Healthcare-associated infections in long-term care facilities: a systematic review and meta-analysis of point prevalence studies

Protocol and registration

The protocol for this systematic review and meta-analysis was registered with PROSPERO (No:CRD42023406844) and developed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (online supplemental file S1).19

Public involvement

LTCF residents and their representatives were not involved in the conduct of this systematic review and meta-analysis.

Information sources and search strategy

On 1 March 2023, an initial systematic search of bibliographic databases MEDLINE (Ovid), Embase (Ovid) and CINANL (EBSCOhost) was conducted to identify potentially eligible publications published between January 1991 and March 2023. This search strategy, repeated on 30 June 2023 with the eligibility criteria extended to include publications up to June 2023, used a combination of keywords and subject headings on ‘long-term care facilities’, ‘infections’ and ‘point prevalence’ (online supplemental file S2). References of short-listed full-text publications, the ECDC HALT project website10 and an unpublished 2016–2022 Australian Aged Care National Antimicrobial Prescribing Survey (NAPS) report (coauthored by NB, RJ, CI, MJM and others) were also reviewed. The annual Australian Aged Care NAPS PPS based on the ECDC HALT project officially commenced in 2016.

Eligibility criteria and study selection

For each publication that was identified as relevant to the area of interest, selection was based on the following inclusion and exclusion criteria (box 1).

Using Covidence,20 the titles and abstracts of all publications resulting from the searches were independently screened for inclusion by two reviewers (NB and SKT). Full-text articles of the preselected publications were assessed before making a final decision on eligibility for inclusion. In case of discrepancy between the two reviewers, the publication was discussed until a consensus was reached.

Data extraction

Standardised data were collated from eligible full-text publications by reviewer NB and cross-checked by reviewer SKT. For each PPS, the following variables, if available, were recorded:

• First author.

• Year of publication.

• Location (country and region).

• Study year (and season).

• Number and type of LTCFs (total no. or mean (range)).

• Number of residents (total no. or mean (range)).

• HAI surveillance definition used.

• Prevalence (%).

• Most frequent HAI (% of all HAIs or prevalence of residents with HAI).

Where prevalence information was available in figure format only, the numerical value was estimated using WebPlotDigitizer V.4.6 (https://automeris.io/WebPlotDigitizer) and clearly labelled in the footnote of table 1.

Quality assessments

The quality of each PPS was independently assessed by two reviewers (NB and SKT) using a modified version of the recommended Joanna Briggs Institute (JBI) Critical Appraisal Tool for Prevalence Studies.21 The JBI tool was developed following extensive peer review and has been approved by the JBI Scientific Board. Nine criteria are described to appraise the methodological quality of a study and the extent to which the possibility of bias in its design, conduct and analysis has been addressed. To enable a detailed assessment, in our modified version (11 criteria), some JBI criteria were subdivided or merged (online supplemental file S3). For all included PPS, results were presented as yes (Y), no (N) or unknown (U) for each criterion and summarised as a percentage for yes (Y) scores.

Data synthesis

A meta-analysis was performed overall and by prespecified subgroups:

• Geographical region (Oceania (Australia), Europe, North America (USA) and Asia (Hong Kong)). European countries were further grouped into Eastern, Western, Northern, and Southern Europe, and the British Isles.22

• Study decade (1991–2000, 2001–2010, 2011–2020, 2021–2023).

• Season (winter vs other seasons).

• Facility type (nursing homes (NHs), residential homes (RH) and mixed/unspecified LTCFs).

• HAI surveillance definition used (McGeer et al23 and Stone et al24 with or without modification).

A PPS was classified as having been conducted in winter if any of their participating LTCFs collected data during any part of winter; in the northern hemisphere from December to February and in the southern hemisphere from June to August. NHs differ from RHs in that medical and skilled nursing care is provided in addition to assistance with daily living activities. HAI surveillance definitions specifically for use by LTCFs were first developed by McGeer et al23 in 1991 and revised by Stone et al24 in 2012.

Where provided by the original publication, the prevalence estimates by facility type were considered as separate prevalences such that a survey could contribute >1 estimate to the meta-analysis. A random intercept logistic regression25 was used to model the average HAI prevalence across the various studies with a categorical indicator for study as the random effect. The R function metaprop() from R package meta26 was used with CIs calculated using the Clopper-Pearson method. A random-effects meta-analysis was performed to allow for heterogeneity in true HAI prevalences, rather than a fixed-effects meta-analysis which assumes any interstudy differences in prevalence is due to sampling error.27 The same method was used for all subgroup analyses, with data restricted to the subgroup of interest, and for the sensitivity analysis which included only studies assessed as high quality (quality score ≥70%). Heterogeneity in study estimates within and between subgroups was examined using Cochran’s Q test, the Higgins’ I² and τ.2,28–30 Maximum-likelihood estimator was used to estimate τ². Funnel plots and Egger’s regression test were used to check for publication bias.31 Publication bias was accounted for using the Duval and Tweedie’s trim-and-fill method.32

All analyses were performed by using RStudio V.2022.12.0. The R script used in our analysis is in online supplemental file S4.

Search output

The initial search of three bibliographic databases identified 1354 publications. After duplicates were removed, the title and abstract of 979 publications were screened against the inclusion and exclusion criteria (box 1). 47 of these publications and 13 additional publications identified by checking references of short-listed full-text publications, the ECDC website10 and an unpublished 2016–2022 Australian Aged Care NAPS PPS report were considered eligible for full-text review (figure 1). Of reviewed publications, 31 were eligible for inclusion, encompassing a total of 123 PPS (74 PPS were incorporated in one of three ECDC HALT reports11–13 which are described separately in online supplemental file S55. Additional information for two PPS33 34 was obtained by contacting the author (Dr A, Eikelenboom-Boskamp, 2023) (online supplemental file S6). 17 publications (13 PPS)35–54 were excluded because the reported findings had mostly been published in ECDC HALT reports.

Request permission

Study characteristics

The study characteristics of the 123 PPS (127 point prevalence estimates), including HAI prevalence and quality ratings, are presented in table 1 and online supplemental files S5 and S7, disaggregated by study year where possible. Four PPS55 56 contributed two point prevalence estimates each, separately for NHs and RHs.

The vast majority of PPS were undertaken in high-income European countries (87.8%, 108/123) and primarily between 2010 and 2019 (79.7%, 98/123). Over half (58.5%, 72/123) of PPS had less than 50 participating LTCFs, and in 41.5% (51/123), the cohort of participating LTCFs was NHs only. Norway (10 PPS between 1997 and 2016/17) and the Netherlands (14 PPS between 2007 and 2016/17) mostly participated with sample sizes ranging from 21 to 540 and 4 to 57 LTCFs, respectively.

Quality assessments

Nine PPS (7.3%, 9/123) were allocated a modified JBI assessed quality score greater than 80% while 30 PPS (24.4%, 30/123) were allocated a score between 70% and 80%. The quality score range of PPS included in the three ECDC HALT reports improved over time: HALT 1 (2010) 36.4%–45.5%, HALT 2 (2013) 45.5%–54.6% and HALT 3 (2016/17) 63.6%–81.8%. Most PPS scored a ‘no’ or ‘unknown’ for having conducted data analysis with sufficient coverage of the identified sample (96.7%, 119/123) and reporting interobserver comparison where data collection was performed by more than one observer (87.8%, 108/123). Less than one-third (27.6%, 34/123) of PPS included at least the estimated minimum sample size for adequate precision of the prevalence estimate (6108 residents). Nearly all PPS scored a ‘yes’ for using valid methods (99.2%, 122/123) and reporting prevalences as % with N or n/N (100%, 123/123) where N=number of included eligible residents and n=number of eligible residents with one or more HAI (online supplemental file S7).

Prevalence of HAIs

The HAI point prevalence ranged from 0% in Cyprus in the 2010 HALT study (online supplemental file S5) to 18.7% in Poland (table 1). The overall pooled point prevalence across 127 prevalence estimates encompassing 709 860 residents was 3.5% (95% CI 3.1% to 4.0%), with heterogeneity statistic I²=98.9% and τ²=0.553. A similar pooled point prevalence was found when the analysis was restricted to PPS with a quality score ≥70% (3.9% (95% CI 3.2 to 4.7%) across 39 prevalence estimates encompassing 221 942 residents.

Visual inspection of the funnel plot and Egger’s regression test (p=0.02) suggested possible publication bias. The pooled HAI point prevalence, corrected for publication bias using Duval and Tweedie’s trim-and-fill method, was 5.4% (95% CI 4.6% to 6.3%; I²=99.2%, τ²=1.045).

Type of HAIs

Of 120 PPS, the most frequently reported HAIs were urinary tract infections (UTIs; 38.9%, 51/131), respiratory tract infections (RTIs; 33.6%, 44/131), and skin and soft tissue infections (SSTIs; 23.7%, 31/131). For three PPS, the most frequently reported HAIs were not specified. For 10 PPS, 2 HAI types were most frequently and equally reported.

Variation in HAI point prevalence

Variation in HAI pooled point prevalence was observed by geographical region (p<0.001; figure 2), though with a notably uneven covariate distribution encompassing a range of overall sample sizes from 5460 residents among surveys conducted in Asia to 211 730 in Oceania. The prevalence by region ranged from 1.2% (95% CI 0.9% to 1.5%) in Oceania to 5.2% (95% CI 4.0% to 6.7%) in Southern Europe. HAI pooled point prevalence decreased over decades from 6.9% (95% CI 0.06% to 0.08%) between 1991 and 2000 to 3.0% (95% CI 0.03% to 0.04%) between 2011 and 2020 and 1.0% (95% CI 1.0% to 1.1%) between 2021 and 2023 (test for subgroup differences: p<0.001; figure 3), noting that the latter comprised two Australian Aged Care NAPS PPS only. The sample size ranged from 16 357 residents for surveys conducted between 1991 and 2000 to 430 756 for surveys conducted between 2011 and 2020. HAI pooled point prevalence was much higher in PPS that used the McGeer et al HAI surveillance definitions without modification (10.6% (95% CI 8.4% to 13.4%)) compared with PPS using McGeer et al definitions with modification (2.4% (95% CI 1.9% to 2.9%)) or the Stone et al definitions with (2.9% (95% CI 2.4% to 3.5%)) or without modification (3.4% (95% CI 1.9% to 6.0%)) (test for subgroup differences: p<0.001; figure 4). The sample size ranged from 6744 residents for studies that used McGeer et al definition without modification to 392 785 for studies that used the Stone et al definition with modification. HAI pooled point prevalence did not differ across facility types (p=0.57) nor by season (p=0.46).

Request permission

Request permission

Request permission

Principal findings

Our comprehensive systematic review and meta-analysis provides valuable insight into HAIs as determined by PPS in LTCFs; literature detailing 123 PPS (127 point prevalence estimates) undertaken in 33 countries between 1991 and mid-2023 was compiled and analysed. The pooled HAI point prevalence (overall global burden) was 3.6%. The majority of PPS reported more than one infection type; most frequently UTIs, RTIs and SSTIs. There was some variation in HAI point prevalence by geographical region, study decade and HAI surveillance definition used; within these subgroups, the highest prevalence was reported for Southern Europe (5.2%), 1991–2000 (6.9%) and McGeer et al without modification (10.6%), respectively.

Our reported high level of LTCF participation in HAI PPS supports the general view that this surveillance method is easy to conduct, at least in high-income countries.8 57 Low-income to mid-income countries may have been under-represented because comparatively their healthcare systems are inadequately funded and developed and HAI surveillance programmes need to be improved.18 58 59 Three PPS (2020–2022 unpublished Australian Aged Care NAPS PPS) only reporting HAI prevalence since the onset of the COVID-19 pandemic may have been due to a reporting lag or it may have been that participation in a PPS was postponed or declined during this time as LTCFs were largely focused on responding to the COVID-19 pandemic.60

We found the highest HAI point prevalence estimate was reported by one Polish NH; 18.7% in 2009.56 Later, 3 and 24 Polish LTCFs (although still small sample sizes) reported their HAI prevalence as 1.9% (HALT 1; 2010)11 and 3.9% (HALT 3; 2016/2017),12 respectively. For both these studies, the Polish LTCFs point prevalence estimates were above the median estimates; 1.5% and 2.1%, respectively. Polish LTCFs did not participate in HALT 2 (2013).

For some LTCF cohorts in Italy and The Netherlands that similarly participated at least annually over a prolonged period, their HAI prevalence decreased from the first to last PPS.34 61 62 A recently published study likewise found, over 10 years (2009–2019) the HAI risk for residents in Dutch LTCFs that participated ≥4 years in PPS was decreased (OR 0.72 (95% CI 0.57 to 0.92)) compared with the first year.63 This large study was excluded from our review because our required variables were not reported for each biannual PPS.

Variation in HAI point prevalence estimates could have been attributable to other factors not specifically explored in our meta-analysis. HALT 3 reported, in addition to facility size, HAI prevalence within an LTCF was associated with the number of residents >85 years, wheelchair-bound, bedridden, disorientated, having a wound other than a pressure sore and/or with a urinary or vascular catheter.13 Other reported factors include differences in the level of training and skills of surveyors, recommendations for and availability of diagnostic testing and surveillance reporting behaviour.12 57 Comparing HAI point prevalence estimates in LTCFs within and between countries should at least take into account these factors to avoid misleading conclusions.

While we found that HAI prevalence differed by study decade and HAI definition used, we acknowledge that these factors are likely correlated because the McGeer et al definition23 was primarily used prior to 2012 and thereafter superseded by the Stone et al definition.24 We expected HAI pooled point prevalence estimates to vary too across seasons (winter vs other seasons); the seasonally high incidence of RTIs during winter in temperate regions and the rainy season in tropical regions is widely recognised.64 Although this was not observed in our study, there was a particularly notable imbalance in sample size between the subgroups, with only 15 prevalence estimates contributing to the pooled HAI prevalence for winter vs 112 for other seasons. A recent study65 suggests future LTCF HAI point prevalence estimates may reflect seasonal spikes of COVID-19 infection during winter despite continual transmission throughout the year. COVID-19 infection is to be reported for the first time by LTCFs participating in the fourth HALT PPS.14

Strengths

A notable strength of our systematic review and meta-analysis, aside from following PRISMA guidelines and using two independent reviewers for study selection, was undertaking a modified JBI assessment of the methodological quality of each PPS. This rigorous assessment assists in determining how the results should be interpreted.

Our modified JBI quality assessment found for the most PPS there was insufficient representative coverage of identified country samples. Many LTCFs within countries volunteered to participate, perhaps resulting in a selection of ‘PPS participants’ with higher awareness and implementation of optimal IPC policies and practices. It is unknown exactly how much non-representation impacted on the extent to which ‘national results’ could be extrapolated to entire countries.

We also found for most PPS interobserver comparisons were not reported. Exceptionally and as part of the HALT 2 and 3 studies,12 13 17 and 20 LTCFs, respectively, from 10 countries or administrations participated in an optional validation study. External ‘gold standard’ teams re-examined resident charts independently from the primary PPS surveyors. The sensitivity of the HALT 2 and 3 HAI data was 79% and 76%, respectively. This may have led to a slight underestimation of HAI prevalence. For both studies, the specificity was nearly 100%.

Limitations

The foremost limitation was that some relevant PPS may have been missed. As is most common for scientific publications,66 we because of resource constraints explicitly excluded non-English language publications. Ideally, systematic reviews and meta-analyses should include these publications if it is determined such publications are of comparable quality to those published in English.67 While we included unpublished Australian Aged Care NAPS PPS reports (easily accessible to the authors), we did not actively seek to include other similar unpublished reports. Current empirical research has shown results may be impacted only for a minority of meta-analyses excluding unpublished publications.68

Our subgroup analyses were limited by uneven covariate distribution and high heterogeneity within subgroups, and thus the results should be interpreted with caution.69 When stratified by geographical region, the number of subgroup-specific prevalence estimates ranged from 2 in Asia to 24 in Southern Europe, and the number of residents ranged from 5460 in Asia to 198 605 in Oceania. An uneven distribution was also observed by study decade, season, facility type and HAI surveillance definition used. Heterogeneity was very high with I²≥88% for all subgroups, indicating a large portion of the observed variation within the subgroup was due to variation in the real prevalences.

Future research

Some results of our systematic review and meta-analysis highlighted subsequent research projects are warranted, including:

• Exploring why LTCF HAI PPS data from low-income to mid-income countries is lacking and determining the feasibility, resources needed and cost-effectiveness for LTCFs in these countries to participate in reliable HAI surveillance systems.

• To support benchmarking (less variance) between and within countries, standardising and validating HAI surveillance methodology. In countries where microbiology laboratory access is limited or non-existent, HAI definitions based on clinical data only may be most appropriate.

• Testing the (long-term) impact on HAI point prevalence estimates of implementing prioritised UTI, RTI and/or SSTI clinical guidelines and quality improvement activities.

• Evaluating exactly why for some repeatedly participating LTCF cohorts, their HAI point prevalence estimates have decreased over time.