We estimate that 1.7 million HAIs occurred in U.S. hospitals in 2002 and were associated with approximately 99,000 deaths. The number of HAIs exceeded the number of cases of any currently notifiable disease,13
and deaths associated with HAIs in hospitals exceeded the number attributable to several of the top ten leading causes of death reported in U.S. vital statistics.14
These estimates are sobering and reinforce the need for improved prevention and surveillance efforts.
These estimates have several limitations. We used 1990s data from hospital-wide surveillance for estimates in 2002 in two areas: infection rates in well-baby nurseries and the distribution of infections by major site. Similar data are not available for a more recent time period. The impact of using old infection rates in well-baby nurseries is minimal because the rate was the lowest among the subpopulations and the total number of infections (19,059) was only 1.1% of the total number of infections. The distribution by major site of infection has a large impact on our estimates because adults and children outside of ICUs accounted for 68.8% of all HAIs. However, there is a lack of data in the United States to suggest that the distribution has changed since the 1990s. In other countries, recent studies provide support for using at least 20% for surgical site infections as a percentage of infections by site.15–17
We may have underestimated the total number of HAIs because surgical site infections are likely underreported in the NNIS system. Most surgical site infections become evident after discharge,18,19
and the completeness and accuracy of post-discharge surveillance is variable in NNIS hospitals. Since we extrapolated from the number of surgical site infections among adults and children outside of ICUs to other infection sites in the same population using NNIS infection rates, the total number might be underestimated. In addition, the decrease in the average hospital length of stay over time might have increased the chance of missing post-discharge surgical site infections.20
There may be other factors, however, such as a higher rate of infections that might result in an overestimate of infections. Therefore, we acknowledge a lack of precision in our estimates.
Infection rates from NNIS hospitals might be different than those in other hospitals. NNIS hospitals are frequently larger, more likely to be affiliated with academic institutions, and located in the mid- and south-Atlantic regions of the United States.21
Rates of HAIs from NNIS hospitals cannot be applied to other healthcare settings.
Finally, our death estimate is limited in that attributable mortality is often difficult to determine from a patient's records. Even for experts, it can be problematic to determine whether patients die from their infection or from their co-morbidities.22
Other methods might be useful to estimate national burden including prevalence surveys and use of surrogate data. Annual prevalence surveys are used to measure the burden of HAIs in many countries. For example, prevalence per 100 admissions was 9.1 in Greece in 1999,15
8.0 in Denmark in 1999,16
7.0 in Spain in 1997,17
5.1 in Norway in 2002,23
and 4.6 in Slovenia in 2001.24
A disadvantage to annual prevalence surveys is that trends might reflect changes in case ascertainment over time rather than true changes in prevalence.25
In addition, data from annual prevalence surveys is less useful for prevention at the facility level. In the United States, prevalence surveys could be used periodically to supplement surveillance data to estimate HAIs in hospitals.
There are several examples of using surrogate data from administrative records for surveillance purposes (e.g., the International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes).26
An evaluation of administrative data to identify injuries in children demonstrated high correlation, sensitivity, and specificity.27
Previously unrecognized differences in sepsis by race were described at the national level using ICD-9-CM codes;28
however, the frequency of HAIs has been more difficult to capture using administrative data. Some HAIs are captured in estimates using the Agency for Healthcare Research and Quality patient safety indicators to estimate patient safety events during Medicare hospitalizations29
and by the mandatory reporting system in Pennsylvania.30
Evaluations of these surrogate systems have so far indicated low sensitivity31,32
and low predictive value.31
An evaluation of the performance of five different measures of bloodstream infections demonstrated improved performance of clinical indicators over administrative indicators.33
In general, the difficulty may be related to the need to determine if the infection is associated with the delivery of healthcare services. Standardized definitions and methods are features associated with the success of HAI surveillance.8
In 1995, CDC estimated that 1.9 million HAIs occurred in U.S. hospitals.34
In 2002, we estimated 1.7 million HAIs. Direct comparison of these estimates should be avoided because both are based on the same hospital-wide surveillance data. However, our estimates of surgical site infections do not depend on hospital-wide data and might be useful to compare. In 1995, we estimated that there were 269,268 surgical site infections, or 2.21% of surgical procedures monitored in NNIS. In 2002, we estimated there were 274,268 surgical site infections, or 1.96% of procedures monitored.
New attention to HAIs and advances in information technology could lead to greater participation of hospitals in organized surveillance efforts.35
At CDC, the evolution of the NNIS system into the NHSN has provided a web-based platform that could help address the need for HAI data at the local, state, and national levels.