based on 3,705 NHs, in which the 2012 mortality rate was estimated at 25% per
year.26 The data we collected showed an increase in mortality in January-March
2015, probably related to the large influenza epidemic that occurred in France at the
time. Because, in this randomized controlled trial, this increase was significantly
larger in control than intervention NHs, even when adjusting for resident and NH
characteristics, we concluded that the HH-related intervention had impacted
mortality. Nevertheless, the influenza epidemic may have had a strong
organizational impact on NHs, including changes in quality of care and care
practices (eg, mandatory masking), changes in resident needs for care, and
increased staff and resident turnover or limitation of family visits, and we did not
have complete data to prove that this impact was similar in the intervention and
control groups. Hence, there is a risk for confounding which should be kept in mind
when interpreting our results in terms of mortality. The same may be said for
antibiotic prescriptions, which are affected by many other Nhspecific factors beyond
the infectious risk, including prescriber behavior.
In addition, although our main result was a significantly lower mortality in
intervention NHs, this impact was not sustained after the end of the study. Indeed, in
intervention NHs, the mortality rate increased from 2.53 deaths per 100 residents per
month inJanuary-March 2015 to 2.87 deaths per 100 residents per monthin April-
June 2015 (vs a decrease from 3.64 to 2.63 deaths per 100 residents per month in
the control group). This may be because of a harvesting effect of the winter
epidemics (including influenza) which occurred in control NHs but was prevented by
our intervention. Such mortality displacement has been observed before in older
populations, for instance when investigating the impact of the 2003 heat wave in
France.27 Although this certainly puts our results into perspective, it should be
considered regarding the 18-month median life expectancy at admission in private
medicalized NHs in France,28 compared with which a 3-month delay in mortality is
nonnegligible.
Finally, a rather unexpected result of this study was the significant positive
association between influenza vaccination coverage among the staff and resident
mortality (detailed data available from the authors on request). Although the
evidence on the benefits of staff vaccination in NHs is very limited,29 to our
knowledge no earlier study has found such an adverse impact. However, correlation
does not imply causation, and there may be a confounding factor; however, none
was identified in our analysis. The increased mortality in NHs with high vaccination
coverage among the staff could also be because of a behavioral reason. Indeed,
some of the NH staff who were vaccinated against influenza may hypothetically have
felt protected against seasonal infections (in general), leading them to be less
compliant with preventive measures and therefore more likely to transmit other
infectious diseases. Finally, because a significant portion of the influenza viruses
which circulated in France in early 2015 were antigenically drifted from the vaccine
component, the 2014-2015 vaccine did not offer the expected high protection.30
Although we did find a protective effect of resident vaccination on hospitalizations, a
differential impact of the vaccine according to age may have led to an increased risk
of infection with the circulating virus in vaccinated staff. Such an increased risk was
suggested during the 2009-2010 pandemic in individuals vaccinated with the
seasonal vaccine.31

Discussion of study methodology and collected data

This study has several limitations. First and foremost, we were not able to
analyze data on our primary end point, that is, the incidence rate of ARI and AGE
reported in the context of clustered case episodes. The underreporting that occurred
in our study is actually consistent with a recent report on the French surveillance
system for gastroenteritis in long-term care, which described a great variation in the
outbreak notification rate between regions, and suggested that the true burden of
outbreaks in NHs was greatly underestimated.32 Interestingly, personal
communication with NHs participating in our study suggests that underreporting was
large in control NHs (>90%), whereas it was far less frequent in intervention NHs.
This indicates that infection surveillance within French NHs might be improved via
interventions targeted toward NH staff to increase their understanding of the
infectious risk and their perception of epidemic reporting. This may actually explain
the 81% sensitivity of the French ARI and AGE surveillance system that was
reported in the 2011-2012 pilot study, during which a number of accompanying
measures were proposed to participating NHs (eg, training sessions, freely available
influenza testing kits).
Second, all-cause mortality is not an accurate indicator of the infectious risk,
and using specific mortality rates because of infectious diseases would have been
preferable. However, determining the true cause of deaths in NH residents often
proves complicated.33In addition, resident mortality clearly peaked during the 2014-
2015 influenza epidemic, suggesting that, at least in our study, variations in all-cause
mortality were correlated to the infectious risk.34 The same argument can be made
for overall antibiotic prescription rates. In the absence of reliable infection data, these
2 secondary outcomes were used as surrogate measures of the impact of our
intervention on the infectious risk.
Third, handrub consumption data (per resident day) is a globalmeasure which
does not provide information on interindividual or temporal variability in HH
compliance. Additionally, if NHs differed in resident characteristics, such as acuity,
dependency, or age, theremay be more expected staff-resident encounters in some
NHs, thereby requesting more handrub solution to reach a similar compliance rate.
For all these reasons, the gold standard recommended by the World Health
Organization for HH assessment is direct observation of HH opportunities and
practices. However, this is a costly and labor-intensive option requiring adequately
trained personnel. In this study, we chose handrub consumption as a simpler
alternative based on routinely available data thatwould not be biased by the
intervention. Because there was no significant baseline difference between
intervention and control resident characteristics (Table 1), we considered that this
measure constituted a reliable proxy of HH compliance, and that the 40% higher
handrub consumption in intervention NHs indeed reflected increased HH
compliance, even though the possibility of staff bringing home some of the pocket-
sized handrub containers cannot be ruled out.

Conclusions
Because of the specificities of long-term care settings, reachingyear-long HH
compliance rates similar to those achieved in acute care settings seems unrealistic.
However, our results underline the importance of increasing HH compliance in NHs
during epidemic periods (eg, the January-March 2015 period in our study). Hence, a
practical strategy for infection prevention specifically adapted to NHs could be
proposed, with potentially high benefits. Because our study underlined the low
sensitivity of the current ARI and AGEFrench surveillance system, this would involve
building an improved surveillance system that would allow the detection of outbreaks
of transmissible diseases within NHs with high sensitivity. This highly sensitive
surveillance system could then be coupled with guidelines for increased HH
practices during epidemic alerts.

Acknowledgments
We thank Caroline Blochet for her help in collecting the data onantibiotic
prescriptions, and Marie Arduin for her work in providing support and training
sessions on hand hygiene in intervention nursing homes.We also thank the staff of
all participating nursing homes.

Table 1
Baseline characteristics of participating nursing homes, January-March 2014, and
influenza vaccination rates among staff and residents for the 2014-2015 season

Nursing home characteristic

No. of beds per NH
Bed occupancy rate
Resident age (y)
Proportion of male residents
Resident dependency score (GMP*)
Resident pathology score (PMP†)
Staff-to-resident ratio
Staff absenteeism rate
Proportion of double rooms
Staff influenza vaccination rate
Resident influenza vaccination rate

NOTE. Values are mean •} SD or as otherwise indicated.

GMP, mean weighted dependency score; NH, nursing home; PMP, mean weighted
pathology score.
*Measure of the average dependency level of the residents of a NH, on a scale of
70-1,000. A high GMP is found in NHs which host highly dependent residents.
†Measure of the average pathology level of the residents of a NH, on a scale of 0-
800. A high PMP is found in NHs which host residents requiring high levels of health
care.

Table 2
Predictors of all-cause mortality in nursing home residents, April 2014-March 2015

Variables, Coefficient,P value

Study group (intervention vs control) −0.27 .008
Influenza vaccine coverage among staff 0.84 .007
Influenza vaccine coverage among residents 0.02 .95
Proportion of double rooms 0.07 .91
Proportion of male residents 1.10 .19
Proportion of residents with high
dependency scores
0.52 .36
Proportion of residents aged ≥85 y −1.29 .16
Paramedical staff per resident ratio −0.02 .005
Study period
(reference: April-June 2014)
July-September 2014 −.05 .61
October-December 2014 −.13 .07
January-March 2015 .30 .004

NOTE. Results are from a generalized estimating equation approach with a Poisson
distribution.