infection control and hospital epidemiology

august 2008, vol. 29, no. 8

original article

Case-Mix Adjustment Approach to Benchmarking Prevalence Rates of Nosocomial Infection in Hospitals in Cyprus and Greece
Evangelos I. Kritsotakis, MSc, PhD; Ioannis Dimitriadis, MD; Maria Roumbelaki, RN, PhD; Emelia Vounou, MD; Maria Kontou, MD; Panikos Papakyriakou, MD, PhD; Maria Koliou-Mazeri, MD; Ioannis Varthalitis, MD, PhD; George Vrouchos, MD, PhD; George Troulakis, MD; Achilleas Gikas, MD, PhD

(See the commentary by Harris and McGregor on pages 693 694) objective. To examine the effect of heterogeneous case mix for a benchmarking analysis and interhospital comparison of the prevalence rates of nosocomial infection. design. setting. Cross-sectional survey. Eleven hospitals located in Cyprus and in the region of Crete in Greece.

methods. The survey included all inpatients in the medical, surgical, pediatric, and gynecology-obstetrics wards, as well as those in intensive care units. Centers for Disease Control and Prevention criteria were used to dene nosocomial infection. The information collected for all patients included demographic characteristics, primary admission diagnosis, Karnofsky functional status index, Charlson comorbidity index, McCabe-Jackson severity of illness classication, use of antibiotics, and prior exposures to medical and surgical risk factors. Outcome data were also recorded for all patients. Case mixadjusted rates were calculated by using a multivariate logistic regression model for nosocomial infection risk and an indirect standardization method. results. The overall prevalence rate of nosocomial infection was 7.0% (95% condence interval, 5.9% 8.3%) among 1,832 screened patients. Signicant variation in nosocomial infection rates was observed across hospitals (range, 2.2%9.6%). Logistic regression analysis indicated that the mean predicted risk of nosocomial infection across hospitals ranged from 3.7% to 10.3%, suggesting considerable variation in patient risk. Case mixadjusted rates ranged from 2.6% to 12.4%, and the relative ranking of hospitals was affected by case-mix adjustment in 8 cases (72.8%). Nosocomial infection was signicantly and independently associated with mortality (adjusted odds ratio, 3.6 [95% condence interval, 2.1 6.1]). conclusion. The rst attempt to rank the risk of nosocomial infection in these regions demonstrated the importance of accounting for heterogeneous case mix before attempting interhospital comparisons. Infect Control Hosp Epidemiol 2008; 29:685 692

Improvement in infection control practices and promotion of patient safety have received increasing attention,1 with emphasis on designing and implementing standardized surveillance programs for nosocomial infection (NI) that produce data accurate enough to meaningfully be used for interhospital comparisons.2 Continuous surveillance programs may be the most accurate method for obtaining such data, but the resources and effort required to implement and maintain such programs have rarely allowed multicenter studies to be performed.3 Therefore, prevalence surveyswhich are easier to perform on

a large scale, less expensive, and less time consumingare often conducted to provide baseline information about the occurrence of NI and to help establish priorities for infection control.3 Using NI prevalence rates to compare hospitals that participate in standardized surveys may be the rst step toward a benchmarking analysis, which is a process of learning about, changing, and improving the effectiveness of infection control practices through comparison with the practices and results of other institutions. Because hospitals may differ considerably in

From the Laboratory of Clinical Bacteriology, Parasitology, Zoonoses, and Geographical Medicine, University of Crete (E.I.K., A.G.), the Infection Control Unit, University Hospital of Heraklion (M.R.), the Department of Oncology, General Hospital of Chania (I.V.), the Intensive Care Unit, Venizelion General Hospital (G.V.), and the Intensive Care Unit, General Hospital of Agios Nikolaos (G.T.), Crete, Greece, the Department of Pneumonology, General Hospital of Larnaca (I.D.), the Department of Internal Medicine, Limassol General Hospital (E.V.), the Department of Internal Medicine, Nicosia General Hospital (M.K.), the Department of Internal Medicine, General Hospital of Pafos (P.P.), the Infectious Diseases and Immunology Unit, Archbishop Makarios Hospital of Nicosia (M.K.-M.), Cyprus. Received February 7, 2008; accepted April 1, 2008; electronically published July 18, 2008. 2008 by The Society for Healthcare Epidemiology of America. All rights reserved. 0899-823X/2008/2908-0001$15.00. DOI: 10.1086/588704

686

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

terms of their patient populations, risk adjustment is essential if researchers are to make meaningful interhospital comparisons in observational studies that do not equalize risks through randomization.4 However, adjustment for case mix is often overlooked or performed only minimally in multicenter, cross-sectional studies of NI.5-7 The objective of this study was to assess the prevalence of and risk factors for NI in 11 hospitals in Cyprus and Greece, and to examine the extent to which the heterogeneous case mix in individual hospitals might hamper a benchmarking analysis. We also evaluated the use of antibiotics on the day of the survey and assessed the impact of NI on mortality.

methods
Setting All public hospitals in Cyprus and in the region of Crete in Greece participated in the survey. Of these, 2 serve as tertiarycare referral centers, and the rest are district general hospitals. Four (36.4%) of these hospitals have more than 400 beds, 3 (27.2%) have 200 400 beds, and 4 (36.4%) have fewer than 200 beds. The infection control team in each hospital comprises a microbiologist, an infectious diseases specialist and at least 1 infection control nurse. None of these hospitals is required to include surveillance as part of their basic infection control program. The University Hospital of Heraklion, a 750bed, tertiary-care institution with an established continuous surveillance program in its surgical and intensive care units (ICUs), served as the coordinating center for this study. The survey was part of a funded research program, created by researchers at the coordinating center with the aim of establishing a program for systematic surveillance and control of NI in Cretan and Cypriot hospitals. Study Design and Data Collection This point-prevalence survey included all inpatients older than 1 year who, on the study day, had been present for at least 24 hours in medical, surgical, pediatric, and/or gynecologyobstetrics wards, as well as ICUs. All surveys were done in November 2006. The information collected for all patients included demographic characteristics; primary admission diagnosis (classied according to the International Statistical Classication of Diseases and Health Related Problems, 10th Revision [ICD-10], second edition8); functional status, determined in accordance with the Karnofsky functional status index9; comorbidity, determined by use of the weighted Charlson cormorbidity index,10; severity of underlying disease, determined in accordance with the McCabe and Jackson classication11; antibiotics received on the survey day; prior surgical intervention; and exposure to invasive devices. Functional status, comorbidities, and disease severity were evaluated at the time of admission. Disease severity classication was determined on the basis of the subjective prognosis of the physicians in charge of the patients. Exposure to invasive devices was assessed for the 7 days

before onset of infection for patients with NI and for the 7 days before the study day for uninfected patients, provided that patients had been exposed to such devices for at least 24 hours during that period. Prior surgery was recorded if it had occurred during the 30 days before the survey day or the onset of infection. NI was dened according to Centers for Disease Control and Prevention criteria.12 Infections were identied by review of nursing and medical records and on the basis of information provided by the physicians and nurses in charge of the patients; this information included the results of cultures performed up to 1 week after the survey day. The observers who gathered study data were infection control practitioners and physicians who had attended training sessions regarding the study protocol. An infectious diseases specialist and an infection control professional from the coordinating center supervised data collection at all hospitals to ensure uniformity and data validity. At the end of the study, all cases of NI in each hospital were discussed and approved for inclusion in the study during a local meeting. Outcome data were recorded for all patients included in the prevalence survey to calculate hospital lengths of stay (LOS) and mortality rates. The duration of perioperative antibiotic prophylaxis was also recorded. All raw data were submitted to the coordinating center for analysis through an Internet-based data-entry system. Statistical Analysis Prevalence rates were calculated as the number of infected patients divided by the total number of patients at the time of the survey, with the exception of the prevalence rate for surgical site infection, which was calculated as the number of infected patients who had undergone surgery divided by the total number of patients who had undergone surgery. Prevalence rates in different patient groups were compared by using the 2 test. Condence intervals (CIs) and P values for prevalence rates were calculated by use of the Fisher exact method. Risk factor analysis was performed in relation to 2 outcomes, nosocomial infection and in-hospital death within 4 months after the survey. Outcome risk was estimated by calculating the odds ratios (ORs) by means of logistic regression. To identify independent risk factors for each outcome while adjusting for potential confounding effects, adjusted odds ratios (aORs) were calculated by multivariate logistic regression analysis. The nal model was developed by adding the factors that were associated with outcome risk (P .20) to univariate analysis in a forward stepwise manner.13 Each continuous covariate (including age, LOS, Charlson index, and Karnofsky index) was examined to determine whether a linear relationship existed with the log odds of each outcome. If this sort of linear trend was not observed, the continuous variable was categorized on the basis of quartiles or quintiles, or categorization was performed in such a way that each category involved a

table 1. Endogenous and Exogenous Risk Factors of Patients Surveyed for Nosocomial Infection in Cretan and Cypriot Hospitals No. (%) of patients, by hospital region Variable Male sex Age 65 yearsa Emergency admission Primary admission diagnosisb Trauma Cancer Circulatory system disease Respiratory system disease Digestive system disease Genitourinary system disease Musculoskeletal system disease Nervous system disease Infection Other Patient locationa,c Medical ward Surgical ward Gynecology-obstetrics Pediatric ward Intensive care unit McCabe-Jackson classicationa,d Nonfatal disease Ultimately fatal disease Rapidly fatal disease Charlson comorbidity indexa,d 01 24 512 Karnofsky functional status indexd 810 57 04 Neutropenia Underwent previous surgical proceduree Use of urinary catheterf Use of peripheral venous accessa,f Use of central venous catheterf Receipt of mechanical ventilationf Receipt of parenteral nutritiona,f Use of nasogastric tubef Receipt of antibiotic therapya,c Length of stay at risk, daysa,g 5 614 15
a b

Crete (n 1,127) 611 (54.2) 625 (55.5) 746 (66.2) 137 (12.2) 160 (14.2) 134 (11.8) 137 (12.2) 124 (11.0) 69 (6.1) 59 (5.2) 56 (5.0) 31 (2.8) 220 (19.5) 531 (47.1) 420 (37.3) 70 (6.2) 55 (4.9) 51 (4.5) 904 (80.5) 165 (14.7) 54 (4.8) 670 (59.4) 326 (28.9) 131 (11.6) 634 (56.4) 320 (28.5) 170 (15.1) 17 (1.5) 342 (30.3) 293 (26.0) 840 (74.5) 57 (5.1) 21 (1.9) 27 (2.4) 50 (4.4) 725 (64.3) 628 (55.7) 357 (31.7) 142 (12.6)

Cyprus (n 705) 369 (52.3) 331 (47.0) 486 (68.9) 97 (13.8) 70 (9.9) 96 (13.6) 77 (10.9) 85 (12.1) 56 (7.9) 41 (5.8) 28 (4.0) 8 (1.1) 147 (20.9) 250 (35.5) 282 (40.0) 68 (9.6) 59 (8.4) 46 (6.5) 603 (86.8) 82 (11.8) 10 (1.4) 472 (67.0) 188 (26.7) 45 (6.4) 381 (54.7) 195 (28.0) 120 (17.2) 13 (1.8) 237 (33.6) 209 (29.6) 393 (55.7) 44 (6.2) 20 (2.8) 3 (0.4) 32 (4.5) 345 (48.9) 406 (57.6) 187 (26.5) 112 (15.9)

Range among the 11 study hospitals, % of patients 28.858.6 13.872.2 49.994.4 3.824.2 1.019.8 0.019.4 2.517.8 8.815.8 3.716.7 0.011.4 0.011.1 0.04.4 8.855.0 10.069.4 8.854.0 0.042.5 0.038.8 0.011.8 76.398.0 1.817.2 0.06.7 54.977.5 15.833.3 2.516.6 41.578.5 10.141.7 5.624.4 0.08.8 16.737.6 8.344.0 40.091.7 0.011.8 0.06.5 0.09.7 0.010.8 38.876.9 49.671.4 13.838.9 4.322.8

Signicantly different distribution at the regional level (P .05). Table includes most common diagnosis categories. c At the time of survey. d At the time of admission. e In the 30 days before onset of infection or the 30 days before the survey day. f Receipt or use for at least 24 hours during the 7 days before onset of infection or the 7 days before the survey day. g Length of stay before onset of rst infection or the survey day.

688

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

table 2.

Univariate and Multivariate Analysis of Risk Factors for Nosocomial Infection (NI) in Cretan and Cypriot Hospitals No. (%) of patients Univariate analysis OR (95% CI) 1.1 (0.81.6) 1.4 (1.02.1) 1.7 (1.12.7) 2.0 (1.23.1) 4.2 (1.99.0) Reference 2.9 (1.94.5) 8.6 (4.815.2) Reference 2.0 (1.32.9) 3.4 (2.05.5) Reference 2.8 (1.94.1) 5.1 (2.211.6) 1.3 (0.91.9) 2.5 (1.73.6) 0.3 (0.20.4) 3.4 (1.95.9) 8.5 (4.416.5) 5.1 (2.112.3) 6.4 (3.810.8) 3.7 (2.55.4) P .465 .052 .011 .003 .001 ... .001 .001 ... .001 .001 ... .001 .001 .157 .001 .001 .001 .001 .001 .001 .001 Multivariate analysis aOR (95% CI) ... ... 2.2 (1.33.7) ... 4.1 (1.710.1) Reference 2.1 (1.33.5) 6.3 (3.212.6) ... ... ... Reference 1.7 (1.12.6) 3.9 (1.59.9) 2.4 (1.53.8) ... 0.3 (0.20.4) ... 3.5 (1.67.5) ... ... 2.0 (1.33.1) P ... ... .002 ... .002 ... .004 .001 ... ... ... ... .022 .004 .001 ... .001 ... .001 ... ... .002

Factor Male sex Age 65 years Emergency admission Primary admission diagnosis Cancer Infection McCabe-Jackson classicationa Nonfatal disease Ultimately fatal disease Rapidly fatal disease Charlson comorbidity index 01 24 512 Karnofsky functional status indexb 810 07 Neutropenia Underwent previous surgical procedurec Use of urinary catheterd Use of peripheral venous accessd Use of central venous catheterd Receipt of mechanical ventilationd Receipt of parenteral nutritiond Use of nasogastric tubed Length of stay at risk 15 days note.
a b c d

With NI (n 129) 73 (56.6) 78 (60.5) 100 (77.5) 27 (20.9) 9 (7.0) 76 (58.9) 33 (25.6) 20 (15.5) 56 (43.4) 47 (36.4) 26 (20.2) 42 (32.6) 87 (67.4) 8 (6.2) 48 (37.2) 60 (46.5) 49 (38.0) 17 (13.2) 15 (11.6) 7 (5.4) 23 (17.8) 44 (34.1)

Without NI (n 1,703) 907 (53.3) 878 (51.6) 1132 (66.5) 203 (11.9) 30 (1.8) 1431 (84.7) 214 (12.7) 44 (2.6) 1086 (63.8) 467 (27.4) 150 (8.8) 973 (57.5) 718 (42.5) 22 (1.3) 531 (31.2) 442 (26.0) 1184 (69.5) 73 (4.3) 26 (1.5) 19 (1.1) 56 (3.3) 210 (12.3)

aOR, adjusted odds ratio; CI, condence interval; OR, odds ratio. Data were missing for 14 patients without NI. Data were missing for 12 patients without NI. In the 30 days before onset of infection or the 30 days before the survey day. Receipt or use for at least 24 hours during the 7 days before onset of infection or the 7 days before the survey day.

sufcient sample size. Adjacent categories that had odds ratios of similar magnitude were combined. Case mixadjusted prevalence rates in participating hospitals were calculated by using a 2-step method that involved a logistic regression model and indirect standardization.14 The rst step of the method used the nal logistic regression model for NI risk to calculate the expected probability of NI for each patient, conditioned on that patients specic risk factors. Expected probabilities for individual patients were aggregated by hospital to determine the mean predicted risk for each hospital. The second step of the method was indirect standardization: standardized ratios were calculated by dividing the observed (unadjusted) prevalence rate by the mean predicted risk in each hospital, and adjusted prevalence rates were determined by multiplying standardized ratios by the observed prevalence rate in the entire study sample. WinPepi, version 4 (freeware), was used to carry out statistical tests and calculate condence intervals. SPSS, version 14 (SPSS), was used to perform forward conditional logistic re-

gression analysis. All statistical tests were 2-tailed, and P values less than .05 were considered signicant.

results
Population Studied A total of 1,832 patients were screened: 1,127 (61.5%) patients in Crete and 705 (38.5%) patients in Cyprus. Table 1 presents endogenous and exogenous risk proles for the surveyed population, revealing considerable variation in case mix both at the regional level and at the level of individual hospitals. Prevalence, Sites of Infection, and Pathogens A total of 145 NIs were detected in 129 patients, for an overall prevalence rate of 7.0% (95% CI, 5.9% 8.3%). At the regional level, a higher NI rate was observed in Cretan hospitals (7.9% [95% CI, 6.4%9.6%]) than in Cypriot hospitals (5.7% [95% CI, 4.1%7.7%]), but the difference was not statistically significant (P .070). Signicant variation in overall prevalence rates was observed among hospitals (range, 2.2%9.6%).

benchmarking infection rates in cyprus and greece

689

table 3. Unadjusted (Observed) and Case MixAdjusted Prevalence Rates for Nosocomial Infection (NI) and Relative Rankings of Cretan and Cypriot Hospitals Hospital location, number Crete, 1 Crete, 2 Cyprus, 5 Crete, 5 Cyprus, 1 Cyprus, 3 Cyprus, 4 Crete, 4 Crete, 3 Crete, 6 Cyprus, 2 note. Observed rank 1 2 3 4 5 6 7 8 9 10 11 Observed prevalence of NI, % (95% CI) 9.6 (7.112.7) 8.9 (5.713.2) 8.6 (3.217.7) 7.5 (3.114.9) 6.9 (4.210.5) 6.3 (2.114.0) 5.9 (2.212.5) 5.5 (1.812.4) 5.0 (2.48.9) 2.8 (0.114.5) 2.2 (0.65.7) Mean predicted risk of NI, % 6.9 6.3 4.9 7.7 10.3 8.0 6.3 8.3 5.9 3.7 6.0 Standardized risk ratio 1.39 1.42 1.77 0.98 0.67 0.79 0.94 0.66 0.84 0.76 0.37 Adjusted prevalence of NI, % 9.7 10.0 12.4 6.9 4.7 5.5 6.6 4.6 5.9 5.4 2.6 Adjusted rank 3 2 1 4 9 7 5 10 6 8 11

CI, condence interval.

The highest overall prevalence rate was observed in the ICUs (28.9% [95% CI, 20.1%38.9%]), followed by the pediatric wards (7.0% [95% CI, 3.1%13.4%]), medical wards (6.8% [95% CI, 5.1% 8.8%]), surgical wards (5.4% [95% CI, 3.9% 7.4%]), and gynecology-obstetrics wards (1.4% [95% CI, 0.2%5.1%]). Among the 145 episodes of NI, the most frequent type was lower respiratory tract infection, including pneumonia (2.2 infections per 100 patients [27.6% of infections]), followed by urinary tract infection (1.9 infections per 100 patients [24.1% of infections]), bloodstream infection (1.5 infections per 100 patients [18.6% of infections]), and surgical site infection (4.0 infections per 100 operations [15.9% of infections]). The other category accounted for an additional 13.8% of infections (1.1 infections per 100 patients). A total of 123 microorganisms were isolated during 97 (66.9%) of 145 episodes of NI. There were 23 polymicrobial NIs (15.9%). The pathogens isolated most frequently included Staphylococcus species (27 isolates [22.0% of all isolates]), Pseudomonas species (23 isolates [18.7%]), Acinetobacter species (14 isolates [11.4%]), Escherichia coli (12 isolates [9.8%]), Enterococcus species (11 isolates [8.9%]), Klebsiella species (10 isolates [8.1%]), and Candida species (10 isolates [8.1%]). Risk Factors for NI Factors associated with NI risk are shown in Table 2. By multivariate analysis, independent risk factors for NI were emergency admission, infection as primary admission diagnosis, McCabeJackson classication of ultimately fatal disease or rapidly fatal disease, impaired functional status (Karnofsky index 0 7), neutropenia, LOS exceeding 15 days at risk for infection, receipt of mechanical ventilation, and having undergone a surgical procedure in the 30 days before onset of infection. Case MixAdjusted Prevalence Rates According to the logistic regression model, the mean predicted risk of NI in individual hospitals ranged from 3.7% to 10.3%,

indicating that patient mix differed across hospitals with respect to the signicant predictors of NI risk. The risk-adjusted prevalence rates in participating hospitals, compared with the observed (unadjusted) rates, are shown in Table 3. The adjusted prevalence rates ranged from 2.6% to 12.4%. Of 11 hospitals, 7 (63.7%) had an adjusted rate greater than the observed prevalence rate. For 8 hospitals (72.7%), the relative ranking of the hospital was affected by case-mix adjustment. Antibiotic Use On the day of survey, 1,070 patients (58.4%) were receiving antibiotics (interhospital range, 38.8%76.9% of patients). There were 358 patients (33.5%) who received 2 antibiotics, and 71 patients (6.6%) received 3 or more. A total of 525 patients (49.1%) received empirical treatment; 351 patients (32.8%) received antibiotics for surgical prophylaxis, and 123 patients (11.5%) received them for bacteriologically documented infection. For 71 patients (6.6%), no justication for antibiotic use was provided. Of the patients who received perioperative prophylaxis, 225 (64.1%) had undergone operations classied as clean. The median duration of perioperative prophylaxis was 5 days (interhospital range, 216 days). The classes of antibiotics administered most commonly included rst- and second-generation cephalosporins (19.3%), extended-spectrum cephalosporins (13.7%), uoroquinolones (13.9%), combinations of penicillins with -lactamase inhibitors (10.9%), imidazoles (9.6%), glycopeptides (7.2%), carbapenems (5.8%), and aminoglycosides (5.1%). Patient Outcomes Of the 1,832 patients surveyed, 1,692 (92.4%) were discharged, 111 (6.1%) died, and 29 (1.6%) remained hospitalized 4 months after the survey. The overall mean LOS was 19.4 days (median, 10 days [interquartile range, 518 days]). The mean LOS after onset of infection for patients who developed NI (27.5 days) was signicantly longer than

690

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

table 4. Univariate and Multivariate Analysis of Risk Factors for Mortality Associated with Nosocomial Infection in Cretan and Cypriot Hospitals No. (%) of patients Factor Male sex Age 65 years Emergency admission Primary admission diagnosisa Cancer Respiratory system disease Genitourinary system disease Digestive system disease McCabe-Jackson classicationb Nonfatal disease Ultimately fatal disease Rapidly fatal disease Charlson comorbidity index 01 24 512 Karnofsky functional status indexc 810 07 Neutropenia Underwent previous surgical procedured Exposed to 3 invasive devices Developed nosocomial infection note.
a b c d

Died (n 111) 60 (54.1) 80 (72.1) 91 (82.0) 35 (31.5) 23 (20.7) 2 (1.8) 7 (6.3) 34 (30.6) 52 (46.8) 25 (22.5) 26 (23.4) 49 (44.1) 36 (32.4) 16 (14.4) 95 (85.6) 6 (5.4) 18 (16.2) 31 (27.9) 36 (32.4)

Survived (n 1,721) 920 (53.5) 876 (50.9) 1,141 (66.3) 195 (11.3) 191 (11.1) 123 (7.1) 202 (11.7) 1,473 (86.3) 195 (11.4) 39 (2.3) 1,116 (64.8) 465 (27.0) 140 (8.1) 999 (58.5) 710 (41.5) 24 (1.4) 561 (32.6) 54 (3.1) 93 (5.4)

Univariate analysis OR (95% CI) 1.0 (0.71.5) 2.5 (1.63.8) 2.3 (1.43.8) 3.6 (2.45.5) 2.1 (1.33.4) 0.2 (0.11.0) 0.5 (0.21.1) Reference 11.6 (7.318.3) 27.8 (15.150.9) Reference 4.5 (2.87.4) 11.0 (6.518.8) Reference 8.4 (4.914.3) 4.0 (1.610.1) 0.4 (0.20.7) 10.6 (6.218.0) 8.4 (5.413.2) P .903 .001 .001 .001 .003 .046 .087 ... .001 .001 ... .001 .001 ... .001 .003 .001 .001 .001
.20).

Multivariate analysis aOR (95% CI) ... ... ... ... 2.3 (1.34.2) ... ... Reference 4.9 (2.98.3) 8.7 (4.317.6) Reference 2.2 (1.33.9) 2.9 (1.55.6) Reference 3.2 (1.85.7) ... ... 4.2 (2.37.5) 3.6 (2.16.1) P ... ... ... ... .006 ... ... ... .001 .001 ... .006 .001 ... .001 ... ... .001 .001

OR, odds ratio; aOR, adjusted odds ratio; CI, condence interval. Table only includes admission diagnosis categories signicantly associated with death in univariate analysis (P Data missing for 14 patients. Data missing for 12 patients. In the 30 days before onset of infection or the 30 days before the survey day.

the mean total LOS for patients without NI (17.4 days)an excess hospital stay of 10.1 days (95% CI, 3.9 16.4 days). The crude mortality rate was higher for patients with NI (27.9%) than for those without NI (4.4%), resulting in a crude OR of 8.4 (95% CI, 5.4 13.2). After controlling for confounding, the adjusted OR relating NI to mortality was 3.6 (95% CI, 2.1 6.1) (Table 4).

discussion
The overall prevalence of patients with NI observed in the Cretan and Cypriot regions (7.0% [95% CI, 5.9 8.3]) appears to be lower than that reported in recent multicenter surveys in other European countries, including Switzerland (10.1% [95% CI, 9.211.0]),6 Scotland (9.5% [95% CI, 8.8 10.2]),15 and the United Kingdom (9.0% [95% CI, 8.8 9.3]).16 The rate reported here appears to be of the same magnitude as that reported in Denmark (8.0% [95%CI, 7.2 8.8]),17 Italy (7.8% [95% CI, 7.3 8.3]),18 Spain (6.9% [95% CI, 6.77.2]),19 and France (6.7% [95% CI, 6.6 6.8])20, and it is higher than the rates reported in hospitals in Norway (5.1% [95% CI, 4.7 5.5]),21 Slovenia (4.6% [95% CI, 4.25.2]),22 and Germany

(3.5% [95% CI, 3.13.9]).23 These discrepancies may represent differences in infection control practices but they may also be explained by differences in how NI was dened, variation in the sensitivity of the methods used to detect NI, emphasis on different sites of infection, and/or surveys that involved different patient populations.3,24 Adjustment for case-mix variation has not been absent from multicenter prevalence studies, but it has been mostly limited to reporting stratied rates according to hospital setting, teaching afliation, and size.15-23 Previous studies have noted that scanty adjustment for the heterogeneity of inpatient risk of infection is a major drawback of comparative studies.5,6,25 Sax et al.6 illustrated the impact of case mix on infection rate by using the example of hospital size: after adjusting for a large number of intrinsic and extrinsic risk factors, they showed that the higher prevalence rate observed in larger hospitals did not suggest an infection control problem but instead was associated with unfavorable case mix. Floret et al.25 stratied patients into distinct risk categories (on the basis of age, immune status, and McCabe-Jackson classication) and demonstrated increases in NI prevalence rates for higher-risk patients, inde-

benchmarking infection rates in cyprus and greece

691

pendent of hospital type or size. The latter approach was, however, limited by the small number and subjective selection of the factors used for risk stratication. In this survey, we found considerable variation in NI rate among hospitals (range, 2.2%9.6%), but we also noted considerable heterogeneity in case mix. Thus, directly attributing variation in infection rates to differences in infection control practices would be questionable, and feedback of unadjusted rates to participating hospitals would not be likely to prompt any corrective interventions to improve current practices. To adjust for case mix, we considered a large number of potential confounding variables and used those that were signicantly associated with NI risk in multivariate analysis. Case-mix adjustment of prevalence rates affected the relative ranking for most hospitals, revealing the importance of accounting for this type of variability before attempting interhospital comparisons. Because the same methodology for data collection was used in all hospitals and the variability of patient exposures to major risk factors was also considered, differences in adjusted prevalence rates may mainly represent differences in infection control practices among hospitals in these regions. Moreover, persistent disparity in adjusted prevalence rates (range, 2.6%12.4%) may indicate that there is room for improvement in current infection control efforts in these regions and deserves further investigation. Among the exposures to invasive devices that were examined in this study, receipt of mechanical ventilation was independently associated with increased risk of NI, suggesting a need for more effective preventive measures against ventilatorassociated infections in these regions.26 Prolonged length of stay was also identied as an independent risk factor for NI in this study. The median duration of hospitalization was 10 days in the study population, and we identied a large proportion of patients with considerably longer stays (75th percentile, 18 days). Moreover, 29 (1.6%) of the patients surveyed remained in the hospital 4 months after the survey. These observations may be partially explained by the absence of long-term care facilities in the study regions, but efforts to reduce LOS should also be considered in these hospitals. The need to harmonize Greek hospitals antibiotic use policies with international guidelines has been emphasized in previous studies.27-29 Similarly, in the hospitals examined in this study, including the Cypriot hospitals, we found a high prevalence of antibiotic use (58.4% of patients received antibiotics), compared with the prevalence seen in most European countries (32% 49%).5,6,15,18,19 Moreover, the frequent use of antibiotic combinations (for 40% of patients) and the extended duration of prophylaxis (median, 5 days) are not consistent with prevailing guidelines.30 The overuse of antibiotics and inappropriately lengthy antibiotic therapy not only have an impact on the economics of these hospitals, but they could also be contributing to the risk of NI and continuing emergence of antibiotic-resistant pathogens. It is an undisputed fact that patients with NI have high crude mortality rates, but the exact contribution of NI to mortality is

understood only partially.31 We attempted to investigate this issue by using multivariate analysis to estimate the independent effect of NI after controlling for a large set of potential confounders. Our results indicate that NI signicantly increases mortality, independent of patients underlying conditions. However, our approach assumes that the composition of the 2 patient groups (infected or not infected at the time of the survey) and their underlying conditions remained constant until discharge. The effect of these assumptions on risk estimates is unknown, but the reported contribution of NI to mortality is certainly an improvement over that of univariate analysis (OR, 3.6 vs 8.4). The selection of case-mix markers in this study was made on the basis of previous work by Pittet et al.5 and Sax et al.6 Measuring case mix, however, is a complex process, and the most rewarding set of indicators is yet to be determined.6,10,32 We evaluated functional status and severity of illness indices at time of admission to ensure that effects precede outcome, but it is possible that these variables may have changed over the course of hospitalization, especially for patients with long lengths of stay. Our approach would probably be enhanced by considering single-site infections and site-specic indicators,33 but the limited number of events in a prevalence survey prohibits site-specic analysis. Moreover, we cannot exclude the possibility that random variation may have affected our results,34 and the rankings provided in Table 3 should be interpreted accordingly. The message for participating hospitals, especially the smaller ones, is to repeat this survey or incorporate a case-mix adjustment approach into prospective surveillance studies to conrm the current ndings. In conclusion, this rst attempt to rank the risk of NI in Cretan and Cypriot hospitals demonstrated the importance of accounting for heterogeneous case mix before attempting interhospital comparison of infection rates. Persistent disparity in case mixadjusted rates among participating hospitals deserves further examination, and repeated prevalence surveys or prospective surveillance studies may be required in these regions to conrm current ndings.

acknowledgments
We thank the following infection control professionals and physicians who participated in data collection and management: A. Messaritaki and C. Tsioutis, from University Hospital of Heraklion; C. Koulas and P. Panagiotou, from Nicosia General Hospital; E. Bolikas, from Venizelion General Hospital of Heraklion; S. Androulaki and G. Pavlidaki, from the General Hospital of Chania; M. Pavlou and A. Aristodemou, from Limassol General Hospital; M. Toumasi-Stavridi and L. Chalatzian, from Archbishop Makarios Hospital of Nicosia; A. Tsourdalaki, E. Tsourdalaki, and P. Drandakis, from the General Hospital of Rethymnon; T. Aristeidou, from the General Hospital of Larnaca; D. Christofaki, from General Hospital of Agios Nikolaos; C. Panagiotou and M. Kapnisi-Andreou, from the General Hospital of Pafos; and A. Drandaki and M. Koufalexi, from the General Hospital of Sitia. Financial support. This study was supported by the European Regional Development Funds through the initiative INTERREG III Greece-Cyprus 2000-2006.

692

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

Potential conicts of interest. relevant to this article.

All authors report no conicts of interest

Address reprint requests to Achilleas Gikas, MD, University Hospital of Heraklion 1352/71110, Crete, Greece (gikas@med.uoc.gr).

references
1. Burke JP. Infection controla problem for patient safety. N Engl J Med 2003;348:651 656. 2. Gaynes R, Richards C, Edwards J, et al. Feeding back surveillance data to prevent hospital-acquired infections. Emerg Infect Dis 2001;7:295298. 3. Humphreys H, Smyth ET. Prevalence surveys of healthcare-associated infections: what do they tell us, if anything? Clin Microbiol Infect 2006;12: 2 4. 4. Richardson D, Tarnow-Mordi WO, Lee SK. Risk adjustment for quality improvement. Pediatrics 1999;103(suppl E):255265. 5. Pittet D, Harbarth S, Ruef C, et al. Prevalence and risk factors for nosocomial infections in four university hospitals in Switzerland. Infect Control Hosp Epidemiol 1999;20:37 42. 6. Sax H, Pittet D; Swiss-NOSO Network. Interhospital differences in nosocomial infection rates: importance of case-mix adjustment. Arch Intern Med 2002;162:24372442. 7. Sartor C, Delchambre A, Pascal L, Drancourt M, De Micco P, Sambuc R. Assessment of the value of repeated point-prevalence surveys for analyzing the trend in nosocomial infections. Infect Control Hosp Epidemiol 2005;26:369 373. 8. World Health Organization. International Statistical Classication of Diseases and Related Health Problems, 10th Revision. 2nd ed. Geneva: World Health Organization; 2006. 9. Mor V, Laliberte L, Morris JN, Wiemann M. The Karnofsky Performance Status Scale: an examination of its reliability and validity in a research setting. Cancer 1984;53:20022007. 10. Needham DM, Scales DC, Laupacis A, Pronovost PJ. A systematic review of the Charlson comorbidity index using Canadian administrative databases: a perspective on risk adjustment in critical care research. J Crit Care 2005;20:1219. 11. McCabe WR, Jackson GG. Gram-negative bacteremia. I. Etiology and ecology. Arch Intern Med 1962;110:847 855. 12. Horan TC, Gaynes RP. Surveillance of nosocomial infections. In: Mayhall CG, ed. Hospital Epidemiology and Infection Control. 3rd ed. Philadelphia: Lippincott Williams & Wilkins; 2004:1659 1702. 13. Hosmer D, Lemeshow S. Applied Logistic Regression. 2nd ed. New York: Wiley; 2000. 14. Kahn HA, Sempos CT. Statistical Methods in Epidemiology. New York: Oxford University Press; 1989. 15. Reilly J, Stewart S, Allardice, et al. Health Protection Scotland. NHS Scotland National Hospital Acquired Infection Prevalence Survey. Final Report 2007. Available at: http://www.hps.scot.nhs.uk/haiic/sshaip/publications/ national-prevalence-study/report/full-report.pdf. Accessed January 30, 2007. 16. Emmerson AM, Enstone JE, Grifn M, Kelsey MC, Smyth ET. The Second National Prevalence Survey of infection in hospitals overview of the results. J Hosp Infect 1996;32:175190.

17. Christensen M, Jepsen OB. Reduced rates of hospital-acquired UTI in medical patients: prevalence surveys indicate effect of active infection control programmes. J Hosp Infect 2001;47:36 40. 18. Zotti CM, Messori Ioli G, Charrier L, et al. Hospital-acquired infections in Italy: a region wide prevalence study. J Hosp Infect 2004;56:142149. 19. Vaqu J, Rossell J, Arribas JL. Prevalence of nosocomial infections in Spain: EPINE study 1990-1997. EPINE Working Group. J Hosp Infect 1999;43(Suppl):105111. 20. The French Prevalence Survey Study Group. Prevalence of nosocomial infections in France: results of the nationwide survey in 1996. J Hosp Infect 2000;46:186 193. 21. Eriksen HM, Iversen BG, Aavitsland P. Prevalence of nosocomial infections in hospitals in Norway, 2002 and 2003. J Hosp Infect 2005;60:40 45. 22. Klavs I, Bufon Luznik T, Skerl M, et al. Prevalance of and risk factors for hospital-acquired infections in Sloveniaresults of the rst national survey, 2001. J Hosp Infect 2003;54:149 157. 23. Gastmeier P, Kampf G, Wischnewski N, et al. Prevalence of nosocomial infections in representative German hospitals. J Hosp Infect 1998;38:37 49. 24. Ruef C. Prevalence of nosocomial infectionswho knows the true rates? Infection 1997;25:203205. 25. Floret N, Bailly P, Bertrand X, et al. Results from a four-year study on the prevalence of nosocomial infections in Franche-Comt: attempt to rank the risk of nosocomial infection. J Hosp Infect 2006;63:393398. 26. Dodek P, Keenan S, Cook D, et al. Evidence-based clinical practice guideline for the prevention of ventilator-associated pneumonia. Ann Intern Med 2004;141:305313. 27. Gikas A, Pediaditis J, Papadakis JA, et al. Prevalence study of hospitalacquired infections in 14 Greek hospitals: planning from the local to the national surveillance level. J Hosp Infect 2002;50:269 275. 28. Gikas A, Roumbelaki M, Pediaditis J, et al. Prevalence of nosocomial infections after surgery in Greek hospitals: results of two nationwide surveys. Infect Control Hosp Epidemiol 2004;25:319 324. 29. Kritsotakis EI, Assithianakis P, Kanellos P, Tzagarakis N, Ioannides MC, Gikas A. Surveillance of monthly antimicrobial consumption rates stratied by patient-care area: a tool for triggering and targeting antibiotic policy changes in the hospital. J Chemother 2006;18:394 401. 30. Mangram AJ, Horan TC, Pearson ML, Silver LC, Jarvis WR. Guideline for prevention of surgical site infection, 1999. Centers for Disease Control and Prevention (CDC) Hospital Infection Control Practices Advisory Committee. Am J Infect Control 1999;27:97132. 31. Rello J. Impact of nosocomial infections on outcome: myths and evidence. Infect Control Hosp Epidemiol 1999;20:392394. 32. Norena M, Wong H, Thompson WD, Keenan SP, Dodek PM. Adjustment of intensive care unit outcomes for severity of illness and comorbidity scores. J Crit Care 2006;21:142150. 33. Gaynes RP. Surgical-site infections (SSI) and the NNIS Basic SSI Risk Index, part II: room for improvement. Infect Control Hosp Epidemiol 2001; 22:266 267. 34. Gastmeier P, Sohr D, Rath A, et al. Repeated prevalence investigations on nosocomial infections for continuous surveillance. J Hosp Infect 2000;45: 4753.

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

commentary

The Importance of Case-Mix Adjustment for Infection Rates and the Need for More Research
Anthony D. Harris, MD, MPH; Jessina C. McGregor, PhD

(See the article by Kritsotakis et al. on pages 685 692)

After years of discussion, hospitals in many countries are now being required to publicly report healthcare-associated infection rates. For example, in the United States, legislation to this effect is pending or has been approved in 32 states.1 The impetus for the public reporting movement is complex, but important driving goals are providing information to patients, health insurance carriers, and health maintenance organizations about hospital performance data and better informing consumers about healthcare services. Indeed, these efforts have been driven not only by consumers but also by entities as diverse as the Centers for Medicare and Medicaid Services, the Healthcare Infection Control Practices Advisory Committee, the Centers for Disease Control and Prevention, and state and federal legislators. However, for the reported data on healthcare-associated infection rates to be meaningful, the rates need to be properly risk adjusted to account for the differences in patient populations that lead to inherently different baseline risks. At present, many groups, including the Centers for Disease Control and Prevention and the Society for Healthcare Epidemiology of America, conclude that the use of infection rates for interhospital comparisons is crude, inaccurate, and potentially awed because of the inability to effectively riskadjust rates using scientically validated methods.2,3 Thus, the problem is that for most healthcare-associated infections, there are no validated and optimal methods of risk adjustment. We believe that properly risk-adjusted healthcare-associated infection rates can serve a number of very important purposes, and thus we are not against the reporting of these rates if they are properly risk adjusted. However, until validated methods of risk adjustment exist for each outcome measure, we favor the reporting of process measures. Properly risk-adjusted healthcare-associated infection rates are critical to the achievement of the following goals: (1) allowing consumers and healthcare maintenance orga-

nizations to be more informed purchasers of healthcare, and (2) comparing infection rates, after controlling for possible patient-level confounders, both between institutions and with benchmark values. Valid benchmarks will serve as the foundation for further scientic inquiry regarding the etiologies of differences in rates between sites and populations, as well as for the design of future interventions to curb infection rates, leading to the ultimate goal of improved patient care. In infection control, risk adjustment generally needs to take account of a number of important confounding variables that affect infection rates. These include patients comorbid conditions and the severity of patients illnesses before they develop infection. Some work has been done on the development and validation of comorbidity risk adjustment, but more is needed.4 Similarly, some work has been done on the development and validation of severity of illness risk-adjustment methods, but more work is needed.5 Against this background, in this issue of the journal, Kritsotakis et al.6 examine the effect of case-mix adjustment on the rankings of 11 hospitals prevalence rates for nosocomial infections. This is the primary aim of their study, and the authors should be commended for addressing this very important topic. Additionally, they aim to analyze the effect of nosocomial infection on mortality. The method used for case-mix adjustment was as follows: rst, logistic regression modeling was performed to analyze risk factors for nosocomial infection, then the identied risk factors were used to adjust nosocomial infection rates by use of indirect standardization. In our opinion, the most fascinating results of the study6 lie in Table 3 of their article. This table outlines the rankings of hospitals according to infection rates, both risk-adjusted and nonrisk-adjusted rates. What is interesting about Table 3 is the large change in the rankings after risk adjustment. These results demonstrate the critical nature of proper risk adjustment. It is important to appreciate that a

From the Department of Epidemiology and Preventive Medicine, University of Maryland, Baltimore, Maryland (A.D.H.), and the Oregon State University College of Pharmacy, Portland, Oregon (J.C.M.). Received May 29, 2008; accepted June 1, 2008; electronically published July 18, 2008 Infect Control Hosp Epidemiol 2008; 29:693 694 2008 by The Society for Healthcare Epidemiology of America. All rights reserved. 0899-823X/2008/2908-0002$15.00. DOI: 10.1086/590471

694

infection control and hospital epidemiology

august 2008, vol. 29, no. 8

less-than-ideal risk adjustment measure may also change the rankings, but not in a valid way. Unfortunately, there is no gold standard or statistical test that can be used to identify a model that is sufciently risk adjusted for differences in case mix. Development of a case-mix adjustment method requires consideration of a wide range of patientlevel variables, as well as validation in large, multiinstitutional data sets in order to achieve a high degree of validity. Limitations of the Kritsotakis et al. study6 include the subjective nature of the severity of illness adjuster used (McCabe-Jackson severity of illness classication), the lack of a separate development and validation data set, the use of regression techniques that did not account for clustering effects within institutions, and the use of point prevalence rates of nosocomial infections instead of incidence rates. Furthermore, the ranking method should allow for institutions to share a single rank if their adjusted rates are not signicantly different from one another. Extensive research has been conducted into the development of case-mix adjustment methods for mortality following acute myocardial infarctions and heart failure.7,8 Although these case-mix adjustment methods cannot be directly applied to infection rates, future research should be guided by the methodologic advancements made in this area of research. In summary, we believe that more research is needed with respect to the development and validation of riskadjustment methods for healthcare-associated infections such as surgical site infection, ventilator-associated pneumonia, and bloodstream infection. Until we have valid riskadjustment methods, the utility of hospital-reported infection rates is questionable with regard to identifying good and bad hospitals, and thus may be easily misinterpreted by the public.

acknowledgments
Potential conicts of interest. evant to this article. Both authors report no conicts of interest rel-

Address reprint requests to Anthony D. Harris, MD, MPH, University of Maryland, Department of Epidemiology and Preventive Medicine, 100 North Greene Street Lower Level, Baltimore, MD 21201 (aharris@epi.umaryland .edu).

references
1. Association for Professionals in Infection Control and Epidemiology. Mandatory reporting of healthcare performance measures. Available at: http://www.apic.org/AM/Template.cfm?Section Mandatory_Reporting. Accessed July 3, 2003. 2. Nosocomial infection rates for interhospital comparison: limitations and possible solutions. A Report from the National Nosocomial Infections Surveillance (NNIS) System. Infect Control Hosp Epidemiol 1991;12:609 621. 3. Wong ES, Rupp ME, Mermel L, et al. Public disclosure of healthcareassociated infections: the role of the Society for Healthcare Epidemiology of America. Infect Control Hosp Epidemiol 2005;26:210 212. 4. McGregor JC, Perencevich EN, Furuno JP, et al. Comorbidity riskadjustment measures were developed and validated for studies of antibiotic-resistant infections. J Clin Epidemiol 2006;59:1266 1273. 5. Fine MJ, Auble TE, Yealy DM, et al. A prediction rule to identify low-risk patients with community-acquired pneumonia. N Engl J Med 1997;336: 243250. 6. Kritsotakis EI, Dimitriadis I, Roumbelaki M, et al. Case-mix adjustment approach to benchmarking prevalence rates of nosocomial infection in hospitals in Cyprus and Greece. Infect Control Hosp Epidemiol 2008;29:685 692 (in this issue). 7. Krumholz HM, Wang Y, Mattera JA, et al. An administrative claims model suitable for proling hospital performance based on 30-day mortality rates among patients with an acute myocardial infarction. Circulation 2006;113: 16831692. 8. Krumholz HM, Wang Y, Mattera JA, et al. An administrative claims model suitable for proling hospital performance based on 30-day mortality rates among patients with heart failure. Circulation 2006;113:16931701.