Q J Med 2001; 94: 223-225
© 2001 Association of Physicians
Clostridium difficile infection, hospital geography and time-space clustering
From the Departments of Medicine and 1 Microbiology, Chelsea and Westminster Hospital, London, UK
 |
Summary |
|---|
 Top Summary IntroductionMethodsResultsDiscussionReferences |
|---|
To analyse spatial and temporal relationships of Clostridium difficile-associated disease in an inner-city hospital, we retrospectively evaluated 283 episodes of confirmed C. difficile diarrhoea in the Chelsea and Westminster Hospital between 1995 and 1998, against a background of relatively stable case mix, antibiotic usage and admission numbers, using Knox analysis to determine the presence of disease clustering in time and space. We found five time-space clusters on four medical wards and between two adjacent units. The clusters were not related to the overall case number on single wards, and were separated in time. Knox time-space analysis provides a simple screening tool to identify disease clusters, assess the efficacy of infection control measures and the influence of hospital geography and traffic. The results support the importance of infection control measures in the prevention of C. difficile-related disease.
|
Introduction |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
Clostridium difficile remains the commonest cause of nosocomial diarrhoea in British hospitals, and broad-spectrum antibiotics are recognized as the main causative factor.1,2 We reviewed all adult cases of toxin-positive diarrhoea in our hospital between 1995 and 1998, with the aim of identifying the main anti-microbial agents involved. The Chelsea and Westminster Hospital practices an integrated admission and care system, i.e. patients were admitted to all medical wards according to bed availability. Since no selection criteria apply (e.g. age-related segregation), a relatively homogenous patient population results, apart from ward CH with a high proportion of elective patients. The number of cases of hospital-acquired diarrhoea, the number of admissions per ward, the case-mix and the antibiotic usage showed only minor variations over these 4 years. Statistical analysis of the data confirmed no significant correlation between any of 27 antibiotics used and the case incidence of hospital-acquired diarrhoea with positive toxin assays.3 In the absence of a significant correlation with antibiotic usage or individual antibiotic prescribing, we investigated our sample for space-time clustering, to evaluate patient-to-patient cross-infection.
|
Methods |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
Data collection
From 1 January 1995 to 31 December 1998, all patients with positive C. difficile toxin assays were identified from the records of our Microbiology Department. We confirmed the information using the clinical case notes, and 283 adult patients, with a history of hospital-acquired diarrhoea, were included.
Statistical tools (time-space cluster analysis)
The technique described by Knox4–6 was used to examine our data set for case clustering, both in time and space. There are eight medical wards on a single hospital floor, leading to 36 ward combinations (e.g. AD-AD, AD-FB). The schematic floor plan is shown in Figure 1
. For the 283 cases, there are 39903 pair permutations. The analysis was conducted in two steps. Firstly, the pairs were grouped according to time interval between cases. Based on the described incubation time of C. difficile of up to 7 days,1 two target time distances (14 and 21 days) were used for the contingency table. Secondly, the sub-sample was investigated according to the spatial distance between the wards, and the Poisson distribution was used to calculate the expected values (p<0.05).
|
|
Results |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
The first part of Knox's cluster analysis revealed six significant pair combinations (Table 1
). The results were virtually identical for 2 and 3 weeks time distances, and only the values for the former are shown. Four of the six pairings were found within single wards, one further significant pairing resulted between two adjacent wards (FB and MC), and one pairing between two distant wards. More detailed enquiries discovered that the distant pairing was related to the transfer of a single patient from a disease cluster on FB to CH (there were only two cases of C. difficile-associated diarrhoea on ward CH during the 4 years).
|
The clusters did not correlate with the incidence on single wards; indeed, the two wards with the highest case numbers between 1995 and 1998 (AD 54 cases and FB 66 cases) showed no significant clustering.
|
Discussion |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
Knox's cluster analysis is a robust mathematical method, which produces statistical information about relationships in patient populations using a minimal clinical data set.4–6 By examining the probability of the observed space-time clusters, the Knox's cluster analysis can produce four different patterns. First, it may reveal an absence of significant clustering, thus making a causal relationship between the encountered cases statistically unlikely. Secondly, it may demonstrate significant clustering in time but not in space. This constellation occurs with a community-based epidemic. The patients admitted to hospital are allocated to vacant beds, and if the hospital has an unselective admission policy, no spatial clustering between wards will occur. Thirdly, significant spatial but no temporal clusters may be found. The sporadic transmission of hepatitis B due to infected health workers in a surgical or dialysis unit would produce such a pattern since the long incubation time of the infection and the sporadic spread would obscure the temporal relationship. Fourthly, the analysis may demonstrate clustering in both time and space. A localized in-hospital epidemic would create such a pattern, e.g. an outbreak of infectious diarrhoea due to food contamination by hospital staff.
In our study, we observed significant clustering in both time and space. We used a population model of C. difficile infection suggested by Starr et al.7 to interpret our findings. This model suggests that there are four categories of in-patients in relation to C. difficile infection: Group I, resistant to disease and not colonized; Group II, resistant, but colonized with the organism; Group III, susceptible to disease and not colonized; and Group IV, susceptible and colonized).
As our patient population and the case-mix remained relatively stable during the study period, the epidemic clusters observed can be explained as the result of a shift between Groups III and IV. This interpretation explains the lack of correlation with antibiotic use in the study period.3 Other units have noted similar trends. In an outbreak at the Hammersmith Hospital in 1995, which coincided with an increased use of cefotaxime, only 42% of the affected patients were treated with this particular drug and there was no major increase in other antibiotic use.7 Indeed the number of cases of C. difficile disease may depend on the overall amount of broad-spectrum antibiotics prescribed rather than the specific agents used. We believe that the use of antibiotics identified a susceptible population of patients, and that infectious spread between individuals at risk produced the clustering observed.
The clustering between two neighbouring wards illustrates the influence of hospital geography and the resulting flow paths of hospital traffic on cross-infection. As shown in the floor plan (Figure 1
), the medical wards are concentrated on a single hospital floor with interconnecting galleries, and most neighbouring wards are connected via internal doors; the layout of each ward is identical. The main in-hospital corridor is located along wards WG to CH. Many hospital staff use the interconnecting doors between wards MC to DE as a short-cut. Furthermore, a small survey confirmed that staff from ward FB in particular, use the access via ward MC to reach the amenities on the lower floors of the building. This traffic may have resulted in the spread of C. difficile between the two wards. It is surprising that no clusters occurred within the wards FB and MC, indicating that ward-internal infectious control measures were adequate. As a consequence of this cluster analysis, we have recommended the closure of this door to all non-emergency hospital traffic.
Our findings provide further support for stringent infection control measures,1,2,8 and illustrate that hospital geography and traffic has to be taken into account. The Knox cluster analysis is a technically simple tool to identify ‘hot spots’ and to assess and target intervention. The minimal case details (case confirmation, date of positive C. difficile toxin assay and ward location) and easy calculations make this a powerful and underused tool to identify disease clusters and act as general screening tool for more detailed microbiological studies.
|
Notes |
|---|
Address correspondence to Dr P. Kroker, Medical Day Unit, Chelsea and Westminster Hospital, Fulham Road, London SW10 9NH. e-mail: pbkroker{at}aol.com
|
References |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionReferences |
|---|
1. Spencer RC. The role of antimicrobial agents in the etiology of Clostridium difficile-associated disease. J Antimicrob Chemother1998; 41(Suppl. C):21–7.
2.
Worsley MA. Infection control and prevention of Clostridium difficile infection. J Antimicrob Chemother1998; 41(Suppl. C):59–66.
3. Audit Report: Clostridium difficile-associated diarrhoea between 1995 and 1998. London, Chelsea and Westminster Hospital, in press.
4. Knox EG. The detection of space-time interactions. Applied Statistics1964; 13:25–9.
5. Knox G. Epidemiology of childhood Leukaemia in Northumberland and Durham. Br J Prev Soc Med1964; 18:17–24.[Web of Science][Medline]
6.
Mantel N. The detection of disease clustering and a generalised regression approach. Cancer Res1967; 27:209–19.
7. Starr JM, Roger TR, Impallomeni M. Hospital-acquired Clostridium difficile diarrhoea and herd immunity. Lancet1997; 349:426–8.[Web of Science][Medline]
8. McFarland LV, Mulligan ME, Kwok RYY, Stamm WE. Nosocomial infection acquisition of Clostridium difficile infection. N Engl J Med1989; 320:204–10.[Abstract]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  I. Thompson Clostridium difficile-associated disease: update and focus on non-antibiotic strategies Age Ageing, January 1, 2008; 37(1): 14 - 18. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||