- Email: [email protected]
Comparison of rates of nosocomial infections in neonatal intensive care units in the United States
Comparison of Rates of Nosocomial Infections in Neonatal Intensive Care Units in the United States ROBERT P. GAYNES, M.D., WILLIAMJ. MARTONE, M.D., DAVIDH. CULVER, Ph.D.,T. GRACEEMORI,R.N., TERESA C. HORAN,M.P.H.,c.I.c., SHAILENN. BANERJEE, Ph.D.,JONATHAN R. EDWARDS, MS., WILLIAMR. JARVIS,M.D.,JAMESS. TOLSON,B.S., TONYA S. HENDERSON, B.S., JAMESM. HUGHES, M.D., AND THE NATIONAL
NOSOCOMIAL
INFECTIONS
SURVEILLANCE
SYSTEM,
To determine nosocomial infection (NI) rates among neonatal intensive care units (NICUs) that are useful for interhospital comparison, we analyzed data reported in 1986-1990 from 35 hospitals that have level III NICUs and used standard National Nosoc0mia.l Infections Surveillance protocols and NI site definitions. Overall rates of NI were calculated as the number of NI per 100 patients (overall NI patient rates) or the number of NI per 1,000 NICU patient-days (overall NI patient-day rates). A strong positive association was found between overall NI patient rates and the neonates’ average length of stay, a marker for duration of exposure to important risk factors. No correlation was found between overall NI patient-day rates and average length of stay. However, a strong positive correlation between overall NI patient-day rates and a measure of device utilization (total devicedays/total patient-days x 100) was found. Additionally, a positive correlation between overall NI patient rates and device utilization was found. Stratification among the three birthweight groups ( < 1,500 g, 1,5002,500 g, > 2,500 g) did not eliminate the need to control for variations in these factors among NICUs. Device-associated, device-day infection rates, calculated as the number of umbilical or central line-associated bloodstream infections per 1,000 umbilical or central line-days and the number of ventilator-associated pneumonias per 1,000 ventilator days, were not correlated with a unit’s site-specific device utilization. These data suggest that calculation of device-asso-
Centers for Disease Control, Atlanta, Georgia. Hospital Infections Program A07, Building 3, Room B16A, Centers for Disease Control, 1600 Clifton Road N.E., Atlanta, Georgia 30333.
3B-192s
September
16, 1991
The American
Journal
of Medicine
Atlanta,
M.s.,
Georgia
ciated NI rates in NICUs using device-days as the denominator helps to control for the duration of exposure to the primary risk factor and will be more meaningful for purposes of interhospital comparison.
N
osocomial infections (NIs) result in considerable morbidity and mortality among neonates, especially those with birthweights less than 1,500 g [l-51. Previous studies have reported rates of NI in neonatal intensive care units (NICUs) that range from 5.2 to 30.4 infections per 100 patients [l-51. However, these rates have been criticized for a variety of reasons: (1) the definitions of NI were not standardized; (2) the average length of stay of patients differed from hospital to hospital; and (3) there were uncontrolled variations in exposure to known risk factors for NI [6,7]. This report examines different methods of calculating NI rates among Level III NICUs, each of which used standard surveillance protocols and nosocomial infection site definitions from the National Nosocomial Infections Surveillance (NNIS) System. The purpose of this report was to determine which NI rates might allow for meaningful interhospital comparison and, thus, provide a basis for improved prevention of NIs in NICUs.
METHODS The surveillance methods of the Centers for Disease Control’s NNIS system have been previously described [81. Briefly, in the high-risk nursery (HRN) surveillance protocol of the NNIS system, all patients in a participating hospital’s Level III NICU are monitored for NI at all body sites for a minimum of 1 calendar month. Standard definitions of NI are used [9]. Infections that are transmitted transplacentally are not considered, but infections that are transmitted via the birth canal or after admission to the NICU are reported as NI. Patients are monitored from the time they enter the NICU until 48 hours after discharge from the unit. Data that are collected because of its potential use in calculating infection rates include the Volume
91 (suppl3B)
CONFERENCE
number of patients in the unit on the first day of the month, the number of admissions during the month, the total number of patient-days, and the total number of device-days for all patients in the NICU and within each of three birthweight groups (< 1,500 g, 1,500-2,500 g, > 2,500 g). Data allowing the calculation of average length of stay for all infants in the NICU and in each of the three birthweight groups are also collected 181. For each hospital unit, the data collected were pooled for all reporting months. Overall NI rates were then calculated by dividing the total number of NIs that occurred by the total number of patients present in the NICU and multiplying by 100 (overall NI patient rate) and by dividing the total number of NIs that occurred by the total number of patient-days and multiplying by 1,000 (overall NI patient-day rate). Overall NI patient rates or overall NI patient-day rates were not calculated for hospital units that did not report at least 50 patients or at least 50 patient-days, respectively. The overall NI patient rates and overall NI patient-day rates were also calculated in each of the three birthweight strata. Device-associated, device-day infection rates were calculated by dividing the total number of deviceassociated infections (either the number of umbilical or central line-associated bloodstream infections (BSI) or the number of ventilator-associated pneumonias) by the total number of device days (umbilical or central line days or ventilator days) and multiplying by 1,000. Hospital units that did not report at least 50 device-days were excluded from analysis. Three factors were examined for their potential confounding influence when performing interhospita1 comparison of these rates. The first factor was average length of stay of neonates. Secondly, we examined a measure termed the “device utilization” which was calculated by dividing the total number of umbilical or central line days plus ventilator-days for a hospital’s NICU by the total number of patient-days and multiplying by 100. When examining device-associated, device-day infection rates, the site-specific device utilization was calculated as follows: for umbilical or central lines, the total number of umbilical or central line days was divided by the total number of patient days and multiplied by 100; for ventilators, the total number of ventilator days was divided by the total number of patient days and multiplied by 100. Finally, we examined the potential confounding role of birthweight ( < 1,500 g, 1,500-2,500 g, and > 2,500 g) when comparing these rates. To see more clearly the effect that the choice of denominator had on infection rates, we compared September
16, 1991
9 50.0 5 'G 22 g
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
r = 0.44 p
40.0
.
i g se 5 1
30.0.
20.0-
5 88 z
10.0.
. .
. l .
.
l
.
‘**
B
9
5.0
10.0
15.0 20.0 25.0 30.0 35.0 40.0 Average Length of Stay (days)
45.0
Figure 1. Overall nosocomial infection rates in the neonatal care units (NICUs) plotted against average length of stay.
50.0
intensivi
the rank order of the hospitals on the deviceassociated, device-day rates (e.g., umbilical or central line-associated BSI rates) already defined with the rank ordering of the hospitals on a corresponding patient rate. Although many hospitals calculate primary BSI rates (number of primary BSIs/ number of patients x loo), we chose to use a device-associated patient rate for comparison. Using a device-associated patient rate, defined here as the number of device-associated infections (BSI or pneumonia) divided by the number of patients present in the NICU, x 100, allows determination of the effect that the choice of denominator had on infection rates without the influence of any changes in the numerator, i.e., the number of infections. Statistical methods employed included the use of Spearman rank correlation coefficients, the Wilcoxon signed rank test, the paired t-test, and linear model analyses [lo].
RESULTS From 1986-1990, 35 hospitals reported HRN component data covering 452 calendar months. Information was collected on 3,206 NI occurring among 24,480 neonates over 221,710 patient-days during this period. When the overall NI patient rate was plotted against the average length of stay, a strong positive correlation was demonstrated (r = 0.44; p = 0.01) (Figure 1). Additionally, the overall NI patient rate was positively correlated with the device utilization (r = 0.34; p = 0.05). In contrast, when the overall NI patient-day rate was plotted against average length of stay, no significant correlation was found (r = -0.08; p = 0.65). Although overall NI patient-day rates were not correlated with average length of stay, a strong positive correlation between these rates and device utilization remained (r = 0.41, p = 0.01). In addition, we found a wide variation among the Level III NICUs in device utilization, ranging from 6 to 168
The American
Journal
of Medicine
Volume
91 (suppl
38)
36-1938
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
Birth Weight 21500 am
igure 2. Umblkal or central We-associated bloodstream lntection rates, by birthweight for neonatal intensive care units. On the top, the rates are for birthweight < 1,500 g; on the bottom, for birthweight 2 1,500 g.
device-days per 100 patient-days. Stratification among the three birthweight groups (< 1,500 g, 1,500-2,500 g, >2,500 g) did not eliminate the need to control for variations in these factors among NICUs. Since interhospital comparison of overall NI patient-day rates were still confounded by variation among NICUs in device utilization, we directed our attention to the device-associated, deviceday infection rates. When the umbilical or central line-associated BSI rates and ventilator-associated pneumonia rates were compared with average length of stay or their respective device-specific measures for device utilization, no significant correlations were found for the NICU overall or in any of the birthweight categories with one exception. Umbilical or central line-associated BSI rates in the < 1,500 g birthweight category were positively correlated with average length of stay (r = 0.56; p = 0.001). Thus, device-associated, device-day infection rates appear to control for variations among NICUs in the extent of device utilization and average length of stay with one possible exception. We further examined the influence of birthweight on the device-associated, device-day infection rates. For umbilical or central line-associated BSI rates, no significant difference was found in the distribution of these rates between the 1,5002,500 g birthweight group and the > 2,500 g group (p >0.05). Therefore, we pooled the data from these two birthweight groups. However, the distribution of umbilical or central line-associated BSI rates in this pooled group differed significantly from the distribution of rates among infants in the < 1,500 g birthweight group (p = O.OOOl), the medians being 5.1 and 14.6 umbilical or central line-associated BSIs per 1,000 umbilical or central line days, respectively (Figure 2). These distributions identified hospitals that were “outliers,” i.e., 3B-194s
September
16,
1991
The American
Journal
of Medicine
hospitals that had rates significantly above rates of similar units. A single hospital with rates of 33.2 umbilical or central line-associated BSIs per 1,000 umbilical or central line days for birthweight < 1,500 g and 24.4 umbilical or central lineassociated BSIs per 1,000 umbilical or central line days for birthweight 2 1,500 g was identified as an outlier and was contacted. The infection control personnel then examined the umbilical or central line-associated BSI rates and established a multidisciplinary committee to investigate the situation in the NICU. When the number of ventilator-associated pneumonias per 1,000 ventilator days was examined, the distributions did not differ significantly among any of the birthweight groups (p = 0.65). Therefore, the rates in the three birthweight groups were pooled for each hospital, yielding a single distribution of ventilator-associated pneumonia rates (Figure 3). One hospital with a rate of 20.8 ventilator-associated pneumonias per 1,000 ventilator days had experienced an outbreak of respiratory syncytial virus (RSV) infection, which accounted for the high rate. Since then, the infection control measures for RSV have been emphasized, particularly during the winter months, with a subsequent drop in the infection rate. The choice of denominator had a dramatic effect on rank ordering of the infection rates by hospital. We compared the rank ordering of the deviceassociated, device-day rates with the rank ordering of the device-associated patient rates. For illustration purposes, the comparison is shown only for the umbilical or central line-associated BSI rates among neonates with birthweight 2 1,500 g (Figure 4), although similar changes in the rank order of rates by hospital were seen for the other birthweight group (< 1,500 g) and for ventilatorassociated pneumonia rates. The rank ordering of the hospitals changed significantly when one compared the umbilical or central line-associated BSI rate, which uses umbilical or central line-days in the denominator, with the umbilical or central line-associated BSI patient rate, which uses patients in the denominator. The Spearman rank correlation coefficient between the hospital ranks was 0.82; if the rank order had not changed, the Spearman rank correlation coefficient would have been 1.0. Hospital B, which was near the median in the distribution of umbilical or central lineassociated BSI patient rates, had a rate that was in the 90th percentile among the umbilical or central line-associated BSI umbilical or central line-day rates (Figure 4). In contrast, hospital G was near the 90th percentile in the distribution of umbilical or central line-associated BSI patient rates, but Volume
91 (suppl
3B)
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
3 + 5
10
15
20
25
30
35
40
45
50
55
Median = 3.3 Number of Ventilator-Associated
igure 3. Ventilator-associated days, for neonatal intensive
Pneumonias Per 1000 Ventilator-Days
pneumonias, per 1,000 care units for all birthweight
ventilatorcategories.
had a rate that was near the median in the distribution of the umbilical or central lineassociated BSI umbilical or central line-day rates. The explanation for hospital G lies in noting that umbilical or central line utilization for this hospital was extremely high (Figure 4). Controlling for exposure to umbilical or central line, hospital G’s umbilical or central line-associated BSI umbilical or central line-day rate approaches the median.
igure 4. Frequency histograms. Top: Umbilical or central line bloodstream infection rates for neonates (birthweight 2 1,500 g) per 100 patients in the high-risk nursery (HRN). The letters A through J represent rates for hospitals whose location is also shown on the other histograms. Middle: This histogram represents umbilical or central line-associated BSls per 1,000 umbilical or central line-days in the HRN (see text for definition of these rates). Bottom: Umbilical or central line (umbilical or central line-days/100 patient-days in HRN for birthweight 2 1,500 g). *Hospital I was excluded from the top histogram. Since < 50 patients were reported from this hospital for the t 1,500 g birthweight group, accurate assessment of the umbilical or central line-associated BSI patient rate was not possible.
COMMENTS Our results suggest that the calculation of overall NI patient-day rates is preferred over overall NI patient rates. Although overall NI patient rates are commonly reported in the literature [l-5], it may be misleading to compare such rates without controlling for differences in average length of stay at different institutions. These data are often unknown. The overall NI patient-day rate at least partially controls for variations in average length of stay. In the NICU, however, even this rate must, at the very least, be adjusted for the NICU’s device utilization. The device utilization appears to be a marker for the type of NICU and may indicate the unit’s invasive practices or the overall average severity of illness of the infants. The wide variation in the device utilization among NICUs in our study suggests a wide variation in invasive practices in NICUs in NNIS hospitals, in severity of illness of patients in Level III NICUs in NNIS hospitals, or both. The device-associated, device-day rates, i.e., the number of umbilical or central line-associated BSIs per 1,000 umbilical or central line days and the number of ventilator-associated pneumonias per 1,000 ventilator days, may allow for more meaningful interhospital comparison of rates. With the exception of the umbilical or central line-associated BSI rates in the < 1,500 g birthweight group,
these device-associated, device-day rates are uncorrelated with average length of stay or device utilization, suggesting that the denominator (devicedays) adequately controls for exposure to the major risk factor for each infection. The residual correlation between umbilical or central line-associated BSI rates and average length of stay in the < 1,500 g birthweight group may indicate that average length of stay serves as a proxy for another risk factor in this extremely low birthweight group. Further adjustment may be needed to eliminate the confounding influence of average length of stay or this other risk factor. This risk factor may relate to a refinement of catheter-days (e.g., duration of catheterization for an individual catheter) or the intrinsic susceptibility to BSI of the neonates of the < 1,500 g birthweight group. For example, one recent study suggested that receipt of lipid emulsions in neonates is an independent risk factor for primary bacteremia with coagulase-negative staphylococci [ill. This may explain, in part, variations in NICUs’ umbilical or central line-associated BSI rates. That same study showed that smaller neonates received more lipid emulsion, which may explain our finding that smaller birthweights had higher umbilical or central lineassociated BSI rates. Alternatively, the altered immune status of extremely low birthweight infants may explain this finding 1121. Whatever the explanation, our data do suggest that stratification
September
16, 1991
The American
Journal
of Medicine
Volume
91 (suppl3B)
3B-19%
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
for extremely low birthweight ( < 1,500 g) is needed when comparing umbilical or central line-associated BSI rates. The use of comparative data, as in Figures 2 and 3, is critical to the interpretation of an NICU’s device-associated, device-day rate. However, knowledge of an outlier status does not define a problem for the hospital. Rather, it suggests an area for investigation by the hospital. One hospital recently reported on the utility of the central line-associated BSI rate in their adult intensive care unit [ 131. Having been identified as an outlier in comparison to other hospitals participating in the NNIS system, the hospital identified and corrected a problem, marking the first published use of deviceassociated, device-day rates in the NNIS system in the prevention of NIs. Finally, our results have general implications in the field of hospital epidemiology. The changes in rank order of the infection rates by hospital in Figure 4 demonstrated significant differences when comparing hospital’s device-associated NI patient rate with device-associated, device-day rates. For example, hospital G’s rate was near the median of one distribution but an outlier in the distribution of umbilical or central line-associated patient rates. This implies that outcome measures of quality of care may require examination of more than merely the infection (or complication) rate. An examination of the appropriateness of exposure to the major risk factor (or factors) for each event monitored is also needed. The umbilical or central line utilization ratio of hospital G (over 80%) suggests that an examination of umbilical or central line utilization practices in that hospital may be necessary. After controlling for exposure to the major risk factor or factors, the relative frequency of adverse events can be assessed. For example, our data suggest that site-specific device-associated infection rates calculated using all NICU patients in the denominator do not adequately control for exposure to the device. Using catheterized or ventilated patients in the denominator for these rates may partially control for the exposure to the device but would not control for duration of the exposure. Therefore, the device-associated, device-day rates
3B-196s
September
16,
1991
The American
Journal
of Medicine
would allow for more meaningful interhospital comparison of rates. Assuring quality medical care has been widely embraced and is a priority for the Joint Commission on Accreditation of Healthcare Organizations [141. Our data suggest that attempts to define rates dealing with the complications of hospitalization will be more useful for interhospital comparison if they examine appropriate use of interventions that increase patient risk, e.g., central intravenous catheterization, and rates of occurrence of adverse events that attempt to control for exposure to the major risk factor or factors. Our analysis suggests that failure to do both may make interhospital comparisons meaningless or even misleading.
REFERENCES 1. Hemming VG, Overall JC, Britt MR. Nosocomial infections in a newborn intensive-care unit. N Engl J Med 1976; 294: 1310-6. 2. Goldmann DA, Durbin WA, Freeman J. Nosocomial infections in a neonatal intensive care unit. J Infect Dis 1981; 144: 449-59. 3. Jarvis WR. Epidemiology of nosocomial infections in pediatric patients. Pediatr Infect Dis 1987; 6: 344-51. 4. Maguire GC, Nordin J, Myers MG, et al. Infections acquired by young infants. Am J Dis Child 1981; 134: 693-8. 5. Hoogkamp-Korstanje JA, Cats B, Senders RC, et al. Analysis of bacterial infections in a neonatal intensive care unit. J Hosp Infect 1982; 3: 275-84. 6. Goldmann DA. Prevention and management of neonatal infections. Infect Dis Clin North Am 1989; 3: 779-813. 7. Fuchs PC. Will the real infection rate please stand up? Infect Control 1987; 8: 235-6. 8. Emori TG, Culver DH, Horan TC, et al. National Nosocomial Infections Surveillance (NNIS) System: Description of surveillance methodology. Am J Infect Control 1991; 19: 19-35. 9. Garner JS, Jarvis WR, Emori TG, Horan TC, Hughes JM. CDC definitions for nosocomial infections, 1988. Am J Infect Control 1988; 16: 128-40. 10. Snedecor GW, Cochran WG: Statistical methods, 7th ed. Ames, IA State University Press, 1980. 11. Freeman J, Goldmann DA, Smith NE, Sidebottom DG, Epstein MF, Platt R. Association of intravenous lipid emulsion and coagulase-negative staphylococcal bacteremia in neonatal intensive care units. N Engl J Med 1990; 323: 301-8. 12. Ballow M, Cates KL, Rowe JC, eta/. Development of the immune system in very low birth weight (less than 1500 g) premature infants: concentrations of plasma immunoglobulin and patterns of infections. Pediatr Res 1986; 20: 899-904. 13. Selva J, Toledo A, Maroney A, Forlenza S. The value of participation in the CD&National Nosocomial Infection Surveillance (NNIS) System in a large teaching hospital. Presented at APIC ‘89: Sixteenth Annual Educational Conference, Reno, May21-26,1989. 14. Joint Commission on Accreditation of Hospitals. The Joint Commission’s: Agenda for Change,” November, 1986.
Volume
91 (suppl
3B)
NOSOCOMIAL
INFECTIONS
SURVEILLANCE
SYSTEM,
To determine nosocomial infection (NI) rates among neonatal intensive care units (NICUs) that are useful for interhospital comparison, we analyzed data reported in 1986-1990 from 35 hospitals that have level III NICUs and used standard National Nosoc0mia.l Infections Surveillance protocols and NI site definitions. Overall rates of NI were calculated as the number of NI per 100 patients (overall NI patient rates) or the number of NI per 1,000 NICU patient-days (overall NI patient-day rates). A strong positive association was found between overall NI patient rates and the neonates’ average length of stay, a marker for duration of exposure to important risk factors. No correlation was found between overall NI patient-day rates and average length of stay. However, a strong positive correlation between overall NI patient-day rates and a measure of device utilization (total devicedays/total patient-days x 100) was found. Additionally, a positive correlation between overall NI patient rates and device utilization was found. Stratification among the three birthweight groups ( < 1,500 g, 1,5002,500 g, > 2,500 g) did not eliminate the need to control for variations in these factors among NICUs. Device-associated, device-day infection rates, calculated as the number of umbilical or central line-associated bloodstream infections per 1,000 umbilical or central line-days and the number of ventilator-associated pneumonias per 1,000 ventilator days, were not correlated with a unit’s site-specific device utilization. These data suggest that calculation of device-asso-
Centers for Disease Control, Atlanta, Georgia. Hospital Infections Program A07, Building 3, Room B16A, Centers for Disease Control, 1600 Clifton Road N.E., Atlanta, Georgia 30333.
3B-192s
September
16, 1991
The American
Journal
of Medicine
Atlanta,
M.s.,
Georgia
ciated NI rates in NICUs using device-days as the denominator helps to control for the duration of exposure to the primary risk factor and will be more meaningful for purposes of interhospital comparison.
N
osocomial infections (NIs) result in considerable morbidity and mortality among neonates, especially those with birthweights less than 1,500 g [l-51. Previous studies have reported rates of NI in neonatal intensive care units (NICUs) that range from 5.2 to 30.4 infections per 100 patients [l-51. However, these rates have been criticized for a variety of reasons: (1) the definitions of NI were not standardized; (2) the average length of stay of patients differed from hospital to hospital; and (3) there were uncontrolled variations in exposure to known risk factors for NI [6,7]. This report examines different methods of calculating NI rates among Level III NICUs, each of which used standard surveillance protocols and nosocomial infection site definitions from the National Nosocomial Infections Surveillance (NNIS) System. The purpose of this report was to determine which NI rates might allow for meaningful interhospital comparison and, thus, provide a basis for improved prevention of NIs in NICUs.
METHODS The surveillance methods of the Centers for Disease Control’s NNIS system have been previously described [81. Briefly, in the high-risk nursery (HRN) surveillance protocol of the NNIS system, all patients in a participating hospital’s Level III NICU are monitored for NI at all body sites for a minimum of 1 calendar month. Standard definitions of NI are used [9]. Infections that are transmitted transplacentally are not considered, but infections that are transmitted via the birth canal or after admission to the NICU are reported as NI. Patients are monitored from the time they enter the NICU until 48 hours after discharge from the unit. Data that are collected because of its potential use in calculating infection rates include the Volume
91 (suppl3B)
CONFERENCE
number of patients in the unit on the first day of the month, the number of admissions during the month, the total number of patient-days, and the total number of device-days for all patients in the NICU and within each of three birthweight groups (< 1,500 g, 1,500-2,500 g, > 2,500 g). Data allowing the calculation of average length of stay for all infants in the NICU and in each of the three birthweight groups are also collected 181. For each hospital unit, the data collected were pooled for all reporting months. Overall NI rates were then calculated by dividing the total number of NIs that occurred by the total number of patients present in the NICU and multiplying by 100 (overall NI patient rate) and by dividing the total number of NIs that occurred by the total number of patient-days and multiplying by 1,000 (overall NI patient-day rate). Overall NI patient rates or overall NI patient-day rates were not calculated for hospital units that did not report at least 50 patients or at least 50 patient-days, respectively. The overall NI patient rates and overall NI patient-day rates were also calculated in each of the three birthweight strata. Device-associated, device-day infection rates were calculated by dividing the total number of deviceassociated infections (either the number of umbilical or central line-associated bloodstream infections (BSI) or the number of ventilator-associated pneumonias) by the total number of device days (umbilical or central line days or ventilator days) and multiplying by 1,000. Hospital units that did not report at least 50 device-days were excluded from analysis. Three factors were examined for their potential confounding influence when performing interhospita1 comparison of these rates. The first factor was average length of stay of neonates. Secondly, we examined a measure termed the “device utilization” which was calculated by dividing the total number of umbilical or central line days plus ventilator-days for a hospital’s NICU by the total number of patient-days and multiplying by 100. When examining device-associated, device-day infection rates, the site-specific device utilization was calculated as follows: for umbilical or central lines, the total number of umbilical or central line days was divided by the total number of patient days and multiplied by 100; for ventilators, the total number of ventilator days was divided by the total number of patient days and multiplied by 100. Finally, we examined the potential confounding role of birthweight ( < 1,500 g, 1,500-2,500 g, and > 2,500 g) when comparing these rates. To see more clearly the effect that the choice of denominator had on infection rates, we compared September
16, 1991
9 50.0 5 'G 22 g
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
r = 0.44 p
40.0
.
i g se 5 1
30.0.
20.0-
5 88 z
10.0.
. .
. l .
.
l
.
‘**
B
9
5.0
10.0
15.0 20.0 25.0 30.0 35.0 40.0 Average Length of Stay (days)
45.0
Figure 1. Overall nosocomial infection rates in the neonatal care units (NICUs) plotted against average length of stay.
50.0
intensivi
the rank order of the hospitals on the deviceassociated, device-day rates (e.g., umbilical or central line-associated BSI rates) already defined with the rank ordering of the hospitals on a corresponding patient rate. Although many hospitals calculate primary BSI rates (number of primary BSIs/ number of patients x loo), we chose to use a device-associated patient rate for comparison. Using a device-associated patient rate, defined here as the number of device-associated infections (BSI or pneumonia) divided by the number of patients present in the NICU, x 100, allows determination of the effect that the choice of denominator had on infection rates without the influence of any changes in the numerator, i.e., the number of infections. Statistical methods employed included the use of Spearman rank correlation coefficients, the Wilcoxon signed rank test, the paired t-test, and linear model analyses [lo].
RESULTS From 1986-1990, 35 hospitals reported HRN component data covering 452 calendar months. Information was collected on 3,206 NI occurring among 24,480 neonates over 221,710 patient-days during this period. When the overall NI patient rate was plotted against the average length of stay, a strong positive correlation was demonstrated (r = 0.44; p = 0.01) (Figure 1). Additionally, the overall NI patient rate was positively correlated with the device utilization (r = 0.34; p = 0.05). In contrast, when the overall NI patient-day rate was plotted against average length of stay, no significant correlation was found (r = -0.08; p = 0.65). Although overall NI patient-day rates were not correlated with average length of stay, a strong positive correlation between these rates and device utilization remained (r = 0.41, p = 0.01). In addition, we found a wide variation among the Level III NICUs in device utilization, ranging from 6 to 168
The American
Journal
of Medicine
Volume
91 (suppl
38)
36-1938
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
Birth Weight 21500 am
igure 2. Umblkal or central We-associated bloodstream lntection rates, by birthweight for neonatal intensive care units. On the top, the rates are for birthweight < 1,500 g; on the bottom, for birthweight 2 1,500 g.
device-days per 100 patient-days. Stratification among the three birthweight groups (< 1,500 g, 1,500-2,500 g, >2,500 g) did not eliminate the need to control for variations in these factors among NICUs. Since interhospital comparison of overall NI patient-day rates were still confounded by variation among NICUs in device utilization, we directed our attention to the device-associated, deviceday infection rates. When the umbilical or central line-associated BSI rates and ventilator-associated pneumonia rates were compared with average length of stay or their respective device-specific measures for device utilization, no significant correlations were found for the NICU overall or in any of the birthweight categories with one exception. Umbilical or central line-associated BSI rates in the < 1,500 g birthweight category were positively correlated with average length of stay (r = 0.56; p = 0.001). Thus, device-associated, device-day infection rates appear to control for variations among NICUs in the extent of device utilization and average length of stay with one possible exception. We further examined the influence of birthweight on the device-associated, device-day infection rates. For umbilical or central line-associated BSI rates, no significant difference was found in the distribution of these rates between the 1,5002,500 g birthweight group and the > 2,500 g group (p >0.05). Therefore, we pooled the data from these two birthweight groups. However, the distribution of umbilical or central line-associated BSI rates in this pooled group differed significantly from the distribution of rates among infants in the < 1,500 g birthweight group (p = O.OOOl), the medians being 5.1 and 14.6 umbilical or central line-associated BSIs per 1,000 umbilical or central line days, respectively (Figure 2). These distributions identified hospitals that were “outliers,” i.e., 3B-194s
September
16,
1991
The American
Journal
of Medicine
hospitals that had rates significantly above rates of similar units. A single hospital with rates of 33.2 umbilical or central line-associated BSIs per 1,000 umbilical or central line days for birthweight < 1,500 g and 24.4 umbilical or central lineassociated BSIs per 1,000 umbilical or central line days for birthweight 2 1,500 g was identified as an outlier and was contacted. The infection control personnel then examined the umbilical or central line-associated BSI rates and established a multidisciplinary committee to investigate the situation in the NICU. When the number of ventilator-associated pneumonias per 1,000 ventilator days was examined, the distributions did not differ significantly among any of the birthweight groups (p = 0.65). Therefore, the rates in the three birthweight groups were pooled for each hospital, yielding a single distribution of ventilator-associated pneumonia rates (Figure 3). One hospital with a rate of 20.8 ventilator-associated pneumonias per 1,000 ventilator days had experienced an outbreak of respiratory syncytial virus (RSV) infection, which accounted for the high rate. Since then, the infection control measures for RSV have been emphasized, particularly during the winter months, with a subsequent drop in the infection rate. The choice of denominator had a dramatic effect on rank ordering of the infection rates by hospital. We compared the rank ordering of the deviceassociated, device-day rates with the rank ordering of the device-associated patient rates. For illustration purposes, the comparison is shown only for the umbilical or central line-associated BSI rates among neonates with birthweight 2 1,500 g (Figure 4), although similar changes in the rank order of rates by hospital were seen for the other birthweight group (< 1,500 g) and for ventilatorassociated pneumonia rates. The rank ordering of the hospitals changed significantly when one compared the umbilical or central line-associated BSI rate, which uses umbilical or central line-days in the denominator, with the umbilical or central line-associated BSI patient rate, which uses patients in the denominator. The Spearman rank correlation coefficient between the hospital ranks was 0.82; if the rank order had not changed, the Spearman rank correlation coefficient would have been 1.0. Hospital B, which was near the median in the distribution of umbilical or central lineassociated BSI patient rates, had a rate that was in the 90th percentile among the umbilical or central line-associated BSI umbilical or central line-day rates (Figure 4). In contrast, hospital G was near the 90th percentile in the distribution of umbilical or central line-associated BSI patient rates, but Volume
91 (suppl
3B)
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
3 + 5
10
15
20
25
30
35
40
45
50
55
Median = 3.3 Number of Ventilator-Associated
igure 3. Ventilator-associated days, for neonatal intensive
Pneumonias Per 1000 Ventilator-Days
pneumonias, per 1,000 care units for all birthweight
ventilatorcategories.
had a rate that was near the median in the distribution of the umbilical or central lineassociated BSI umbilical or central line-day rates. The explanation for hospital G lies in noting that umbilical or central line utilization for this hospital was extremely high (Figure 4). Controlling for exposure to umbilical or central line, hospital G’s umbilical or central line-associated BSI umbilical or central line-day rate approaches the median.
igure 4. Frequency histograms. Top: Umbilical or central line bloodstream infection rates for neonates (birthweight 2 1,500 g) per 100 patients in the high-risk nursery (HRN). The letters A through J represent rates for hospitals whose location is also shown on the other histograms. Middle: This histogram represents umbilical or central line-associated BSls per 1,000 umbilical or central line-days in the HRN (see text for definition of these rates). Bottom: Umbilical or central line (umbilical or central line-days/100 patient-days in HRN for birthweight 2 1,500 g). *Hospital I was excluded from the top histogram. Since < 50 patients were reported from this hospital for the t 1,500 g birthweight group, accurate assessment of the umbilical or central line-associated BSI patient rate was not possible.
COMMENTS Our results suggest that the calculation of overall NI patient-day rates is preferred over overall NI patient rates. Although overall NI patient rates are commonly reported in the literature [l-5], it may be misleading to compare such rates without controlling for differences in average length of stay at different institutions. These data are often unknown. The overall NI patient-day rate at least partially controls for variations in average length of stay. In the NICU, however, even this rate must, at the very least, be adjusted for the NICU’s device utilization. The device utilization appears to be a marker for the type of NICU and may indicate the unit’s invasive practices or the overall average severity of illness of the infants. The wide variation in the device utilization among NICUs in our study suggests a wide variation in invasive practices in NICUs in NNIS hospitals, in severity of illness of patients in Level III NICUs in NNIS hospitals, or both. The device-associated, device-day rates, i.e., the number of umbilical or central line-associated BSIs per 1,000 umbilical or central line days and the number of ventilator-associated pneumonias per 1,000 ventilator days, may allow for more meaningful interhospital comparison of rates. With the exception of the umbilical or central line-associated BSI rates in the < 1,500 g birthweight group,
these device-associated, device-day rates are uncorrelated with average length of stay or device utilization, suggesting that the denominator (devicedays) adequately controls for exposure to the major risk factor for each infection. The residual correlation between umbilical or central line-associated BSI rates and average length of stay in the < 1,500 g birthweight group may indicate that average length of stay serves as a proxy for another risk factor in this extremely low birthweight group. Further adjustment may be needed to eliminate the confounding influence of average length of stay or this other risk factor. This risk factor may relate to a refinement of catheter-days (e.g., duration of catheterization for an individual catheter) or the intrinsic susceptibility to BSI of the neonates of the < 1,500 g birthweight group. For example, one recent study suggested that receipt of lipid emulsions in neonates is an independent risk factor for primary bacteremia with coagulase-negative staphylococci [ill. This may explain, in part, variations in NICUs’ umbilical or central line-associated BSI rates. That same study showed that smaller neonates received more lipid emulsion, which may explain our finding that smaller birthweights had higher umbilical or central lineassociated BSI rates. Alternatively, the altered immune status of extremely low birthweight infants may explain this finding 1121. Whatever the explanation, our data do suggest that stratification
September
16, 1991
The American
Journal
of Medicine
Volume
91 (suppl3B)
3B-19%
CONFERENCE
ON NOSOCOMIAL
INFECTIONS
/ GAYNES
ET AL
for extremely low birthweight ( < 1,500 g) is needed when comparing umbilical or central line-associated BSI rates. The use of comparative data, as in Figures 2 and 3, is critical to the interpretation of an NICU’s device-associated, device-day rate. However, knowledge of an outlier status does not define a problem for the hospital. Rather, it suggests an area for investigation by the hospital. One hospital recently reported on the utility of the central line-associated BSI rate in their adult intensive care unit [ 131. Having been identified as an outlier in comparison to other hospitals participating in the NNIS system, the hospital identified and corrected a problem, marking the first published use of deviceassociated, device-day rates in the NNIS system in the prevention of NIs. Finally, our results have general implications in the field of hospital epidemiology. The changes in rank order of the infection rates by hospital in Figure 4 demonstrated significant differences when comparing hospital’s device-associated NI patient rate with device-associated, device-day rates. For example, hospital G’s rate was near the median of one distribution but an outlier in the distribution of umbilical or central line-associated patient rates. This implies that outcome measures of quality of care may require examination of more than merely the infection (or complication) rate. An examination of the appropriateness of exposure to the major risk factor (or factors) for each event monitored is also needed. The umbilical or central line utilization ratio of hospital G (over 80%) suggests that an examination of umbilical or central line utilization practices in that hospital may be necessary. After controlling for exposure to the major risk factor or factors, the relative frequency of adverse events can be assessed. For example, our data suggest that site-specific device-associated infection rates calculated using all NICU patients in the denominator do not adequately control for exposure to the device. Using catheterized or ventilated patients in the denominator for these rates may partially control for the exposure to the device but would not control for duration of the exposure. Therefore, the device-associated, device-day rates
3B-196s
September
16,
1991
The American
Journal
of Medicine
would allow for more meaningful interhospital comparison of rates. Assuring quality medical care has been widely embraced and is a priority for the Joint Commission on Accreditation of Healthcare Organizations [141. Our data suggest that attempts to define rates dealing with the complications of hospitalization will be more useful for interhospital comparison if they examine appropriate use of interventions that increase patient risk, e.g., central intravenous catheterization, and rates of occurrence of adverse events that attempt to control for exposure to the major risk factor or factors. Our analysis suggests that failure to do both may make interhospital comparisons meaningless or even misleading.
REFERENCES 1. Hemming VG, Overall JC, Britt MR. Nosocomial infections in a newborn intensive-care unit. N Engl J Med 1976; 294: 1310-6. 2. Goldmann DA, Durbin WA, Freeman J. Nosocomial infections in a neonatal intensive care unit. J Infect Dis 1981; 144: 449-59. 3. Jarvis WR. Epidemiology of nosocomial infections in pediatric patients. Pediatr Infect Dis 1987; 6: 344-51. 4. Maguire GC, Nordin J, Myers MG, et al. Infections acquired by young infants. Am J Dis Child 1981; 134: 693-8. 5. Hoogkamp-Korstanje JA, Cats B, Senders RC, et al. Analysis of bacterial infections in a neonatal intensive care unit. J Hosp Infect 1982; 3: 275-84. 6. Goldmann DA. Prevention and management of neonatal infections. Infect Dis Clin North Am 1989; 3: 779-813. 7. Fuchs PC. Will the real infection rate please stand up? Infect Control 1987; 8: 235-6. 8. Emori TG, Culver DH, Horan TC, et al. National Nosocomial Infections Surveillance (NNIS) System: Description of surveillance methodology. Am J Infect Control 1991; 19: 19-35. 9. Garner JS, Jarvis WR, Emori TG, Horan TC, Hughes JM. CDC definitions for nosocomial infections, 1988. Am J Infect Control 1988; 16: 128-40. 10. Snedecor GW, Cochran WG: Statistical methods, 7th ed. Ames, IA State University Press, 1980. 11. Freeman J, Goldmann DA, Smith NE, Sidebottom DG, Epstein MF, Platt R. Association of intravenous lipid emulsion and coagulase-negative staphylococcal bacteremia in neonatal intensive care units. N Engl J Med 1990; 323: 301-8. 12. Ballow M, Cates KL, Rowe JC, eta/. Development of the immune system in very low birth weight (less than 1500 g) premature infants: concentrations of plasma immunoglobulin and patterns of infections. Pediatr Res 1986; 20: 899-904. 13. Selva J, Toledo A, Maroney A, Forlenza S. The value of participation in the CD&National Nosocomial Infection Surveillance (NNIS) System in a large teaching hospital. Presented at APIC ‘89: Sixteenth Annual Educational Conference, Reno, May21-26,1989. 14. Joint Commission on Accreditation of Hospitals. The Joint Commission’s: Agenda for Change,” November, 1986.
Volume
91 (suppl
3B)