Jm. Mylotte, ANALYSIS OF INFECTION-CONTROL SURVEILLANCE DATA IN A LONG-TERM-CARE FACILITY - USE OF THRESHOLD TESTING, Infection control and hospital epidemiology, 17(2), 1996, pp. 101-107
OBJECTIVE: To describe long-term trends in nosocomial infection rates
and the threshold testing method for evaluating nosocomial surveillanc
e data in a long-term-care facility (LTCF). DESIGN: Descriptive epidem
iology of prospectively collected infection control surveillance data
and application of threshold testing for detecting possible outbreaks.
Threshold testing uses the binomial distribution to calculate probabi
lities of infection frequency at selected endemic levels (mean number
of infections per month) and compares these probabilities to observed
infection frequency. SETTING: One hundred twenty-bed LTCF located with
in a public, university-affiliated hospital. PATIENTS AND METHODS: The
study period was 1987 through 1994. Yearly endemic levels of specific
types of infection were calculated and threshold levels were determin
ed using a previously published method. In this study, a probability o
f P=.01 was chosen to determine the threshold at a specific endemic le
vel; if the observed number of infections in a month reached or exceed
ed the threshold level, the likelihood that this occurred by chance al
one was 1% or less. INTERVENTIONS: None. RESULTS: The overall mean nos
ocomial infection rate ranged from three to five episodes per 1,000 re
sident care days per year; mean yearly rates from 1990 onward were hig
her and more stable than those from 1987 to 1989. The most common infe
ctions identified were lower respiratory tract, skin and soft tissue,
urinary tract, and conjunctivitis. For each of these infections, thres
hold levels were calculated periodically, using only the monthly frequ
ency of infection. Despite variations in the yearly mean endemic level
of various infections, threshold levels were stable except for skin a
nd soft-tissue infection. CONCLUSIONS: Threshold testing for analysis
of infection control surveillance data in the LTCF setting is straight
forward and does not require knowledge of statistics, special computer
software, or calculation of rates; given the stable population in the
typical LTCF, threshold testing can be based on variations in the mon
thly count of infections and provides an objective evaluation of surve
illance data and a method to identify when outbreaks may be occurring.