JSTOR

Methods

The study was performed in the emergency and medical‐surgical ICU of a 550‐bed teaching hospital within the Spanish National Health Service that serves as a referral institution for a population of 350,000. The ICU, which had 8 beds in 1995 and had grown to 12 beds by 2000, cares for critically ill medical, surgical, neurosurgical, and trauma patients.

In May 1995 the hospital’s Nosocomial Infection Surveillance System, created in 1994, added a new component consisting of prospective surveillance in ICUs. The surgical ICU was chosen for the implementation of this component. The definitions of nosocomial pneumonia, primary and secondary bloodstream infection (BSI), and urinary tract infection (UTI) used were identical to those used by the Centers for Disease Control and Prevention (CDC).^5,10,11 The new component began to be used in May 1995.

The same 2 nurses collected information daily about potential extrinsic and intrinsic risk factors and isolation procedures for all new patients admitted to the ICU for a period of 48 hours or more. They reviewed each patient’s clinical history every 48 hours and consulted the patient’s doctor or the supervising nurse when questions arose. When the patient was discharged, the data form was checked and coded; when errors or inconsistencies were detected, the clinical history was reviewed. The first 6 months of surveillance were considered the training period for data collection methods and use of the established criteria. Thus, information collected during this period was not used for the present study.

At monthly intervals, site‐specific NI incidence densities were determined by calculating the device‐days contribution for each patient and the rates of use of medical devices.^5,9 Descriptive statistics were calculated for all variables; when data exhibited a log‐normal distribution, we chose the geometric mean as a measure of centralization. Information analysis and report preparation were performed in 2 time frames. At monthly intervals, we calculated the Poisson probability distribution to determine whether the NI case count was equal to the number of expected cases,¹⁴ which was established as the mean number of NI cases during the previous year, with extreme values excluded. When the observed case count was greater than the expected count, the procedures and circumstances were reviewed to determine whether there had been any deviation from the established protocols. When deemed necessary, an investigation was undertaken.¹⁵ A monthly report was sent to the supervisor and the director of the ICU.

Each year, we created a report that presented the time trend of monthly NI rates, the mean annual rate for each infection site, and a summary of the incidents occurring throughout the year. This report was discussed in a meeting with the entire unit.

Using the information collected through 2000, we compared the yearly and monthly density rates of NI incidence by using Poisson regression analysis. The monthly numbers of cases of pneumonia, BSI, and UTI were defined as dependent variables, with year and month fitted as dummy variables and the year 1996 and the month January used as reference values. Finally, annual trends were tested with an alternative Poisson regression model fitting a single linear trend.^13,16,17 Analyses were performed with Stata (StataCorp LP), release 7.0.

Results

The ICU’s mean monthly volume was 14 patients; this number generated an average of 160 days of ICU hospitalization per month and 7503 days of ICU hospitalization from 1996 through 2000. During this period, 546 patients (361 men, 185 women) were stayed in the ICU for a period of 48 hours or more; of these, 248 experienced NIs, and 169 died.

The incidence density of UTIs associated with indwelling urinary catheters over the entire time series was 8.4 UTIs per 1000 urinary‐catheter days, ranging from 4.4 infections per 1000 urinary‐catheter days in 1996 to 6.4 infections per 1000 urinary‐catheter days in 2000. The incidence density of pneumonia associated with mechanical ventilation for the overall period was 21 cases per 1000 mechanical ventilator–days, ranging from 26 infections per 1000 mechanical ventilator–days in 1996 to 16.6 infections per 1000 mechanical ventilator–days in 2000. The overall incidence density of BSIs associated with central lines was 30 BSIs per 1000 central line–days, ranging from 39 infections per 1000 central line–days in 1996 to 26 infections per 1000 central line–days in 2000. The organisms most frequently isolated were Candida albicans, Pseudomonas aeruginosa, and Enterococcus faecalis for UTIs; Pseudomonas aeruginosa for pneumonia; and Staphylococcus epidermidis and coagulase‐negative Staphylococcus for BSIs (Tables 1 and 2).

Table 1.  Accumulated Incidence of Device‐Related Nosocomial Infections, by Year

Open New Window

Table 2.  Distribution of Major Pathogens Isolated, by Infection Type, 1994 Through 2000

Open New Window

The monthly frequency of mechanical ventilator use and catheter use and the trend for monthly NI rates are shown in Figures 1 through 6. The monthly analyses showed that the observed values surpassed the Poisson probability threshold at 8 time points: November 1997, January and September 1998, and September 1999 for UTIs; September 1998 and December 2000 for pneumonia; and August 1999 and July and August 2000 for BSIs. The Poisson regression trend analysis of NI rates adjusted by year and month (Table 3) showed no changes in the rate of UTIs over the period 1996 through 2000. The overall trends for BSIs and pneumonia indicated decreases in the rates of these infections during this same time period. Variations according to month of the year were evident primarily for pneumonia and BSIs.

(87 KB)

Figure 1.  Mechanical ventilation by month, 1996 through 2000.

Open New Window

(87 KB)

Figure 2.  Central line use by month, 1996 through 2000.

Open New Window

(23 KB)

Figure 3.  Monthly incidence density of mechanical ventilator–associated pneumonia, 1996 through 2000.

Open New Window

(21 KB)

Figure 4.  Monthly incidence density of central line–associated bloodstream infection, 1996 through 2000.

Open New Window

(23 KB)

Figure 5.  Monthly incidence density of urinary catheter–associated urinary tract infection, 1996 through 2000.

Open New Window

(55 KB)

Figure 6.  Incidence density of nosocomial infection, by year. BSI = bloodstream infection; UTI = urinary tract infection.

Open New Window

Table 3.  Poisson Regression Trend Analysis of Site‐Specific Nosocomial Infections, Adjusted by Year and Month

Open New Window

Discussion

The results of this study were obtained with an NI surveillance system in an ICU. The system was based on the approach used by other surveillance systems: review of medical history at the time of admission, follow‐up every 48 hours by Kardex screening, subsequent processing of information, and preparation of monthly and annual reports. Although Poisson regression is a well‐established method of trend analysis,^13,16,18 our literature search yielded no reports of studies that used Poisson distributions for the monthly analysis of hospital infection rates. The method is recommended when studies involve incidence rates and when the probability of occurrence of an event is small (<0.10) and the population is large (>100).^14,19 The Epi Info program (CDC) allows this calculation to be performed simply and establishes the limits of random variation and the probability that the values found are or are not higher than those expected. The statistical tools allow automated analysis at a predefined frequency, as well as automated preparation of reports.

The impact of the use of devices (eg, urinary and central catheters and mechanical ventilation) on the frequency of UTIs, BSIs, and pneumonia in ICUs has been well documented.¹,^3‐10 The incidence density of NIs, adjusted by days of device use, allows us to assess the risk of infection associated with the use of devices. An understanding of the adjusted incidence density trends for NIs allows us to compare data over time and establishes a framework for comparison of the incidence of NIs in our ICU with that in other similar ICUs. The initial data collected in the ICU provide the basis for internal comparisons over time, regardless of whether the data are comparable to data from other hospitals in Spain or from hospitals in other countries. One of the main objectives of a surveillance program is to create this frame of reference. Although there were some monthly and annual variations in UTI rates, no clear trends were detected.

The trend toward a decrease in the incidence of pneumonia and BSIs over the study period (1996 through 2000) indicates the impact of the implementation of the surveillance system. Introducing a surveillance system has an effect, in itself, on the frequency of NIs^20,21 and is related to the use of criteria and standards and to the “observer effect” on department professionals.

For the monthly rate analysis, using the mean value of the previous year as the expected value resulted in successively lower threshold values. This outcome explains the increase in the number of times that the expected monthly threshold was exceeded.

The fact that we did not have initial reference values for our population was a determining factor in our choice of the mean NI incidence density from the previous year as the expected NI value. Because it is an overall estimation, this value does not consider seasonal variations. If previous values had been available for each month, we could have established the monthly expected value as the mean NI incidence density for the same month in previous years. This value would have allowed easy adjustment for seasonal variations related to the disease or to staff rotations. Now that we have the results from the present series, we can contemplate this change in the manner of comparison.

Examination of the NI results at staff meetings showed that the times when the threshold values were exceeded seemed to coincide with vacation periods (ie, when substitute workers were caring for the patients). For this reason, month and year were introduced as dummy variables in the Poisson regression trend analysis. Although the risk of pneumonia seemed to be highest in February and the risk of BSIs seemed to be highest in May, a finding indicating a seasonal effect, our results are not conclusive. Other studies^21‐23 have shown that staff rotation and reduced staffing during vacation periods may affect the NI levels; these studies have produced estimates indicating that the BSI risk is 2.75‐fold greater when a patient is cared for by a substitute nurse. Staff rotation also varies according to the season.

We compared our NI rates with those of other hospitals in Spain. The rates determined by the ENVIN⁴ (Estudio Nacional de Vigilancia e Infección Nosocomial en Servicios de Medicina Intensiva) study (5.6 BSIs per 1000 central line–days; 17.8 cases of pneumonia per 1000 mechanical ventilator–days; 5.48 UTIs per 1000 urinary catheter–days) are lower than ours; this difference could be explained by the type of ICU (medical only) included in the ENVIN study and by the difference in follow‐up time. In our study, follow‐up was initiated after a 48‐hour stay in the ICU and was continued until discharge; both primary and secondary BSIs were included. In the ENVIN study, follow‐up began after a 24‐hour stay in the ICU and continued for as long as 30 days. These variations in the definition of the study period could explain the differences in observations between the 2 studies.

At the international level, the National Nosocomial Infection Surveillance System (NNISS) Report for 2001⁵ classifies ICUs according to the types of patients admitted and the degree of complexity associated with their care. In hospital complexes such as ours, the ICUs handle a mixture of surgical, neurosurgical, and trauma patients; thus, direct comparison of our results with those of the NNISS is not feasible. Our values were higher than the NNISS values and were close to those reported by Urli and colleagues,⁸ although the values reported in the literature vary widely, ranging from 4.9 to 17.4 BSIs per 1000 central line–days; 4.4 to 46 cases of pneumonia per 1000 mechanical ventilator–days; and 4.62 to 28 UTIs per 1000 urinary catheter–days.^2,3,6‐9

The frequency of device use seems not to have changed substantially over the study period (1996 to 2000). A plot of the results suggests seasonal changes, but no detailed study of these variations has been performed. Apart from the use of values adjusted for days at risk because of device use and type of ICU, studies such as ours must establish the point up to which comparison of results between countries is possible when other potential risk factors (eg, those related to the type of health care system, the staff ratio per bed, the staff rotation, and other variables) have not been completely defined.

Our surveillance system covers events that rarely occur; thus, Poisson distribution analysis is applicable because it is particularly suitable for infrequent events. If the resulting variable follows a Poisson distribution, Poisson regression can be used to analyze time trends that are affected by person‐time variation along different periods. If the observed number of events is within the range of expected values, the Poisson distribution allows monthly estimation of trends,^12,13 and such estimates allow the implementation of control measures as soon as an increase in NI rates is detected. Nevertheless, the ability of the surveillance system to detect events may be achieved only at the detriment of other attributes, such as simplicity or timeliness.¹⁰

In our study, events were detected in real time, and the monthly report was a means of feedback for doctors and nurses. This feedback resulted in a decrease in the number of NIs. The impact of staff rotation and departmental overload on the NI rate requires further study, which our department will soon undertake.

Our study showed a progressive reduction in all site‐specific infection rates; the reduction was more pronounced for BSIs than for pneumonia or UTIs. Indeed, it is likely that the high rate of BSIs in our ICU at the beginning of the study was reduced by a decrease in the use of intravascular devices and by the adoption of guidelines for BSI prevention. We use such devices more frequently than US hospitals do,⁵ and our use of these devices seems to follow a seasonal pattern.

A precise knowledge of the specific patterns of disease occurrence in a surgical ICU facilitates proper application of research criteria and identification of changes in the risk for NIs over periods of time. With Poisson regression analysis, increases in the number of cases are detected more effectively, the causes of these increases can be analyzed, and appropriate measures for correcting these increases can be put into place as soon as possible.