Antimicrobial misuse in patients with positive blood cultures

I

CLINICAL STUDIES

Antimicrobial Wm. CLAIEIORNE DUNAGAN, III, Pharm.D., ED CASABAR, SPITZNAGEL. Ph.D. st. Louis,

Misuse in Patients with Positive Blood Cultures M.D., ROBERTS. Pharm.D., MARC

WOODWARD, Ph.D., GERALD Ph.D., CANDACEA.

D. SMITH,

MEDOFF, M.D., JAMES L. GRAY, LAWRENZ, Pharm.D., EDWARD

M~SSOU~

PURPOSE: Inappropriate antimicrobial use was examined among a randomly and prospectively selected cohort of patients with at least one positive result of blood cultures. This misuse was then analyzed with respect to hospital charges and length of stay (LOS). PATIENTSANDMETHODS: Thestudyconsistedof 70 patients (average age, 58.5 years) who had not undergone bone marrow transplantation. Patient charts were reviewed daily for the following information: clinical signs and symptoms of infection, pertinent laboratory data, culture results, detailed data on each antimicrobial in every antimicrobial regimen and their appropriateness, hospital charges, LOS, diagnostic and procedure codes, and discharge status. Three severity of illness variables were generated. Inappropriate antimicrobial use was described according to one of 12 categories. RESULTS: The percent of antimicrobial misuse, defined as the proportion of days of administration of antimicrobials on which one or more antimicrobials were judged inappropriate, was found to be 22.3%. After adjustment for severity of illness and diagnosis, this-average inappropriateness correlated with 4.2 additional hospitalization days and $5,368additional hospital charges. CONCLUSION: Our results cannot distinguish among several possible reasons for these associations, including direct causality (e.g., toxicity and prolonged hospitalization for antimicrobial use) and indirect links such as inappropriate utilization of other resources and influences of severity of illnesson antimicrobial use not accounted for in our equations. Nevertheless, the magnitude of the association gives import to the desirability of further studies.

From the Infectious Disease Division, Department of Medicine, Washington University School of Medicine (WmCD and GM); Health Administration Program, Washington University School of Medicine (RSW); Department of Pharmacy, Barnes Hospital (JLG and EC): Strategic Planning, Christian Health Services (MDS): Department of Pharmacy, Providence Hospital (CAL); and Mathematics Department, Washington University (ES). This research was supported by a grant from Barnes Hospital to the Washington University School of Medicine. Requests for reprints should be addressed to Wm. Claiborne Dunagan, M.D., Infectious Disease Division, Department of Medicine, Washington University School of Medicine, 660 South Euclid, BOX 8051. St. Louis, Missouri 63110. Manuscript submitted November 21. 1988. and accepted in revised form June 1, 1989.

tudies indicate that about one third of hospitalized S patients receive antimicrobial therapy and up to 65% of such coursesare judged inappropriate in [l-5],

somerespect [2-51. For example, Maki and Schuna [4] found that approximately 26%of patients in a university hospital were given antimicrobial agents, but 41% of these courseswere believed to be either unnecessary or otherwise inappropriate. The consequences of inappropriate antimicrobial use have been lessclearly demonstrated but probably include increased drug costs, longer hospitalizations, unnecessary toxicity, and the selection of resistant organisms [5]. Thus, programs to decrease the inappropriate useof antimicrobials have been implemented in many institutions, including our own [6,7]. For the most part, however, these programs have emphasized reducing hospital pharmacy expenditures by controlling the useof comparatively expensive antimicrobials and have not carefully documented effects on medical care. The indirect cost implications of medically inappropriate antimicrobial therapies are particularly poorly understood. Eisenberg and colleagues [8] did demonstrate that nephrotoxicity associated with aminoglycosideuse increased hospital length of stay (LOS) and hospital charges by an average of $2,501 for affected patients. The substitution of the monobactam aztreonam for an aminoglycoside in a multi-drug regimen may well decreasethe incidence of nephrotoxicity and, thus, the indirect cost of hospitalization, but offsetting indirect costs could be incurred as a result of therapeutic failure if substitution is not appropriate in a particular situation. In general, there is little information in the literature about the cost and outcome implications of antimicrobial therapies that are medically inappropriate, but not inappropriately expensive. Questions about whether errors or oversights in clinical antimicrobial judgments have significant impact on hospital charges, LOS, or outcome remain largely unanswered. This study therefore examined inappropriate antimicrobial use among a prospectively sampled cohort of patients who had at least one positive blood culture result. This misuse was then analyzed with respect to hospital charges and LOS. Although a causal relationship was not established, the degree of antimicrobial usebelieved to be inappropriate wasfound to correlate significantly with both hospital chargesand LOS, even when attempts were made to control for the severity of the patient’s underlying diseaseand acute physiologic status.

PATIENTS AND METHODS Patient Selection All non-bone marrow transplant recipients in Barnes Hospital who had at least one positive blood

September

1989

The American

Journal

of Medicine

Volume

87

253

lNAPPROPRlATE

ANTIMICROBIAL

USE / DUNAGAN

ET AL

Data Collection Pertinent information was collected from each patient’s hospital record at the time of admission to the study. Background information included the patient’s age and sex, the suspected sites of infection, a record of any intensive care unit (ICU) admission, and data pertinent to assessing the severity of illness. Clinical signs and symptoms of infection, pertinent laboratory data, culture results, and detailed information on each antimicrobial agent in every antimicrobial regimen and their appropriateness were collected by a daily review of patient charts. For the purposes of this study, a regimen was defined as the list of all antimicrobials and their corresponding dosages prescribed for a patient at any given time. A regimen was defined as ending on the discontinuation or the addition of any antimicrobial or on a change in the dosage of any of the component drugs. Finally, hospital charges, LOS, diagnostic and procedure codes, and the patient’s discharge status were recorded. Sites of infection included the bloodstream, urinary tract, lower respiratory tract, abdomen, wound, skin, head and neck, bone, and endovascular structures. In five patients, single blood culture isolates were considered contaminants. Four of these patients had other sites of infection. The one patient with no other clinical evidence of infection was listed as having “no site” of infection. Three severity of illness variables were generated. The first categorized patients in one of three severity of illness classes in a manner similar to that of McCabe and Jackson [9]. “Rapidly fatal” illnesses were defined as those with expected mortality within one year and “ultimately fatal” diseases as those with expected death between one and four years. Other conditions were considered nonfatal. The second severity of iilness measure was the patient’s APACHE-II (acute physiologic and chronic health evaluation) score [lo] at the time antimicrobial therapy was initiated. The third measure recorded admission to an ICU. Other well-known severity of illness indicators [ll] could not

be implemented for lack of appropriately trained personnel during the data collection stage of the study. The APACHE-II severity of illness classification system was originally designed to stratify acutely ill patients according to their risk of death and has been validated among ICU patients. Although the current application follows Solomkin et al [12] and Johnson et al [13], who used APACHE to predict mortality among patients irrespective of ICU admission, the APACHEII score has not been thoroughly validated as a means of stratifying all patients according to their severity of illness. In calculating the APACHE-II score, the most abnormal vital signs during the 24 hours preceding antimicrobial therapy were selected. Transiently abnormal values that clearly reflected situations readily corrected prior to antimicrobial institution (e.g., hypotension due to dehydration that resolved with fluid resuscitation) were excluded. Laboratory data used were the last values prior to antimicrobial therapy. In one case, data were not available until 24 hours after initiation of antimicrobial therapy. A number of patients did not have arterial blood gas measurements. In these cases, the arterial partial pressure of oxygen was presumed to be 70 mm Hg (9.31 kPa) or greater. Some patients had received antimicrobial agents prior to entering the study. Oral antimicrobials and antimicrobials discontinued more than 72 hours prior to the start of antimicrobial regimens being reviewed were disregarded in calculating the APACHE-II score. If patients entered the study in the middle of an antimicrobial course, values for the APACHE-II score were taken from data collected before that course of therapy. Each antimicrobial agent was judged appropriate or inappropriate according to a modification of the criteria of Kunin and co-workers 151.A new assessment of each drug was made whenever new information became available. For example, broad-spectrum antimicrobial coverage might be appropriate before the results of cultures were available, but inappropriate after a bacterial pathogen was identified and antimicrobial sensitivities were known. The following categories were used to describe inappropriate antimicrobial use: 1. Agree with the choice of antimicrobial, but the dosage was not appropriately adjusted for serum levels. Acceptable levels were as follows: gentamicin, trough less than 2 CrglmL and peak = 4 to 6 PgImL; amikacin, trough = 5 to 10 pg/mL and peak = 20 to 30 pg/mL; and vancomycin, trough = 5 to 10 rg/mL and peak = 20 to 35 ccg/mL. Peak levels were defined as the one-hour postinfusion level for each drug. Trough levels were defined as the serum level immediately prior to the next dose. 2. Agree with the choice of antimicrobial, but serum levels were not obtained. 3. Agree with the choice of antimicrobial, but a loading dose was needed. 4. Agree with the choice of antimicrobial, but disagree with the duration of therapy chosen. 5. Agree with the choice of antimicrobial, but the dosage was not altered for the patient’s current renal or hepatic function. 6. Agree with the choice of antimicrobial, but the dosage was inappropriate based on the literature.

254

87

culture result between November 24, 1986, and April 4, 1987, were eligible for the study. These inclusion criteria were selected because patients with these characteristics could be objectively identified by review of microbiology records and because positive blood cultures often represented readily diagnosable conditions for which appropriate antimicrobial therapy was relatively well established. Patients in the bone marrow transplant unit were excluded from the study because of the complexity of their condition. Patients were selected from a list of names reported to the research coordinator by the clinical microbiology laboratory at Barnes Hospital each weekday during the period of the study. The randomization table consisted of the integers 1 through 17 permuted repeatedly, The choice of 17 was based on a review that determined that the number of new patients with a positive blood culture would never exceed 17 on any given day. On each day of sampling, the research coordinator chose from an alphabetized list the patient whose position appeared first on the list of permuted numbers. If a permuted number exceeded the number of patients in the list, the next permuted number was substituted until a patient was selected.

September

1989

The American

Journal

of Medicine

Volume

INAPPROPRIATE

7. Disagree with the choice of antimicrobial because it was ineffective against the isolated pathogen based on culture and sensitivity tests. 8. Disagree with the choice of antimicrobial because of toxicity or allergy. 9. Disagree with the need for an antimicrobial. 10. Disagree with the choice of antimicrobial because an equally or more effective drug was available at a lower cost. 11. Disagree with the choice of antimicrobial becauseof the deteriorating clinical status of the patient. 12. Other reason for inappropriate useof antimicrobial. Each day, data were reviewed by the research coordinator (C.A.L.), a doctor of pharmacy with specific training for this task, and by a physician from the Infectious DiseaseDivision (G.M.), and a judgment of appropriateness was made for each antimicrobial. Coded data were subsequently reviewed by two physicians from the Infectious DiseaseDivision (G.M. and W.C.D.) and by a clinical pharmacist with a doctorate in pharmacy (E.C.). Differences of opinion were resolved by discussion and, if necessary, re-review of charts. No intervention in patient care was included in the study design. The charts were reviewed without communicating with the physicians caring for the patient. A provision was made to allow intervention only in casesin which inappropriate treatment was believed to endanger the patient directly; such action was necessary on only one occasion. Analysis Because most patients received multiple antimicrobials aspart ofseveral different regimens, the percentageof antimicrobial usejudged inappropriate could be expressed in a number of different ways. Previous studies have judged antimicrobial use inappropriate without considering the number of days patients were exposed to such inappropriate drugs [l-4,14]. Because we were interested in determining whether there was a correlation between antimicrobial misuse and both LOS and hospital charges,we believed a more continuous representation of antimicrobial inappropriateness would be useful. Therefore, we defined inappropriate antimicrobial useasthe proportion of days of administration of antimicrobials during which the patient received one or more antimicrobial agents inappropriately. Two specifications of the “Charge” and “LOS” variables were considered. The first examined the actual charges and LOS directly. The second included an additional control for diagnosis by examining the difference between each patient’s actual value and the value the hospital expected for the patient’s diagnosisrelated group (DRG). The specific regression equations to be estimated using patient observations are described in the Appendix. Seven patients whose antimicrobial medications were believed inappropriate on the basis of improper duration were excluded from statistical analyses. For these patients, “inappropriate” antimicrobial days were those additional days on which the antimicrobials should have been given but were not. Although the appropriate length of therapy could have been estimated in each case, such an estimate would have

ANTIMICROBIAL

USE / DUNAGAN

ET AL

TABLEI DescriptiveStatistics Patient count Percent male Average age

70 42.9 58.5 years

Infection site counts Bloodstream Urinary tract lower respiratory Abdomen Skin Head and neck Endovasculature Bone Wound No site/prophylactic

;: 17 ; i 2 :

Severity of illness Some ICU days Underlying disease Nonfatal Rapidly fatal Ultimately fatal APACHE-II score (average) Mortality

25.7% 38.6% 40.0% 21.4% 12.66 14.3%

Antimicrobial therapies Total regimen count Total antimicrobial count Total days receivingantimicrobials Average antimicrobials per day Total “antimicrobial days”*

220 432 749.5 1.824 1,367

Average LOS Average hospital charge Total “antimicrobial days” equals the product of the total days receivin antrmrcrobrals trmes the average number of antrmrcrobrals per day

been subject to uncertainties concerning the patient’s response to, and the potential sequelae of, continued parenteral medication. In addition, two patients were excluded from the regression analyses because their LOS was greater than 150 days. Since the percentage of inappropriate antimicrobial use was high in both these patients, their inclusion would have skewed results in favor of higher correlations between antimicrobial inappropriateness and the dependent variables.

RESULTS Patient Characteristics Data on the remaining 70 patients indicated that 30 (42.9%) were men; the patients had an average age of 58.5 years (Table I). Because of the selection criteria, the bloodstream (bacteremia) was the most frequent infection site. Five patients had blood isolates that were considered contaminants, and one of these patients had no other site of infection. Other infection sites appeared typical of what is usually seen among inpatients in a large teaching hospital. Antimicrobials and Appropriateness These 70 patients were treated with 432 antimicrobials delivered aspart of 220 regimens. These patients experienced a total of 749.5 days of antimicrobial therapies with an average of 1.824 antimicrobials per day, giving a product of 1,367 “antimicrobial days” (Table I). Table II details the breakdown of antimicrobial usage for the study. Gentamicin accounted for 24.7% of

September

1989 The American

Journal

of Medicine

Volume

87

255

INAPPROPRIATE

ANTIMICROBIAL

TABLE II Total and Inappropriate

USE / DUNAGAN

Antimicrobial

Days

ET AL

by Antimi-

crobial Total Antimicrobial

Gentamicin Cefazolin Vancomycin Ampicillin Ticarcillin Clindamycin Ceftazidime ThlP/SMZ Ceftrtaxone lmipenem Metronidazole Penicillin Amikacin Nafcillin Erythromycin Cephalexin Cefoxitin Totals

Days

Inappropriate Antimlcrobial Days % of Own Days* Count

count

% of Total

:g”i::

‘% 69.0 59.0

24.7 21.8 14.9 8.0 6.8 5.0 4.3

39.0 17.0 49.5 19.0 19.0 7.0 1.0

11.5 5.7 24.4 17.3 20.5 10.1 1.7

40.0 30.0 25.0 25.0 24.0 23.0 13.0 8.0 8.0 1.0

2.2 2.9 1.8 1.8 1.8 1.7 1.0 0.6 0.6 0.1

10.0 2.0 0.0 0.0 0.0 7.0 0.0 kx l:o

335.: 0:o 0.0 0.0 30.4 0.0 0.0 100.0 100.0

1.367.0

100.0

179.5

13.1

203.0

dP/SMZ = trimethoprim/sulfamethoxazole. Percent of each anWnkrob!al’s total appropriate.

days of therapy

that was considered

total “antimicrobial days.” Cefazolin and vancomycin accounted for 21.8% and 14.9% of “antimicrobial days,” respectively. Other antimicrobial agents each represented less than 10% of total “antimicrobial days.” The inappropriate use of particular agents is also detailed in Table II. Among the antimicrobials accounting for more than 20 “antimicrobial days,” notably misuseddrugs included vancomycin (24%), ampicillin (17%), ticarcillin (21%), ceftriaxone (33%), and amikacin (30.4%). Gentamicin usewas errant for only 12%of days. Ceftazidime, metronidazole, and a number of less frequently employed agents were seldom judged inappropriate. Overall, only 13.1% of “antimicrobial days” were judged inappropriate. Table III summarizes the categories of antimicrobial misuse.Among all antimicrobial days, 4.5% were deemed unnecessary. The dosagefailed to take renal or hepatic dysfunction into consideration on 2.4% of antimicrobial days, while on 1.7% of antimicrobial days, the dosagewas inappropriate basedon available serum levels. Appropriate serum levels were not obtained for 1.5%of the antimicrobial days. For 1.2%of antimicrobial days, a more effective drug was believed to be available. Each of the other reasonsfor inappropriate use accounted for 1% or less of antimicrobial days. Only 0.8%of antimicrobial days represented unnecessarily costly therapy. length of Stay The percentage of days during which one or more antimicrobial agents were judged inappropriate correlated significantly with LOS (Table IV). One result, Equation 1, predicted that patients whose regimens included at least one inappropriate antibiotic every day (100% inappropriate use) had an LOS 22.42 days longer (p = 0.0177) than patients with no inappropriate antibiotic in any regimen. A second result adjusted for differences in DRG256

September

1989

The American

Journal

of Medicine

expected LOS. Equation 2 predicted that patients whose regimens included at least one inappropriate antibiotic every day had an LOS 18.79days longer (p = 0.0426) than patients with no inappropriate antibiotic in any regimen. (The actual observations and a regression line basedon the Intercept and Percent Inappropriate coefficients in Equation 2 are represented graphically in Figure 1.) Both of these results were obtained in equations that controlled for the three measuresof severity of illness. Multiplying the average level of inappropriateness (22.3%) by the 18.79 days longer stay for 100% inappropriateness gave an extra 4.2 days longer stay expected for an average patient. Expressed another way, a patient with an average percent of days of inappropriate antimicrobial therapy (22.3%) had an average increase in LOS of 4.2 days over the DRG-predicted LOS for a patient without any inappropriate days. Charges Antimicrobial inappropriateness alsocorrelated significantly with surprisingly large increases in charges (Table IV). One result, Equation 3, predicted that patients whose regimens included at least one inappropriate antibiotic every day (100% inappropriate use) had actual hospital charges $28,249above (p = 0.0089) those of patients with no inappropriate antibiotic in any regimen. A second result adjusted for differences in DRGexpected hospital charges. Equation 4 predicted that patients whose regimens included at least one inappropriate antibiotic every day had charges $24,073 above (p = 0.0201) those of patients with no inappropriate antibiotic in any regimen. Both of these results were obtained in equations that controlled for three measuresof severity of illness. Multiplying the average level of inappropriateness (22.3%) by the $24,073 additional charges for 100% inappropriateness gave an extra $5,368 charges expected for an average patient. Expressed another way, a patient with an average percentage of days of inappropriate antimicrobial therapy (22.3%) had an increase in charges of $5,368 over the DRG-predicted charges for a patient without any inappropriate days. In addition, two of the severity of illness measures Hospital

~ TABLE Ill Reasons for Antimicrobial

Inappropriateness Antimicrobial Count

Appropriate Dose inappropriate for serum level Serum level not obtained No loading dose Wrongduration Renal/hepatic dose needed Dose inappropriate per literature More effective drug available Toxicity or allergy Antimicrobial not needed Less expensive drug available Patient’s condition deteriorating Other Total Percentages aregreater than 100 because priate for more than one reason. l

Volume 87

1,188.5 23.0 21.0 0.0 0.0 32.5 13.0 16.0 6% 11.0 0.0 2.0 1.371.0 some antimicrobialswere

Days %’ 86.9 1.7 1.5 0.0 0.0 2.4 1.0 1.2 0.2 c 0:o 0.1 100.3 inappro-

correlated highly with hospital charges. Admission to an ICU was associated with $24,222 in actual charges (p = 0.0033) and $21,582 in charges above the DRGpredicted values (Equations 3 and 4). Similarly, a rapidly fatal underlying disease correlated with charge increases above the DRG-predicted charges of approximately $17,700 (p = 0.0433). Neither an ultimately fatal disease nor the patient’s preantimicrobial APACHE-II score demonstrated any significant correlation with hospital charges.

Outcome Fortunately, inappropriate antimicrobial use did not correlate with mortality. The number of deaths in the study was small. Regression analysis with death as the dependent variable revealed significant correlations only with an ICU stay and a rapidly fatal underlying illness. All 10 deaths occurred in patients with rapidly fatal diseases, and seven patients were admitted to an ICU at some point in their hospitalization. There was some concern that patient deaths could have affected results by shortening the LOS for some patients or, on the other hand, by being associated with illnesses requiring very long hospitalizations. Regressions were therefore recalculated excluding those patients who died. There were only small differences between these results and those listed before. In general, the coefficients in these modified equations were larger and more significant.

COMMENTS

Figure

In reporting the results of their study of antimicrobial misuse at the University of Wisconsin Hospital, Maki and Schuna [4] summarized the results of 10 other usage studies in addition to their own. In seven of these studies, the appropriateness of antimicrobial TABLE IV Regression Results*

Independent

Variables

Intercept Standard error Significance level Percent inappropriate Standard error Significance level ICU Standard error Significance level Ultimately fatal Illness Standard error Significance level Rapidly fatal illness Standard error Significance level APACHE-II score Standard error Significance level Age Standard error Significance level Adjusted R* F Probability of larger F

EQ #l: Actual LOS 13.72 7.32 0.0657 22.42 9.20 0.0177 10.44

Dependent Variables EQ #2: EQ #3: EQ #4: LOS Actual Charges above DRG- Hospital above DRGPredlcted Charges Predlcted 6.87 7.21 0.3449 18.79 9.07 0.0426 f3:3;

$3,049 8,316 0.7152 $20,249 10,450 0.0089 $24,222

0.2427 1.57 9.35 0.8670 1;:;;

0.0033 $5.157 10,779 0.6340 $16,927 8,875 0.0612 $353 699 0.6157 $79 204 0.7005 0.3450 6.88 0.0001

6.96 0.1388 2.81 9.49 0.7682 11.09 7.81 0.1608 .19 .62 0.7545 0.26 0.18 0.1582 0.2358 4.45 0.0009

7,910

0.1468 .21 .61 0.7325 0.24 0.18 0.1747 0.1939 3.69 0.0034

Coefficients that are significant at the 0.05 level are printed

($1,067) 8,028 0.8947 sy; 0.6201

s2y93; O.iO64 !kE oi705 “;g O.d433 $321 675 0.6355 KY 0.7264 0.3793 6.21 0.00001 in bold.

1. Actual

and predicted

length of stay (LOS).

use was judged. Rates of inappropriateness ranged from a low of 41% in Maki and Schuna’s [4] own study to a high of 87%in a study by Roberts and Visconti [2]. In these and other studies, however, antimicrobial therapy was judged appropriate or inappropriate without regard to the number of antimicrobials or the duration of treatment [3,5,15]. We believed that this “all or none” representation of appropriateness was not optimal for quantitative analysis. Therefore, “inappropriateness” was defined in our study in a manner that would reflect the duration of inappropriate therapy. Using this definition, the overall percentage of regimen days that were inappropriate for our sample of patients with a positive blood culture result was 22.3%. A figure more comparable to that in previous studies can be generated from our data by calculating the percentage of patients who had some inappropriate antimicrobial therapy. This value is 38.6%, a number similar to Maki and Schuna’s [4] 41%. However, since we judged appropriateness on a daily basis and were very stringent in applying our criteria for inappropriateness,these numbers may not be directly comparable. Although no causality was demonstrated, our results suggestedthat inappropriate antimicrobial therapies may be associated with substantial increases in LOS and hospital charges. In our study population, the average 22.3%inappropriate antimicrobial usewas associated with an increased LOS of 4.2 hospital days and increased charges of $5,368. There are a number of potential explanations for the striking associations reported here; several are illustrated in Figure 2. Clearly, the cost of the individual agents used inappropriately could not account for the

September

1989

The American

Journal

of Medicine

Volume

87

257

INAPPROPRIATE

ANTIMICROBIAL

USE / DUNAGAN

ET AL 1

Figure 2. Possible their

relationship

factors involved in inappropriate to increased LOS and hospital

antibiotic charges.

use and

question did reflect a noncausal association with increased severity of illness. Finally, another possibleexplanation for our results was that antimicrobial misuse may have served as a “marker” for otherwise inefficient or inappropriate care. For example, the physician who prescribed antimicrobials incorrectly may have been more likely (1) to have prescribed other drugs inappropriately, (2) to have ordered unnecessary laboratory tests, or (3) to have failed to respond to other objective data in a standard fashion (arrows G and H). The association uncovered by our data was likely to have represented some combination of all these factors. Thus, the preeminent policy question was to define the factors that were significant contributors to the increased LOS and charges and that were amenable to intervention. Regardlessof the mechanismsthat were operational, the data reported here were inadequate to establish the nature of these associations. Rather, this study defined a quantitative method for representing inappropriate antimicrobial use, described the association of inappropriate antimicrobial use with increased LOS and hospital charges, and, ultimately, provided a rationale for further investigations of these important correlations.

substantial associated charge increases (arrow B). A second possibility was that inappropriate therapy directly or indirectly led to medical problems (e.g., aminoglycoside nephrotoxicity or catheter-related superinfection) that, in turn, increased LOS and hospital charges(arrows C and D). Alternatively, the inappropriate use of antimicrobials may have reflected some feature of the patient or patient care processthat was associatedwith, but not directly due to, inappropriate antimicrobial use.In particular, severity of illnessmay have played a significant role in the inappropriate use of antimicrobial agents (arrow E). This could, of course, have led directly to toxicity and complications, prolonging hospital stay and increasing charges (arrows A through D). However, increased LOS and charges were likely to occur in severely ill patients regardlessof inappropriate antimicrobial use (arrow F), and the direct effects of misused antimicrobials may have been comparatively small. In the data analyses, we attempted to control for severity of illness and thereby eliminate the possibility that our inappropriateness variable actually measured severity of illness. Although an ICU stay and rapidly fatal conditions did correlate with hospital charges, neither these nor any of the insignificant severity of illness variables eliminated the correlations between inappropriateness, LOS, and charges. One interpretation of these data is that all these measuresof severity of illness were flawed. For example, although the preantimicrobial APACHE-II score may have well represented the patient’s acute status at the time antimicrobial therapy was initiated, it may not have been a good indicator of the patient’s overall severity of illness. It is also likely that the useof DRGestimated values as a means of controlling for diagnosis was inadequate to adjust for severity of illness. Many authors, including Horn and colleagues [16], have noted the marked heterogeneity of patients within the sameDRG category, although there is considerable debate about whether severity of illness or other factors are more important sources of variability [16,17]. It is quite possible that despite our best efforts, a significant component of the correlation in

1. Scheckler WE. Bennett JV: Antimicrobial use in seven community hospitals, JAMA 1970; 213: 264-267. 2. Roberts AW, Vrsconti JA: The rational and irrational use of systemic antimicrobial drugs. Am J Hosp Pharm 1972: 29: 828-834. 3. Castle M. Willert CM, Cate TR. Osterhout use at at Duke Duke University University -.S: Antimicrobial ” . ..microbial use Medical Center. JAMA 1977; 237: 2819-2833. _--.!833. 4. Maki OG. Schuna AA: A study of antimicrobial misuse in a university hospital. Am ttimicrobial

258

87

September

1989

The American

Journal

of Medicine

Volume

APPENDIX The following ordinary least-squares multiple regression[18] equation was usedto estimate the partial correlations between antimicrobials used inappropriately and hospital charges and LOS. LOS, CHG = be + biPIN + bsICU + bsRF + b.,UF + bsAPACHE + bsAGE where LOS = hospital LOS (actual or difference from expected value for DRG); CHG = hospital charges (actual or difference from expected value for DRG); bc = intercept term, an average dependent variable value before adjustments for the regressionvariables; PIN = the proportion of all the patient’s antimicrobials that were inappropriate; bi = the change in dependent variable attributable to 100% inappropriate antimicrobials; ICU = a binary variable set to 1 if the patient entered the ICU, and 0 otherwise; bs = the increase in the dependent variable attributable to an ICU stay; RF = a binary variable set to 1 if the patient’s condition was initially judged “rapidly fatal”; bs = the change in the dependent variable associatedwith the patient’s RF status; UF = a binary variable set to 1 if the patient’s condition was initially judged “ultimately fatal”; bq = the change in the dependent variable associated with the patient’s UF status; APACHE = the APACHE-II index of the severity of illness; bs = the change in the dependent variable associated with the APACHE score; AGE = the patient’s age; and bs = the change in the dependent variable associatedwith each additional year of age.

REFERENCES

INAPPROPRIATE J Med Sci 1978: 275: 271-282. 5. Kunin CM, Tupasi T, Craig WA: Use of anlimicrobials: a brief exposition of the problem and some tentative solutions. Ann Intern Med 1973; 79: 555-560. 6. Woodward R’S, Medoff G. Smith MD, Gray JL: Antimicrobial cost savings from formulary restrictions and physician monitoring in a medical-school-affiliated hospital. Am J Med 1987: 83: 817-823. 7. Powers DA: Antimicrobial surveillance in a VAMC teaching hospital-resulting cost avoidance. Drug Intel1 Clin Pharm 1986; 20: 803-805. 8. Eisenberg JM. Koffer H. Glick HA, ef al: What is the cost of nephrotoxicity associated with aminoglycosides? Ann Intern Med 1987; 107: 900-909. 9. McCabe WR. Jackson GG: Gram-negative bacteremia. Arch Intern Med 1962: 110: 847-855. 10. Knaus WA, Draper EA. Wagner DP. Zimmerman JE: APACHE II: a severity of disease classification system. Crit Care Med 1985: 13: 881-892. 11. Horn SD, Horn RA: Reliability and validity of the Severity of Illness Index. Med Care 1986: 24: 159-178. 12. Sotomkin JS. Fant WK. Rivera JO, Alexander JW: Randomized trial of imi-

ANTIMICROBIAL

USE / DUNAGAN

ET AL

penem/cilastatin versus gentamicin and clindamycin in mixed flora infections. Am J Med 1985; 78 (suppl 6A): 85-91. 13. Johnson MH. Gordon PW, Fitzgerald FT: Stratification of prognosis in granulocytopenic patients with hematologtc malignancies using the APACHE-II severity of illness score. Crit Care Med 1986: 14: 693697. 14. Appleby DH, John JF Jr: Use, misuse, and cost of parenteral cephalosporins at a county hospital. South Med J 1980: 73: 1473-1475. 15. Jones SR. Barks J. Bratton T. efatThe effect of an educational program upon hospital antimicrobial use. Am J Med Sci 1977; 273: 79-85. 16. Horn SD. Buckley G. Sharkey PD. Chambers AF. Horn F!A. Schramm GJ: Interhospital differences in patient severity: problems for prospective payment based on diagnosis related groups (DRGs). N Engl J Med 1985: 313: 20-24. 17. Smits HL. Fetter RB. McMahon LF Jr: Variations in resource use within DRGs: the “severity” issue. Health Care Financing Review (Annual Supplement) 1984: 71-78. 18. Johnston J: Econometric methods. New York: McGraw-Hill, 1972; 121185.

September

1989

The American

Journal

of Medicine

Volume

87

259