Hospitalization Decision in Febrile Intravenous Drug Users JEFFREY H. SAMET, M.D., ABBY SHEVITZ, M.D., JANET FOWLE, M.D., DANIEL E. SINGER, M.D., Boston, Massachusetts PURPOSE: To determine the frequency and to identify predictive factors of occult major illness in febrile intravenous drug users (MXJs) presenting to an emergency room. PATIENTS AND METHODS: A prospective follow-up study was performed involving a consecutive series of 296 presentations of febrile IVDUs to a public hospital emergency room. Follow-up information was obtained for 283 presentations (95.6%). Physicians’ initial assessment was obtained for 204 presentations (70%). Illness was classified as major or minor using explicit criteria. Frequency of occult major ilhmss was determined among patients without obvious major illness on presentation. Risk factors for occult major illness were determined. RESULTS: Occult major illness occurred in 11 patients (4%). This represented 11% of the 103 presentations without obvious major illness on presentation. Pneumonia and cellulitis occurred in 128 of 180 patients (71%) with obvious major ilhmss on presentation. Racteremia was present in seven of 11 patients with occult major illness. Physician predictions were not sufficiently sensitive to provide the basii of the hospitalization decision in febrile IVDUs. The best combination of features suggesting major illness were last use of intravenous drugs less than 5 days and fever greater than 388°C (102.0’F) (sensitivity 64%, specificity 77%). CONCLUSION: Clinical tests and physician assessments are unable to distinguish occult major illness from minor illness among febrile IVDUs at presentation. Occult major illness is best identified by blood culture. If patient follow-up is unreliable, then hospitalization of febrile IVDUs, while awaiting blood culture results, remains a wise policy. From the Section of General Internal Medicine, Department of Medrcine, and Thorndike Memorial Laboratory, Boston City Hosprtal and Boston Universitv School of Medicine. Boston. Massachusetts: and the General Internal Medicine Unrt. Department of Medictne. Massachusetts General Hosprtal and Harvard Medical School, Boston, Massachusetts. Dr. Samet was supported by a training grant for faculty development in general internal medrcrne by the Health Resources Services Admimstration (2 028 PE51006-04). Dr. Singer was partially supported as a Henry J. Kaiser Family Foundation Scholar in General Internal Medicine. Requests for reprints should be addressed to Jeffrey H. Samet, M.D.. Section of General Internal Medicine, Boston City Hospital, 818 Harrison Avenue, Boston, Massachusetts 02118. Manuscript submitted December 11, 1989. and accepted in revised form March 30. 1990. drug use is a prevalent social and mediIsonsntravenous cal disease affecting an estimated 1.2 million perin the United States and associated with many medical complications [l-5]. Fever may be the initial clinical feature of minor illness or life-threatening disease in the intravenous drug user (IVDU) [6]. The presentation of a febrile IVDU is common in some municipal hospitals [7]. The possibility of occult life-threatening disease makes the presentation of an IVDU with fever a difficult clinical problem. This difficulty has led to the recommendation to hospitalize all febrile IVDUs [7]. The necessity of this costly policy is uncertain. Life-threatening diseases described in IVDUs include infective endocarditis, osteomyelitis, and pneumonia [8-lo]. The frequency of endocarditis among febrile IVDUs presenting to an emergency room was found to be 13% in one series [7]. This disease’s presentation may range from fulminant to occult. Although occult life-threatening disease is the chief motivation for universal hospitalization of this population, the frequency of this threat is uncertain. The usefulness of clinical and laboratory information and initial physician impression in the detection of occult life-threatening disease among those febrile IVDUs with apparent minor illness at presentation is uncertain. This information is essential for consideration of a policy concerning selective hospitalization of febrile IVDUs. To address these issues, we performed a prospective follow-up study of a consecutive series of febrile IVDUs presenting to a public hospital emergency room. PATIENTS AND METHODS Patients Between January 11,1988, and January 10,1989, we enrolled all persons who presented to the Boston City Hospital emergency room (ER) with admitted illicit use of intravenous drugs within the previous 6 months and a documented temperature of 378°C (lOO.O’F) or greater. Initial identification of appropriate patients was performed by ER physicians, and daily review of medical admissions was carried out to identify any appropriate patients not enrolled in the ER. Hospitalization of febrile IVDUs was the accepted ER practice during the study period and occurred in 95% of the study patients. Comprehensive accrual of patients in the ER and review of all medical admissions enabled us to enroll essentially all febrile IVDUs. The study protocol was approved by the Human Subjects Institutional Review Board. Data Collection A standard data form was completed at the time of enrollment by the ER physician responsible for the patient (70%) or, if unavailable, by one of the study investigators (30%). This form contained the following: I2 historical or demographic questions (age, sex, race, duration of parenteral drug use, last use of intraJuly 1990 The American Journal of Medicine Volume 89 53 HOSPITALIZATION IN FEBRILE IVDUs / SAMET ET AL venous drugs, history of endocarditis, antibiotic use within 48 hours, hospitalization within 3 months or within 12 months, history of heart murmur, history of cardiac valvular disease, reason for seeking care), 12 specifically defined symptoms (dyspnea, hemoptysis, pleuritic chest pain, cough, vomiting, diarrhea, weight loss, headache, arthralgias, myalgias, night sweats, and sore throat), 12 specifically defined physical signs (temperature, pulse, systolic blood pressure, meningismus, Roth spots, cardiac murmur, petechiae, pustules, Janeway lesions, Osler’s nodes, splenomegaly, and hepatomegaly), and three questions concerning physician’s assessment (percent likely to be bacteremit, percent likely to have infective endocarditis, and primary diagnoses). All forms were reviewed within 1 week by one of the investigators to ensure comprehensiveness and uniformity of data collection. A separate standard follow-up data form was completed by one of the study investigators on each patient. This form contained the following information: from the ER-complete blood count, creatinine, blood urea nitrogen, serum bicarbonate, erythrocyte sedimentation rate; within 72 hours of presentation-albumin, bilirubin, serum glutamic oxaloacetic transaminase @GOT), hepatitis B surface antigen and surface antibody; within 24 hours of presentation-urinalysis and chest radiographic results; during hospitalization-all culture results; and discharge diagnoses. Follow-up data were obtained until the time of discharge by review of patient charts and computerized laboratory and culture results. For analysis, temperature was corrected to equal the oral temperature. As rectal or tympanic membrane temperature determinations reflect core body temperature, each of these observed temperatures was reduced by 0.44”C [ll]. If the route of temperature determination was not specified, then the more commonly used oral route was assumed. All data were entered onto an RBASE System V database development and management system [12]. After entry, all data on the computerized database were checked against the original forms and appropriate corrections were made. Definition of Outcomes Diseases that were believed by the investigators to require hospitalization are classified in this study as “major illness” and those believed not to require hospitalization, as “minor illness.” After completion of the data collection for the entire cohort, patients were categorized into one of three outcome groups: Group l-major illness that was clinically obvious on presentation to the ER; Group 2-minor illness; Group 3major illness that was not clinically apparent (occult) on presentation to the ER. Criteria for major illness included any of the following findings: (1) a new radiographic pulmonary infiltrate, (2) a physician’s determination of cellulitis based on physical evidence of warmth, erythema, and induration, or (3) endocarditis as diagnosed by strict criteria: definite if pathologic confirmation; probable if two or more positive blood cultures and a new regurgitant murmur or peripheral or pulmonary embolic phenomena; possible if two or more positive blood cultures and a known abnormal cardiac valve [13]. Additionally, all IVDUs with two or more positive blood cultures, no known underlying cardiac valvular dis54 July 1990 The American Journal of Medlclne Volume 89 ease, and no discernable focus of infection were defined as having presumed endocarditis. A number of other diagnoses not meeting the aforementioned criteria but requiring hospitalization were also designated as major illness (Appendix). Minor illness (Group 2) included presumed viral syndrome (uncomplicated headache, myalgias, cough, diarrhea, or pharyngitis), streptococcal pharyngitis (throat culture positive for group A /3-hemolytic streptococci), hepatitis B (elevation of SGOT greater than 40 mU/mL and positive serology for hepatitis B surface antigen without complications of ascites, coagulopathy, or altered mental status), and no established diagnosis with spontaneous resolution of the fever (empiric antibiotics administered less than 48 hours). Several other diagnoses not requiring hospitalization were also designated as minor illness (Appendix). Patients with major illness were assigned to Group 1 (obvious on presentation) if any of the following criteria were satisfied: (1) new infiltrate on initial chest radiograph, (2) ER physician’s diagnosis of cellulitis, or (3) diagnosis of another major illness by the ER physician (e.g., abscess, arthritis, meningitis) and a discharge diagnosis of the same major illness. Patients with major illness were assigned to Group 3 (occult at presentation) if the etiology of the fever as ultimately diagnosed was not known at the time of presentation. In the rare event that an ER diagnosis was a major illness that differed from the actual discharge major illness, the patient was classified into Group 1, since a need for admission based on a major illness was obvious in the ER. Patients had incomplete follow-up if no diagnosis was clear at presentation and the clinical course was not known. For purposes of tabulating overall disease frequency, the following rules were followed: only one diagnosis was assigned to each patient; pneumonia was defined as a new radiographic pulmonary infiltrate and blood cultures negative for Staphylococcus aureus. The assessment of final diagnosis was based on all available information. In the event of dual diagnoses, the priority for diagnosis used for tabulating overall disease frequency was in decreasing order: infective endocarditis, septic arthritis, acute renal insufficiency, pneumonia, and cellulitis. Methods of Analysis Comparisons of differences in variables between patient subgroups were performed using the chi-square test or Fisher’s exact test for categoric variables and Student’s t-test for continuous variables [14]. A twotailed p value <0.05 was considered statistically significant. When appropriate, continuous variables were analyzed as categoric variables using clinically meaningful cut-points. We analyzed all variables in the database to compare patients in Group 2 (minor illness) to those in Group 3 (occult major illness). These analyses were carried out in two ways: using all presentations, and using only the first presentation of those patients who had more than one presentation to the ER during the study period. We describe results based on all febrile IVDU presentations, since using only first presentations of patients with multiple ER visits during the study period did not substantially alter the study results. Relative risks and 95% confidence intervals were calculated using the SAS command “proc freq” [15]. HOSPlTALlZATlON RESULTS Description of Cohort Figure 1 provides an overview of the study population. During the 12-month period, 296 IVDUs presented to the ER with a temperature greater than 37PC (lOO.O”F). Follow-up was inadequate in 13 patients who left the hospital against medical advice or were transferred to other facilities. Thus, 283 presentations (95.6% of all presentations) were evaluated. Of these, 180 (64%) had a major illness that was obvious in the ER (Group l), and 103 (36%) had no obvious major illness in the ER (Groups 2 and 3). Of the latter group, 92 (89%) were ultimately diagnosed with minor illness (Group 2) and 11 (11%) were diagnosed with major illness (Group 3). The 296 presentations involved 244 different patients. Table I summarizes demographic and clinical characteristics of the 244-patient cohort. Spectrum of Illness Figure 2 illustrates the distribution of major diseases in which the need to hospitalize the febrile IVDU was apparent at presentation (Group 1). Two diagnoses, pneumonia and cellulitis, were evident on the basis of physical examination or a chest radiograph. These two infectious diseases accounted for 71% (128 of 180) of Group 1. Infective endocarditis and abscess each comprised 6%. Several other diagnoses each accounted for fewer than 4% of Group 1 (Appendix). The 11 patients with occult major illness on presentation (Group 3) are described in Table II. Seven of these 11 patients were bacteremic and diagnosed with endocarditis. Two patients were ultimately diagnosed with pneumonia after initial clear chest radiographs. One patient with nonspecific clinical characteristics consistent with a viral syndrome developed disseminated intravascular coagulation. One patient was found by venogram to have a deep femoral venous thrombosis after more than 7 days of hospitalization for fever without an evident source. Among the 92 patients with minor illness (Group 2), the most common diagnosis was presumed viral syndrome in 44 patients (47%). No established diagnosis with spontaneous resolution of fever occurred in 18 patients (20%). Streptococcal pharyngitis and hepatitis B accounted for four and three patients, respectively. The other diagnoses are included in the Appendix. Predictors of Major Illness Among the 103 febrile IVDUs without apparent major illness at presentation, we identified three univariate predictors of occult major illness (Table III). These included last use of intravenous drugs less than 5 days, fever greater than 38.8”C (102.O”F), and proteinuria greater than trace by dipstick. Temperature greater than 38.8”C (102.0°F) and last use of intravenous drugs less than 5 days were both present in seven of the 11 patients in Group 3 (sensitivity 64%). These same two features were present in 21 of 92 patients in Group 2 (specificity 77%). The other combinations of significant univariate predictors yielded very low sensitivities. Other clinical and laboratory features that did not predict major illness included duration of intravenous drug use, prior history of endocarditis, recent hospitalization, presence of a cardiac murmur, serum bicarbonate, and erythrocyte sedimentation rate. We did 296 IN FEBRILE IVDUs / SAMET ET AL Total # cf presentations to ER I -- 2 1 3 Presentations follow-up 8 3 Presentations evaluated 180(64%) 1 o3(369/O) with Inadequate Presentations wth apparent ma,or Illness (Group 1) Prewntations wlthout Presentations minor illness (Group 2) with apparent 1 1 ~4%/x’:~:.., Presentations wth inapparent major illness (Group 3) Figure 1. Overview of the study population, presentations of febrile intravenous drug emergency users. room TABLEI Demographicand ClinicalCharacteristicsof 244 Febrile IntravenousDrug UsersPresentingto an EmergencyRoom Characteristic Se;,;;e243)* Male Ra;h$;thnicity 82 (33.7%) 161 (66.3%) (n = 240) Black Hispanic Other History of endocarditis (n = 215) Hospitalization within the past 1 year (n = 203) Age (mean years) (n = 226) Duration of intravenous drug use (median years) (n = 215) Time since last use of intravenousdrugs(mediandays)(n = 241) 46 (19.2%) 154 (64.2%) 3; $\.’ 24(li.&) 103 (50.7%) 33.6(range, 19-64) 10 (range, O-40) 1 (range, O-180) * “n” values are the number of different patients about whom the specific characteristic is known TABLEII Diagnosisand Sourceof BacteremiaIn Patientswith OccultMajor Illness Patient Diagnosis 1 Infective endocarditis (possible) Infective endocarditis (definite) Infective endocarditis (presumed) Infective endocarditis (possible) Infective endocarditis (probable) Infective endocarditis (presumed) Infective endocarditis (possible) Pneumonia Pneumonia Disseminated intravascular coagulation Deep venous thrombosis 2 3 4 5 6 7 9” 10 11 July 1990 The American Bacteremia Journal Group G @-hemolytic streptococcus Staphylococcus aureus Staphylococcus aureus Staphylococcus aureus Staphylococcus aureus Staphylococcus aureus Streptococcus viridans None None None None of Medicine Volume 89 55 HOSPITALIZATION IN FEBRILE IVDUs /SAMET ET AL sentations (Group 3) were associated with positive blood cultures, while only one of the 91 minor illness presentations (Group 2) was bacteremic. This single bacteremic patient in Group 2 had a painful exudative pharyngitis and showed growth of group A f)-hemolytic streptococci on both blood and throat cultures. Cellulitis (37%) \ :: ;-~'- c 11 : , 67 \ \ Infective -~~:·~'7 endocarditis (6%) ~~~Abscess (6%) ~~:::.::..\~~:;~~> -~. ·"-"-~~'-.'Y" Other apparent major illnesses (17%) Figure 2. Distribution of major illnesses at presentation in febrile intravenous drug users (Group 1, n = 180). Other apparent major illnesses are listed in the Appendix. TABLE Ill Significant Univariate Predictors of Major Illness (Among 103 Febrile Intravenous Drug Users without Apparent Major Illness at Presentation) Predictors RR* 95% Cit Historical Last use of intravenous drugs <5 days 6.30* 1.05-37.79 Symptoms and signs Cough Headache Sore throat Temperature >38.8°C Meningismus 2.32 0.40 1.09 4.76* 0.66 0.56-9.52 0.13-1.30 0.30-3.93 1.52-14.89 0.04-10.13 Laboratory results White blood cell count> 104 jmm3 Neutrophils >70% Creatinine> 1.3 mgjdL Albumin <3.5 gjdL Proteinuria > trace§ 1.69 1.40 3.33 2.44 4.44* 0.53-5.36 0.42-4.64 0.53-20.97 0.79-7.52 1.27-15.5 • RR denotes relative risk. t Cl denotes confidence interval. The relative risk and their 95% confidence intervals were calculated using SAS ··proc freq."" "<0.05. §Proteinuria scaling is qualitative based on standard dipstick procedure. not analyze human immunodeficiency virus serology as a predictor since this information was available in only 28% of all presentations. Physician Predictions Physicians' estimates of the probability of bacteremia and endocarditis were obtained for 204 presentations (70%). Table IV lists the numbers of patients with bacteremia and endocarditis stratified by the ER physician's estimate of likelihood to have these diseases. Cases were distributed through all quintiles of risk. Although the actual risk of bacteremia or endocarditis tended to increase with higher physician risk category, the relationship was not smooth and was poorly calibrated (i.e., the actual risks were not near the midpoints of the physician risk category). The lowest quintile included 24% of all patients with bacteremia and 37% of all patients with endocarditis, indicating substantial inaccuracy. Bacteremia Blood cultures were obtained in 276 of 283 presentations (98%). Seven of the 11 occult major illness pre56 July 1990 The American Journal of Medicine Volume 89 COMMENTS Our study examined the possibility of more selective hospitalization for febrile IVDUs. We utilized a cohort over three times the size of the previously studied group of febrile IVDUs, and an analytic framework focused on the most critical clinical problem, the identification of major illness among those patients without obvious major illness on presentation [7]. We provide data on which a policy concerning triage of febrile IVDUs can be based. The majority of febrile IVDUs presented with obvious major illness in the ER, and the risk of occult major illness was low. Our results may be used as a basis for two very different approaches. The fiscally conservative would argue that since nearly 90% of patients with no apparent major illness would be appropriately triaged, none should be admitted to the hospital. Outpatient follow-up would be adequate. The medically conservative would suggest that all febrile IVDUs should be hospitalized since greater than 10% of those without obvious major illness may have occult major illness. Outpatient followup is often poor in this particular population. Since bacteremia was present in 64% of cases of occult major illness, the hospitalization decision in febrile IVDUs could depend on the possibility of reliably informing patients of blood culture results. This approach raises the key questions of the clinical importance of occult bacteremia in this setting and the risk of waiting 24 to 48 hours to institute therapy. Although there is uncertainty about both these questions, the clinical meaning of occult bacteremia is generally assumed to indicate serious illness in this population. Institution of antibiotic therapy is not without risk, since central intravenous access is frequently required. Bacteremia was found in patients in Group 3, in part because its presence alone defined "presumed endocarditis." Nonetheless, the occurrence of bacteremia in an IVDU without obvious major illness does change the hospitalization decision. In this study, bacteremia was a striking discriminator between those with minor illness and those with occult major illness (Groups 2 and 3). However, because blood culture information is not available for 24 to 48 hours, it is not useful for initial triage purposes. If a technology that determined bacteremia within a matter of hours could be developed, this might improve triage decisions for febrile IVDUs. Although other clinical or laboratory features (i.e., last use of intravenous drugs less than 5 days, temperature greater than 38.8°C [102.0°F], and proteinuria greater than trace [by dipstick]) were significantly associated with occult major illness, the associations were not sufficiently strong to direct the triage decision. Physicians' estimates of the probability of bacteremia or endocarditis revealed the uncertainty of this clinical problem. Although physicians' higher risk categories did include greater proportions of patients with true disease (Table IV), their assessments were HOSPITALIZATION TABLEIV EmergencyRoom PhysicianPredictionof Bacteremiaand Endocarditis Physician Prediction (% likely to have) Number of Patients with Bacteremia O-20% 21-40% 41-60% Percent of Prediction Category Actually with 9% (7/79)f 5% (2/44)t 29% (10/35)t 61-80% 81-100% 18x(5/28)7 29% (5/17)t 29 (loo)* Total Endocarditir - \Ol”G.J O-20% 21-40% 41-60% U-80% 81-100% 2 (13)* Total 16 (100)’ II, )/,olO11,115553 015 pa 4% (6/ 140)* 8% (2/25)t 21% (5/24)f 9%(1/11)~ 50% (2/4)* LmlL~~~c.. r Values rn parentheses are number of pabents wrth bacteremra rn the specific risk qurntile/number of all patients rn specrfic risk quintile. t Values rn parentheses are number of patients with endocarditis in the specrfic risk qurntile/number of all patients in specrfic rusk quintrle. IN FEBRILE IVDUs / SAMET ET AL warfarin, Pneumocystis carinii pneumonia (PCP) with a negative chest radiograph but positive sputum stain, PCP with a negative chest radiograph treated presumptively with a full course of therapy, hip fracture, small bowel obstruction, coma, pelvic inflammatory disease, pneumonia with a negative chest radiograph but with evidence of consolidation (egophony) on physical examination, hepatitis with ascites, biliary colic, new-onset seizures, disseminated gonococcal infection, and multiple trauma. The following diagnoses were designated as minor illness, in addition to presumed viral syndrome, streptococcal pharyngitis, uncomplicated hepatitis B, and no established diagnosis with spontaneous resolution of fever (values in parentheses are the number of patients involved, if two or more): heroin withdrawal (4), sinusitis (3), adenopathy (2), acquired immunodeficiency syndrome with no other source of fever (2), vaginitis, venous insufficiency, buttock hematoma, prostatitis, superficial thrombophlebitis, secondary syphilis, otitis externa, acute gouty arthritis, uncomplicated pyelonephritis (without vomiting, leukocytosis), mild possible pelvic inflammatory disease, chronic nonhealing right middle lobe infiltrate, gonococcal urethritis, and urinary tract infection. ACKNOWLEDGMENT fraught with error. In particular, physician assessments were too insensitive to provide the basis of the hospitalization decision. This finding confirms the previous experience in a small physician cohort (n = 40) 171. In conclusion, the need for hospitalization is apparent at presentation in the majority of febrile IVDUs. Although we found three clinical or laboratory features associated with infrequent occult major illness, it is unlikely that these or other clinical tests currently available at initial patient evaluation will be adequate to guide triage of febrile IVDUs. Development of a rapid assay for bacteremia may improve triage decisions for these patients and reduce unnecessary hospitalizations. At the present time, the hospitalization decision in the febrile IVDUs rests primarily on the need for patient follow-up after blood culture results are known. If 24- and 48-hour follow-up is not possible, then hospitalization of the febrile IVDU remains a wise policy. APPENDIX: MAJOR AND MINOR ILLNESSES The following diagnoses did not meet the criteria for major illness but were considered to require hospitalization (values in parentheses are the number of patients involved, if two or more): deep venous thrombosis (4), septic arthritis (4), pyelonephritis (3), acute renal insufficiency (2) (new elevation of serum creatinine greater than 1.5 mg/dL [130 rmol/L]), cryptococcal meningitis (2), altered mental status (2), bleeding diathesis (hematocrit = 21%) in a patient receiving We thank the house staff in the Departments of Medrcine, Emergency Medicine. and Surgery at Boston City Hospital for their active parbcipabon in this study. We also thank Donald Craven, M.D., John Noble, M.D.. Kenneth Freedberg, M.D., and John Rrch, M.D., for therrcntical revrew of themanuscnpt, andTerry Field, M.P.H., for her assrstance with the statrstrcal analyses. REFERENCES 1. Hahn RA, Onorato IM. Jones TS. Dougherty J: Prevalence of HIV infectron among intravenous drug users in the Unrted States. JAMA 1989; 261: 2677-2684. 2. Hussey HH. Katz S: lnfectrons resultingfrom narcotic addiction. Am J Med 1950; 9: 186-193. 3. Louria DB. Hensle R. Rose J: The major medical complications of herorn addictron. Ann Intern Med 1967; 67. l-22. 4. Saprra JD: The narcotrc addtct as a medical patrent. Am J Med 1968; 45. 555588. 5. Whrte AG: Medical drsorders in drug addrcts. JAMA 1973; 223: 1469-1471. 6. Tuazon CU, Cardella TA, Sheagren JN: Staphylococcal endocarditis in drug users: clinrcal and microbrologic aspects. Arch Intern Med 1975; 135: 1555-1561. 7. Marantz PR. Linzer M, Femer CJ, Fernstein SA. Kozin AM, Fnedland GH: Inabrlityy to predict diagnosis in febrile rntravenous drug abusers. Ann Intern Med 1987; 106: 823-828. 8. Sheagren JN: Endocarditis complrcating parenteral drug abuse. In: Remrngton JS, Swartz MN, eds. Current cknrcal topics rn infectious diseases, ~012. New York: McGraw-Hill, 1981; 211-233. 9. Holtzman RS, Bishko F: Osteomyelitis in heroin addicts. Ann Intern Med 1971; 75: 693-696. 10. O’Donnell AE. Pappas LS: Pulmonary compkcatrons of Intravenous drug abuse. Chest 1988; 94: 251-253. 11. Shinozaki T, Deane R. Perkins FM: Infrared tympanic thermometer: evaluation of a new clinrcal thermometer. Crit Care Med 1988; 16: 148-150. 12. RBASE System V. Version 1.0; Microrim. Inc.. Redmond, Washrngton, July 1986. 13. VonReyn CF. Levy BS, Arbeit RD, Fnedland G. Crumpacker CS: lnfectrve endocarditis: an analysis based on strict case definitions, Ann Intern Med 1981; 94(part 1): 505-518. 14. Kuzma JW: Chapter 8: Estimatron of population means, and Chapter 12: The chr-square test. In: Kuzma JW. Basic stabstics for the health sciences. Palo Alto: Mayfield, 1984; 89-100, 141-155. 15. SAS lnstttute Inc: SAS/STATguide for personal computers, versron 6 edition. Gary. North Carokna: SAS Institute. 1987; 1028. July 1990 The American Journal of Medicine Volume 89 57
