Spatial and temporal analyses to investigate infectious disease transmission within healthcare settings.

BACKGROUND: Healthcare-associated infections (HCAIs) cause significant morbidity and mortality worldwide, and outbreaks are often only identified after they reach high levels. A wide range of data is collected within healthcare settings; however, the extent to which this information is used to understand HCAI dynamics has not been quantified. AIM: To examine the use of spatiotemporal analyses to identify and prevent HCAI transmission in healthcare settings, and to provide recommendations for expanding the use of these techniques.
METHODS: A systematic review of the literature was undertaken, focusing on spatiotemporal examination of infectious diseases in healthcare settings. Abstracts and full-text articles were reviewed independently by two authors to determine inclusion.
FINDINGS: In total, 146 studies met the inclusion criteria. There was considerable variation in the use of data, with surprisingly few studies (N = 22) using spatiotemporal-specific analyses to extend knowledge of HCAI transmission dynamics. The remaining 124 studies were descriptive. A modest increase in the application of statistical analyses has occurred in recent years.
CONCLUSION: The incorporation of spatiotemporal analysis has been limited in healthcare settings, with only 15% of studies including any such analysis. Analytical studies provided greater data on transmission dynamics and effective control interventions than studies without spatiotemporal analyses. This indicates the need for greater integration of spatiotemporal techniques into HCAI investigations, as even simple analyses provide significant improvements in the understanding of prevention over simple descriptive summaries.

Introduction

Healthcare-associated infections (HCAIs) are problematic worldwide, with a recent report by the World Health Organization estimating hospital-wide prevalence in high-income countries at 8%. In addition to causing significant, yet preventable, morbidity and mortality in countries with centrally-funded and managed healthcare systems, such as the UK National Health Service, HCAIs increase waiting times and reduce availability of resources to provide care to the population.

HCAIs present a unique challenge as active transmissions are often only identified after numerous patients have been infected. Additionally, the wide range of HCAI facilitators (e.g. procedures, environment) and increasingly susceptible patients complicate transmission dynamics, making prospective identification and control exceedingly difficult. When multiple cases of an infection occur within a hospital, it is difficult to differentiate a true nosocomial transmission from unrelated cases, and cohorting patients by risk group may lead to assumptions of a common source but molecular analyses often demonstrate lack of transmission.

Sophisticated spatiotemporal analyses can be used to confirm clustering statistically over time and/or space, which would increase confidence in assuming the relatedness of cases. These methods can also be used to control for the effects of cohorting and other patient characteristics that may give the spurious impression of clustering or transmission when it has not occurred. This, in turn, would provide better information on where interventions could be targeted most effectively, and when or where to anticipate outbreaks. These methods may also be useful in more rapid identification of a problem, as even small clusters (e.g. two or three cases) can be detected. Even the introduction of more simplified analytical methods to evaluate spatial and temporal relationships could be beneficial. One example is the Knox test, which has been used widely to detect time–space clusters since the 1960s. The null hypothesis in Knox testing would be that all HCAI cases are independent, and the test returns the number of pairs of cases that are deemed to cluster in time and/or space. The tool is simple to apply as it only requires information on cases, not controls or susceptible individuals, and can work on a minimal clinical dataset.

Nowadays, researchers are using geographic information systems (GIS) to further extend understanding of spatiotemporal clustering and transmission. These are computer-based programs that combine cartography, statistical analysis and database technology to layer databases on top of a predefined map. They have been applied in a range of ecological investigations of disease, and to determine whether there is a spatial association between disease risk and environmental pollution. In this study, GIS and spatial analysis were employed to investigate the risk of breast and lung cancer in a small region. After identification of significant clusters, it was possible to identify local risk factors specific to each cancer type, providing evidence of potential environmental contamination. The use of similar techniques to create hospital maps, on which infection data can be displayed and analysed, could increase understanding of local transmission and risk, and provide rapid dissemination of information through visualization.

With healthcare systems worldwide under pressure to improve patient safety whilst cutting costs, use of the existing infrastructure of routinely collected data, which are often overlooked for HCAI investigation and research, is an innovative solution. Frequently, investigations of HCAIs provide a basic epidemiological description of cases over time by providing an epidemic curve, or show how cases are distributed across wards using a diagram. However, hospital databases contain laboratory results, building management data and floor plans, and information on patient admissions and movement that could easily be incorporated into more detailed analyses to improve understanding of local HCAI epidemiology. Use of interdisciplinary tools may increase the ability to identify transmission prospectively and implement preventive measures.

The aims of this review were to determine the extent of use of spatiotemporal analyses for identifying and preventing HCAI transmission, and to provide recommendations for expanding the use of GIS and spatiotemporal statistical analyses within healthcare settings.

Methods

A systematic review of the literature on spatiotemporal examination of infectious diseases in healthcare settings between January 1961 and June 2013 was conducted using the following search terms: infection (e.g. HCAI, nosocomial, etc.); healthcare settings (e.g. hospital, intensive care, etc.); and time/space (e.g. space–time, spatial epidemiology, etc.). Potential synonyms for each search term (e.g. infection, healthcare settings and time/space) were identified and combined using Boolean operators.

To ensure comprehensive capture of the literature, BIOSIS, Cochrane Review, CSA, DARE, Embase, HEED, JSTOR, PubMed, Science Direct and Web of Science were searched for all indexed publications. Additionally, Google Scholar was searched for indexed and grey literature using the above search terms. All papers, reports, abstracts and letters were included in the initial search.

Inclusion/exclusion criteria

Inclusion/exclusion was conducted in two stages: abstract/title review and full-text review. All identified titles/abstracts were reviewed independently by two authors to ensure reliability in full-text retrieval. Papers were retrieved if they mentioned time or space in the abstract, or no abstract was provided and the title did not provide enough information to assess inclusion.

Full-text papers were reviewed independently by two authors and included if they were: (a) published post-1961; (b) written in English; (c) examined potential transmission in more than three patients; (d) provided more than a simple report of cases over time periods exceeding three months (i.e. not routine national surveillance reports); and (e) discussed time/space as a specific aim or discussion point of the study, rather than a simple mention in the results. Any studies on which the reviewers did not agree were discussed and a consensus was reached.

Data extraction

The methodologies of all included studies were reviewed and categorized into either descriptive or analytical studies of time/space, and further ‘subtyped’ based on the data and analyses employed. Studies were classified as ‘case reporting’ if they only used temporal or spatial data as an overview (i.e. an epidemic curve). ‘Basic descriptive epidemiology’ studies examined how the cases were linked by describing their locality in time and/or space and possible exposure events, but did not use statistical methods to determine the probability that they were linked. ‘Basic descriptive epidemiology with molecular data’ studies focused on the genetic diversity of the organisms, including a description of the distribution of the strains in space and/or time. ‘Statistical spatiotemporal analysis’ studies used statistical methods to explore HCAI distribution in time or space. Studies were classified as ‘statistical spatiotemporal analysis with molecular data’ if they combined molecular data with statistical analyses to investigate the dissemination of strains across time or space. The final category, ‘spatiotemporal analysis using GIS’, included studies that used GIS within hospital settings. Findings were synthesized to evaluate the actual use of spatiotemporal analysis on infectious diseases in healthcare settings. Meta-analysis was considered but was not deemed to be feasible due to the heterogeneity of research designs and outcome measures; as such, the findings were synthesized qualitatively.

Results

In total, 43,819 titles/abstracts were identified during the literature search (Figure 1 ). Of these, 584 met the inclusion criteria for full-text retrieval. Four of the included studies were not available, and 146 were included in the review. Most of the studies excluded did not meet the spatiotemporal criteria; 171 provided some spatiotemporal results but did not discuss this information as it was not the focus of the study. Others examined annual trends, were case studies or review papers, or focused on non-healthcare settings. Six studies were excluded for being written in a language other than English.

Figure 1

Flowchart of article inclusion.

Characteristics of included studies

Included studies were predominantly descriptive in their reporting of spatiotemporal data (85%), and varied greatly in encompassing a range of settings, populations and HCAIs. Half of the studies were carried out hospital-wide (53%), 42% (N = 61) were performed on specific wards, and only seven (5%) were based within nursing homes. The majority of studies were retrospective (72%), often using data to investigate outbreaks after the event had ended. Five (3%) studies combined retrospective analyses and prospective interventions to enhance surveillance, whilst the remaining 25% of studies were prospective.

Numerous HCAIs were investigated, with the most common group being bacteria (62%), such as meticillin-resistant Staphylococcus aureus (MRSA) (11%) and Clostridium difficile (10%). Studies focusing solely on viral or fungal infections accounted for 29% of studies, including norovirus (3%), severe acute respiratory syndrome (3%) and Aspergillus spp. (7%). Among the 74 molecular studies in this review, only five incorporated spatiotemporal analyses to understand the transmission dynamics.

The studies were separated into two types, ‘descriptive’ and ‘analytical’, based on their use of spatiotemporal-specific statistical analyses. To enable clearer comparisons between the groups and to evaluate the variation in exploitation of clinical data, the studies were further classified into six subtypes (Table I ). These are described in detail below.

Table I

Categories of studies identified in the review

Study type	Study subtype	Number of papers (%)
Descriptive (N = 124, 85%)	Case reporting	29 (20)
	Basic descriptive epidemiology	26 (18)
	Basic descriptive epidemiology with molecular data	69 (47)
Analytical (N = 22, 15%)	Statistical spatiotemporal analysis	13 (9)
	Statistical spatiotemporal analysis with molecular data	5 (3)
	Spatiotemporal analysis using geographic information systems	4 (3)

Descriptive studies

Descriptive studies primarily focused on summarizing outbreak investigations, environmental assessments and cluster identification. For the vast majority of these studies, causes or sources of outbreaks, cases or clustering were the primary aim. However, simple qualitative descriptions were not sufficient in most cases to confirm or refute identified sources or clusters.

Case reporting studies

Basic descriptions of time and space were common (20%, N = 29) (Table II ). Two-thirds of these studies provided a retrospective temporal description of the incidence of cases over time (i.e. an epidemic curve). Many of these studies examined outbreaks15, 16, 17, 18, 19, 20, 21, 22 and evaluations of intervention strategies,23, 24, 25 while others attempted to identify factors associated with potential nosocomial transmission (i.e. healthcare worker carriage,26, 27 direct contact with cases,28, 29 inadequate cleaning of medical equipment30, 31 and the physical layout of hospital utilities32, 33). Only four studies described the spatial distribution of cases to show the impact of hospital renovations or the layout of cases across specialities,16, 27, 34, 35 with the majority of studies describing temporal trends in cases.36, 37 Most case reporting studies examined bacteria (62%); however, the most informative studies were those examining organisms such as Aspergillus spp. and Legionella spp., where environmental contamination is considered to be the primary risk factor.15, 35, 38, 39, 40, 41, 42

Table II

Characteristics of descriptive studies included in the review (N = 124)

Temporal/spatial focus	First author	Year	Setting	Organism	Molecular methods	Aim	Study design
Case reporting
S	Arnow PM	1978	Renal ward	Aspergillus spp.	N/A	Describe outbreak	Retrospective
S	Baird SF	2011	Haematology ward	Non-tuberculous mycobacterium	N/A	Outbreak investigation	Retrospective
S	Panwalker AP	1986	Whole hospital	Mycobacterium gordonae	N/A	Describe IC measures	Retrospective
S	Patterson JE	1998	Geriatric ward	Hib	N/A	Detect clusters	Retrospective
T	Addiss DG	1991	Nursing home	Bordetella pertussis	N/A	Describe outbreak	Retrospective
T	Alonso-Echanove J	2001	Whole hospital	Mycobacterium tuberculosis	N/A	Describe outbreak	Retrospective
T	Arnow PM	1991	Whole hospital	Aspergillus spp.	N/A	Assess environmental contamination	Prospective
T	Arnow PM	1998	Haematology ward	Bloodstream infection	N/A	Describe outbreak	Retrospective
T	Arnow PM	1982	Whole hospital	Legionella pneumophila	N/A	Describe outbreak	Retrospective
T	Bayat A	2003	Intensive care unit	Multiple	N/A	Describe outbreak	Retrospective
T	Belani A	1986	Paediatric ward	Staphylococcus aureus	N/A	Describe outbreak	Retrospective
T	Bonilla HF	1997	Whole hospital	VRE	N/A	Describe incidence	Prospective
T	Cartmill TDI	1994	Haematology ward	Clostridium difficile	N/A	Describe impact of IC measures	Prospective
T	Emont SL	1993	Nursing home	Gastroenteritis	N/A	Describe incidence	Retrospective
T	Fourneret-Vivier A	2006	Whole hospital	Aspergillus spp.	N/A	Describe incidence	Prospective
T	Fowler SL	1998	Paediatric ward	Candida lusitaniae	N/A	Describe transmission	Retrospective
T	Haley CE	1979	Whole hospital	Legionella pneumophila	N/A	Outbreak investigation	Both
T	Klimowski LL	1989	Whole hospital	Aspergillus spp.	N/A	Describe incidence	Retrospective
T	Lai KK	1998	Whole hospital	VRE	N/A	Assess impact of IC measures	Prospective
T	Larson JL	2003	Whole hospital	Mycobacterium tuberculosis	N/A	Outbreak investigation	Retrospective
T	Ofner-Agostini M	2006	Multiple hospitals	SARS	N/A	Review IC policies	Retrospective
T	Pegues CF	2001	Whole hospital	Aspergillus spp.	N/A	Describe incidence	Retrospective
T and S	Abulrahi HA	1997	Whole hospital	Plasmodium falciparum	N/A	Determine transmission route	Prospective
T and S	Davies BI	1999	Whole hospital	Streptococcus pyogenes	N/A	Outbreak investigation	Retrospective
T and S	Deutscher M	2011	Whole hospital	Group A streptococcus	N/A	Identify risk factors	Retrospective
T and S	Helms CM	1983	Whole hospital	Legionella pneumophila	N/A	Outbreak investigation	Retrospective
T and S	MacDonald KS	1993	Whole hospital	Clostridium difficile	N/A	Describe incidence	Retrospective
T and S	McGrathEJ	2011	Paediatric ward	Acinetobacter spp.	N/A	Determine transmission route	Retrospective
T and S	WangH	2013	Whole hospital	Listeria monocytogenes	N/A	Describe clinical outcomes	Retrospective
Basic descriptive epidemiology
S	Lentino JR	1982	Whole hospital	Aspergillus spp.	N/A	Detect clusters	Retrospective
T	Bowen KE	1995	Whole hospital	Clostridium difficile	N/A	Describe incidence	Retrospective
T	Buchbinder N	2011	Paediatric ward	Influenza A (H1N1)	N/A	Describe impact of IC measures	Retrospective
T	Burney MI	1980	Whole hospital	CCHF	N/A	Outbreak investigation	Retrospective
T	Burwen DR	2001	Paediatric ward	Aspergillus spp.	N/A	Outbreak investigation	Retrospective
T	Degail MA	2012	Whole hospital	Human metapneumovirus	N/A	Outbreak investigation	Retrospective
T	Fretz R	2009	Whole hospital	Norovirus	N/A	Outbreak investigation	Retrospective
T	Gastmeier P	2003	Paediatric ward	Klebsiella pneumoniae	N/A	Describe a cluster of cases	Prospective
T	Gomersall CD	2006	Intensive care unit	SARS	N/A	Describe incidence	Prospective
T	Kaplan JE	1982	Nursing home	Norovirus	N/A	Evaluate transmission	Retrospective
T	Kimura AC	2005	Paediatric ward	Ralstonia pickettii	N/A	Outbreak investigation	Retrospective
T and S	Alam NK	2005	Whole hospital	Salmonella enterica	N/A	Describe cluster	Retrospective
T and S	Auerbach SB	1992	Nursing home	Group A streptococcus	N/A	Outbreak investigation	Retrospective
T and S	Barrett SP	1988	Whole hospital	MRSA	N/A	Describe incidence	Retrospective
T and S	Bitar CM	1987	Whole hospital	MRSA	N/A	Describe outbreak	Retrospective
T and S	Chen YC	2004	A&E	SARS	N/A	Outbreak investigation	Retrospective
T and S	Faustini A	2004	Whole hospital	Necrotizing enterocolitis	N/A	Outbreak investigation	Retrospective
T and S	Foulke GE	1989	Intensive care unit	Clostridium difficile	N/A	Describe use of IC measures	Retrospective
T and S	Goldmann DA	1981	Paediatric ward	Multiple	N/A	Describe outbreak	Prospective
T and S	Lai KK	2001	Transplant ward	Aspergillus spp.	N/A	Detect clusters	Retrospective
T and S	Mody LR	2001	Whole hospital	Clostridium difficile	N/A	Describe incidence	Prospective
T and S	Pavlov I	2009	Whole hospital	Clostridium difficile	N/A	Detect clusters	Retrospective
T and S	Pegues DA	1993	Nursing home	Gastroenteritis	N/A	Outbreak investigation	Retrospective
T and S	Strabelli TMV	2006	Paediatric ward	Enterococcus faecalis	N/A	Detect clusters	Retrospective
T and S	Turcios-Ruiz RM	2008	Paediatric ward	Norovirus	N/A	Outbreak investigation	Retrospective
T and S	Warren D	1989	Whole hospital	Keratoconjunctivitis	N/A	Outbreak investigation	Retrospective
Basic descriptive epidemiology with molecular data
S	Abdallah IM	2006	Whole hospital	Multiple	RAPD	Investigate strain distribution	Retrospective
S	Aita J	1996	Whole hospital	Mycobacterium tuberculosis	RFLP	Outbreak investigation	Retrospective
S	Hoefnagels-Schuerman A	1997	Whole hospital	MRSA	PFGE	Outbreak investigation	Prospective
S	Katsoulidou A	1999	Haematology ward	Hepatitis C virus	PCR	Outbreak investigation	Retrospective
S	Pegues DA	2002	Intensive care unit	Aspergillus spp.	RFLP	Detect clusters	Retrospective
S	Vazquez JA	1993	Whole hospital	Candida albicans	REA	Investigate strain distribution	Prospective
S	Venezia RA	1994	Intensive care unit	Legionella pneumophila	PFGE	Identify source	Retrospective
S	Witte W	2001	Multiple hospitals	MRSA	PCR	Investigate strain distribution	Prospective
S	Zervos MJ	1987	Whole hospital	Enterococcus faecalis	Plasmid typing	Describe incidence	Prospective
T	Adachi JA	2009	Intensive care unit	Pseudomonas aeruginosa	PFGE	Assess impact of molecular typing	Retrospective
T	Adams G	1981	Paediatric ward	Herpes simplex virus 1	REF	Describe outbreak	Retrospective
T	Alfieri N	1999	Intensive care unit	Stenotrophomonas maltophilia	RFLP	Outbreak investigation	Both
T	Allander T	1995	Haematology ward	Hepatitis C virus	PCR/NASeq	Detect clusters	Retrospective
T	Assadian O	2002	Paediatric ward	Serratia marcescens	PCR	Describe outbreak	Retrospective
T	Aumeran C	2008	Whole hospital	VRE	PFGE	Describe use of IC measures	Prospective
T	Baddour LM	1999	Whole hospital	Enterococcus faecium	CHEF	Describe outbreak	Retrospective
T	Belmares J	2009	Whole hospital	Clostridium difficile	REA	Describe incidence	Retrospective
T	Ben Abdeljelil J	2011	Paediatric ward	Candida albicans	PFGE	Investigate strain distribution	Retrospective
T	Ben Abdeljelil J	2012	Paediatric ward	Candida albicans	PFGE	Outbreak investigation	Retrospective
T	Brillowska-Dabrowska A	2009	Haematology ward	Candida parapsilosis	RAPD	Assess impact of molecular typing	Retrospective
T	DavinRegli A	1996	Intensive care unit	Enterobacter aerogenes	RAPD	Outbreak investigation	Prospective
T	Eyre DW	2012	Multiple hospitals	Clostridium difficile/MRSA	SNV analysis	Outbreak investigation	Retrospective
T	Falk PS	2000	Burns ward	VRE	PFGE	Outbreak investigation	Retrospective
T	Geis S	2013	Haematology ward	Respiratory syncytial virus	RT-PCR	Outbreak investigation	Retrospective
T	Gray J	2012	Paediatric ward	Klebsiella pneumoniae	PFGE	Detect clusters	Retrospective
T	Harvala H	2012	Haematology ward	Parainfluenza type 3	RT-PCR	Outbreak investigation	Retrospective
T	Helali NE	2005	Whole hospital	Staphylococcus aureus	PFGE	Outbreak investigation	Both
T	Helweg-Larsen J	1998	Whole hospital	Pneumocystis carinii	PCR	Detect clusters	Retrospective
T	Hong KB	2012	Paediatric ward	Acinetobacter baumannii	MLST	Outbreak investigation	Retrospective
T	Kakis A	2002	Whole hospital	Group A streptococcus	M typing/T agglutination	Outbreak investigation	Retrospective
T	Layton MC	1993	Dermatology ward	MRSA	PFGE	Detect clusters	Retrospective
T	L'Ecuyer PB	1996	Multiple hospitals	Salmonella senftenberg	PFGE	Outbreak investigation	Retrospective
T	Le Gal S	2012	Renal ward	Pneumocystis spp.	RFLP	Detect clusters	Retrospective
T	Loudon KW	1994	Haematology ward	Aspergillus fumigatus	RAPD	Detect clusters	Retrospective
T	McAdams RM	2008	Paediatric ward	MRSA	PFGE	Detect clusters	Retrospective
T	McFarland LV	1989	Whole hospital	Clostridium difficile	Immunoblot	Describe incidence	Prospective
T	Peta M	2006	Intensive care unit	Enterococcus faecium	PFGE	Outbreak investigation	Both
T	Rupp ME	2001	Paediatric ward	VRE	PFGE	Outbreak investigation	Prospective
T	Sardan YC	2004	Whole hospital	Klebsiella oxytoca	AP-PCR	Outbreak investigation	Retrospective
T	Zoltanski J	2011	Paediatric ward	ARGNB	PFGE	Describe incidence	Prospective
T and S	Abb J	2004	Whole hospital	MRSA	PFGE	Investigate strain distribution	Prospective
T and S	Abbo A	2005	Whole hospital	Acinetobacter baumannii	PFGE	Describe incidence	Retrospective
T and S	Arnold KE	2006	Nursing home	Group A streptococcus	RFLP	Outbreak investigation	Retrospective
T and S	Boyce JM	1993	Whole hospital	MRSA	Plasmid typing	Assess impact of IC measures	Retrospective
T and S	Byers KE	2001	Whole hospital	VRE	PFGE	Assess impact of IC measures	Prospective
T and S	Carneiro MAS	2007	Haematology ward	Hepatitis C virus	RT-PCR	Investigate strain distribution	Prospective
T and S	Chen LF	2011	Haematology ward	Influenza A (H1N1)	RT-PCR	Outbreak investigation	Retrospective
T and S	Culebras E	2010	Whole hospital	Acinetobacter baumannii	RAPD	Describe outbreak	Retrospective
T and S	Cuny C	1993	Whole hospital	MRSA	Phage typing	Outbreak investigation	Retrospective
T and S	Debast SB	1996	Intensive care unit	Acinetobacter baumannii	PCR fingerprinting	Outbreak investigation	Retrospective
T and S	Diab-Elschahawi M	2012	Intensive care unit	Candida parapsilosis	Microsatellite typing/repPCR	Outbreak investigation	Prospective
T and S	Dijkshoorn L	1993	Intensive care unit	Acinetobacter spp.	DNA-DNA hybridization	Investigate strain distribution	Retrospective
T and S	Englund JA	1991	Whole hospital	Respiratory syncytial virus	EIA	Evaluate possible transmission	Prospective
T and S	Fawley WN	2001	Whole hospital	Clostridium difficile	RAPD	Investigate strain distribution	Prospective
T and S	Ferroni A	1998	Whole hospital	Pseudomonas aeruginosa	PFGE	Outbreak investigation	Retrospective
T and S	Fisher GM	1986	Whole hospital	Multiple	Plasmid typing	Investigate strain distribution	Retrospective
T and S	Graindorge A	2010	Intensive care unit	Burkholderia cenocepacia	RFLP	Describe outbreak	Retrospective
T and S	Kassis C	2011	Whole hospital	MRSA	PFGE	Outbreak investigation	Retrospective
T and S	Kondili LA	2006	Renal ward	Hepatitis B/hepatitis C	PCR	Outbreak investigation	Retrospective
T and S	Levidiotou S	2002	Intensive care unit	Acinetobacter baumannii	RAPD	Describe outbreak	Retrospective
T and S	Lin YC	2007	Paediatric ward	MRSA	PFGE	Assess HCW carriage	Retrospective
T and S	Lowe C	2012	Intensive care unit	Klebsiella oxytoca	PFGE	Outbreak investigation	Retrospective
T and S	Lutz BD	2003	Whole hospital	Aspergillus spp.	RAPD	Detect clusters	Retrospective
T and S	Marx A	1999	Nursing home	Gastroenteritis	RT-PCR	Determine transmission route	Retrospective
T and S	Morter S	2011	Whole hospital	Norovirus	Nucleic acid sequencing analysis	Outbreak investigation	Prospective
T and S	Traub WH	1998	Intensive care unit	Pseudomonas aeruginosa	PFGE	Investigate strain distribution	Prospective
T and S	Widmer AF	1993	Intensive care unit	Pseudomonas aeruginosa	CHEF	Evaluate possible transmission	Retrospective
T and S	Xia Y	2012	Intensive care unit	Acinetobacter baumannii	PCR	Outbreak investigation	Retrospective
T and S	Yoon YK	2009	Paediatric ward	VRE	PFGE	Assess impact of IC measures	Both

S, spatial; T, temporal; N/A, not applicable; Hib, Haemophilus influenzae type B; VRE, vancomycin-resistant enterococci; SARS, severe acute respiratory syndrome; CCHF, Crimean–Congo haemorrhagic fever; MRSA, meticillin-resistant Staphylococcus aureus; ARGNB, antibiotic-resistant Gram-negative bacteria; RAPD, random amplification of polymorphic DNA; RFLP, restriction fragment length polymorphism; PFGE, pulsed-field gel electrophoresis; AP-PCR, arbitrarily primed polymerase chain reaction; repPCR, repetitive element palindromic polymerase chain reaction; CHEF, clamped homogeneous electric field electrophoresis; EIA, enzyme immunoassay; MLST, multi-locus sequence typing; PCR, polymerase chain reaction; REA, restriction endonuclease analysis; REF, restriction endonuclease fingerprinting; RT-PCR, reverse transcriptase polymerase chain reaction; A&E, accident and emergency; HCW, healthcare worker; IC, infection control; NASeq, nucleic acid sequencing analysis; SNV, single nucleotide variant analysis.

Basic descriptive epidemiology

Investigations that described the temporal or spatial distribution of cases were categorized as basic epidemiology (18%, N = 26) (Table II). A number of studies provided a retrospective evaluation of the incidence of cases by assessing temporal links between patients, while 58% of studies combined spatial and temporal elements to varying degrees in their evaluations. The main organisms considered were bacteria (46%); however, the studies that evaluated Aspergillus spp. focused on spatial data to the greatest extent. Indeed, the only study that focused solely on the spatial element examined fungal contamination of the hospital environment.

Some studies combined infection information with building ‘schematics’ to explain the physical layout of the ward or to display the location of patients.45, 46, 47, 48, 49, 50, 51, 52, 53 However, these studies did not investigate the importance of the geographical distribution of cases, as has been highlighted in studies that have looked at the impact of construction on the incidence of fungal infections.43, 54, 55, 56 In addition to evaluating the distribution of cases, some investigators used graphics to visualize the connections between cases, and attempted to identify possible clusters58, 59, 60, 61, 62 or provide evidence of potential transmission.63, 64

A number of studies used timelines to evaluate how cases were linked.65, 66, 67 Chen et al. visualized the spread of severe acute respiratory syndrome within an emergency department. By combining temporal data with patient locations, the researchers identified distinct ‘clusters’ of cases, and prevented further dissemination of the disease by quarantining contacts of these individuals. Whilst the outcome was positive, the assumed clusters were based purely on description of the patients' locations at certain times, and the lack of statistical analysis meant that the clusters were not proven to be statistically significant.

Basic descriptive epidemiology with molecular data

Most descriptive studies incorporated molecular data (47%; N = 69) (Table II), presumably because a molecular link provided more evidence of clustering or transmission than purely describing potential clusters. Numerous studies combined spatiotemporal and molecular data to attempt to develop a better understanding of strain dissemination69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81 or potential sources82, 83, 84, 85 within the institutions in which they were conducted. Integration of molecular and temporal data enabled investigators to highlight potential links between patients, and identify potential transmission events,86, 87, 88, 89, 90, 91, 92 with greater substantiation than simple descriptive studies. In one study, researchers were able to differentiate between two consecutive outbreaks of Stenotrophomonas maltophilia on their intensive care unit by visualizing the temporal distribution of isolates identified using restriction fragment length polymorphism (RFLP); however, it is possible that these two distinct outbreaks could have encompassed several smaller events with the same RFLP type introduced multiple times.

Descriptive studies that included molecular data covered a range of applications including outbreak investigations,94, 95, 96, 97, 98, 99, 100, 101, 102 cluster identification and improving control interventions.104, 105 However, the most common applications of molecular typing were in identifying the probable sources of an infection106, 107, 108, 109, 110, 111, 112, 113, 114 and whether transmission had occurred.77, 115, 116, 117, 118, 119, 120, 121 Many studies evaluated the distribution of strains over time or space,122, 123, 124, 125, 126 aiming to establish epidemiological links between cases, but were limited by lack of statistical analysis. The geographical layout of cases was used in some studies to suggest potential factors associated with their distribution.127, 128, 129, 130, 131, 132, 133 The study by Witte et al. mapped the distribution of MRSA strains at national level in Germany to compare changes in resistance phenotypes with various local prescription practices across regions, but there were no statistical analyses to support or refute these qualitative observations.

Analytical studies

Analytical studies tended to focus more on predictive modelling of future outbreaks and determining the impact of various changes within the healthcare setting, rather than describing an outbreak. They used a wide range of statistical modelling techniques, indicating a number of options for looking at spatiotemporal clustering. Interestingly, a number of the descriptive studies focused on identifying the source of an outbreak, and found that they were unable to do so conclusively. In contrast, GIS was shown to enable fast identification of possible sources during an outbreak, and enabled a targeted investigation that led to the source being discovered. This was possible using clinical data that are collected routinely, and required little additional data retrieval.

Statistical spatiotemporal analysis

Thirteen studies (9%) conducted statistical analyses of temporal and spatial data (Table III ). All were undertaken in hospital settings, 85% (N = 11) were published in 2000 or later, and 62% (N = 8) focused on bacterial infections.

Table III

Characteristics of analytical studies included in the review (N = 22)

Temporal/spatial focus	First author	Year	Setting	Organism	Molecular methods	Statistical methods	Aim	Study design
Statistical spatiotemporal analysis
T	Aldeyab MA	2009	Whole hospital	Clostridium difficile	N/A	ARIMA time series	Assess impact of AB use	Retrospective
T	Aldeyab MA	2008	Whole hospital	MRSA	N/A	ARIMA time series	Assess impact of AB use	Retrospective
T	Bertrand X	2012	Whole hospital	MRSA	N/A	ARIMA time series	Assess impact of IC measures	Retrospective
T	Birnbaum D	1984	Whole hospital	Multiple	N/A	Outbreak threshold levels	Detect outbreaks	Prospective
T	Charvat H	2010	Whole hospital	Multiple	N/A	Monte Carlo/time interval distance modelling	Detect clusters	Retrospective
T	Haley RW	1982	Paediatric ward	Staphylococcus aureus	N/A	Multi-variate statistical model	Assess impact of IC measures	Retrospective
T	Polgreen PM	2010	Multiple hospitals	Clostridium difficile/influenza	N/A	Auto-regressive time series analysis	Characterize incidence	Retrospective
T	Vernaz N	2008	Whole hospital	Clostridium difficile	N/A	ARIMA time series	Assess impact of AB use	Prospective
T and S	Kong F	2012	Whole hospital	MRSA	N/A	Nested tri-level hierarchical log regression models	Quantify infection risk	Retrospective
T and S	Kroker P	2001	Whole hospital	Clostridium difficile	N/A	Knox regression analysis	Detect clusters	Retrospective
T and S	Rushton SP	2010	Intensive care unit	Multiple	N/A	Monte Carlo	Investigate spread of infection	Retrospective
T and S	Starr JM	2009	Whole hospital	Clostridium difficile	N/A	Monte Carlo Markov chain	Assess impact of IC measures	Retrospective
T and S	Yu ITS	2005	Whole hospital	SARS	N/A	Cox regression analysis	Outbreak investigation	Retrospective
Statistical spatiotemporal analysis with molecular information
T	de Celles MD	2012	Surgical ward	MDRAB	repPCR	Stochastic transmission model	Assess impact of molecular typing	Retrospective
T	Gandhi NR	2013	Whole hospital	XDRTB	RFLP	Network analysis	Investigate transmission	Retrospective
T	Nübel U	2013	Paediatric ward	MRSA	SNP analysis	Bayesian skylines	Identify risk factors	Retrospective
T and S	Kumar VS	2006	Whole hospital	MDRGN	Not stated	Monte Carlo/SatScan	Detect clusters	Retrospective
T and S	Rexach, CE	2005	Whole hospital	Clostridium difficile	AP-PCR	Knox regression analysis	Investigate transmission	Retrospective
Spatiotemporal analysis using GIS
S	Kistemann, T	2000	Whole hospital	Salmonella enteritidis	N/A	Geographical distribution	Outbreak investigation	Retrospective
S	Kruger, H	2002	Multiple hospitals	Multiple	N/A	Mapping of resistant isolates	Mapping distribution	Prospective
T and S	Kho, A	2006	Whole hospital	MRSA/VRE	N/A	Visualization of HCW movement	Identify risk factors	Prospective
T and S	Kwan, MYW	2009	Whole hospital	Multiple	N/A	IC	Investigate spread of infection	Prospective

T, temporal; S, spatial; N/A, not applicable; MRSA, meticillin-resistant Staphylococcus aureus; BSI, bloodstream infection; SARS, severe acute respiratory syndrome; MDRAB, multi-drug-resistant Acinetobacter baumannii; XDRTB, extensively-drug-resistant tuberculosis; SNP, single nucleotide polymorphism; MDRGN, multi-drug-resistant Gram-negative bacteria; VRE, vancomycin-resistant enterococci; repPCR, repetitive element palindromic polymerase chain reaction; AP-PCR, arbitrarily primed polymerase chain reaction; RFLP, restriction fragment length polymorphism; AB, antibiotic; HCW, healthcare worker; IC, infection control.

The temporal studies (N = 8) tended to be retrospective and employed time-series analysis (e.g. weekly aggregated measures plotted over time) to demonstrate if antibiotic prescription had an effect on the incidence of MRSA135, 136, 137 and C. difficile, or if control measures for multiple organisms reduced the incidence. Without the incorporation of temporal analysis into these investigations, the impact of the interventions may have been masked by other factors, such as seasonality. Additionally, temporal analysis was used to examine ways to improve infection control measures,140, 141 while Haley and Bregman used multi-variate statistical models to assess the temporal associations between infections and overcrowding, providing evidence that handwashing compliance is reduced markedly under these conditions.

Spatiotemporal studies (N = 5) typically aimed to model infections retrospectively to investigate outbreaks and to detect clustering.8, 143 By using modelling techniques, others were able to estimate the potential effect of interventions, beyond which descriptive studies could use the results to advocate their incorporation into standard control measures.

Statistical spatiotemporal analysis with molecular data

Only 3% (N = 5) of studies combined the use of molecular typing with spatial or temporal analyses (Table III), which has the benefit of molecular differentiation and statistical evidence in confirming transmission. All of these studies were undertaken in 2005 or later, and analysed the retrospective distribution of bacteria while attempting to establish links between isolates.

The earliest study in this category compared the effectiveness of molecular typing with spatiotemporal analysis. Polymerase chain reaction was used to characterize toxin genes in C. difficile isolates, which were mapped to a grid representing each ward, and analysed statistically for clustering by Knox test. This identified a single ward cluster compared with four clusters detected by molecular fingerprinting analysis, leading the investigators to conclude that the Knox test was less effective for identifying nosocomial transmission than molecular fingerprinting. However, most studies have shown that in order to gain the most from available data, spatiotemporal and molecular analyses should be used in combination.

The remaining studies evaluated potential transmission routes or attempted to gain a better understanding of outbreaks. Nuebel et al. applied whole-genome sequencing of MRSA in a neonatal intensive care unit to compare accumulated sequence variation in the isolates, and used Bayesian skyline analysis to reveal epidemiological links between patients, healthcare workers and parents. They concluded that integration of epidemiological mapping and genomic data was necessary to understand MRSA transmission. Similarly, Gandhi et al. performed a retrospective study to investigate epidemiological links between extensively-drug-resistant tuberculosis patients in South Africa by combining RFLP analysis and social network data to build transmission networks among genotypically similar patients. Their findings showed that the epidemic was highly clonal, and network analysis indicated transmission across a network with high levels of interconnectedness. de Celles et al. tried to estimate the variability in transmission between different multi-drug-resistant Acinetobacter baumannii clonal groups using data on carriage on a surgical ward. They identified three clonal complexes by performing molecular fingerprinting, and applied stochastic transmission models to estimate transmission rates for each complex. Results suggested that one of the clones had enhanced transmissibility compared with the other two clones, and further explained local epidemic dynamics. Finally, Kumar et al. optimized cluster identification by organizing multi-drug-resistant Gram-negative bacteria isolates from admitted patients into co-resistance groups, and using schematics of the ward layouts in a Monte Carlo simulation. They concluded that this was ‘a powerful way to quickly identify outbreaks’, and early detection is critical with the decreasing number of effective treatment regimens available.

Spatiotemporal analysis using GIS

GIS was used in only 3% (N = 4) of studies identified in this review, demonstrating its limited uptake in the investigation of HCAIs (Table III); all of these studies were undertaken in 2000 or later. The studies were conducted hospital-wide, and all but one described a prospective application.

Kistemann et al. employed GIS for a retrospective investigation of a salmonella outbreak, the source of which could not be identified by biological testing as food samples had been discarded. By mapping the distribution of cases across the hospital site and using analytical tools in GIS, the researchers identified that the sole link between cases was food delivery from a central kitchen. This led to an investigation of food production and the source was discovered. Kruger and Steffen used GIS to undertake geostatistical analysis of local antibiotic resistance to act as an early warning system for the emergence of drug-resistant strains, enabling doctors to alter their prescription practices. Kho et al. developed and implemented GIS software that enabled them to demonstrate inappropriate patient placement and insufficient hand hygiene in 14% of healthcare provider–patient contacts. Kwan et al. incorporated GIS successfully in a wide range of hospital-based investigations. Using GIS as the central repository for spatial and temporal data of infectious disease cases, the collected data were queried and analysed to identify disease clusters. The results were then communicated to the appropriate personnel, helping decision makers to target control efforts.

Discussion

This review highlights numerous (N = 146) studies focusing on spatiotemporal investigations of infectious diseases within healthcare settings; however, very few of these (N = 22) used appropriate statistical methods to confirm transmission or clustering. This suggests that spatiotemporal data are regularly collected in healthcare settings to examine the potential for clustering, but confirmation using statistical analysis is infrequent, which introduces the risk of misinterpretation and hence development of less effective interventions and management of the problem. Of note, most of the published descriptive analyses were also retrospective, and in the absence of further statistical testing, provide little information for future prevention or prediction activities. Similarly, while half of all identified studies included molecular techniques for differentiating clusters, many of these were older techniques used to determine very large differences in bacteria, and are not conclusive. Only 7% of studies that included molecular data also used statistical analyses to provide more quantitative evidence of transmission clusters.

Underuse of data

This review found that while the collection of spatiotemporal data has been integral to HCAI prevention activities for decades, the use of spatiotemporal statistical analysis is relatively new to the study of HCAIs in comparison with infectious diseases occurring within the community. Most of the studies identified in this review used spatial and temporal information to provide a qualitative description of disease occurrence by time/space, and in contrast to those that employed more sophisticated analyses, were limited in the scope of their findings. Naturally, the ways in which spatial or temporal data are used within investigations of HCAIs varied greatly, and presentation depended upon the aim of each study. However, the large amount of data collected was often not used to its full potential, and opportunities to gain a more thorough understanding of the problem were missed.

The ability of descriptive studies to identify any significant influences of infectious disease dynamics is limited. Several studies discussed the significance of the geographical distribution of cases without undertaking any analyses,48, 64, 68 which is a serious issue as sharing the same geographical space does not prove that transmission has occurred.19, 133 The smaller the scale at which populations are studied, the greater the possibility that ‘clustering’ of cases could have occurred due to a confounding factor (e.g. cohorting of high-risk patients). This highlights a missed opportunity to learn more about the spread of organisms within healthcare settings, and to develop more effective intervention strategies based upon transmission dynamics within that particular setting.

Maximizing data usage

It is extremely important that hospitals are able to understand the local HCAI epidemiology to inform their routine practice, rather than generalizing evidence from other settings. A major stumbling block can be the perception that these analyses may involve active data collection; however, an abundance of existing datasets could be used. Examples of disparity in data usage are the studies by Kroker et al. and Mody et al., in which they attempted to identify potential clusters of C. difficile within their hospitals.8, 62 Both studies were published in 2001 and used clinical patient notes and laboratory results for C. difficile toxin assays. Mody et al. defined a cluster arbitrarily as one or more cases occurring within 21 days of a previous case on the same nursing unit, and used this to identify temporal clusters within their dataset; however, they were unable to suggest potential factors related to the observed pattern. Kroker et al. employed the Knox test to identify time–space clusters, which highlighted hospital geography and traffic between wards as significant factors, and enabled them to adapt their infection control procedures.

Incorporation of molecular data into investigations can have a profound impact on the effectiveness of any outbreak response. Recent advances in molecular biology, such as rapid benchtop sequencing, have led to a revolution in the detail that can be gained from clinical samples. While many hospitals only perform basic identification of micro-organisms due to the resources available, this data, when available, can be useful to enhance current investigations. A study by Adams et al. in 1981 investigated nosocomial infections on a paediatric intensive care unit, and was able to distinguish that there had been two separate outbreaks involving separate strains of herpes simplex virus type 1. The initial investigation, which had not included molecular data, concluded that all cases belonged to a single outbreak, and thus limited the impact of their initial control measures.

Benefits of incorporating statistical testing

Molecular data can be invaluable in ruling out a link between cases; however, as emphasized in the study by Helweg-Larsen et al., clusters of infections can occur for many reasons and are often caused by factors other than nosocomial spread. Therefore, incorporation of true temporal or spatial analysis, to eliminate similarity of micro-organisms by chance, alongside molecular techniques could lead to a better understanding of the true transmission dynamics, as inappropriate assumptions are often made about clustering when based solely on molecular data.

Prior to 2000, only two studies were identified that had performed spatial or temporal analysis; in the last decade, this number has increased to 11. This may be due, in part, to the development of novel statistical methods and further advancement of user-friendly statistical and GIS software; however, the majority of the statistical methods in these studies have been widely employed in other fields since before the 1980s. The increased use of electronic databases in hospitals for storage of medical information has created a rich source of epidemiologically and clinically relevant information, allowing more detailed analyses to be performed.

The major aim of this review was to identify how spatiotemporal analyses have been used previously, and to suggest how they can be employed to benefit practices within healthcare settings. The Knox test is a simple analysis that can be used to identify clustering, and methods using outbreak thresholds are common within infection control reporting. However, the ideal situation is to design control programmes based upon the dynamics and processes observed within the local institution.

As randomized controlled trials can be difficult or costly to undertake in clinical settings, some authors have employed predictive modelling techniques to build their own evidence base. For example, Rushton et al. used a statistical approach to investigate clustering and patterns of spread of a number of organisms within an intensive treatment unit. They obtained data from pre-existing datasets including numbers of infected patients, admission details, duration of stay and bed movement, whilst they estimated some additional variables from evaluating nurse–patient contacts. They identified variation in the degree of clustering of different organisms, and tested the impact of potential control interventions in the model. The findings suggest that bed movement and staff–patient contacts have to be controlled, and control strategies may need to be organism-specific.

Recommendations

Spatiotemporal analysis can distil a much greater amount of relevant information from data collected on HCAIs than purely descriptive studies. Analysis is key to furthering understanding of the epidemiology and dynamics of transmission of these organisms. The underuse of spatial and temporal data may be due to the primary focus of studies on retrospective actions in response to an outbreak, and this ‘fire-fighting’ approach may be propagated by institutional goals. However, new sophisticated techniques allow for greater adaptability to the current challenges in health care, including increased cohorting of at-risk patients, spread of resistance and the corresponding decrease in the number of available effective antimicrobials. A focused approach on development of understanding of HCAI epidemiology is likely to lead to identification of significant risk factors and better prevention.

Molecular typing technology is quickly moving from research to clinical settings, and it is becoming more common for detailed molecular analyses to be undertaken to investigate nosocomial infections. From this review, it is clear that the uptake of statistical analyses is the limiting step in moving towards modern sophisticated analyses of HCAIs. Few healthcare workers have the training to develop statistical models or perform in-depth analyses, and this is where collaboration with academic institutions can be exploited to improve the understanding of local disease dynamics without massively increasing the costs to hospitals. These collaborations would provide rich datasets for researchers to use, while enabling clinicians to employ cutting-edge methodologies that will inform their routine practice.

One possible intermediary step would be to further the use of GIS for HCAI investigations, as it enables a wide range of analyses to be undertaken within one piece of software in which staff could be trained. Whilst the initial implementation may be time and cost intensive, the benefits are clear from the few studies identified in this review. The combination of HCAI datasets from numerous sources in one system and subsequent visualization can enable healthcare workers to incorporate up-to-date infection data into their clinical practice. As hospitals move to combine databases and increase the level of electronic recording, this presents an ideal time to incorporate GIS into these systems and create a fully-integrated hospital information system.

With movement of patients between care structures, differentiation between infection control in primary and secondary care is becoming more difficult. The small number (N = 7, 5%) of studies based within nursing homes in this review suggests that less attention has been given to care units outside of hospitals. However, these could act as reservoirs of infections, and regular re-admission of their residents could lead to further hospital transmission. In addition to providing an analytical toolkit for spatial clustering, GIS could improve the understanding of this relationship by enabling healthcare providers to consider the impact of their local community.

The future potential applications in healthcare settings are ever expanding as more sophisticated molecular, statistical and computational techniques become increasingly commonplace. Publication of analytical studies on HCAI in major clinical journals rather than specialized niche journals, as observed in this review, could increase awareness of these techniques and widen their use.

Limitations

In structuring the search strategy for the review, the authors endeavoured to ensure the inclusion of studies from as broad a range as possible. However, due to the great variation in the terminology used across the numerous clinical and scientific fields, it is possible that a few studies were missed. Further, the heterogeneity within the evidence base precluded meta-analysis of the findings. Finally, publication bias cannot be fully avoided in a review.

Conclusion

To truly understand and stem the growing problem of HCAIs worldwide, a multi-disciplinary approach is required. This is dependent on the skills and technology available to those investigating the problem, and is likely to require collaboration between experts. This review suggests that greater integration of spatiotemporal techniques into HCAI investigations could prove invaluable in highlighting previously unobserved patterns of infections, and maximizing the understanding of disease dynamics. Given that infections within a small contained area, such as a hospital, have greater potential for misclassification of clustering, it is necessary to use both molecular techniques and the appropriate spatiotemporal statistical techniques to maximize the accuracy of the study findings. Given the expanding technology of information systems (e.g. electronic medical databases), advancement in molecular and statistical techniques, development of analytical platforms that enable greater access to non-experts and increased interdisciplinary collaboration, the potential for using pre-existing data to prevent future avoidable infections and improve patient safety can become a reality.

Conflict of interest statement

None declared.

Funding sources

Funding for this study was provided by the UK Clinical Research Council (UK-CRC G0800777) and the Imperial College Healthcare Trust NIHR Biomedical Research Centre. Grahame Davis is supported by funding from the UK Clinical Research Council (UK-CRC G0800777). Dr. Nick Sevdalis is affiliated with the Imperial Centre for Patient Safety and Service Quality (www.cpssq.org), which is funded by the National Institute for Health Research. Dr. Lydia Drumright is supported by a National Institute of Health Research Career Development Award (NIHR CDF-2011-04-017).