Background
The effects of hospital-acquired Clostridium difficile infection (CDI) on patient outcomes are incompletely understood. We conducted this study to determine the independent impact of hospital-acquired CDI on in-hospital mortality after adjusting for the time-varying nature of CDI and baseline mortality risk at hospital admission.
Methods
This retrospective observational study used data from the Ottawa Hospital (Ottawa, Ontario, Canada) data warehouse. Inpatient admissions with a start date after July 1, 2002, and a discharge date before March 31, 2009, were included. Stratified analyses and a Cox multivariate proportional hazards regression model were used to determine if hospital-acquired CDI was associated with time to in-hospital death.
Results
A total of 136 877 admissions were included. Hospital-acquired CDI was identified in 1393 admissions (overall risk per admission, 1.02%; 95% confidence interval [CI], 0.97%-1.06%). The risk of hospital-acquired CDI significantly increased as the baseline mortality risk increased: from 0.2% to 2.6% in the lowest to highest deciles of baseline risk. Hospital-acquired CDI significantly increased the absolute risk of in-hospital death across all deciles of baseline risk (pooled absolute increase, 11%; 95% CI, 9%-13%). Cox regression analysis revealed an average 3-fold increase in the hazard of death associated with hospital-acquired CDI (95% CI, 2.4-3.7); this hazard ratio decreased with increasing baseline mortality risk.
Conclusions
Hospital-acquired CDI was independently associated with an increased risk of in-hospital death. Across all baseline risk strata, for every 10 patients acquiring the infection, 1 person died.
Hospital-acquired Clostridium difficile infection (CDI) has become an important problem in many health care systems.1 The prevalence of hospital-acquired CDI has doubled in the past 2 decades.2,3 Hospital-acquired CDI likely has a major impact on patients and the health system; however, these effects are incompletely understood. On an individual patient basis, symptoms attributable to CDI may range from mild abdominal pain and diarrhea to toxic megacolon requiring colectomy to death.4 Studies have reported that 1.5% to 5.7% of hospitalized patients acquiring CDI die3,5-8 and surviving patients have increased length of stay and costs.6,9,10
Previous studies have examined the independent association between hospital-acquired CDI and mortality.6,9-13 Each of these studies has reported increases in mortality with varying degrees of magnitude. However, 2 important questions remain unanswered. First, the impact of adjusting for the time-varying nature of hospital-acquired CDI on the association between CDI and mortality is unknown. Second, the influence of using clinical data, in addition to administrative data, to account for baseline risk factors has not been explored.
We conducted this cohort study to determine the independent association of hospital-acquired CDI on in-hospital mortality after accounting for the time-varying nature of CDI and mortality risk at hospital admission using a validated index.
This was a retrospective observational study using data from the Ottawa Hospital data warehouse. The Ottawa Hospital is a tertiary-care teaching facility in Ottawa, Ontario, Canada, that has 2 general acute care inpatient campuses containing 1100 beds. It is the largest adult referral center for a population of 1.1 million.
The Ottawa Hospital data warehouse contains information from the important operational information systems of The Ottawa Hospital. These systems include the patient registration system, the clinical data repository (containing laboratory, pharmacy, radiology, and clinical notes), and the discharge abstract database (which summarizes demographic, diagnostic, and procedural information for each inpatient admission). This study was approved by the Ottawa Hospital research ethics board.
Study eligibility criteria
We included inpatient admissions at The Ottawa Hospital with a start date after July 1, 2002, and a discharge date before March 31, 2009. The study start date was selected to ensure that we had laboratory data for all admissions; the study end date represented the most complete admissions from the discharge abstract database at the time of analysis.
We excluded inpatient admissions where the length of stay was less than 3 days because hospital-acquired CDI requires a positive test result no earlier than 3 days into the admission.14 We also excluded admissions for patients younger than 15 years and those admitted to an obstetrical service because the baseline risk prediction method for in-hospital mortality that we used in this study was not validated in these populations.
For all included admissions, we abstracted the following information from the data warehouse: age, sex, admission and discharge dates, admitting hospital service, diagnoses coded using International Statistical Classification of Diseases, 10th Revision (ICD-10), and all laboratory test results. For each admission, we measured health care utilization in the year prior to the admission by calculating the number of inpatient admissions and emergency department visits to The Ottawa Hospital and the total number of inpatient days in The Ottawa Hospital.
Determining hospital-acquired cdi status
A hospital-acquired CDI case was defined as laboratory confirmation of a positive C difficile toxin assay from liquid stool obtained at least 72 hours after admission. Positive toxin assays that occurred within 2 months of a previous positive test result were not included. In other words, the 2 positive samples were counted as 1 case with the date of the first positive test result used. These criteria are mandated by the Ontario Provincial Infectious Diseases Advisory Committee.14 If 2 positive test results occurred during a single admission and were more than 2 months apart, only the first positive test result was included in our analysis.
We used electronic laboratory reports of C difficile toxin assays, instead of ICD-10 diagnostic codes, to determine hospital-acquired CDI status. The accuracy of the ICD-10 code for CDI is poor, with sensitivity and specificity of 70% and 99%, respectively.15,16 We applied text searching techniques to the electronic laboratory reports to identify all reports of positive test results for the included admissions. We then compared the date and time of the positive toxin assay with the admission date and time to determine if the positive C difficile toxin assay indicated a hospital-acquired case. We validated the results from our electronic algorithm by comparing with reports of positive results manually classified by infection control staff. Compared with a gold standard classification by our hospital's Department of Infection Control, our algorithm had a sensitivity and specificity of 100%.
The primary outcome was time to in-hospital death. This outcome was determined using the discharge abstract database housed within the data warehouse. Richards et al17 validated the coding of this outcome in the discharge abstract database by reabstracting data from 1936 randomly selected medical charts from 18 hospitals across Canada; no discrepancy was found between the discharge abstract database and the medical chart review.
We summarized characteristics of included admissions by hospital-acquired CDI status. An important characteristic that we examined was baseline mortality risk.
We calculated the baseline mortality risk for each admission using a regression model developed by Escobar et al18 and validated in our hospital's population.19 This model calculates a patient's risk of dying in-hospital using data available at the time of admission. This feature of the model is important because it ensures that the baseline mortality risk is not confounded by complications arising during the admission. The model includes age, sex, comorbidities (measured with the Charlson comorbidity score20 using weights from Schneeweiss et al21), acuity of admission, admitting hospital service, and severity of acute disease (summarized using the Laboratory-based Acute Physiology Score [LAPS]18).
To investigate the association between baseline mortality risk and hospital-acquired CDI, we divided the population into deciles of baseline mortality risk and calculated the proportion of admissions within each decile that were classified as a hospital-acquired CDI case. We used the Cochran-Armitage test22 to determine if there was a significant trend in the risk of infection by baseline mortality risk.
We calculated the unadjusted association between hospital-acquired CDI and in-hospital mortality using an odds ratio. The adjusted association between hospital-acquired CDI and in-hospital mortality was assessed using 2 methods. First, we performed a stratified analysis in which we divided the population into deciles of baseline mortality risk and calculated the absolute and relative risk of mortality associated with hospital-acquired CDI within each stratum. For both absolute and relative risk, we created a forest plot and estimated the overall risk using a fixed-effects model. Second, we used a Cox multivariate proportional hazards regression model23 with patient observation starting at admission and being censored at discharge. This approach was selected to accommodate the time-varying nature of hospital-acquired CDI. Baseline mortality risk, hospital-acquired CDI (expressed as a time-dependent covariate24), and the interaction between these 2 variables were the primary predictors forced into the model. We included the interaction term to evaluate the presence of effect modification. We also included other covariates that were available within the data warehouse for evaluation as possible predictors of mortality. These covariates were number of admissions to our hospital in the previous year, number of emergency department visits to our hospital in the previous year, total number of inpatient days in our hospital in the previous year, and year of admission (to account for trends over time). Because inclusion of these covariates was considered exploratory and not based on specific clinical hypotheses, they were subjected to stepwise backward elimination from the model. We chose an α level of .10 for retention in the model because the consequences of retaining an unimportant variable in our model were considered less severe than those of excluding an important variable, given our large sample size.25 We used fractional polynomial functions to determine the best linear or nonlinear form for continuous covariates.26-28 Covariates were centered about their means to reduce the risk of multicollinearity.29 We accounted for clustering due to multiple hospitalizations of the same patient within the study period using a marginal approach with robust sandwich estimates for the covariance matrix.30 The estimated regression coefficients and covariance parameters from the model were used to generate a plot of the adjusted hazard ratio (HR) and 95% confidence intervals (CIs) associated with hospital-acquired CDI as a function of baseline mortality risk. We used SAS statistical software (version 9.1.3; SAS Inc, Cary, North Carolina) for all analyses.
Description of study cohort
A total of 136 877 admissions were included (Table) from 89 086 unique patients (29.8% with repeated admissions). Approximately half of the admissions were female patients, and their mean age was 63 years. Chronic diseases were common in the cohort with a mean Charlson score of 2.4; only 37% of admissions had a Charlson score of 0. Three-quarters of patients were admitted urgently. Thirty-five percent of patients had 1 or more admission to The Ottawa Hospital in the year prior to their study admission. The median baseline mortality risk was 1.8% (interquartile range [IQR], 0.4%-7.6%).
Risk of hospital-acquired cdi
Hospital-acquired CDI was identified in 1393 admissions for an overall risk of 1.02% per admission (95% CI, 0.97%-1.06%). Only 27 patients (0.03%) had repeated hospital-acquired CDI across their multiple admissions. Compared with uninfected patients, patients with hospital-acquired CDI were older, had more comorbidities, inpatient admissions, and emergency department visits in the previous year, and were more likely to be admitted urgently (Table). In addition, patients with hospital-acquired CDI were more acutely ill at admission (having a higher LAPS) and had a higher median baseline mortality risk (8.0% vs 1.8%).18
The risk of hospital-acquired CDI significantly increased as the baseline mortality risk increased (χ2 value = 15 679.6; P < .001) (Figure 1). Overall, the absolute risk of hospital-acquired CDI was low. Approximately 2.6% of the admissions in the decile with the highest baseline mortality risk acquired CDI, compared with only 0.2% of the admissions in the decile with the lowest risk.
Timing of hospital-acquired cdi
Most cases of hospital-acquired CDI (75%) were identified on or after the seventh day of the admission. Only 2.8% and 5.5% of CDI cases were detected on the third and fourth day of the encounter, respectively. The median number of days from admission to CDI case detection was 12 days (IQR, 7-24 days). For patients with hospital-acquired CDI who died, the median number of days from CDI detection to death was 14 days (IQR, 5.5-31.0 days).
The unadjusted association between hospital-acquired CDI and in-hospital death was OR, 4.62 (95% CI, 4.06-5.25). The stratified analysis showed that hospital-acquired CDI was significantly associated with an increased risk of in-hospital death, regardless of the baseline mortality risk. The absolute risk difference (Figure 2A) ranged from a minimum of 4% (95% CI, −4% to 11%) in the second risk decile to a maximum of 17% (95% CI, 11%-23%) in the eighth risk decile. There was moderate heterogeneity across the deciles (tests for heterogeneity: χ2 value, 15.89; P = .07; I2 = 43%). The pooled absolute risk difference estimate was 11% (95% CI, 9%-13%). The relative risk of death (Figure 2B) was also significantly increased regardless of the baseline mortality risk, although the relative risk decreased from the lowest to the highest baseline mortality risk deciles. Within the lowest-risk decile, the relative risk of death associated with hospital-acquired CDI was 45.70 (95% CI, 11.35-183.98), while in the highest-risk decile it was 1.29 (95% CI, 1.11-1.50). There was considerable heterogeneity across the deciles (χ2 value, 153.31; P < .001; I2 = 94%). The pooled relative risk estimate was 1.99 (95% CI, 1.81-2.19).
The results of the Cox model show that baseline mortality risk (HR, 1.07; 95% CI, 1.06-1.07), hospital-acquired CDI (HR, 2.98; 95% CI, 2.42-3.65), and the interaction between these 2 variables (HR, 0.67; 95% CI, 0.61-0.73) was significant (P < .001). Regression coefficients associated with baseline mortality risk were transformed so that the HRs represent the effect associated with an approximately 10% relative increase in baseline mortality risk. Other covariates retained in the model were year of admission (HR, 0.96; 95% CI, 0.95-0.97), as well as transformations of the number of admissions to our hospital in the previous year (HR, 1.15; 95% CI, 1.05-1.26) and the total number of inpatient days in our hospital in the previous year (HR, 1.03; 95% CI, 1.00-1.05). The only variable that was offered to the model and excluded was the number of emergency department visits to our hospital in the previous year. This variable was excluded because it was collinear with the number of admissions to our hospital in the previous year.
The regression coefficient associated with hospital-acquired CDI suggests that, for an “average” patient (a patient with covariate values set at mean levels), hospital-acquired CDI increased the hazard of dying 3-fold (95%CI, 2.42-3.65). Moreover, the interaction term of baseline mortality risk and hospital-acquired CDI suggests that the HR associated with hospital-acquired CDI decreased with increasing baseline mortality risk; in particular, the HR decreased by approximately 4% with each 10% relative increase in baseline mortality risk. Thus, hospital-acquired CDI had a greater relative impact on death for patients with a lower baseline mortality risk (Figure 3).
Our study has several important findings. Overall, the risk of acquiring CDI during a nonobstetrical hospital admission is 1 in 100. This risk is lower in patients who have a low baseline mortality risk. For patients with a baseline mortality risk at or below the median of 1.8%, the risk of acquiring CDI ranges from 1 in 120 to 1 in 775. In contrast to the low risk of acquiring CDI in-hospital, we found that the clinical impact of CDI is significant. There was a statistically significant and independent association of CDI with the risk of death in-hospital. Using a simple stratified analysis, we found that across deciles of baseline mortality risk, hospital-acquired CDI increased the absolute risk of death by approximately 10%. This means that across all baseline risk strata, for every 10 patients acquiring the infection, 1 person died. Our model-based approach, accounting for hospital-acquired CDI acquisition as a time-varying covariate and adjusting for year of admission and number of inpatient admissions and hospital days in the previous year, found an average 3-fold increase in the hazard of death associated with CDI acquisition.
We found the relative impact of hospital-acquired CDI on in-hospital mortality to decrease with increasing baseline mortality risk. We speculate that this occurs because patients with a high baseline mortality risk already have many factors that predispose them to a poor outcome. Acquiring CDI, in this context, will have a small relative impact on a patient's overall sickness. In contrast, a healthy patient who acquires CDI will experience a larger relative impact.
Taken together, these findings highlight the need for and the challenge of overcoming an important barrier to sustaining efforts to minimize the spread of CDI. Our findings have important implications for health care providers, investigators, and hospital managers responsible for reducing the incidence of CDI. Health care providers need to work diligently to minimize the spread of CDI despite the absence of any immediate clinical feedback to guide their actions. Careful selection of antibiotic usage31 and hand hygiene32 are 2 practices in which physicians are notoriously noncompliant with guidelines. Health care providers must also be aware of the high level of risk that hospital-acquired CDI poses to ill patients as well as to comparatively healthy patients. Given the low risk of hospital-acquired CDI, investigators should assess the impact of interventions by measuring clinical outcomes for sufficiently large patient populations. Finally, hospital managers should focus their interventions specifically toward patients at the highest risk of dying because these patients have the highest risk for hospital-acquired CDI. In addition, managers must work to minimize the role of environmental contamination in the spread of infection.
Our results are, in many ways, consistent with those of previous studies that have examined the association between hospital-acquired CDI and mortality.6,9-13 We believe we have quantified the impact in a valid manner given our analytical methods, the clinical data we had available, and the size of our cohort. An important strength of our approach was its ability to account for the time-varying nature of CDI. To our knowledge, this had not been done by any of the previous studies. Failing to account for the time in hospital before CDI acquisition may lead to an inflated estimate of its impact on mortality.33 Another strength of our study was its ability to explore the risks of hospital-acquired CDI while accounting for baseline mortality risk using clinical data. It is also noteworthy that nearly all of the previous studies had been case-control designs. These factors may have led to some nuanced differences between our study and prior investigations.
There are limitations to our work. First, we may not have identified all CDI cases. Our institution uses an enzyme immunofluorescence assay to detect C difficile toxin on submitted stool samples. The sensitivity of this test is estimated to be 70% to 90%.34,35 We may also have missed cases that were diagnosed following discharge or detected solely by colonoscopy or sigmoidoscopy. It is also possible that community-acquired cases of CDI may have been misclassified as hospital-acquired. Because misclassification was likely to be nondifferential across our entire cohort, we do not expect that it resulted in a biased estimate of hospital-acquired CDI impact, which was the main objective of the study. Second, we did not evaluate the impact of treatments for CDI. It is possible that the sicker patients may have been treated more aggressively for their CDI and, therefore, not had the same sequelae. We feel this to be unlikely, but we will analyze the impact of CDI treatment in future investigations. Third, we did not account for CDI strain in our analysis. Our hospital does perform surveillance for the NAP1 strain, which has been shown to be associated with CDI severity. Our surveillance data show that NAP1 strains account for approximately 20% of isolates, and this proportion seems to be gradually increasing over time. While we are not certain whether it was NAP1 isolates that were disproportionately attributable for deaths, we did include year in our model to account for the evolving pattern of isolates and did not find a significant direct association.
In summary, we have demonstrated that hospital-acquired CDI has a major impact on patients, with an average 3-fold increase in the hazard of death despite its low absolute infection rate. We recommend further research to better understand the modifiable factors predicting who will acquire CDI in-hospital and who will do poorly when such an event occurs. This proposed research will guide institutions and health care providers to select the most appropriate strategy for reducing the infection and its impact.
Correspondence: Alan J. Forster, MD, FRCPC, MSc, Clinical Quality and Performance Management, Ottawa Hospital, Civic Campus, ASB 1-1053 Carling Ave, Ottawa, ON K1Y 4E9, Canada (aforster@ohri.ca).
Accepted for Publication: March 26, 2010.
Author Contributions:Study concept and design: Oake, van Walraven, Wilson, and Forster. Acquisition of data: Oake, van Walraven, and Forster. Analysis and interpretation of data: Oake, Taljaard, van Walraven, Wilson, Roth, and Forster. Drafting of the manuscript: Oake, van Walraven, Wilson, and Forster. Critical revision of the manuscript for important intellectual content: Oake, Taljaard, van Walraven, Wilson, Roth, and Forster. Statistical analysis: Taljaard, van Walraven, and Forster. Obtained funding: Oake, van Walraven, Wilson, and Forster. Administrative, technical, and material support: Roth. Study supervision: van Walraven, Wilson, and Forster.
Financial Disclosure: None reported.
Funding/Support: This study was funded by the Canadian Patient Safety Institute and the Ottawa Hospital. The Ottawa Hospital Data Warehouse was funded by the Canadian Foundation for Innovation, the Ontario Innovation Trust, and the Ministry of Research and Innovation.
Additional Information: Dr Forster is a career scientist with the Ontario Ministry of Health and Longterm Care. Dr Wilson holds a Canada Research Chair in Public Health Policy.
1.Goorhuis
AVan der Kooi
TVaessen
N
et al. Spread and epidemiology of
Clostridium difficile polymerase chain reaction ribotype 027/toxinotype III in the Netherlands.
Clin Infect Dis 2007;45
(6)
695- 703
PubMedGoogle ScholarCrossref 2.Ricciardi
RRothenberger
DAMadoff
RDBaxter
NN Increasing prevalence and severity of
Clostridium difficile colitis in hospitalized patients in the United States.
Arch Surg 2007;142
(7)
624- 631
PubMedGoogle ScholarCrossref 3.Zilberberg
MDShorr
AFKollef
MH Increase in adult
Clostridium difficile -related hospitalizations and case-fatality rate, United States, 2000-2005.
Emerg Infect Dis 2008;14
(6)
929- 931
PubMedGoogle ScholarCrossref 4.Sunenshine
RHMcDonald
LC
Clostridium difficile -associated disease: new challenges from an established pathogen.
Cleve Clin J Med 2006;73
(2)
187- 197
PubMedGoogle ScholarCrossref 5.Dobson
GHickey
CTrinder
J
Clostridium difficile colitis causing toxic megacolon, severe sepsis and multiple organ dysfunction syndrome.
Intensive Care Med 2003;29
(6)
1030
PubMedGoogle Scholar 6.Miller
MAHyland
MOfner-Agostini
MGourdeau
MIshak
MCanadian Hospital Epidemiology Committee; Canadian Nosocomial Infection Surveillance Program, Morbidity, mortality, and healthcare burden of nosocomial
Clostridium difficile -associated diarrhea in Canadian hospitals.
Infect Control Hosp Epidemiol 2002;23
(3)
137- 140
PubMedGoogle ScholarCrossref 8.Miller
MGravel
DMulvey
M
et al. Health care-associated
Clostridium difficile infection in Canada: patient age and infecting strain type are highly predictive of severe outcome and mortality.
Clin Infect Dis 2010;50
(2)
194- 201
PubMedGoogle ScholarCrossref 9.Song
XBartlett
JGSpeck
KANaegeli
ACarroll
KPerl
TM Rising economic impact of
Clostridium difficile -associated disease in adult hospitalized patient population.
Infect Control Hosp Epidemiol 2008;29
(9)
823- 828
PubMedGoogle ScholarCrossref 10.Pépin
JValiquette
LCossette
B Mortality attributable to nosocomial
Clostridium difficile -associated disease during an epidemic caused by a hypervirulent strain in Quebec.
CMAJ 2005;173
(9)
1037- 1042
PubMedGoogle ScholarCrossref 11.Kenneally
CRosini
JMSkrupky
LP
et al. Analysis of 30-day mortality for
Clostridium difficile -associated disease in the ICU setting.
Chest 2007;132
(2)
418- 424
PubMedGoogle ScholarCrossref 12.Dubberke
ERButler
AMReske
KA
et al. Attributable outcomes of endemic
Clostridium difficile -associated disease in nonsurgical patients.
Emerg Infect Dis 2008;14
(7)
1031- 1038
PubMedGoogle ScholarCrossref 13.Loo
VGPoirier
LMiller
MA
et al. A predominantly clonal multi-institutional outbreak of
Clostridium difficile -associated diarrhea with high morbidity and mortality.
N Engl J Med 2005;353
(23)
2442- 2449
PubMedGoogle ScholarCrossref 14.Provincial Infectious Diseases Advisory Committee (PIDAC), Best Practices Document for the Management of Clostridium difficile in All Health Care Settings. Toronto Ontario Ministry of Health and Long-term Care2009;
15.Scheurer
DBHicks
LSCook
EFSchnipper
JL Accuracy of
ICD-9 coding for
Clostridium difficile infections: a retrospective cohort.
Epidemiol Infect 2007;135
(6)
1010- 1013
PubMedGoogle ScholarCrossref 16.Dubberke
ERReske
KAMcDonald
LCFraser
VJ
ICD-9 codes and surveillance for
Clostridium difficile -associated disease.
Emerg Infect Dis 2006;12
(10)
1576- 1579
PubMedGoogle ScholarCrossref 17.Richards
JBrown
AHoman
C The data quality study of the Canadian discharge abstract database. Presented at: The Statistics Canada Symposium: Achieving Data Quality in a Statistical Agency: a Methodological Perspective September 12, 2001 Ottawa, Ontario, Canada
18.Escobar
GJGreene
JDScheirer
PGardner
MNDraper
DKipnis
P Risk-adjusting hospital inpatient mortality using automated inpatient, outpatient, and laboratory databases.
Med Care 2008;46
(3)
232- 239
PubMedGoogle ScholarCrossref 19.van Walraven
CEscobar
GJGreene
JDForster
AJ The Kaiser Permanente inpatient risk adjustment methodology is valid in an external patient population.
J Clin Epidemiol 2010;63
(2)
798- 803
PubMedGoogle ScholarCrossref 20.Quan
HSundararajan
VHalfon
P
et al. Coding algorithms for defining comorbidities in
ICD-9-CM and
ICD-10 administrative data.
Med Care 2005;43
(11)
1130- 1139
PubMedGoogle ScholarCrossref 21.Schneeweiss
SWang
PSAvorn
JGlynn
RJ Improved comorbidity adjustment for predicting mortality in Medicare populations.
Health Serv Res 2003;38
(4)
1103- 1120
PubMedGoogle ScholarCrossref 22.Armitage
PBerry
GMatthews
JNS Statistical Methods in Medical Research. 4th ed. Malden, MA Blackwell Science Ltd2005;
23.Allison
PD Estimating Cox-Regression Models With PROC PHREG: Survival Analysis Using the SAS System. Cary, NC SAS Institute Inc2000;111- 184
24.Fisher
LDLin
DY Time-dependent covariates in the Cox proportional-hazards regression model.
Annu Rev Public Health 1999;20145- 157
PubMedGoogle ScholarCrossref 25.Kennedy
WBancroft
TA Model building for prediction in regression based upon repeated significance tests.
Ann Math Stat 1971;42
(4)
1273- 1284
Google ScholarCrossref 26.Royston
PAltman
DG Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling.
Appl Stat 1993;43
(3)
429- 467
Google ScholarCrossref 27.Sauerbrei
WRoyston
P Building multivariable prognostic and diagnostic models: transformation of the predictors by using fractional polynomials.
J R Stat Soc Ser A 1999;162
(pt 1)
71- 94
Google ScholarCrossref 28.Sauerbrei
WMeier-Hirmer
CBenner
ARoyston
P Multivariable regression model building by using fractional polynomials: description of SAS, STATA and R programs.
Comput Stat Data Anal 2006;50
(12)
3464- 3485
Google ScholarCrossref 29.Armitage
PBerry
GMatthews
JNS Modelling Continuous Data: Statistical Methods in Medical Research. 4th ed. Malden, MA Blackwell Science Ltd2002;312- 377
30.Gharibvand
LLiu
L Analysis of Survival Data With Clustered Events. Cary, NC SAS Global Forum2009;237
31.Kisuule
FWright
SBarreto
JZenilman
J Improving antibiotic utilization among hospitalists: a pilot academic detailing project with a public health approach.
J Hosp Med 2008;3
(1)
64- 70
PubMedGoogle ScholarCrossref 32.Cantrell
DShamriz
OCohen
MJStern
ZBlock
CBrezis
M Hand hygiene compliance by physicians: marked heterogeneity due to local culture?
Am J Infect Control 2009;37
(4)
301- 305
PubMedGoogle ScholarCrossref 33.van Walraven
CDavis
DForster
AJWells
GA Time-dependent bias was common in survival analyses published in leading clinical journals.
J Clin Epidemiol 2004;57
(7)
672- 682
PubMedGoogle ScholarCrossref 34.Merz
CSKramer
CForman
M
et al. Comparison of four commercially available rapid enzyme immunoassays with cytotoxin assay for detection of
Clostridium difficile toxin(s) from stool specimens.
J Clin Microbiol 1994;32
(5)
1142- 1147
PubMedGoogle Scholar 35.Alfa
MJSwan
BVanDekerkhove
BPang
PHarding
GK The diagnosis of
Clostridium difficile -associated diarrhea: comparison of Triage
C. difficile panel, EIA for Tox A/B and cytotoxin assays.
Diagn Microbiol Infect Dis 2002;43
(4)
257- 263
PubMedGoogle ScholarCrossref