If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Corresponding author. Address: Department of Clinical Microbiology, University College London Hospital NHS Foundation Trust, 65 Whitfield Street, London W1T 4EU, UK. Tel.: +44 7403 568687; fax: +44 2034 778992.
Department of Clinical Microbiology, University College London Hospital NHS Foundation Trust, London, UKDepartment of Public Health and Primary Care, Institute of Public Health, The University of Cambridge, Cambridge, UK
Clostridium difficile infection (CDI) is one of the most important healthcare-associated infections, causing considerable mortality. Numerous severity scores have been proposed to identify patients with CDI at risk of mortality, but a systematic review of the evidence upon which these are based has never been published. Such a review could permit future development of scores that better predict mortality.
Aim
A systematic review of the published literature investigating clinically useful risk markers for mortality in CDI.
Methods
We searched MEDLINE 1950 to present, Web of Science with conference proceedings 1899 to present and BIOSIS Citation Index 1969 to present using PubMed and Web of Knowledge. Potential risk markers that had been evaluated by at least four studies were extracted.
Findings
Twenty-six studies, of 1617 initially identified, met inclusion criteria. The majority were retrospective cohort studies, mostly based in the USA. Older age, higher white blood cell count (WBC), higher creatinine level, lower albumin levels and, to a lesser extent, corticosteroid use were most frequently associated with mortality. Presence of fever, haemoglobin/haematocrit level, diarrhoea severity, presence of renal disease, diabetes, cancer, or nasogastric tube use did not appear to be associated with mortality.
Conclusion
Our results support the use of age, WBC, serum creatinine, serum albumin level and possibly pre-existing corticosteroid use as potentially useful risk markers for mortality in CDI. Our results do not support the use of fever, haemoglobin/haematocrit, diarrhoea severity and several comorbidities as useful risk markers, raising questions about their inclusion in CDI severity scores.
Clostridium difficile infection (CDI) is the most common form of hospital-acquired infectious diarrhoea, and represents one of the most important causes of healthcare-associated infection.
Substantial rises in the reported incidence of CDI were seen in the early parts of this century across the USA, Canada, and Europe, and there is evidence suggesting that the incidence continues to rise among elderly patients in the USA.
Rates of Clostridium difficile infection among hospitalised patients aged ≥65 years, by age group – National Hospital Discharge Survey, United States, 1996–2009.
A recent review of mortality due to CDI found pooled attributable mortality of 8.03% in studies reported since the year 2000, compared with 3.64% in those reported before 2000.
Vancomycin has been shown to be superior to metronidazole in patients with severe CDI, and new therapies are on the horizon, which may offer advantages for patients with severe disease.
As such, the ability to accurately and promptly identify patients who are at risk of dying may permit intervention to reduce the substantial level of mortality attributed to CDI. Numerous severity scores for CDI have been created for this purpose; however, to our knowledge, a systematic synthesis of the evidence upon which these scores are based has never been published.
Such a review would add additional objectivity to such scoring systems, and may permit future development of more accurate and evidence-based severity scores to identify patients at risk of mortality. We aimed to systematically review the literature to identify previous studies investigating clinically useful markers of risk for mortality in patients with CDI.
Methods
Search strategy
The PubMed and Web of Knowledge databases were searched, which included MEDLINE 1950 to present, Web of Science with conference proceedings 1899 to present and BIOSIS Citation Index 1969 to present. Results were limited to publications in English, on research in humans, and published from 1978 onwards. Search terms included ‘Clostridium difficile’ [MeSH], ‘pseudomembranous colitis’, ‘pseudomembranous enterocolitis’ [MeSH], ‘mortality’ [MeSH], ‘risk adjustment’ [MeSH], ‘severity of illness index’ [MeSH] and ‘severity’. The date of the final search was 13 September 2011. A separate hand search of the bibliographies of articles identified in the primary search was undertaken to identify additional relevant studies. The search was performed independently by two authors (M.G.B. and J.C.S.). Any disagreements in article selection were resolved through discussion, and a third author (E.G.K.) was available to resolve disagreement.
Inclusion criteria
Studies fulfilling the following criteria were included: (1) those which investigated the relationship between demographic, clinical or laboratory markers and all-cause or disease-specific mortality; (2) risk markers that could be assessed early in the course of CDI, and assessments made before the onset of CDI-related complications; (3) CDI diagnosis was based on stool toxin assay, colonoscopy or sigmoidoscopy, or histopathological diagnosis; (4) hospital-based. All study types were eligible for inclusion. Conference or meeting abstracts were eligible for inclusion if they met the above criteria, there was no corresponding full text article available, and if they contained the following minimum data set: (1) number of participants; (2) number of deaths occurring in the cohort; (3) mean/median age of the cohort; (4) results of statistical tests to support conclusions.
Exclusion criteria
Studies were excluded if they had <100 participants, or if they were limited to one patient group, e.g. patients with inflammatory bowel disease, transplant patients, ICU patients.
Data extraction and interpretation
Data were extracted on to a pre-designed spreadsheet. Recorded study characteristics included year of publication, geographic location, study design, method of CDI diagnosis, risk markers assessed, timing of assessment, and primary outcome variable. Participant characteristics included total number, age, and sex. Study results included number of deaths, risk markers associated with outcome (with P < 0.05), risk markers not associated with outcome (P > 0.05), and magnitude of association. Results from multivariate analyses were favoured over results from univariate analyses. Risk markers from the included studies that could not be assessed early in the course of illness, e.g. ribotype, were not extracted. Risk markers that had been evaluated by at least four separate studies were included in the review. The judgement as to whether a parameter may be a useful risk marker for mortality was based on the number of studies demonstrating an association versus the number not—the size—of those studies, and the magnitude of association.
Results
Search results
Results of the database search are given in Figure 1. The initial search returned 1617 papers. Screening of titles and abstracts resulted in 1527 exclusions, leaving 90 articles to source in full text. Review of full text articles resulted in a further 65 exclusions. One additional relevant article was found in the hand search of bibliographies, leaving 26 papers in total for the review, including one conference abstract.
Characteristics of the studies identified by the literature search are given in Table I. Only two studies were performed outside of the USA, UK or Canada, with the USA accounting for the majority. Twenty-two of 26 studies based their analyses on retrospective data, mostly relying on medical and laboratory records, and administrative data. Most studies used enzyme immunoassays (EIAs) for CDI diagnosis. Cohort size varied from 108 to 2571 participants, with 17 of 26 studies having <250 participants. The mean/median age of cohorts was 46.8–82.1 years, with more than half having a mean/median age of ≥65 years. There was a female predominance in most cohorts. Variations on all-cause and CDI-specific mortality were used as outcome measures, and several studies used composite outcomes. All-cause mortality was the most commonly used mortality metric, with only six studies using disease-specific mortality. The mortality experience of the cohorts varied from 3.5% to 38.0%. In all except one of the studies using composite outcomes, deaths accounted for the majority of these endpoints.
Table ICharacteristics of studies evaluating potential risk markers for mortality in CDI
Outcome included any of: all-cause mortality prior to completion of CDI treatment, ICU admission, perforation, megacolon, severe hypokalaemia or lower GI bleeding requiring transfusion.
Health care-associated Clostridium difficile infection in adults admitted to acute care hospitals in Canada: a Canadian Nosocomial Infection Surveillance Program study.
Univariate and multivariate analysis of risk factors for severe Clostridium difficile-associated diarrhoea: importance of co-morbidity and serum C-reactive protein.
CDI, Clostridium difficile infection; N, sample size; EIA, enzyme immunoassay; CA, cytotoxin assay; CC, cytotoxigenic culture; PCR, polymerase chain reaction; TA, toxin assay, but actual method not stated; CS, colonoscopy/sigmoidoscopy; C, culture; PA, pathological diagnosis; LA, latex agglutination.
Where a time period for mortality is not specified in the table it reflects this not being specified as a criterion for mortality in the corresponding study. Unless otherwise stated, location of mortality was not limited to in-hospital deaths.
a Included relapse episodes in their analysis, so was based on 365 episodes of CDI from 305 patients.
b Outcome included any of: all-cause mortality prior to completion of CDI treatment, ICU admission, perforation, megacolon, severe hypokalaemia or lower GI bleeding requiring transfusion.
c Inclusion in cohort required positive EIA plus confirmatory culture.
Results of the studies identified in the review are given in Table II, Table III. Six potential clinical risk markers had been evaluated by at least four studies. Pepin et al. reported results from two separate cohorts, one collected from 1991 to 2002 and one from 2003 to 2006; these were regarded as two distinct sets of results.
The majority of potential clinical risk markers were assessed close to the time of CDI diagnosis. For many potential risk markers, various cut-off values were used to define risk groups in the different cohorts. Most studies presented odds ratios (ORs) and 95% confidence intervals derived from logistic regression models, but some only reported P-values, particularly for parameters not associated with outcome. Most parameters that were associated with outcome were assessed in multivariate models, whereas a larger proportion of those not associated with outcome were excluded from analysis at the univariate stage.
Table IIPotential clinical risk markers for mortality in CDI, evaluated by four or more studies
∗Studies finding a significant association between risk marker and outcome.
Pepin et al. (1) refers to the 1991–2002 cohort, and Pepin et al. (2) refers to the 2003–2006 cohort. The ORs reported reflect the odds of the outcome occurring in patients in which the risk marker is present. Where ‘per unit/% increase’ is stated in the cut-off column, the corresponding OR reflects the increase in odds of outcome per unit/% change in the risk marker. Where ‘Mean’ or ‘Median’ is stated in the cut-off column, the corresponding study tested the difference in mean/median value of the risk marker between those experiencing the outcome and those not experiencing the outcome.
ORs and 95% CIs have been rounded to one decimal place and P-values to two significant figures. Due to this, some of the 95% CIs reported as significant appear to include the null value of 1.0, whereas in their original form they did not.
a OR/P-value generated from univariate, rather than multivariate, analysis.
b Baseline defined as lowest creatinine level in first 24 h of admission.
c Baseline creatinine level definition not reported.
d Definition of 5% increase in haematocrit not reported.
Health care-associated Clostridium difficile infection in adults admitted to acute care hospitals in Canada: a Canadian Nosocomial Infection Surveillance Program study.
Defined as presence of human immunodeficiency virus (HIV), leukaemia, lymphoma, organ transplant, neutropenia, immunosuppressive drug use or systemic corticosteroids for >1 month.
Defined as presence of human immunodeficiency virus (HIV), leukaemia, lymphoma, organ transplant, neutropenia, immunosuppressive drug use or systemic corticosteroids for >1 month.
Defined as presence of HIV, diabetes, transplant and on immunosuppressants, solid organ or haematological malignancy, haemodialysis, or corticosteroid use ≥10 mg.
Defined as presence of solid organ or haematopoietic stem cell transplant, immunoglobulin deficiencies, immunosuppressive drug use or severe autoimmune syndromes.
Pepin et al. (1) refers to the 1991–2002 cohort, and Pepin et al. (2) refers to the 2003–2006 cohort. The ORs reported reflect the odds of the outcome (mortality or composite score) occurring in patients in which the risk marker is present. Where ‘per year/point increase’ is stated in the cut-off column, the corresponding OR reflects the increase in odds of outcome per year/point change in the risk marker. Where ‘Mean’ or ‘Median’ is stated in the cut-off column, the corresponding study tested the difference in mean/median value of the risk marker between those experiencing the outcome and those not experiencing the outcome.
∗Studies finding a significant association between risk marker and outcome.
ORs and 95% CIs have been rounded to one decimal place and P-values to two significant figures. Due to this, some of the 95% CIs reported as significant appear to include the null value of 1.0, whereas in their original form they did not.
a OR/P-value generated from univariate, rather than multivariate, analysis.
b This is a hazard ratio, with 95% CI.
c The original ORs for Das et al. and Gravel et al. were 1.04 (1.03, 1.05) and 1.02 (1.01, 1.04), respectively.
d This is a relative risk, with 95% CI.
e Defined as presence of human immunodeficiency virus (HIV), leukaemia, lymphoma, organ transplant, neutropenia, immunosuppressive drug use or systemic corticosteroids for >1 month.
f Defined as presence of HIV, diabetes, transplant and on immunosuppressants, solid organ or haematological malignancy, haemodialysis, or corticosteroid use ≥10 mg.
g Defined as presence of solid organ or haematopoietic stem cell transplant, immunoglobulin deficiencies, immunosuppressive drug use or severe autoimmune syndromes.
h Defined as presence of HIV, leukaemia, lymphoma, chemotherapy in last two months, or renal or stem cell transplantation.
i Defined as presence of HIV, on transplant medications or chemotherapy.
In 10 of 17 cohorts white blood cell count (WBC) was associated with outcome. Of the 10, seven used 20 × 109/L as the cut-off level. Studies not demonstrating an association tended to be smaller and use lower cut-offs. ORs in cohorts in which WBC was associated with outcome were often relatively large, the majority being ≥3.0. Serum creatinine level was associated with outcome in six of 13 cohorts. Studies showing an association with outcome tended to use a higher cut-off, most commonly 200 μmol/L, and a rise in creatinine level of >50% from baseline was significant in two studies. ORs were again often relatively large, several being >4.0. Serum albumin level was significantly associated with outcome in seven studies. Cut-off levels of <25–35 g/L were most frequently used. Only one of nine cohorts evaluating temperature identified an association with outcome. Haemoglobin/haematocrit and diarrhoea severity were assessed by six and four studies, respectively, and were not associated with outcome.
Results of studies: demographic/comorbidity factors
Ten different potential demographic or comorbidity risk markers had been evaluated by at least four studies (Table III). In nine of 19 cohorts increasing age was associated with outcome, with three of these studies having >1000 participants. Age >65–75 years was the most common cut-off to be associated with outcome. Those not demonstrating an association were often smaller, and tended to compare mean age between survivors and non-survivors. Hospital-acquisition of disease was defined in a variety of ways, with one of eight cohorts reporting an association with outcome. Three of four studies evaluating the Charlson Index identified a significant association, including the large study of Das et al.
Pre-existing immunocompromise was associated with outcome in two of nine cohorts. Corticosteroid use was assessed by five studies, with three finding an association with outcome, again including the study of Das et al., which was designed specifically to investigate steroid use and CDI outcome. Only one out of five and seven cohorts, respectively, reported nasogastric tube use and pre-existing renal disease to be associated with outcome. Sex, diabetes, and cancer were not significantly associated with outcome.
Discussion
Summary of main results
There has been considerable interest in potential risk markers for mortality in CDI in recent years, reflected by the fact that more than two-thirds of the studies identified in this review have been published since 2009. We have identified a moderately large number of studies on this topic, with several factors found on multiple occasions to be significantly associated with mortality. The following parameters had the most evidence to support their use as markers of risk for mortality in CDI when assessed at or near the time of diagnosis: age, most likely with a cut-off between >65 and 75 years; WBC, with a cut-off of >20 × 109/L; serum creatinine, possibly with a cut-off of >200 μmol/L; and serum albumin, most likely with a cut-off of <25–35 g/L. These parameters all have the potential to be useful components of a risk score, as they can be assessed cheaply, objectively and in a timely manner early in the course of CDI. Of the lesser-studied parameters, corticosteroid use and Charlson Index appeared to show some promise as potential risk markers.
By contrast, temperature, haemoglobin level/haematocrit, sex, severity of diarrhoea, hospital-acquired disease, presence of renal disease, diabetes, cancer and nasogastric tube use did not appear to be associated with outcome in patients with CDI. Several of these parameters, along with others such as abdominal tenderness and radiological findings, have been incorporated into various severity scores for CDI; however, we have found little evidence to support their use, raising questions about the ongoing inclusion of these parameters in established CDI severity scores.
Whereas age, WBC, serum creatinine and serum albumin each had a large number of cohorts demonstrating significant associations with outcome, each parameter had several studies in which no association was found. The negative studies tended to be smaller in size, so may have lacked power to detect an association. The major exception to this was the study of Das et al., in which WBC and serum creatinine were not associated with outcome.
However, Das et al. used the patient's highest WBC and creatinine levels during their entire admission, rather than the level at or around the time of diagnosis, which may have reduced the applicability of the results in this context. Negative studies for WBC, creatinine and albumin also tended to use lower cut-off levels, suggesting that these levels may be less useful for discriminating between high- and low-risk patients. Charlson Index was consistently associated with outcome, but was designed to quantify comorbidity for research purposes rather than in clinical practice, and involves calculating a score based on 19 comorbidities, so would be unlikely to be useful clinically as part of a risk score for patients with CDI.
Negative findings for several of the potential risk markers may have been affected by study methodologies. Type II error relating to small sample size and low power may mean that associations between risk markers and mortality were missed. Severity of diarrhoea is likely to be difficult to assess from retrospective record review, which may in part explain the lack of association seen. The definition of hospital-acquired disease varied considerably between studies, again making it difficult to conclude based on this evidence that it is not associated with mortality. Similarly, whereas several of the negative studies examining immunocompromise may have lacked sufficient power, it is difficult to draw conclusions regarding its usefulness as a risk marker given the heterogeneity in the definitions of immunocompromise used.
Quality of the evidence
The studies identified in this review have several potential limitations. The majority of studies were retrospective, with investigators relying on routinely collected data. As a result, due to limited available data, many analyses were only able to examine a select few potential risk markers. This also affected reporting of comorbidities, which were extracted directly from records, without prospective diagnostic criteria. A minority of studies reported significant associations based on univariate analyses, and therefore will not have accounted for the effect of one risk marker on another in their models. Most studies employed single stage stool testing procedures, often using an EIA, without confirmatory testing. EIAs have been criticized both for poor sensitivity and specificity when used as a sole means of diagnosis, with the latter resulting in false-positive results.
European Society of Clinical Microbiology and Infectious Diseases (ESCMID): data review and recommendations for diagnosing Clostridium difficile-infection (CDI).
Consequently, many of these cohorts are likely to be contaminated with patients without true CDI. This may have been particularly problematic in the retrospective studies that used laboratory records of toxin-positive patients to assemble their cohorts, where inclusion in the cohort did not also require the presence of diarrhoea. Most studies used all-cause mortality as their endpoint, rather than CDI-specific mortality. As a result, the risk markers identified by these studies may have less relevance in aiding treatment decisions for CDI. However, it is understandable that most studies have used all-cause mortality, because death certification of CDI is known to be poor.
Death certificates provide a poor estimation of attributable mortality due to Clostridium difficile when compared to a death review panel using defined criteria.
Although several studies employed composite endpoints in their analyses, deaths accounted for the majority of outcomes in most of these studies. It is therefore likely that their results are still relevant with respect to the outcome of mortality alone. Furthermore, given that 17 of 26 studies had mortality as their sole outcome variable, it seems likely that the potential risk markers identified in this review are applicable to the outcome of mortality. We did not objectively score studies for methodological quality, but our strict inclusion and exclusion criteria will have selected for studies with more robust methodologies. The baseline characteristics we extracted from studies also permitted an assessment of the quality of included studies with respect to study design, inclusion criteria, size and outcome measure. As such, we believe the overall quality of the majority of studies presented in this review to be acceptable.
Potential limitations of the review process
There was marked variability between studies in how risk markers were assessed, which cut-offs were used and how associations were reported. As a result a formal meta-analysis of results was impossible. Pooled statistics for each risk marker would have permitted objective interpretation of the evidence at hand, rather than the more subjective, narrative approach we were required to take, and also would have mitigated problems relating to individual studies lacking power. The inability to meta-analyse also meant that it was impossible to formally test for publication bias with funnel plots or other tests. It seems likely that publication bias has affected the results of this review, given the large number of small studies reporting significant associations, but it is difficult to determine the likely magnitude of this effect. We attempted to reduce the impact of publication bias by applying a comprehensive search strategy, and by excluding studies with <100 participants.
External validity
Observed relationships between risk markers and mortality may have been influenced by local CDI epidemiology, particularly given the problems experienced in the USA, UK and Canada with the virulent NAP1/B1/027 strain of C. difficile at the times many of the studies in this review were conducted.
This may limit the generalizability of these studies to other countries and time periods, as the relationship between risk markers and mortality may differ in epidemic versus endemic CDI. However, it seems likely that these results would be broadly applicable to similar health systems in other developed countries, particularly since the studies in this review were performed in general hospital populations.
Conflict of interest statement
None declared.
Funding sources
None.
References
Rupnik M.
Wilcox M.H.
Gerding D.N.
Clostridium difficile infection: new developments in epidemiology and pathogenesis.
Rates of Clostridium difficile infection among hospitalised patients aged ≥65 years, by age group – National Hospital Discharge Survey, United States, 1996–2009.
Health care-associated Clostridium difficile infection in adults admitted to acute care hospitals in Canada: a Canadian Nosocomial Infection Surveillance Program study.
Univariate and multivariate analysis of risk factors for severe Clostridium difficile-associated diarrhoea: importance of co-morbidity and serum C-reactive protein.
European Society of Clinical Microbiology and Infectious Diseases (ESCMID): data review and recommendations for diagnosing Clostridium difficile-infection (CDI).
Death certificates provide a poor estimation of attributable mortality due to Clostridium difficile when compared to a death review panel using defined criteria.
00164-8&id=gr1.jpg){kind=link}