THE JOURNAL OF INFECTIOUS DISEASES. VOL. 138, NO.6. DECEMBER 1978
© 1978 by The University of Chicago. 0022-1899178/3806-0019$00.88
Risk Factors for Nosocomial Infection Jonathan Freeman and John E. McGowan, Jr.
From the Hospital Epidemiology Unit, Channing Laboratory, Department of Medicine, Harvard Medical School; Boston City Hospital; and the Peter Bent Brigham Hospital Division of Affiliated Hospitals Center', Inc., Boston, Massachusetts; and the Department of Medicine, Emory University School of Medicine, Atlanta, Georgia
P~tients admitted to hospitals acquire new infections that were not present or incubating at the time of admission at a rate sufficient to present major concerns in terms of excess morbidity, mortality, personal distress, and cost [I, 2]. Recent surveys of the frequency of occurrence of these hospital-acquired or nosocomial infections have recorded large apparent differences among hospitals and patient groups [3-26]. These differences have occasioned much discussion [27-31]. The use of different criteria for nosocomial infection, different survey methods, and different measures of the Received for publication March 24, 1978, and in revised form August 30, 1978. This study was supported by research grant no. HD-03693 from the National Institute of Child Health and Human Development, training grant no. TOI AI-0068 from the National Institute of Allergy and Infectious Diseases, and special fellowship no. 5 F03 HL54288-0l from the Medical Foundation, Inc. This paper was presented in part at a meeting of the American Federation for Clinical Research, Washington, D.C., May 2,1977. This study was prepared in honor of Dr. Edward H. Kass on his 60th birthday. Please address requests for reprints to Dr. Jonathan Freeman, Channing Laboratory, 180 Longwood Avenue, Boston, Massachusetts02115.
Occurrence of such infections have all contributed to the difficulty in understanding and comparing the available data. No accounting has been made of the effect of various lengths of hospital stay on observed rates of infection. No formal numerical study has been made of the host risk factors for nosocomial infection, nor of the effect of differential distribution of the factors determining susceptibility to infection in the patient populations that have been surveyed, although the importance of such underlying factors is universally recognized. Several epidemiologic measures-prevalence, incidence, "attack rate," and number of infections per 100 discharges-have been used in surveys of nosocomial infection. The confusion in the medical literature regarding the application of these different epidemiologic measures was clearly brought out in two recent reviews [32, 33]Most surveys of prevalence [3, 5-15] have defined the point prevalence of nosocomial infection as the proportion of all patients screened at one time who are identified as having active nosocomial infection. In these point prevalence surveys, each patient is counted only once and is designated either as having one or more nosocom-
811
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
Studies of nosocomial infection are difficult to evaluate because of differences in the relative susceptibility of patients to the acquisition of such infections, the use of different methods of surveillance, and the frequent failure to distinguish between measurements of incidence and of prevalence. A standardized approach to these variables has been tested at a large municipal hospital. The systematic identification of potential risk factors for nosocomial infection allows the evaluation of the individual components of risk, valid epidemiologic comparisons between hospital populations, and a more accurate estimate of the potential cost-effectiveness of activities for the control of infection. The data indicate that it is feasible to calculate the relative risk of nosocomial infection for each patient, using basic criteria obtainable at the bedside, supplemented with other generally available information. The risk of infection must be calculated per day rather than per admission to separate the effect of long hospital stay from the effect of high daily risk. Certain underlying diseases, procedures, hospital services, and categories of age, sex, race, and urgency of admission were all found to be significant risk factors for nosocomial infection.
812
major factor in explaining the differences in the risk of acquiring nosocomial infection. It is widely agreed that increased risk is associated with the increasing frequency of use of procedures or drugs that suppress host resistance and with the increasing age of the population within hospitals, but there is no systematic quantitative approach for estimation of the relative importance of these patient-related variables on the risk of acquiring nosocomial infection. A risk factor for nosocomial infection is simply an indicator of risk, or a factor that is associated with nosocomial infection. Such a risk indicator need not be a cause of infection or even precede nosocomial infection. Studies of one cause or one effect of nosocomial infection must control for other causes or other effects of nosocomial infection that may confound comparisons. Thus even effects of infection may be confounding variables. Since this survey was undertaken to identify as many risk indicators or potential confounding variables as possible, factors both preceeding and following infection are included. Some information concerning host factors as risk factors for nosocomial infection has been drawn from nursing treatments and procedures in the form of "Kardex clues" [34]. Underlying disease has been used as a risk factor for mortality in studies of bacteremia [35, 36]. This system has been adapted for use on one hospital medical service [37]. Some quantitation of host risk factors would seem to be essential for comparison of rates of acquisition of nosocomial infection among patients or among hospitals. Such quantitation of factors that can be identified at the bedside prior to the onset of nosocomial infection might also be useful in prediction and prevention of infection. The purpose of this study was to make a systematic survey of risk factors for nosocomial infection in a municipal general hospital that has been the subject of repeated bed-to-bed surveys by specially trained physicians for over a decade [3, 5, 6, 11, 13]. Information used in this study included the observations recorded by the surveying physician at the bedside, as well as computer-generated summaries and line lists of certain standard information obtained for all hospital discharges. After the results of this initial analysis became
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
iaI infections active at the time of survey, or as not having a nosocomial infection active at the time of survey. Patients with prior nosocomial infections who have already finished successful treatment at the time of survey are not recorded as having active nosocomial infections. The proportion of patients with nosocomial infections active at the time of survey is the point prevalence or prevalence proportion. Patients with nosocomial infections acquired on a previous hospitalization are counted as infected if the infections are still active at the time of survey. One may strictly define the incidence of nosocomial infection as the fraction of the population at risk who acquire new nosocomial infections per unit of time exposed [32, 33]. The unit of time that is universally used by hospitals is the hospital day. If this group of patients is restricted to those without previous nosocomial infection, then this definition corresponds to the "force of morbidity" or the "first bout incidence density" and would exclude patients with nosocomial infections acquired during a previous hospitalization from inclusion in the incidence for the current hospitalization. Because this rate representing strict incidence is difficult to interpret clinically, the results of incidence surveys usually have been expressed as "attack rates" or as the number of new infections occurring per 100 discharges [16-26]. The "attack rate" of nosocomial infection is simply the proportion of all patients admitted or discharged who acquire one or more new infections during that episode of hospitalization. The "attack rate" can be interpreted as the average probability of becoming infected during admission. In the studies cited, first nosocomial infections have not been separated from subsequent nosocomial infections, so the "attack rate" in percent is calculated from the number of nosocomial infections per 100 discharges by dividing by the multiplicity of infection, or the number of nosocomial infections per infected patient. Occasionally, both prevalence and some measure related to incidence have been recorded inthe same population [7-9]. Published reports generally include results stratified by hospital service, procedure, age, or other demographic variables, implying recognition of the effect of these factors on the acquisition of nosocomial infection. The variation in susceptibility of the host has been recognized as a
Freeman and McGowan
813
Risk Factors for Nosocomial Infection
available, two additional studies were performed. Using the same data, we investigated the intercorrelation of all the risk factors for nosocomial infection using matched pairs in a case-control study. Finally, the accuracy of the diagnoses of nosocomial infections recorded at discharge by the personal physicians of the patients in the study was investigated. This factor is relevant in relation to the new requirements for infection-related information by accreditation agencies [38]. Methods
10.
Estimation of risk ratio for identification of risk factors. The prevalence odds ratio was used as the crude relative measure of the effect of each variable investigated as a risk factor or risk indicator for nosocomial infection [41, 42]. Under the circumstances of the survey, the hospital was
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
Design of the primary study. The characteristics of patients with nosocomial infections that differentiate them from other patients were studied with use of data obtained from a bed-tobed prevalence survey conducted by one of us (J. E. M.) at Boston City Hospital, Boston, Mass., in January and February 1973 [13]. The analysis for identification of these characteristics or risk factors was not performed until all of the data became available. The activities of Boston City Hospital are monitored by the Massachusetts Hospital Association, which every six months provides member hospitals with computer-generated summaries of demographic and hospital variables for all discharged patients. This service is similar to that provided on a national basis by the Commission on Professional and Hospital Activities as part of the Professional Activity Study [39]. From these summaries information was obtained concerning the nature of admission (emergency or nonemergency), hospital service, outcome of hospitalization (death, transfer, or discharge), age, sex, race, primary discharge diagnosis, secondary discharge diagnosis, and primary operative procedure, both for the patients studied and for all those discharged from Boston City Hospital from January through June 1973. Additional information concerning the nature and duration of all infections, the prevalence of endotracheal tubes, bladder catheters, and iv catheters, and the frequency of consultation with members of the Infectious Disease U nit was obtained at the time of the prevalence survey [13]. Monthly totals of admissions, discharges, and number of days of hospital care rendered were obtained from the administrative offices of the hospital.
Grouping of diagnoses. The Massachusetts Hospital Association codes all discharge diagnoses according to the International Classification of Diseases [40]. These codes were converted into the corresponding numbers of the Hospital Adaptation of the International Classification of Diseases as used in the Professional Activity Study [39]. The days of hospital care rendered by Boston City Hospital taken from the six-month summary were grouped according to the 45 subgrou ps of primary diagnoses used in the Professional Activity Study. For each of the 45 diagnostic subgroups that contained more than one case of nosocomial infection recorded during the prevalence survey [13], the risk ratio for that subgroup was calculated in the manner detailed below. A high-risk stratum was compiled from the diagnostic subgroups that had the highest daily risk of nosocomial infection. Successivegroups of diagnoses were added in order of decreasing risk until patients in the high-risk stratum accounted for about half of all hospital days of care at Boston City Hospital. All diagnostic SUbgroups not included in the highrisk stratum were assigned to the low-risk stratum. The names and codes of the diagnostic subgroups used in the Professional Activity Study that comprise the high-risk stratum include the following: malignancy and hematologic, 140-209 and 280-289; myocardial infarction and other heart, 390-398 and 410-429; cerebrovascular and other vascular, 430-438 and 440-458; pneumonia and bronchtis, 480-491; cholecystitis and calculus and other gastrointestinal, 560-577; genitourinary, 580-607; signs and symptoms, 780-797; fractures and other trauma, 800-959; and adverse effects, 960-999. The low-risk stratum comprised the other 31 diagnostic SUbgroups of the Professional Activity Study not listed above. For simplicity in the final display of the diagnoses that form the high-risk stratum, closely related subgroups arbitrarily have been listed together to reduce the number of separately named groups to fewer than
Freeman and McGowan
814
Table 1. Sample calculation of prevalence odds ratio for determination of risk factors for nosocomial infection from results of a point prevalence survey of 645 hospitalized patients.
Nosocomial infection With Without Total
Risk factor: endotracheal tube With 7 (a) 4 (c) 11 (a+c)
Without 90 (b) 544 (d) 634 (b + d)
Total 97 (a +b) 548 (c + d) 645 (a + b + C + d)
NOTE. Data are given as number of subjects. Prevalence odds ratio = (a X d)/(b x c) = (7 x 544)/(4 x 90) = 10.6 (P 0.10). The patients with nosocomial infections who were included in the line lists appear to be representative of all patients with nosocomial infections active during the prevalence survey with respect to all eight variables for which data were available. Estimation of risk ratio using matched pairs for study of intercorrelation of risk factors. The risk ratio was used as the crude relative measure of the effect of each variable investigated. The risk ratio for matched pairs in a case-control study was calculated in the standard manner [44]. Statistics. The X2 test for either independent or paired samples, with a continuity correction when appropriate, was used for hypothesis testing for the comparison of sample proportions. Durations of infection were compared by use of a ttest [42]. Results
Identification of risk factors for nosocomial infection. Of the 645 patients screened by bedside examination during the 1973 point prevalence survey at Boston City Hospital, 97 (15%) were found to have active nosocomial infections [13]. Eleven of 13 attributes studied for their association with nosocomial infection are described in table 2. These factors are arbitrarily divided into two groups on the basis of time of occurrence and are listed in descending order of relative risk. Risk factors that are labeled "early" are traditionally viewed as potential causes of nosocomial infection, and those labeled "late" as the effects of nosocomial infection. Many of the risk factors can be either a cause or an effect of infection, depending on the circumstances. The relative risk of nosocomial infection per
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
assumed to be in a steady state and the rate at which patients entered the prevalence pool of active nosocomial infections was therefore equal to the rate at which patients left the prevalence pool through recovery or death. In this steady state, the prevalence odds ratio is identical to the ratio of incidences when the durations of infection are equal [43]. An odds ratio of unity implies that the variable investigated as a risk factor carries the average risk of nosocomial infection per hospital day. Since risk per day rather than risk per admission is inherent in the use of the prevalence odds ratio under the circumstances defined above [43], the words "per day" are understood and not repeated with every use of the risk ratio. The frequency of occurrence of nosocomial infection is low enough to satisfy the rare disease assumption [41] (prevalence, ~0;15) [3-15]. An example of the calculation of the prevalence odds ratio is presented in table I. This method was used to obtain all risk ratios presented in tables 2-4. Matching scheme for study of intercorrelation of risk factors. The line lists provided by the Massachusetts Hospital Association were searched to locate the 97 patients with nosocomial infection from the 1973 survey [13], and one control was matched to each case by finding the patient on the list who had the nearest date of discharge and the same primary discharge diagnosis and operative procedure. These controls selected from the line lists were not necessarily included in the original study population. Of the 97 patients identified as having active nosocomial infections in the 1973 survey, 85
815
Risk Factors for Nosocomial Infection
Table 2.
Factors associated with nosocomial infection in 85 patients, grouped by time of occurrence of factor. Percentage of infected patients with factor Risk ratio
Factor
7.2 26.8 35.1 74.1 35.3 68.2 64.7
10.6 5.9 3.4 3.4 2.9 2.0 1.7
67.1
1.6
25.9 71.1 12.9
10.6 5.8 4.0
"The risk ratio is calculated as described in table 1. The average relative risk per day for the entire study population is defined as 1.0. All differences were significant (P
© 1978 by The University of Chicago. 0022-1899178/3806-0019$00.88
Risk Factors for Nosocomial Infection Jonathan Freeman and John E. McGowan, Jr.
From the Hospital Epidemiology Unit, Channing Laboratory, Department of Medicine, Harvard Medical School; Boston City Hospital; and the Peter Bent Brigham Hospital Division of Affiliated Hospitals Center', Inc., Boston, Massachusetts; and the Department of Medicine, Emory University School of Medicine, Atlanta, Georgia
P~tients admitted to hospitals acquire new infections that were not present or incubating at the time of admission at a rate sufficient to present major concerns in terms of excess morbidity, mortality, personal distress, and cost [I, 2]. Recent surveys of the frequency of occurrence of these hospital-acquired or nosocomial infections have recorded large apparent differences among hospitals and patient groups [3-26]. These differences have occasioned much discussion [27-31]. The use of different criteria for nosocomial infection, different survey methods, and different measures of the Received for publication March 24, 1978, and in revised form August 30, 1978. This study was supported by research grant no. HD-03693 from the National Institute of Child Health and Human Development, training grant no. TOI AI-0068 from the National Institute of Allergy and Infectious Diseases, and special fellowship no. 5 F03 HL54288-0l from the Medical Foundation, Inc. This paper was presented in part at a meeting of the American Federation for Clinical Research, Washington, D.C., May 2,1977. This study was prepared in honor of Dr. Edward H. Kass on his 60th birthday. Please address requests for reprints to Dr. Jonathan Freeman, Channing Laboratory, 180 Longwood Avenue, Boston, Massachusetts02115.
Occurrence of such infections have all contributed to the difficulty in understanding and comparing the available data. No accounting has been made of the effect of various lengths of hospital stay on observed rates of infection. No formal numerical study has been made of the host risk factors for nosocomial infection, nor of the effect of differential distribution of the factors determining susceptibility to infection in the patient populations that have been surveyed, although the importance of such underlying factors is universally recognized. Several epidemiologic measures-prevalence, incidence, "attack rate," and number of infections per 100 discharges-have been used in surveys of nosocomial infection. The confusion in the medical literature regarding the application of these different epidemiologic measures was clearly brought out in two recent reviews [32, 33]Most surveys of prevalence [3, 5-15] have defined the point prevalence of nosocomial infection as the proportion of all patients screened at one time who are identified as having active nosocomial infection. In these point prevalence surveys, each patient is counted only once and is designated either as having one or more nosocom-
811
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
Studies of nosocomial infection are difficult to evaluate because of differences in the relative susceptibility of patients to the acquisition of such infections, the use of different methods of surveillance, and the frequent failure to distinguish between measurements of incidence and of prevalence. A standardized approach to these variables has been tested at a large municipal hospital. The systematic identification of potential risk factors for nosocomial infection allows the evaluation of the individual components of risk, valid epidemiologic comparisons between hospital populations, and a more accurate estimate of the potential cost-effectiveness of activities for the control of infection. The data indicate that it is feasible to calculate the relative risk of nosocomial infection for each patient, using basic criteria obtainable at the bedside, supplemented with other generally available information. The risk of infection must be calculated per day rather than per admission to separate the effect of long hospital stay from the effect of high daily risk. Certain underlying diseases, procedures, hospital services, and categories of age, sex, race, and urgency of admission were all found to be significant risk factors for nosocomial infection.
812
major factor in explaining the differences in the risk of acquiring nosocomial infection. It is widely agreed that increased risk is associated with the increasing frequency of use of procedures or drugs that suppress host resistance and with the increasing age of the population within hospitals, but there is no systematic quantitative approach for estimation of the relative importance of these patient-related variables on the risk of acquiring nosocomial infection. A risk factor for nosocomial infection is simply an indicator of risk, or a factor that is associated with nosocomial infection. Such a risk indicator need not be a cause of infection or even precede nosocomial infection. Studies of one cause or one effect of nosocomial infection must control for other causes or other effects of nosocomial infection that may confound comparisons. Thus even effects of infection may be confounding variables. Since this survey was undertaken to identify as many risk indicators or potential confounding variables as possible, factors both preceeding and following infection are included. Some information concerning host factors as risk factors for nosocomial infection has been drawn from nursing treatments and procedures in the form of "Kardex clues" [34]. Underlying disease has been used as a risk factor for mortality in studies of bacteremia [35, 36]. This system has been adapted for use on one hospital medical service [37]. Some quantitation of host risk factors would seem to be essential for comparison of rates of acquisition of nosocomial infection among patients or among hospitals. Such quantitation of factors that can be identified at the bedside prior to the onset of nosocomial infection might also be useful in prediction and prevention of infection. The purpose of this study was to make a systematic survey of risk factors for nosocomial infection in a municipal general hospital that has been the subject of repeated bed-to-bed surveys by specially trained physicians for over a decade [3, 5, 6, 11, 13]. Information used in this study included the observations recorded by the surveying physician at the bedside, as well as computer-generated summaries and line lists of certain standard information obtained for all hospital discharges. After the results of this initial analysis became
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
iaI infections active at the time of survey, or as not having a nosocomial infection active at the time of survey. Patients with prior nosocomial infections who have already finished successful treatment at the time of survey are not recorded as having active nosocomial infections. The proportion of patients with nosocomial infections active at the time of survey is the point prevalence or prevalence proportion. Patients with nosocomial infections acquired on a previous hospitalization are counted as infected if the infections are still active at the time of survey. One may strictly define the incidence of nosocomial infection as the fraction of the population at risk who acquire new nosocomial infections per unit of time exposed [32, 33]. The unit of time that is universally used by hospitals is the hospital day. If this group of patients is restricted to those without previous nosocomial infection, then this definition corresponds to the "force of morbidity" or the "first bout incidence density" and would exclude patients with nosocomial infections acquired during a previous hospitalization from inclusion in the incidence for the current hospitalization. Because this rate representing strict incidence is difficult to interpret clinically, the results of incidence surveys usually have been expressed as "attack rates" or as the number of new infections occurring per 100 discharges [16-26]. The "attack rate" of nosocomial infection is simply the proportion of all patients admitted or discharged who acquire one or more new infections during that episode of hospitalization. The "attack rate" can be interpreted as the average probability of becoming infected during admission. In the studies cited, first nosocomial infections have not been separated from subsequent nosocomial infections, so the "attack rate" in percent is calculated from the number of nosocomial infections per 100 discharges by dividing by the multiplicity of infection, or the number of nosocomial infections per infected patient. Occasionally, both prevalence and some measure related to incidence have been recorded inthe same population [7-9]. Published reports generally include results stratified by hospital service, procedure, age, or other demographic variables, implying recognition of the effect of these factors on the acquisition of nosocomial infection. The variation in susceptibility of the host has been recognized as a
Freeman and McGowan
813
Risk Factors for Nosocomial Infection
available, two additional studies were performed. Using the same data, we investigated the intercorrelation of all the risk factors for nosocomial infection using matched pairs in a case-control study. Finally, the accuracy of the diagnoses of nosocomial infections recorded at discharge by the personal physicians of the patients in the study was investigated. This factor is relevant in relation to the new requirements for infection-related information by accreditation agencies [38]. Methods
10.
Estimation of risk ratio for identification of risk factors. The prevalence odds ratio was used as the crude relative measure of the effect of each variable investigated as a risk factor or risk indicator for nosocomial infection [41, 42]. Under the circumstances of the survey, the hospital was
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
Design of the primary study. The characteristics of patients with nosocomial infections that differentiate them from other patients were studied with use of data obtained from a bed-tobed prevalence survey conducted by one of us (J. E. M.) at Boston City Hospital, Boston, Mass., in January and February 1973 [13]. The analysis for identification of these characteristics or risk factors was not performed until all of the data became available. The activities of Boston City Hospital are monitored by the Massachusetts Hospital Association, which every six months provides member hospitals with computer-generated summaries of demographic and hospital variables for all discharged patients. This service is similar to that provided on a national basis by the Commission on Professional and Hospital Activities as part of the Professional Activity Study [39]. From these summaries information was obtained concerning the nature of admission (emergency or nonemergency), hospital service, outcome of hospitalization (death, transfer, or discharge), age, sex, race, primary discharge diagnosis, secondary discharge diagnosis, and primary operative procedure, both for the patients studied and for all those discharged from Boston City Hospital from January through June 1973. Additional information concerning the nature and duration of all infections, the prevalence of endotracheal tubes, bladder catheters, and iv catheters, and the frequency of consultation with members of the Infectious Disease U nit was obtained at the time of the prevalence survey [13]. Monthly totals of admissions, discharges, and number of days of hospital care rendered were obtained from the administrative offices of the hospital.
Grouping of diagnoses. The Massachusetts Hospital Association codes all discharge diagnoses according to the International Classification of Diseases [40]. These codes were converted into the corresponding numbers of the Hospital Adaptation of the International Classification of Diseases as used in the Professional Activity Study [39]. The days of hospital care rendered by Boston City Hospital taken from the six-month summary were grouped according to the 45 subgrou ps of primary diagnoses used in the Professional Activity Study. For each of the 45 diagnostic subgroups that contained more than one case of nosocomial infection recorded during the prevalence survey [13], the risk ratio for that subgroup was calculated in the manner detailed below. A high-risk stratum was compiled from the diagnostic subgroups that had the highest daily risk of nosocomial infection. Successivegroups of diagnoses were added in order of decreasing risk until patients in the high-risk stratum accounted for about half of all hospital days of care at Boston City Hospital. All diagnostic SUbgroups not included in the highrisk stratum were assigned to the low-risk stratum. The names and codes of the diagnostic subgroups used in the Professional Activity Study that comprise the high-risk stratum include the following: malignancy and hematologic, 140-209 and 280-289; myocardial infarction and other heart, 390-398 and 410-429; cerebrovascular and other vascular, 430-438 and 440-458; pneumonia and bronchtis, 480-491; cholecystitis and calculus and other gastrointestinal, 560-577; genitourinary, 580-607; signs and symptoms, 780-797; fractures and other trauma, 800-959; and adverse effects, 960-999. The low-risk stratum comprised the other 31 diagnostic SUbgroups of the Professional Activity Study not listed above. For simplicity in the final display of the diagnoses that form the high-risk stratum, closely related subgroups arbitrarily have been listed together to reduce the number of separately named groups to fewer than
Freeman and McGowan
814
Table 1. Sample calculation of prevalence odds ratio for determination of risk factors for nosocomial infection from results of a point prevalence survey of 645 hospitalized patients.
Nosocomial infection With Without Total
Risk factor: endotracheal tube With 7 (a) 4 (c) 11 (a+c)
Without 90 (b) 544 (d) 634 (b + d)
Total 97 (a +b) 548 (c + d) 645 (a + b + C + d)
NOTE. Data are given as number of subjects. Prevalence odds ratio = (a X d)/(b x c) = (7 x 544)/(4 x 90) = 10.6 (P 0.10). The patients with nosocomial infections who were included in the line lists appear to be representative of all patients with nosocomial infections active during the prevalence survey with respect to all eight variables for which data were available. Estimation of risk ratio using matched pairs for study of intercorrelation of risk factors. The risk ratio was used as the crude relative measure of the effect of each variable investigated. The risk ratio for matched pairs in a case-control study was calculated in the standard manner [44]. Statistics. The X2 test for either independent or paired samples, with a continuity correction when appropriate, was used for hypothesis testing for the comparison of sample proportions. Durations of infection were compared by use of a ttest [42]. Results
Identification of risk factors for nosocomial infection. Of the 645 patients screened by bedside examination during the 1973 point prevalence survey at Boston City Hospital, 97 (15%) were found to have active nosocomial infections [13]. Eleven of 13 attributes studied for their association with nosocomial infection are described in table 2. These factors are arbitrarily divided into two groups on the basis of time of occurrence and are listed in descending order of relative risk. Risk factors that are labeled "early" are traditionally viewed as potential causes of nosocomial infection, and those labeled "late" as the effects of nosocomial infection. Many of the risk factors can be either a cause or an effect of infection, depending on the circumstances. The relative risk of nosocomial infection per
Downloaded from http://jid.oxfordjournals.org/ at New York University on May 16, 2015
assumed to be in a steady state and the rate at which patients entered the prevalence pool of active nosocomial infections was therefore equal to the rate at which patients left the prevalence pool through recovery or death. In this steady state, the prevalence odds ratio is identical to the ratio of incidences when the durations of infection are equal [43]. An odds ratio of unity implies that the variable investigated as a risk factor carries the average risk of nosocomial infection per hospital day. Since risk per day rather than risk per admission is inherent in the use of the prevalence odds ratio under the circumstances defined above [43], the words "per day" are understood and not repeated with every use of the risk ratio. The frequency of occurrence of nosocomial infection is low enough to satisfy the rare disease assumption [41] (prevalence, ~0;15) [3-15]. An example of the calculation of the prevalence odds ratio is presented in table I. This method was used to obtain all risk ratios presented in tables 2-4. Matching scheme for study of intercorrelation of risk factors. The line lists provided by the Massachusetts Hospital Association were searched to locate the 97 patients with nosocomial infection from the 1973 survey [13], and one control was matched to each case by finding the patient on the list who had the nearest date of discharge and the same primary discharge diagnosis and operative procedure. These controls selected from the line lists were not necessarily included in the original study population. Of the 97 patients identified as having active nosocomial infections in the 1973 survey, 85
815
Risk Factors for Nosocomial Infection
Table 2.
Factors associated with nosocomial infection in 85 patients, grouped by time of occurrence of factor. Percentage of infected patients with factor Risk ratio
Factor
7.2 26.8 35.1 74.1 35.3 68.2 64.7
10.6 5.9 3.4 3.4 2.9 2.0 1.7
67.1
1.6
25.9 71.1 12.9
10.6 5.8 4.0
"The risk ratio is calculated as described in table 1. The average relative risk per day for the entire study population is defined as 1.0. All differences were significant (P
![[The use of incidence density in assessing the risk factors for nosocomial infection].](/assets/img/hold.png)
