Odds ratios, with 95% confidence intervals (CIs) in parentheses, for nosocomial infection before (solid squares) and after (solid circles) adjustment for case mix, Second Swiss National Prevalence Study. Hospitals were stratified as small (reference), intermediate, and large.
Sax H, Pittet D, and the Swiss-NOSO Network. Interhospital Differences in Nosocomial Infection RatesImportance of Case-Mix Adjustment. Arch Intern Med. 2002;162(21):2437-2442. doi:10.1001/archinte.162.21.2437
Nosocomial infection rates are used to assess patient safety and the effectiveness of health care systems, but adjustment for case-mix, a key factor for benchmarking, is often overlooked.
To perform a nationwide prevalence study of nosocomial infection and evaluate the impact of hospital size on infection rates.
One-week–period prevalence study in 18 acute care hospitals ranging from small primary to large tertiary care institutions. All adult inpatients in medical, surgical, and intensive care units hospitalized at time of study were included. Infection prevalence and case-mix determinants were calculated according to hospital size. After each factor was tested for its significance on the occurrence of nosocomial infection, all factors were introduced in a multivariate model with hospital size as the main explanatory variable and nosocomial infection as the dependent variable.
Among 4252 patients, 429 developed 470 nosocomial infections, for an overall prevalence of 10.1% (intensive care units, 29.7%; medical, 9.3%; surgical, 9.2%; and mixed wards, 14.1%). Unadjusted prevalence rates were 6.1% in small, 10.0% in intermediate, and 10.9% in large hospitals (P =
.007). Increased comorbidity (odds ratio, 1.80), cancer (1.68), trauma (1.75), neutropenia (4.66), antibiotic exposure (6.64), history of intensive care unit stay (2.14), referral from another hospital (1.87), intubation for 24 hours or more (2.09), and prolonged stay (3.35) were independently associated with nosocomial infection (all P<.05), but hospital size was not.
Higher infection rates observed in larger hospitals were partly associated with unfavorable case mix. Unadjusted rates may lead to erroneous assumptions for health care prioritization.
HEALTH CARE–associated infections constitute a major challenge of modern medicine and are considered one of the most accurate indicators of the quality of patient care. Interhospital benchmarking of nosocomial infection (NI) rates is being performed with the aim of improving the effectiveness of health care and promoting patient safety.1
The National Nosocomial Infection Surveillance (NNIS) system in the United States propagates standardized infection rates that serve as a quality reference for many hospitals.2,3 In response to the public's claim for increased transparency and freedom of choice, hospitals' crude infection and other adverse event rates are even being published in the lay press, both in Europe and in the United States.4- 6 In Switzerland, NI rates have been progressively integrated into several quality assurance programs initiated by hospital administration associations and private insurance companies.7 However, the major drawback with these innovative measures is their scanty adjustment for differences in patient characteristics. Therefore, we undertook a nationwide survey to determine the prevalence of NI in Swiss hospitals and to investigate the impact of hospital size after adjustment for case mix and other hospital characteristics.
The study was conducted in April 1999 simultaneously in 18 Swiss acute care hospitals recruited on a voluntary basis by their affiliation with the Swiss-NOSO Network.8
All adult inpatients present at study day on surgical, medical, and intensive care wards were included, with the exception of those in dermatology, ophthalmology, ear-nose-throat, gynecology-obstetrics, bone marrow transplant, and burn units, and long-term care wards (median stay, >30 days). Modified Centers for Disease Control and Prevention criteria were applied to define NI by means of a standardized form as described previously.9,10 In brief, asymptomatic urinary tract colonization was not considered an infection, and a physician's diagnosis of pneumonia was accepted as an additional criterion together with radiologic findings and clinical signs.10
Observers were infection control practitioners who attended at least 3 specific 1-day training sessions. Detailed documentation, including the study protocol, a standardized case report form, written definitions for all study variables, practical exercises, code lists, and modified Centers for Disease Control and Prevention definitions of NI9,10 in observers' native language, was provided. Infection control practitioners from centers with and without experience in prevalence studies were grouped as study teams and supervised by an infection control physician at each center.
Variables collected for all patients included demographics, comorbidity using the Charlson index,11 admission diagnosis classified according to the International Classification of Diseases, 10th Revision,12 and the McCabe and Jackson classification.13 Exposures to antibiotic or antacid medication and intravascular, endotracheal, and urinary devices were recorded quantitatively (number of device-days within 7 days before infection for patients with NI and study day for those without). For surgical patients, the type of surgery and the NNIS surgical site infection risk index14 were recorded, the latter being a composite score consisting of wound classification,15 American Society of Anesthesiologists physical status classification,16 and "T" time (the 75th percentile of the mean duration for the intervention in question).14
Prevalence of NI is presented as prevalence of infected patients (defined as the number of infected patients divided by the total number of patients hospitalized at time of study) and prevalence of infections (defined as the number of NIs divided by the total number of patients hospitalized at time of study).
Patients were stratified by hospital size to test the influence of this factor on infection rates: small, less than 150 patients included; intermediate, 150 to 300; and large, more than 300.
Infection prevalence and case-mix determinants were then calculated for these patient groups. After each factor was tested for its significance on the occurrence of NI, all factors were introduced in a multivariate model with hospital size as the main explanatory variable and NI as the dependent variable.
Patient groups were compared with χ2 test for categorical variables and with t or Wilcoxon rank sum tests for continuous variables. The χ2 test for trends was used to compare patient characteristics according to hospital size. We used odds ratios and their corresponding 95% confidence intervals as a measure of association between explanatory variables and NI. Variables associated with a P value of less than .1 in the univariate analysis were entered in a stepwise logistic regression model to investigate their independent effect on NI.
Statistical analysis was done with Intercooled Stata 6.0 (Stata Corp, College Station, Tex). All tests were 2-tailed, and P values less than .05 were considered statistically significant.
Of the 4252 patients included in the study, 429 experienced 470 NIs, resulting in an overall prevalence of 10.1% infected patients; the prevalence of NIs was 11%. The most frequent infections were surgical site (23.2% of all NIs), lower respiratory tract (22.8%), urinary tract (21.3%), bloodstream (11.5%), eyes or ear-nose-throat (6.4%), gastrointestinal tract (5.1%), and soft tissue (3.6%). Prevalence of infected patients by ward distribution was as follows: intensive care units, 29.7%; medical, 9.3%; surgical, 9.2%; and mixed wards, 14.1%.
Eighteen hospitals contributed a mean of 236 (SD, 160) patients to the study. Numbers of patients included, total hospital beds, annual admissions, and patient-days were significantly different in the 3 size groups (Table 1).
Patient characteristics varied considerably between hospital size groups (Table 2). Most important, patients admitted in larger centers had a greater number of comorbidities (P = .009) and suffered more frequently from ultimately fatal conditions (P<.001), according to the McCabe and Jackson classification (Table 2). Cancer (P<.001), nervous system disease (P = .001), and urogenital disease (P = .01) as admission diagnoses were more frequent in larger hospitals, as were referral from another hospital (P = .02), exposure to intensive care during hospital stay (P<.001), and surgical intervention (P = .001). Solid organ or bone marrow recipients were found only in intermediate and large hospitals, with a predominance in the latter (P = .001). Smaller hospitals featured more patients with respiratory (P = .01), digestive tract (P = .04), and miscellaneous (P<.001) diseases, and those admitted as emergencies (P<.001). In intermediate-size hospitals, 4.8% of patients stayed in "mixed" wards, eg, wards that harbored medical and surgical patients in parallel, and stepdown units. Central venous and urinary catheter use, mechanical ventilation, antacid medication, and antibiotic use were more frequent in larger hospitals (all P<.001). Short hospitalization was much more common in small than in intermediate or large hospitals (P<.001).
The NI prevalence in small, intermediate, and large hospitals was 6.1%, 10.0%, and 10.9%, respectively (P = .007); as shown in Figure 1, unadjusted odds ratios for infection were significantly higher in large and intermediate institutions than in small hospitals.
Variables associated with NI are shown in Table 3. By multivariate analysis, factors independently associated with NI were cancer or trauma as admission diagnosis, an increased Charlson Comorbidity Index, neutropenia, antibiotic exposure before infection or study, a history of intensive care unit stay, referral from another hospital, a hospital stay of longer than 14 days at time of study or infection, and intubation for 24 hours or more (Table 4). Short intubation was inversely associated with NI.
Most important, hospital size was not an independent risk factor for infection after adjustment for case-mix (Figure 1).
Case-mix adjustment has not been abundant in multicenter prevalence studies to date and has been mostly limited to reporting stratified rates by hospital setting, teaching affiliation, and size.10,17- 28 In the present study, the impact of case mix on infection rates is well illustrated with the example of hospital size. Unadjusted rates were significantly lower in smaller than in larger hospitals. After case-mix adjustment, however, small hospitals did not have a lower risk of NI per se but harbored less-sick patients.
Hospital size was naturally used as a surrogate marker of case mix in several multicenter prevalence studies.18- 25 In the German Nosokomiale Infektionen in Deutschland–Erfassung und Prävention (NIDEP; Nosocomial Infection in Germany–Surveillance and Prevention) study in 1994, hospitals larger than 600 beds harbored significantly more infected patients (4.4%) than did those with less than 200 beds (2.3%).24 Similarly, a positive correlation between hospital size and urinary tract and surgical site infection was found among 106 Belgian hospitals.19 With some exceptions22,23,25 or lack of reporting,18,20 most studies described a relationship between increased hospital size and enhanced unadjusted infection rates. In fact, the only previous study to adjust for case mix by means of multivariate analysis was from the Estudio de Prevalencia de Infecciones Nosocomiales en España (EPINE; Prevalence Study of Nosocomial Infections in Spain) study group.22
Hospital size has equally been used to stratify NI rates in incidence studies.29- 32 Secular trends in nosocomial primary bloodstream infection in the NNIS system in the United States showed a clear-cut association with the hospital bed size and teaching affiliation.30 In their last report, NNIS rates for combined medical-surgical intensive care units were stratified by hospital type, since major teaching hospitals yielded higher device utilization ratios and infection rates than all other types.31,32 As with the NNIS system,14 adjustment for case mix is now common in prospective surveillance. Still, however, it is usually performed for only a few variables specific to the infection site, and a more appropriate risk adjustment by multivariate analysis is advocated.33,34
Case-mix indicators are overlapping in their significance, and the most rewarding set of markers is yet to be determined. Surgical intervention is a known risk factor for NI, primarily in the surgical site.10 Patients have a short average length of hospital stay, which results in an underreporting of surgical site infections of usually around 50%.35 In our model, accordingly, the impact of surgical intervention on the overall infection rate was counterbalanced by the introduction of length of stay. Similarly, the McCabe and Jackson classification, a known marker for patients at risk for NI,36 did not hold up as an independent determinant in our study.
Our analysis is restricted to global infection rates but exhibits two problems. First, significance in terms of outcome might not be equal among NI, and second, adjustment is confined to indicators that are associated with global rates. When single-site infections are investigated, adjustment may be enhanced by including site-specific indicators.34 However, a limited number of events in prevalence precludes NI-stratified analysis, leaving it to prospective surveillance. Moreover, the foremost aim of a national study is to be able to assess hospital rates across the country on the unvaried background of the national health system and to establish priorities for infection control.
Nosocomial infections are one of the most important indicators of quality of care. Benchmarking among hospitals requires meticulous adjustment for case mix, and failure to adequately adjust for infection-associated factors can only hinder quality improvement. In this study, hospital size turned out to be a surrogate marker, but not a risk factor for NI. Prevalence studies are increasingly used to close the gap left by cutting back on costly hospitalwide prospective surveillance. In particular, nationwide prevalence studies might be used to produce benchmark data and set priorities in public health care for the future. Hence, it is increasingly vital to have adjustment at hand for cross-sectional NI surveillance.
In conclusion, higher crude infection rates in the present prevalence study suggested a health care quality problem in large compared with small Swiss hospitals. Further analysis demonstrated this difference to be due to case mix instead. Assessment of hospital performance is a complex task, and the impact of case mix should not be underestimated.
Accepted for publication April 3, 2002.
This study was supported by the Swiss Society of Infectiology, the Swiss Society of Hospital Hygiene, the Federal Office of Public Health, and educational grants from the following companies: Wyeth-Lederle, Zug, Switzerland; Aventis Pharma Deutschland GmbH, Bad Soden, Germany; Bayer AG Pharma, Zürich, Switzerland; B. Braun Medical, Emmenbrücke, Switzerland; Bristol-Myers-Squibb, Baar, Switzerland; Henkel-Ecolab, Issy Les Moulineaux, France; Johnson-Johnson, Spreitenbach, Switzerland; Merck Sharp & Dohme Chibret, Glattbrugg, Switzerland; Pfizer, Zürich; Hoffmann-La Roche, Basel, Switzerland; and GlaxoSmithKline, Schönbühl, Switzerland.
We thank Rosemary Sudan for her editorial assistance and the following infection control professionals or physicians who participated in data collection and management: Christoph Akapko, RN; Anna Alexiou, RN; Regula Amiet, RN; Liliane Blanchard, RN; Nicole Bongi, RN; Guitte Bossueat, RN; Dominique Bruttin, RN; Marie-Noëlle Chraïti, RN; Nadia Colaizzi (database manager); Carlo Colombo, RN; Marc Dangel, RN; Luca Donati, MD; Marie-Christine Eisenring, RN; Mario Franciolli, MD; Josiane Galeazzi, RN; Heidi Giger, RN; Anna Gruber, RN; Nicole Henry, RN; Pascale Herrault, RN; Daniela Hutter, RN; Barbera Jaussi, RN; Françoise Kennes, RN; Vittoria La Rocca, RN; Annik Maziéro, RN; Pier-Marie Mazza, RN; Rita Monotti, MD; Monique Mottaz, RN; Lea Mottini, RN; Pijrio Poggioli, RN; Siegrid Pranghofer, MD; Pia Raselli, RN; Valerie Sauvan, RN; Marie-Louise Schaufelberger, RN; Sylvia Schneeberger, RN; Sylvie Touveneau, RN; Gabrielle Tripet, RN; Claudia Van Shrieck, RN; Nicole Weissbrodt, RN; and Gianni Zucchinali, RN. We also thank Pierre-Alain Raeber, MD, Federal Office of Public Health, Bern, and Hans Siegrist, MD, Cantonal Microbiology Laboratories, La Chaux-de-Fonds, Switzerland (Swiss-NOSO Network).
Cantonal Hospital, Lugano: Enos Bernasconi, MD. Cantonal Hospital, Fribourg: Christian Chuard, MD. Regional Hospital, Nyon: Laurent Christin, MD. Cantonal Hospital, St Gallen: Gerhard Eich, MD. Cantonal Hospital, Neuchâtel: Philippe Erard, MD. University Hospital, Lausanne: Patrick Francioli, MD; Christiane Petignat, MD; Philippe Sudre, MD, MSc. University of Geneva Hospitals, Geneva: Stephane Hugonnet, MD, MSc. University Hospital, Bern: Kathrin Muehlemann, MD, MSc, PhD. University Hospital, Zurich: Christian Ruef, MD. University Hospitals, Basel: Andrej Trampuz, MD; Andreas Widmer, MD, MS. Central Institute of the Valais Hospitals, Sion: Nicolas Troillet, MD, MSc.
Corresponding author and reprints: Didier Pittet, MD, MS, Infection Control Program, Department of Internal Medicine, University of Geneva Hospitals, 1211 Geneva 14, Switzerland (http://www.swiss-noso.ch).