- Email: [email protected]
Utility of routine admission chest radiographs in patients with acute gastrointestinal hemorrhage admitted to an intensive care unit
Utility of Routine Admission Chest Radiographs in Patients with Acute Gastrointestinal Hemorrhage Admitted to an Intensive Care Unit Kenneth Tobin, DO, Jeff Klein, MD, Celia Barbieri, MS, John E. Heffner, MD, Phoenix, Arizona
PURPOSE: To determine the diagnostic yield of routine admission chest radiographs in patients with acute gastrointestinal (GI) hemorrhage and clinical predictors of radiographic abnormalities. PATIENTS AND METHODS: The study was a retrospective series of 202 adult patients with GI hemorrhage admitted to intensive care units at an academic medical center. Routine admission chest radiographs were obtained in 161 patients. These radiographs were reviewed by a study radiologist blinded to the study purpose. The radiologist scored radiographic abnormalities into categories of “minor” or “major,” “ new” or “previously known,” and “with an intervention” or “without an intervention.” Nominal logistic regression explored the data for clinical features that identified patients with major new radiographic abnormalities with or without an intervention. RESULTS: Minor radiographic abnormalities were noted in 23 (14.3%) patients, of whom 17 ( 10.6%) patients had “new” (previously unknown) abnormalities. No minor abnormality prompted a therapeutic or diagnostic intervention. Major radiographic abnormalities were detected in 21 (13.0%) patients, of whom 19 (11.8%) had new findings. Major new findings prompted interventions in only 9 (5.6%) of patients. A history of lung disease and an abnormal lung physical examination predicted major new radiographic findings (P = 0.0001, sensitivity 79%, negative predictive value 96%). These variables also identified major new abnormalities that prompted interventions (P = 0.007, sensitivity 89%, negative predictive value 99%). Use of the logistic regression model to select patients for admission chest radiographs
From the Universitv of Arizona Health Scfences Center, St. Joseph’s Hospital and Medical Center, Phoenix, Arizona. Rearrests for reprints should be addressed to John E. Heffner, MD, Professor of Clinical Medicine, University of Arizona Health Sciences Center, Department of Medicine, St. Joseph’s Hospital and Medical Center, 350 West Thomas Road, Phoenix, Arizona 85001-2071. Manuscript submitted May 10, 1996 and accepted in revised form July 12, 1996.
01996 by Excerpta All rights reserved.
Medica,
Inc.
decreased charges from $1,068 to $580 for each detected major new radiographic abnormality and from $2,254 to $1,087 for major new radiographic abnormalities that prompted an intervention. CONCLUSION: These data indicate that routine chest radiographs have a low yield in detecting major new radiographic abnormalities in patients with acute GI hemorrhage. Clinical criteria, available at the time of admission, may be useful for selecting patients for chest radiographic evaluations. Am ./ Med. 1996:101:349-356 cute gastrointestinal (GI) hemorrhage is a comon diagnosis for patients admitted to medical intensive care units (ICUs) . In most medical centers, the majority of these patients undergo routine evaluation with an admission chest radiograph. Most expert reviews and textbook discussions of the management of acute GI hemorrhage, however, do not include a chest radiograph in the initial evaluative studies. Also, clinical data suggest that the diagnostic yield of routine admission chest radiographs is insufficient to justify their acquisition in adults undergoing hospitalization on general wards without signs or symptoms of a cardiopulmonary disorder. ‘8’ These studies, however, have rarely been performed in patients being admitted to an ICU for a specific admitting diagnosis. It has been our experience that admission chest radiographs have a low diagnostic yield in patients admitted to the ICU with a primary diagnosis of GI hemorrhage. We hypothesized that admission chest radiographs for most patients with acute GI hemorrhage have a low yield for identifying clinically important, previously unrecognized flndings and that these studies can be performed selectively in patients who appear at higher risk for cardiopulmonary disorders. To address this hypothesis, we reviewed the hospilzdizdons of patients admitted to our adult ICUs with a primary diagnosis of acute GI hemorrhage to determine the diagnostic yield and clinical importance of admission chest radiographs. We also determined whether clinical features available to clinicians at the time of presentation can identify patients who are most likely to have abnormal chest radiographic findings.
Am
0002-9343/96/$15.00 PII SOOO2-9343(96)00228-S
349
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/lOBIN
ET AL TABLE II Categorization of Chest Radiograph Minor Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Minor abdominal finding Major Abnormalities Definite pleural effusion Subdiaphragmatic air Pulmonary edema Lobar collapse Lobar infiltrate Infiltrate likely pneumonia Cavity Nasogastric tube in the airway Misplaced central line Vascular redistribution Major abdominal abnormality Rib fracture Mediastinal abnormality Cardiomegaly Pulmonary mass Pulmonary nodules Apical fibrosis
TABLE I Study Variables Available on Admission That Were Entered into Predictive Analysis Age Sex History of an acute respiratory condition (cough, dyspnea, changed or increased sputum, chest pain) History of lung disease COPD Asthma Pneumonia Lung Cancer Bronchiectasis Interstitial pneumonitis History of cardiac disease Congestive heart failure Myocardial infarction Valvular disease History of gastrointestional disease Hemetemesis Meiena Hematochezia Pulse Temperature Blood pressure SAPS” Presence of organ failure Abnormal chest examination Rales Rhonchi Wheezes Abnormal cardiac examination Murmurs
cian, house staff, and nursing notes were examined by a research specialist trained in diagnostic coding to extract information pertaining to patients’ history, physical examination, and laboratory findings (Table I). The earliest recorded vital signs were used for analysis. The laboratory test results recorded were those that were available at the time of hospital admission; the first of duplicate studies performed on patients were recorded. Simplified acute physiology scores (SAPS) were calculated on the basis of admission data.3
:i Jugular venous distension Other Abnormal abdominal findings Ascites Rigidity Masses Organomegaly Other Nasogastric bloody aspirate Hematocrit a SAPS denotes
simplified
acute
Scoring of Admission Chest Radiographs physiology
score.
Patients and Study Location The study was performed at St. Joseph’s Hospital and Medical Center, a 576-bed, tertiary care, innercity teaching hospital in Phoenix, Arizona. Medical records coded for gastrointestinal hemorrhage were identified for review. Adult ( > 18 years) patients admitted to the ICUs (32 beds) between December 1991 and November 1993 with a primary diagnosis of acute GI hemorrhage were eligible for study enrollment. The study was approved by the Human Investigative Review Board by expedited review.
Data Collection Medical records were reviewed for data collection without foreknowledge of the content of the patients’ chest radiographic reports. Attending physi350
October
1996
Abnormalities
The American
Journal
of Medicines
Volume
Admission chest radiographs were reviewed by the study pulmonary radiologist (JK) who was blinded both to the study purpose and to the results of the radiologists’ dictated reports in the medical record. All of the chest radiographs were portable studies performed in the anteroposterior projection. Dictated reports in the medical record were not used to categorize chest radiographs because of their marked variation in descriptive terms, frequent use of indecisive language, and potential expectation bias that would occur if the reading radiologists had conferred with clinicians, thereby gaining prior knowledge of the patients’ clinical statuses before the radiologic interpretation.’ Radiographs were categorized by the study radiologist as “normal” or “abnormal.” Abnormal radiographs were further categorized as “major abnormalities” or “minor abnormalities”-on the basis of predetermined criteria (Table II). Previous chest radiographs were examined if available to allow further categorization of radiographic abnormalities as “new” (previous ra101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/lOBIN
diographs did not show the abnormality) or “known” (previous radiographs showed the abnormality). If previous radiographs were unavailable, abnormalities noted on the admission radiographs were categorized as new findings. Progress notes, consultation dictations, and physician order sheets were reviewed to determine if diagnostic or therapeutic interventions or consultations were initiated because of abnormalities noted on the admission chest radiographs. To limit investigator bias,’ interventions were defined by the following preestablished criteria: if a radiographic abnormality prompted a pulmonary consultation, new therapy (antibiotics, chest tube insertion, bronchodilators, incentive spirometry), or initiation of an additional diagnostic study (bronchoscopy, thoracentesis, sputum examination, chest ultrasound, chest computed tomography (CT) scan, repeat chest radiograph). An intervention was also considered to be present if the clinical approach to the GI hemorrhage was altered because of the radiographic abnormality (pneumoperitoneum, pneumomediastinum, findings suggestive of a metastatic malignancy). Abnormal radiographs were then categorized as ‘with an intervention” or ‘without an intervention.” To define the accuracy of the study’s estimated predictive value for routine admission chest radiographs, we reviewed the medical records of patients coded for GI hemorrhage who did not have a chest radiograph at the time of admission. Radiographs obtained in these patients after the first day of hospitalization were examined to determine the presence of radiographic abnormalities that might have been detectable by an admission chest radiograph had one been performed. Physician and nursing notes were also examined for the presence of delayed cardiopulmonary signs or symptoms. An absence of abnormalities on postadmission chest radiographs and cardiopulmonary signs or symptoms during the entire hospitalization suggested a low likelihood that an admission chest radiograph would have revealed major new findings.
Cost Analysis The charge basis cost of a chest radiograph was determined to be $126, which included the hospital ($103) and radiologists’ components ($23). The charge for identifying abnormal chest radiographs in each category of abnormality was calculated by dividing the charge for the 161 routine chest radiographs ($20,286) by the number of abnormal radiographs. The logistic model was evaluated by calculating the charge of performing all necessary radiographs ($126 multiplied by the number of radiographs determined necessary by the model) and
ET Al.
dividing this value by the number of abnormal radiographs in each radiographic category.
Statistical
Analysis
Univariate analysis was performed to compare study variables with radiographic outcomes (presence or absence of chest radiograph abnormalities). Continuous variables were compared using Student’s t test for variables with a normal distribution and Wilcoxon’s rank-sum test for nonparametric variables. All tests were unpaired and two-tailed. Categorical variables were compared using the chisquare statistic or Fisher’s exact test. Nominal logistic regression was performed using JMP software (JMP version 3.1; SAS Institute, Car-y, North Carolina). Variables were added into the model if they were found to be significant (P < 0.10) by univariate analysis. Continuous variables were dichotomized using receiver operating characteristic (ROC) curves to determine decision thresholds and entered into the model as dummy variables.4 Terms were kept in the model if they were significant or were a main-effect term necessary for a significant interaction term. Adjusted odds ratios (AOR) and 95% confidence intervals (CI) were calculated for significant independent variables using standard methods. Statistical significance was determined by a P value less than 0.05. All results are expressed either as mean -+ SD or a percentage of the proportion of patients with a specific outcome variable.
RESULTS Demographics 202 patients admitted to the ICUs with a diagnosis of acute GI hemorrhage, 161 had a routine admission chest radiograph and were entered into t.he study. Characteristics of the 161 patients with admission chest radiographs are shown in Table III. One htmdred (62.1%) of the 161 patients with admission chest radiographs had endoscopic evidence of an upper GI source of bleeding, and 16 (9.9%) patients had radiographic or endoscopic confirmation of a lower GI bleeding site. Thirteen patients (8.1%) had no identifiable bleeding site despite a complete endoscopic evaluation. Endoscopy was not performed in the remaining 32 (19.9%) patients. None of the patients had undergone insertion of an endotracheal tube or central venous catheter prior to the admission chest radiograph.
Radiographic
Outcomes
Table Iv shows the radiographic outcomes in the 161 study patients who underwent a routine admission chest radiograph. Minor radiographic abnormalities were detected in 23 (14.3%) patients; the abnormalities in 17 (10.6%) of these patients were October
1996
The American
Journal
of Medicines
Volume
101
351
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
TABLE
ET AL
III
Characteristic
Abnormalities
Clinical Features Associated with Radiographic Abnormalities Admission Chest Major New Radiographs Abnormalities n= 161 n = 19
Age (yr) Sex (male/female) Acute respiratory symptoms History of lung disease no. (%I History of cardiac disease no. (%) History of GI disease no. (%I Hemetemesis no. (%) Melena no. (%) Hematochezia no. (%) Pulse (beats/min) Temperature KY Blood pressure (sys/dia mm Hg) Abnormal chest examination Rales Rhonchi Wheezes Abnormal cardiac examination Murmurs s3 s4 Jugular venous distension Other Abnormal abdominal findings Ascites Rigidity Masses Organomegaly Other Nasogastric bloody aspirate SAPSb Hematocrit (%I
63.3 2 17.5” 82/79 12 (7.5) 46 (28.6) 80 (49.7) 118 (73.3) 64 (39.8) 58 (36.0) 44 (27.3) 98.5 -+ 19.6 36.8 + 0.9 127 + 28/63 + 17 39 (24.2) 26 (16.2) 9 (5.6) 11 (6.8) 66 (41 .O) 45 (28.0) 2 (1.2) 2 (1.2) 1 (0.6) 16 (9.9) 62 (38.5) 14 (8.7) 0 (0) 3 (1.9) 19 (11.8) 46 (28.6) 24 (14.9) 8.5 2 4.3 27.6 + 9.3
70.0 + 18.6 5/l 4 3 (15.8) 12 (63.2) 9 (47.4) 15 (79.0) 2 (10.5) 5 (26.3) 10 (52.6) 98.0 t 19.3 36.6 -+ 1.1 121 + 31/59 2 18 11 (57.9) 6 (31.6) 2 (10.5) 3 (15.8) 9 (47.4) 7 (36.8) 0 (0) 0 (0) 0 (0) 3 (15.8) 5 (26.3) 1 (5.3) 0 (0) 1 (5.3) 4 (21.1) 7 (36.8) 19 (100) 10.8 2 4.5 26.4 -i- 10.3
Major New with Interventions n=9 71.8 2 18.8
a7 l(ll.l) 6 (66.7) 4 (44.4) 6 (66.71 l(11.1) 3 (33.31 3 (33.3) 93.3 + 17.0 36.8 2 0.9 123 2 3#8/53 k 16 6 (66.6) 4 (44.4) 1 (11.1) l(ll.l) 5 (55.6) 4 (44.4) 0 (0) 0 (0) 0 (01 2 (22.2) 2 (22.2) l(ll.l) 0 (0) 0 (0) l(ll.l) 3 (33.31 3 133.3) 11.2 -c 4.3 22.1 2 7.5
a Mean i- SD; b SAPS denotes simplified acute physiology score.
new, not having been previously documented. No interventions were initiated to treat or evaluate any minor new radiographic finding. Major radiographic abnormalities were detected in 21 (13.0%) patients, of whom 19 (11.8%) patients had major new abnormalities. Major new radiographic abnormalities prompted a diagnostic or therapeutic intervention in 9 (5.6%) patients. No interventions were initiated in the 2 patients who had major known radiographic abnormalities. The ROC analysis determined a cutoff value for SAPS of 210 for identifying patients with major new radiographic findings. Univariate analysis found a significant association of major new radiographic abnormalities with SAPS ~10 (P = 0.012), presence of an abnormal lung physical examination (P = 0.0003), and a history of lung disease (P = 0.0004). Logistic regression determined that presence of an abnormal lung physical examination (AOR = 4.8; 95% CI = 1.7 to 14.0) and a positive history of lung disease (AOR = 4.7; 95% CI = 1.7 to 14.0) were independent explanatory variables of an abnormal admission chest 352
October
1996
The American
Journal
of MedicineE
Volume
radiograph with major new abnormalities. For flnding patients with major new abnormalities, the logistic model had a sensitivity of 79%,specificity of 62%, a positive predictive value of 22%, and a negative predictive value of 96%. Of the 4 patients with major new abnormalities who were not identified by the model, 1 had radiographic evidence of early pulmonary edema and 3 had mediastinal abnormalities consistent with hiatal hernias. None of the 3 patients with mediastinal abnormalities underwent further diagnostic studies for the radiographic abnormalities. Univariate analysis found a significant association of major new radiographic abnormalities that prompted an intervention with the presence of an abnormal lung physical examination (P = 0.007) and a history of lung disease (P = 0.02). Logistic regression confirmed the independent significance of the presence of an abnormal lung physical examination (AOR = 5.9; 95% CI = 1.4 to 29.8) and at history of lung disease (AOR = 4.4; 95% CI = 1.1 to 22.5) as explanatory variables. Predictive values for the lo101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
TABLE
ET AL
IV Distribution
of Radiographic Abnormalities No. Patients (%) 17 (10.6)
Characteristic Minor New Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Minor abdominal finding Minor Known Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Major New Abnormalities Definite pleural effusion Pulmonary edema Lobar infiltrate infiltrate likely pneumonia Major abdominal abnormality Mediastinal abnormality Cardiomegaly Pulmonary mass Pulmonary nodules Major Known Abnormalities Definite pleural effusion Cardiomegaly
6 (3.7)
19 (11.8)
2 (1.2)
gistic model using these 2 explanatory variables were as follows: sensitivity 89%,specificity 60%,positive predictive value 12%, and negative predictive value 99%. The 1 patient who was not identified by the model had radiographic evidence of early pulmonary edema and the intervention was diuretic therapy. Of the 41 patients who did not have an admission chest radiograph, 5 patients had a “routine” chest radiograph within several days of admission (3 minor abnormalities detected) and 2 had a preoperative or postoperative chest radiograph (no abnormalities detected). Two patients had a first chest radiograph performed for the sudden onset of dyspnea several days after admission; both patients were found to have radiographic infiltrates compatible with pneumonia. The remaining 32 patients did not have a chest radiograph and did not experience cardiopulmonary symptoms during their hospitalization. These observations indicate that the absence of chest radiographs for these 41 patients most likely did not cause an important underestimation of the diagnostic yield of admission chest radiographs as calculated for the 161 evaluable patients.
No. Abnormalities 17 3 7 5 2 6 1 3 2 23 3 7 1 1 1 6 2 1 1 2 1 1
normal lung physical findings or the presence of a past history of lung diseasewould have generated 69 radiographs ($8,694 charges) and detected 15 of the 19 patients with major new radiographic abnormalities and 8 of 9 patients with major new abnormalities with interventions. The charge basis for finding these abnormalities using the logistic model would be $580 per major new abnormality and $1,087 per major new abnormality that prompted an intervention.
DISCUSSION
Routine admission chest radiographs are frequently performed to evaluate patients hospitalized for a variety of medical conditions. The diagnostic value of admission chest radiographs is incompletely defined, however, and their routine use is controversial, being supported by inconsistent data.* Previous investigations indicate that routine chest radiographs can be omitted in patients less than 20 years of age who do not have cardiopulmonary symptoms.5 Others recommend the exclusion of all routine admission chest radiographs because the majority of findings are previously known, chronic, or stable, and detection of these findings does usually not alter the patient’s care.“’ Few data exist regardRadiograph Charge Analysis ing the relative yield of routine chest radiographs for The charge basis for tiding major radiographic different admission diagnoses or the potential cost abnormalities was $1,068 per major new abnormality and $2,254 per major new abnormality that savings that might be derived from different ordering prompted an intervention. Selection of patients for strategies. Even fewer data are available regarding admission chest radiographs by the presence of ab- the value of routine chest radiographs for patients October
1996
The American
Journal
of Medicine@
Volume
101
353
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
admitted to a medical ICU with a specific admission diagnosis. It is estimated that ICU patients account for 65% of all routine radiographic studies performed in hospitalized patients.6 The high cost of caring for critically ill patients requires careful consideration of the value of routine radiographic evaluations, especially considering that these studies are considered to be generally overused.2~7,s We selected patients with a primary admission diagnosis of GI hemorrhage for the present investigation because these patients have a lower incidence of multiorgan dysfunction compared with other patients commonly admitted to our ICU. We found that 80% of these patients had an admission portable chest radiograph performed, but only 13.0% had major radiographic abnormalities detected. Minor abnormalities were found in 14.3% of patients, but led to no alterations of care. The observed major abnormalities were previously undocumented in 11.8% of patients, and prompted a diagnostic or therapeutic intervention in only 5.6Ohof patients undergoing routine admission chest radiographs. The study further demonstrated that the low diagnostic yield could be improved by selecting patients for an admission chest radiograph on the basis of the presence of abnormal chest physical findings or a history of lung disease. This strategy decreased the number of admission chest radiographs by 57% and the total charges associated with identifying major new radiographic abnormalities and major new abnormalities that prompted an intervention by 46% ($1,068 to $580 per abnormality) and 52% ($2,254 to $1,087 per abnormality), respectively. The high negative predictive value of the logistic model for major new abnormalities (negative predictive value 96%) and major new abnormalities that prompted an intervention (negative predictive value 99%) allowed most patients to be identified whose care might be altered by an admission chest radiograph. Only 4 (2.5%) of 161 patients had major new radiographic abnormalities in the absence of a history of lung disease or abnormal chest physical findings. Furthermore, these clinical features failed to identify only 1 (0.6%) patient who had major new radiographic abnormalities that prompted an intervention. These findings correspond to previous investigations that symptoms or signs of chest disease are important predictors of abnormalities on routine chest radiographs obtained upon hospital admission to general ward settings.‘,’ Other clinical variables, such as the presence of fever, advanced age, SAPS scores, or history of cardiac disease did not contribute to the predictive model. Some studies show that the incidence of abnormalities on routine radiographs increases with 364
October 1996 The American Journal
of Medicine@
Volume
ET AL
but age,1X5Xg-11
well-designed studies of routine chest radiographs in elderly patients are lacking.2 Previous investigations of ICU patient populations have given limited insight into the diagnostic yield of admission chest radiographs for specific, well-defined clinical conditions. Most of these studies limit their assessments to the potential benefit of daily monitoring chest radiographs, post-event radiographs, or radiographs obtained after the completion of invasive procedures, such as endotracheal intubation or central venous catheterization.‘i’2-14 Investigations that include an assessment of admission radiographs frequently combine these results with those from routine daily and post-procedure studies for a pooled analysis. “,16 Furthermore, most studies include patients with a broad range of diagnoses and do not clearly analyze subgroups of diagnoses except for those with respiratory disorders6~‘3~‘7~1s and unstable cardiac conditions.‘3 Differences in case mix and pooled data analyses have resulted in the impression that 40% to 50% of patients admitted to an ICU have acute radiographic abnormalities justifying routine admission radiographs in all critically ill patients.16r1’ In contrast, the present study indicates that patients with a specific admission diagn.osis of GI hemorrhage have a lower diagnostic ,yield from chest radiographs compared with other general categories of ICU patients. This low yield corresponds to recent findings by Wosornu and coworkers.” These investigators limited their investigation to patients with cardiac conditions admitted to a coronary care unit and found a low incidence of abnormalities detected by routine admission chest radiographs. It appears that admission chest radiographs may not be necessary in all patients admitted to an ICU who have a primary diagnosis, such as GI hemorrhage, wherein the incidence of coexisting intrathoracic disease is low in the absence of suspicious clinical findings. Our investigation was designed to conform to the American College of Radiology’s definition of efficacy for radiographic testing.‘l This definition stratifies efficacy on the basis of (1) diagnostic efficacy (the utility of a test in assisting in diagnosis), (2) therapeutic efficacy (the effect of a test result on patient management), and (3) outcome efficacy (the ultimate contribution of the test and its result to patient morbidity and mortality). Both diagnostic and therapeutic efficacy were low for admission chest radiographs in the study population. We were unable to define outcome efficacy because of the difficulties noted by others” in linking the test result causally with clinical outcome. Furthermore, outcome studies of radiographic interventions require extremely large patient populations entered into ran101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
domized prospective investigations that are impractical for the study of most admission diagnoses.’ Other strengths of the present investigation include the heterogenous sources of GI hemorrhage in the upper and lower GI tract in patients with broad age ranges and SAPS scores, supporting the generalizability of the results. It should be emphasized, however, that only patients with a primary diagnosis of acute GI hemorrhage were included in the study so that our conclusions cannot be applied to patients with coexisting immunosuppression or acute conditions associated with a secondary admission diagnosis of GI hemorrhage. The study also was purposely biased toward tiding benefit from chest radiographs because it defined major radiographic abnormalities as new if previous radiographs were unavailable in the inpatient files; some of these patients may have had outpatient radiographs that might have demonstrated stable abnormalities. Also, the study evaluated the 41 patients who did not have admission chest radiographs to determine their potential impact on the calculated predictive values of an initial radiographic evaluation. The low incidence of cardiopulmonary symptoms and abnormalities noted on chest radiographs obtained after the day of admission in these patients indicate that their exclusion had little effect on our estimates of the admission chest radiograph’s diagnostic yield. An additional strength was the study design, which limited both investigator bias (by using clear definitions of the analyzed interventions) and expectation bias (by blinding the study radiologist to the study purpose and clinical features of the patients) .’ The major limitation of the study is its retrospective design. Incomplete chart documentation of physician decisions may have caused us to undervalue the admission chest radiograph’s contribution to patient care. However, even if all 19patients with major new abnormalities had undergone interventions, the proportion of patients undergoing interventions would have remained low (11.8%). Furthermore, most of these 19 patients were identified by the logistic model. In determining the diagnostic accuracy of the logistic model, we cannot exclude that clinicians first evaluated the chest radiograph, which might have biased their physical examination findings and documentation of the pulmonary history. For these reasons, the value of clinical flndings in selecting patients with acute GI hemorrhage for admission chest radiographs requires confirmation in prospective studies. In conclusion, the study demonstrated that admission chest radiographs performed in the evaluation of patients admitted to ICUs with a primary diagnosis of acute GI hemorrhage have insufficient clinical value to support their routine use in all patients re-
ET AL
gardless of clinical findings. Selecting patients for admission chest radiographs on the basis of a history of chest conditions or an abnormal chest examination identifies most patients with major new radiographic abnormalities, decreases the cost of diagnostic testing, lowers patient exposure to ionizing radiation, and avoids unnecessary evaluations of incidental radiographic findings that turn out to have no clinical significance.
ACKNOWLEDGMENT The authors
thank
Lee Brown,
MD, for his careful
review
of the imanuscript
REFERENCES 1. Hubbell
FA, Greenfield S, Tyler JT, et al. The impact of routine admisslon chest x-ray films on patient care. NEJM. 1985;312:209-213. 2. Tape TG, Mushlin Al. The utility of routine chest radiographs. Ann intern Med.
1986;104:663-670. 3. Le Gall J-R, Loirat P, Alperovitch A, et al. A simplified acute physiology score for ICU patients. Crit Care Med. 1984;12:975-977. 4. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143:29--36. 5. Sage1 SS, Evens RG, Forrest JV, et al. Efficacy of routine lateral chest radiographs in a hospital-based population.
screening and NEJM. 1974;
291:1001-1004. 6. Hall JB, White SR, Karrison T. Efficacy of daily routine chest radiographs in intubated, mechanically ventilated patients. Crit Care Med. 1991; 19:689-693. 7. Hall
FM. Overutilizatlon
of radlological
examinations.
Radiology.
1976;
120:443-448. 8. Griner
PF, Glaser
RJ. Misuse
of laboratory
tests and diagnostic
procedures.
NEJM. 1982;307:1336-1339. 9. Rucker L, Frye EB, Staten MA. Usefulness of screening roentgenograms preoperative patients. JAMA. 1983;250:3209-3211. 10. Rees AM, Roberts CJ, Bligh AS, Evans KT. Routine preoperative radiography in non-cardiopulmonary surgery. BMJ. 1976; 1:1333-1335.
in chest
11. Tornebrandt K, Fletcher R. Pre-operative chest x-rays in elderly patients. Anesthesia. 1982;37:901-902. 12. Greenbaum DM, Marschall KE. The value of routine dally chest x-rays in intubated patients In the medical Intensive care unit. Crit Care Med. 1982;10:29-30. 13. Strain DS, Kinasewitz chest
x-rays
in the medical
GT, Vereen LE, George RB. Value of routine daily intensive care unit. Crit Care Med. 1985; 13:534-
536. 14. Gray P, Sulltvan G, Ostryzniuk P, et al. Value of postprocedural chest diographs in the adult intensive care unit. Crit Care Med. 1992;20:1513-1518. 15. Henschke Cl, Pasternack diagnostic efficacy. Radiology. 16. Janower ML, Jennas-Nocera
ra-
GS, Schroeder S. Bedside chest radiography: 1983; 149:23-26. Z, Mukai J. Utility and efficacy of portable
chest radiographs. AJR. 1984; 142:265-267. 17. Bekemeyer WB, Crapo RO, Calhoon S, et al. Efficacy of chest radiography in a respiratory intensive care unit. A prospective study. Chest. 1985;88:691-
696. 18.
Emerman
CL, Cydulka
RK. Evaluation
of high-yield
criteria
for chest
radi-
ography in acute exacerbation of chronic obstructive pulmonary disease. Ann Emerg Med. 1993;22:680-684. 19. Heffner JE, Klein J. Chest imaging in critically ill patients: analysis of clinical value. Respir Care. 1994;39:51-62. 20. Wosornu D, Pitt M, Deattie JM, Murray RG. Routine chest radlographs in admissions to coronary care. Lancet. 1995;345:62. 21. Loop JW, Lusted LB. American College of Radiology diagnostic efficacy studies. Am J Radiol. 1978;131:173-179. 22. Loder RE. Routine pre-operative chest radiography: 1977 c’zmpared with
1955 at Peterborough 974.
October
1996
Distrrct
The American
General
Journal
Hospital.
of Medicines’
Anaesthesia.
Volume
1977;33:972-
101
355
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
ET AL
23. Catchlove BR, Wilson RM, Spring S, Hall J. Routine investigations in elective surgical patients: their use and cost effectiveness in a teaching hospital. Med JAust. 1979;2:107-110. 24. Delahunt vestigations. 25. Roberts audtt
B, Turnbull PRG. How cost effecbve are routine preoperative N Z Med J. 1980;92:431-432. CJ, Fowkes FGR, Ennis WP, Mitchell M. Possible impact
on chest
x-ray
requests
from
surgtcal
wards.
Lancet.
inof
1983;2:446-
448.
356
October
1996
The American
Journal
of Medicine8
Volume
26. Mindelzun supine patient.
RE, McCort Gastrointest
JJ. The cupola sign of pneumoperitoneum Radiol. 1986; 11:283-285.
27. lshikawa T, Saeki M, Tsukune Y, et al. Detection of paraesophageal by plain films. AJR. 1985;144:701-704, 28. Cole TJ, Turner MA. Manifestations of gastrointestinal disease radiographs. Radiographics. 1993; 13:1013-1034. 29. Gedgaudas-McClees gastrointestinal pathology.
101
RK, Torres WE, Colvin RS, et al. TIloracic Radio! C/in N Am. 1984;22:563-589.
rn the varices on chest findings
in
PURPOSE: To determine the diagnostic yield of routine admission chest radiographs in patients with acute gastrointestinal (GI) hemorrhage and clinical predictors of radiographic abnormalities. PATIENTS AND METHODS: The study was a retrospective series of 202 adult patients with GI hemorrhage admitted to intensive care units at an academic medical center. Routine admission chest radiographs were obtained in 161 patients. These radiographs were reviewed by a study radiologist blinded to the study purpose. The radiologist scored radiographic abnormalities into categories of “minor” or “major,” “ new” or “previously known,” and “with an intervention” or “without an intervention.” Nominal logistic regression explored the data for clinical features that identified patients with major new radiographic abnormalities with or without an intervention. RESULTS: Minor radiographic abnormalities were noted in 23 (14.3%) patients, of whom 17 ( 10.6%) patients had “new” (previously unknown) abnormalities. No minor abnormality prompted a therapeutic or diagnostic intervention. Major radiographic abnormalities were detected in 21 (13.0%) patients, of whom 19 (11.8%) had new findings. Major new findings prompted interventions in only 9 (5.6%) of patients. A history of lung disease and an abnormal lung physical examination predicted major new radiographic findings (P = 0.0001, sensitivity 79%, negative predictive value 96%). These variables also identified major new abnormalities that prompted interventions (P = 0.007, sensitivity 89%, negative predictive value 99%). Use of the logistic regression model to select patients for admission chest radiographs
From the Universitv of Arizona Health Scfences Center, St. Joseph’s Hospital and Medical Center, Phoenix, Arizona. Rearrests for reprints should be addressed to John E. Heffner, MD, Professor of Clinical Medicine, University of Arizona Health Sciences Center, Department of Medicine, St. Joseph’s Hospital and Medical Center, 350 West Thomas Road, Phoenix, Arizona 85001-2071. Manuscript submitted May 10, 1996 and accepted in revised form July 12, 1996.
01996 by Excerpta All rights reserved.
Medica,
Inc.
decreased charges from $1,068 to $580 for each detected major new radiographic abnormality and from $2,254 to $1,087 for major new radiographic abnormalities that prompted an intervention. CONCLUSION: These data indicate that routine chest radiographs have a low yield in detecting major new radiographic abnormalities in patients with acute GI hemorrhage. Clinical criteria, available at the time of admission, may be useful for selecting patients for chest radiographic evaluations. Am ./ Med. 1996:101:349-356 cute gastrointestinal (GI) hemorrhage is a comon diagnosis for patients admitted to medical intensive care units (ICUs) . In most medical centers, the majority of these patients undergo routine evaluation with an admission chest radiograph. Most expert reviews and textbook discussions of the management of acute GI hemorrhage, however, do not include a chest radiograph in the initial evaluative studies. Also, clinical data suggest that the diagnostic yield of routine admission chest radiographs is insufficient to justify their acquisition in adults undergoing hospitalization on general wards without signs or symptoms of a cardiopulmonary disorder. ‘8’ These studies, however, have rarely been performed in patients being admitted to an ICU for a specific admitting diagnosis. It has been our experience that admission chest radiographs have a low diagnostic yield in patients admitted to the ICU with a primary diagnosis of GI hemorrhage. We hypothesized that admission chest radiographs for most patients with acute GI hemorrhage have a low yield for identifying clinically important, previously unrecognized flndings and that these studies can be performed selectively in patients who appear at higher risk for cardiopulmonary disorders. To address this hypothesis, we reviewed the hospilzdizdons of patients admitted to our adult ICUs with a primary diagnosis of acute GI hemorrhage to determine the diagnostic yield and clinical importance of admission chest radiographs. We also determined whether clinical features available to clinicians at the time of presentation can identify patients who are most likely to have abnormal chest radiographic findings.
Am
0002-9343/96/$15.00 PII SOOO2-9343(96)00228-S
349
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/lOBIN
ET AL TABLE II Categorization of Chest Radiograph Minor Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Minor abdominal finding Major Abnormalities Definite pleural effusion Subdiaphragmatic air Pulmonary edema Lobar collapse Lobar infiltrate Infiltrate likely pneumonia Cavity Nasogastric tube in the airway Misplaced central line Vascular redistribution Major abdominal abnormality Rib fracture Mediastinal abnormality Cardiomegaly Pulmonary mass Pulmonary nodules Apical fibrosis
TABLE I Study Variables Available on Admission That Were Entered into Predictive Analysis Age Sex History of an acute respiratory condition (cough, dyspnea, changed or increased sputum, chest pain) History of lung disease COPD Asthma Pneumonia Lung Cancer Bronchiectasis Interstitial pneumonitis History of cardiac disease Congestive heart failure Myocardial infarction Valvular disease History of gastrointestional disease Hemetemesis Meiena Hematochezia Pulse Temperature Blood pressure SAPS” Presence of organ failure Abnormal chest examination Rales Rhonchi Wheezes Abnormal cardiac examination Murmurs
cian, house staff, and nursing notes were examined by a research specialist trained in diagnostic coding to extract information pertaining to patients’ history, physical examination, and laboratory findings (Table I). The earliest recorded vital signs were used for analysis. The laboratory test results recorded were those that were available at the time of hospital admission; the first of duplicate studies performed on patients were recorded. Simplified acute physiology scores (SAPS) were calculated on the basis of admission data.3
:i Jugular venous distension Other Abnormal abdominal findings Ascites Rigidity Masses Organomegaly Other Nasogastric bloody aspirate Hematocrit a SAPS denotes
simplified
acute
Scoring of Admission Chest Radiographs physiology
score.
Patients and Study Location The study was performed at St. Joseph’s Hospital and Medical Center, a 576-bed, tertiary care, innercity teaching hospital in Phoenix, Arizona. Medical records coded for gastrointestinal hemorrhage were identified for review. Adult ( > 18 years) patients admitted to the ICUs (32 beds) between December 1991 and November 1993 with a primary diagnosis of acute GI hemorrhage were eligible for study enrollment. The study was approved by the Human Investigative Review Board by expedited review.
Data Collection Medical records were reviewed for data collection without foreknowledge of the content of the patients’ chest radiographic reports. Attending physi350
October
1996
Abnormalities
The American
Journal
of Medicines
Volume
Admission chest radiographs were reviewed by the study pulmonary radiologist (JK) who was blinded both to the study purpose and to the results of the radiologists’ dictated reports in the medical record. All of the chest radiographs were portable studies performed in the anteroposterior projection. Dictated reports in the medical record were not used to categorize chest radiographs because of their marked variation in descriptive terms, frequent use of indecisive language, and potential expectation bias that would occur if the reading radiologists had conferred with clinicians, thereby gaining prior knowledge of the patients’ clinical statuses before the radiologic interpretation.’ Radiographs were categorized by the study radiologist as “normal” or “abnormal.” Abnormal radiographs were further categorized as “major abnormalities” or “minor abnormalities”-on the basis of predetermined criteria (Table II). Previous chest radiographs were examined if available to allow further categorization of radiographic abnormalities as “new” (previous ra101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/lOBIN
diographs did not show the abnormality) or “known” (previous radiographs showed the abnormality). If previous radiographs were unavailable, abnormalities noted on the admission radiographs were categorized as new findings. Progress notes, consultation dictations, and physician order sheets were reviewed to determine if diagnostic or therapeutic interventions or consultations were initiated because of abnormalities noted on the admission chest radiographs. To limit investigator bias,’ interventions were defined by the following preestablished criteria: if a radiographic abnormality prompted a pulmonary consultation, new therapy (antibiotics, chest tube insertion, bronchodilators, incentive spirometry), or initiation of an additional diagnostic study (bronchoscopy, thoracentesis, sputum examination, chest ultrasound, chest computed tomography (CT) scan, repeat chest radiograph). An intervention was also considered to be present if the clinical approach to the GI hemorrhage was altered because of the radiographic abnormality (pneumoperitoneum, pneumomediastinum, findings suggestive of a metastatic malignancy). Abnormal radiographs were then categorized as ‘with an intervention” or ‘without an intervention.” To define the accuracy of the study’s estimated predictive value for routine admission chest radiographs, we reviewed the medical records of patients coded for GI hemorrhage who did not have a chest radiograph at the time of admission. Radiographs obtained in these patients after the first day of hospitalization were examined to determine the presence of radiographic abnormalities that might have been detectable by an admission chest radiograph had one been performed. Physician and nursing notes were also examined for the presence of delayed cardiopulmonary signs or symptoms. An absence of abnormalities on postadmission chest radiographs and cardiopulmonary signs or symptoms during the entire hospitalization suggested a low likelihood that an admission chest radiograph would have revealed major new findings.
Cost Analysis The charge basis cost of a chest radiograph was determined to be $126, which included the hospital ($103) and radiologists’ components ($23). The charge for identifying abnormal chest radiographs in each category of abnormality was calculated by dividing the charge for the 161 routine chest radiographs ($20,286) by the number of abnormal radiographs. The logistic model was evaluated by calculating the charge of performing all necessary radiographs ($126 multiplied by the number of radiographs determined necessary by the model) and
ET Al.
dividing this value by the number of abnormal radiographs in each radiographic category.
Statistical
Analysis
Univariate analysis was performed to compare study variables with radiographic outcomes (presence or absence of chest radiograph abnormalities). Continuous variables were compared using Student’s t test for variables with a normal distribution and Wilcoxon’s rank-sum test for nonparametric variables. All tests were unpaired and two-tailed. Categorical variables were compared using the chisquare statistic or Fisher’s exact test. Nominal logistic regression was performed using JMP software (JMP version 3.1; SAS Institute, Car-y, North Carolina). Variables were added into the model if they were found to be significant (P < 0.10) by univariate analysis. Continuous variables were dichotomized using receiver operating characteristic (ROC) curves to determine decision thresholds and entered into the model as dummy variables.4 Terms were kept in the model if they were significant or were a main-effect term necessary for a significant interaction term. Adjusted odds ratios (AOR) and 95% confidence intervals (CI) were calculated for significant independent variables using standard methods. Statistical significance was determined by a P value less than 0.05. All results are expressed either as mean -+ SD or a percentage of the proportion of patients with a specific outcome variable.
RESULTS Demographics 202 patients admitted to the ICUs with a diagnosis of acute GI hemorrhage, 161 had a routine admission chest radiograph and were entered into t.he study. Characteristics of the 161 patients with admission chest radiographs are shown in Table III. One htmdred (62.1%) of the 161 patients with admission chest radiographs had endoscopic evidence of an upper GI source of bleeding, and 16 (9.9%) patients had radiographic or endoscopic confirmation of a lower GI bleeding site. Thirteen patients (8.1%) had no identifiable bleeding site despite a complete endoscopic evaluation. Endoscopy was not performed in the remaining 32 (19.9%) patients. None of the patients had undergone insertion of an endotracheal tube or central venous catheter prior to the admission chest radiograph.
Radiographic
Outcomes
Table Iv shows the radiographic outcomes in the 161 study patients who underwent a routine admission chest radiograph. Minor radiographic abnormalities were detected in 23 (14.3%) patients; the abnormalities in 17 (10.6%) of these patients were October
1996
The American
Journal
of Medicines
Volume
101
351
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
TABLE
ET AL
III
Characteristic
Abnormalities
Clinical Features Associated with Radiographic Abnormalities Admission Chest Major New Radiographs Abnormalities n= 161 n = 19
Age (yr) Sex (male/female) Acute respiratory symptoms History of lung disease no. (%I History of cardiac disease no. (%) History of GI disease no. (%I Hemetemesis no. (%) Melena no. (%) Hematochezia no. (%) Pulse (beats/min) Temperature KY Blood pressure (sys/dia mm Hg) Abnormal chest examination Rales Rhonchi Wheezes Abnormal cardiac examination Murmurs s3 s4 Jugular venous distension Other Abnormal abdominal findings Ascites Rigidity Masses Organomegaly Other Nasogastric bloody aspirate SAPSb Hematocrit (%I
63.3 2 17.5” 82/79 12 (7.5) 46 (28.6) 80 (49.7) 118 (73.3) 64 (39.8) 58 (36.0) 44 (27.3) 98.5 -+ 19.6 36.8 + 0.9 127 + 28/63 + 17 39 (24.2) 26 (16.2) 9 (5.6) 11 (6.8) 66 (41 .O) 45 (28.0) 2 (1.2) 2 (1.2) 1 (0.6) 16 (9.9) 62 (38.5) 14 (8.7) 0 (0) 3 (1.9) 19 (11.8) 46 (28.6) 24 (14.9) 8.5 2 4.3 27.6 + 9.3
70.0 + 18.6 5/l 4 3 (15.8) 12 (63.2) 9 (47.4) 15 (79.0) 2 (10.5) 5 (26.3) 10 (52.6) 98.0 t 19.3 36.6 -+ 1.1 121 + 31/59 2 18 11 (57.9) 6 (31.6) 2 (10.5) 3 (15.8) 9 (47.4) 7 (36.8) 0 (0) 0 (0) 0 (0) 3 (15.8) 5 (26.3) 1 (5.3) 0 (0) 1 (5.3) 4 (21.1) 7 (36.8) 19 (100) 10.8 2 4.5 26.4 -i- 10.3
Major New with Interventions n=9 71.8 2 18.8
a7 l(ll.l) 6 (66.7) 4 (44.4) 6 (66.71 l(11.1) 3 (33.31 3 (33.3) 93.3 + 17.0 36.8 2 0.9 123 2 3#8/53 k 16 6 (66.6) 4 (44.4) 1 (11.1) l(ll.l) 5 (55.6) 4 (44.4) 0 (0) 0 (0) 0 (01 2 (22.2) 2 (22.2) l(ll.l) 0 (0) 0 (0) l(ll.l) 3 (33.31 3 133.3) 11.2 -c 4.3 22.1 2 7.5
a Mean i- SD; b SAPS denotes simplified acute physiology score.
new, not having been previously documented. No interventions were initiated to treat or evaluate any minor new radiographic finding. Major radiographic abnormalities were detected in 21 (13.0%) patients, of whom 19 (11.8%) patients had major new abnormalities. Major new radiographic abnormalities prompted a diagnostic or therapeutic intervention in 9 (5.6%) patients. No interventions were initiated in the 2 patients who had major known radiographic abnormalities. The ROC analysis determined a cutoff value for SAPS of 210 for identifying patients with major new radiographic findings. Univariate analysis found a significant association of major new radiographic abnormalities with SAPS ~10 (P = 0.012), presence of an abnormal lung physical examination (P = 0.0003), and a history of lung disease (P = 0.0004). Logistic regression determined that presence of an abnormal lung physical examination (AOR = 4.8; 95% CI = 1.7 to 14.0) and a positive history of lung disease (AOR = 4.7; 95% CI = 1.7 to 14.0) were independent explanatory variables of an abnormal admission chest 352
October
1996
The American
Journal
of MedicineE
Volume
radiograph with major new abnormalities. For flnding patients with major new abnormalities, the logistic model had a sensitivity of 79%,specificity of 62%, a positive predictive value of 22%, and a negative predictive value of 96%. Of the 4 patients with major new abnormalities who were not identified by the model, 1 had radiographic evidence of early pulmonary edema and 3 had mediastinal abnormalities consistent with hiatal hernias. None of the 3 patients with mediastinal abnormalities underwent further diagnostic studies for the radiographic abnormalities. Univariate analysis found a significant association of major new radiographic abnormalities that prompted an intervention with the presence of an abnormal lung physical examination (P = 0.007) and a history of lung disease (P = 0.02). Logistic regression confirmed the independent significance of the presence of an abnormal lung physical examination (AOR = 5.9; 95% CI = 1.4 to 29.8) and at history of lung disease (AOR = 4.4; 95% CI = 1.1 to 22.5) as explanatory variables. Predictive values for the lo101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
TABLE
ET AL
IV Distribution
of Radiographic Abnormalities No. Patients (%) 17 (10.6)
Characteristic Minor New Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Minor abdominal finding Minor Known Abnormalities Possible pleural effusion Segmental or subsegmental atelectasis Nasogastric tube coiled in the stomach Major New Abnormalities Definite pleural effusion Pulmonary edema Lobar infiltrate infiltrate likely pneumonia Major abdominal abnormality Mediastinal abnormality Cardiomegaly Pulmonary mass Pulmonary nodules Major Known Abnormalities Definite pleural effusion Cardiomegaly
6 (3.7)
19 (11.8)
2 (1.2)
gistic model using these 2 explanatory variables were as follows: sensitivity 89%,specificity 60%,positive predictive value 12%, and negative predictive value 99%. The 1 patient who was not identified by the model had radiographic evidence of early pulmonary edema and the intervention was diuretic therapy. Of the 41 patients who did not have an admission chest radiograph, 5 patients had a “routine” chest radiograph within several days of admission (3 minor abnormalities detected) and 2 had a preoperative or postoperative chest radiograph (no abnormalities detected). Two patients had a first chest radiograph performed for the sudden onset of dyspnea several days after admission; both patients were found to have radiographic infiltrates compatible with pneumonia. The remaining 32 patients did not have a chest radiograph and did not experience cardiopulmonary symptoms during their hospitalization. These observations indicate that the absence of chest radiographs for these 41 patients most likely did not cause an important underestimation of the diagnostic yield of admission chest radiographs as calculated for the 161 evaluable patients.
No. Abnormalities 17 3 7 5 2 6 1 3 2 23 3 7 1 1 1 6 2 1 1 2 1 1
normal lung physical findings or the presence of a past history of lung diseasewould have generated 69 radiographs ($8,694 charges) and detected 15 of the 19 patients with major new radiographic abnormalities and 8 of 9 patients with major new abnormalities with interventions. The charge basis for finding these abnormalities using the logistic model would be $580 per major new abnormality and $1,087 per major new abnormality that prompted an intervention.
DISCUSSION
Routine admission chest radiographs are frequently performed to evaluate patients hospitalized for a variety of medical conditions. The diagnostic value of admission chest radiographs is incompletely defined, however, and their routine use is controversial, being supported by inconsistent data.* Previous investigations indicate that routine chest radiographs can be omitted in patients less than 20 years of age who do not have cardiopulmonary symptoms.5 Others recommend the exclusion of all routine admission chest radiographs because the majority of findings are previously known, chronic, or stable, and detection of these findings does usually not alter the patient’s care.“’ Few data exist regardRadiograph Charge Analysis ing the relative yield of routine chest radiographs for The charge basis for tiding major radiographic different admission diagnoses or the potential cost abnormalities was $1,068 per major new abnormality and $2,254 per major new abnormality that savings that might be derived from different ordering prompted an intervention. Selection of patients for strategies. Even fewer data are available regarding admission chest radiographs by the presence of ab- the value of routine chest radiographs for patients October
1996
The American
Journal
of Medicine@
Volume
101
353
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
admitted to a medical ICU with a specific admission diagnosis. It is estimated that ICU patients account for 65% of all routine radiographic studies performed in hospitalized patients.6 The high cost of caring for critically ill patients requires careful consideration of the value of routine radiographic evaluations, especially considering that these studies are considered to be generally overused.2~7,s We selected patients with a primary admission diagnosis of GI hemorrhage for the present investigation because these patients have a lower incidence of multiorgan dysfunction compared with other patients commonly admitted to our ICU. We found that 80% of these patients had an admission portable chest radiograph performed, but only 13.0% had major radiographic abnormalities detected. Minor abnormalities were found in 14.3% of patients, but led to no alterations of care. The observed major abnormalities were previously undocumented in 11.8% of patients, and prompted a diagnostic or therapeutic intervention in only 5.6Ohof patients undergoing routine admission chest radiographs. The study further demonstrated that the low diagnostic yield could be improved by selecting patients for an admission chest radiograph on the basis of the presence of abnormal chest physical findings or a history of lung disease. This strategy decreased the number of admission chest radiographs by 57% and the total charges associated with identifying major new radiographic abnormalities and major new abnormalities that prompted an intervention by 46% ($1,068 to $580 per abnormality) and 52% ($2,254 to $1,087 per abnormality), respectively. The high negative predictive value of the logistic model for major new abnormalities (negative predictive value 96%) and major new abnormalities that prompted an intervention (negative predictive value 99%) allowed most patients to be identified whose care might be altered by an admission chest radiograph. Only 4 (2.5%) of 161 patients had major new radiographic abnormalities in the absence of a history of lung disease or abnormal chest physical findings. Furthermore, these clinical features failed to identify only 1 (0.6%) patient who had major new radiographic abnormalities that prompted an intervention. These findings correspond to previous investigations that symptoms or signs of chest disease are important predictors of abnormalities on routine chest radiographs obtained upon hospital admission to general ward settings.‘,’ Other clinical variables, such as the presence of fever, advanced age, SAPS scores, or history of cardiac disease did not contribute to the predictive model. Some studies show that the incidence of abnormalities on routine radiographs increases with 364
October 1996 The American Journal
of Medicine@
Volume
ET AL
but age,1X5Xg-11
well-designed studies of routine chest radiographs in elderly patients are lacking.2 Previous investigations of ICU patient populations have given limited insight into the diagnostic yield of admission chest radiographs for specific, well-defined clinical conditions. Most of these studies limit their assessments to the potential benefit of daily monitoring chest radiographs, post-event radiographs, or radiographs obtained after the completion of invasive procedures, such as endotracheal intubation or central venous catheterization.‘i’2-14 Investigations that include an assessment of admission radiographs frequently combine these results with those from routine daily and post-procedure studies for a pooled analysis. “,16 Furthermore, most studies include patients with a broad range of diagnoses and do not clearly analyze subgroups of diagnoses except for those with respiratory disorders6~‘3~‘7~1s and unstable cardiac conditions.‘3 Differences in case mix and pooled data analyses have resulted in the impression that 40% to 50% of patients admitted to an ICU have acute radiographic abnormalities justifying routine admission radiographs in all critically ill patients.16r1’ In contrast, the present study indicates that patients with a specific admission diagn.osis of GI hemorrhage have a lower diagnostic ,yield from chest radiographs compared with other general categories of ICU patients. This low yield corresponds to recent findings by Wosornu and coworkers.” These investigators limited their investigation to patients with cardiac conditions admitted to a coronary care unit and found a low incidence of abnormalities detected by routine admission chest radiographs. It appears that admission chest radiographs may not be necessary in all patients admitted to an ICU who have a primary diagnosis, such as GI hemorrhage, wherein the incidence of coexisting intrathoracic disease is low in the absence of suspicious clinical findings. Our investigation was designed to conform to the American College of Radiology’s definition of efficacy for radiographic testing.‘l This definition stratifies efficacy on the basis of (1) diagnostic efficacy (the utility of a test in assisting in diagnosis), (2) therapeutic efficacy (the effect of a test result on patient management), and (3) outcome efficacy (the ultimate contribution of the test and its result to patient morbidity and mortality). Both diagnostic and therapeutic efficacy were low for admission chest radiographs in the study population. We were unable to define outcome efficacy because of the difficulties noted by others” in linking the test result causally with clinical outcome. Furthermore, outcome studies of radiographic interventions require extremely large patient populations entered into ran101
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
domized prospective investigations that are impractical for the study of most admission diagnoses.’ Other strengths of the present investigation include the heterogenous sources of GI hemorrhage in the upper and lower GI tract in patients with broad age ranges and SAPS scores, supporting the generalizability of the results. It should be emphasized, however, that only patients with a primary diagnosis of acute GI hemorrhage were included in the study so that our conclusions cannot be applied to patients with coexisting immunosuppression or acute conditions associated with a secondary admission diagnosis of GI hemorrhage. The study also was purposely biased toward tiding benefit from chest radiographs because it defined major radiographic abnormalities as new if previous radiographs were unavailable in the inpatient files; some of these patients may have had outpatient radiographs that might have demonstrated stable abnormalities. Also, the study evaluated the 41 patients who did not have admission chest radiographs to determine their potential impact on the calculated predictive values of an initial radiographic evaluation. The low incidence of cardiopulmonary symptoms and abnormalities noted on chest radiographs obtained after the day of admission in these patients indicate that their exclusion had little effect on our estimates of the admission chest radiograph’s diagnostic yield. An additional strength was the study design, which limited both investigator bias (by using clear definitions of the analyzed interventions) and expectation bias (by blinding the study radiologist to the study purpose and clinical features of the patients) .’ The major limitation of the study is its retrospective design. Incomplete chart documentation of physician decisions may have caused us to undervalue the admission chest radiograph’s contribution to patient care. However, even if all 19patients with major new abnormalities had undergone interventions, the proportion of patients undergoing interventions would have remained low (11.8%). Furthermore, most of these 19 patients were identified by the logistic model. In determining the diagnostic accuracy of the logistic model, we cannot exclude that clinicians first evaluated the chest radiograph, which might have biased their physical examination findings and documentation of the pulmonary history. For these reasons, the value of clinical flndings in selecting patients with acute GI hemorrhage for admission chest radiographs requires confirmation in prospective studies. In conclusion, the study demonstrated that admission chest radiographs performed in the evaluation of patients admitted to ICUs with a primary diagnosis of acute GI hemorrhage have insufficient clinical value to support their routine use in all patients re-
ET AL
gardless of clinical findings. Selecting patients for admission chest radiographs on the basis of a history of chest conditions or an abnormal chest examination identifies most patients with major new radiographic abnormalities, decreases the cost of diagnostic testing, lowers patient exposure to ionizing radiation, and avoids unnecessary evaluations of incidental radiographic findings that turn out to have no clinical significance.
ACKNOWLEDGMENT The authors
thank
Lee Brown,
MD, for his careful
review
of the imanuscript
REFERENCES 1. Hubbell
FA, Greenfield S, Tyler JT, et al. The impact of routine admisslon chest x-ray films on patient care. NEJM. 1985;312:209-213. 2. Tape TG, Mushlin Al. The utility of routine chest radiographs. Ann intern Med.
1986;104:663-670. 3. Le Gall J-R, Loirat P, Alperovitch A, et al. A simplified acute physiology score for ICU patients. Crit Care Med. 1984;12:975-977. 4. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143:29--36. 5. Sage1 SS, Evens RG, Forrest JV, et al. Efficacy of routine lateral chest radiographs in a hospital-based population.
screening and NEJM. 1974;
291:1001-1004. 6. Hall JB, White SR, Karrison T. Efficacy of daily routine chest radiographs in intubated, mechanically ventilated patients. Crit Care Med. 1991; 19:689-693. 7. Hall
FM. Overutilizatlon
of radlological
examinations.
Radiology.
1976;
120:443-448. 8. Griner
PF, Glaser
RJ. Misuse
of laboratory
tests and diagnostic
procedures.
NEJM. 1982;307:1336-1339. 9. Rucker L, Frye EB, Staten MA. Usefulness of screening roentgenograms preoperative patients. JAMA. 1983;250:3209-3211. 10. Rees AM, Roberts CJ, Bligh AS, Evans KT. Routine preoperative radiography in non-cardiopulmonary surgery. BMJ. 1976; 1:1333-1335.
in chest
11. Tornebrandt K, Fletcher R. Pre-operative chest x-rays in elderly patients. Anesthesia. 1982;37:901-902. 12. Greenbaum DM, Marschall KE. The value of routine dally chest x-rays in intubated patients In the medical Intensive care unit. Crit Care Med. 1982;10:29-30. 13. Strain DS, Kinasewitz chest
x-rays
in the medical
GT, Vereen LE, George RB. Value of routine daily intensive care unit. Crit Care Med. 1985; 13:534-
536. 14. Gray P, Sulltvan G, Ostryzniuk P, et al. Value of postprocedural chest diographs in the adult intensive care unit. Crit Care Med. 1992;20:1513-1518. 15. Henschke Cl, Pasternack diagnostic efficacy. Radiology. 16. Janower ML, Jennas-Nocera
ra-
GS, Schroeder S. Bedside chest radiography: 1983; 149:23-26. Z, Mukai J. Utility and efficacy of portable
chest radiographs. AJR. 1984; 142:265-267. 17. Bekemeyer WB, Crapo RO, Calhoon S, et al. Efficacy of chest radiography in a respiratory intensive care unit. A prospective study. Chest. 1985;88:691-
696. 18.
Emerman
CL, Cydulka
RK. Evaluation
of high-yield
criteria
for chest
radi-
ography in acute exacerbation of chronic obstructive pulmonary disease. Ann Emerg Med. 1993;22:680-684. 19. Heffner JE, Klein J. Chest imaging in critically ill patients: analysis of clinical value. Respir Care. 1994;39:51-62. 20. Wosornu D, Pitt M, Deattie JM, Murray RG. Routine chest radlographs in admissions to coronary care. Lancet. 1995;345:62. 21. Loop JW, Lusted LB. American College of Radiology diagnostic efficacy studies. Am J Radiol. 1978;131:173-179. 22. Loder RE. Routine pre-operative chest radiography: 1977 c’zmpared with
1955 at Peterborough 974.
October
1996
Distrrct
The American
General
Journal
Hospital.
of Medicines’
Anaesthesia.
Volume
1977;33:972-
101
355
CHEST RADIOGRAPHS FOR ACUTE GI HEMORRHAGE/TOBIN
ET AL
23. Catchlove BR, Wilson RM, Spring S, Hall J. Routine investigations in elective surgical patients: their use and cost effectiveness in a teaching hospital. Med JAust. 1979;2:107-110. 24. Delahunt vestigations. 25. Roberts audtt
B, Turnbull PRG. How cost effecbve are routine preoperative N Z Med J. 1980;92:431-432. CJ, Fowkes FGR, Ennis WP, Mitchell M. Possible impact
on chest
x-ray
requests
from
surgtcal
wards.
Lancet.
inof
1983;2:446-
448.
356
October
1996
The American
Journal
of Medicine8
Volume
26. Mindelzun supine patient.
RE, McCort Gastrointest
JJ. The cupola sign of pneumoperitoneum Radiol. 1986; 11:283-285.
27. lshikawa T, Saeki M, Tsukune Y, et al. Detection of paraesophageal by plain films. AJR. 1985;144:701-704, 28. Cole TJ, Turner MA. Manifestations of gastrointestinal disease radiographs. Radiographics. 1993; 13:1013-1034. 29. Gedgaudas-McClees gastrointestinal pathology.
101
RK, Torres WE, Colvin RS, et al. TIloracic Radio! C/in N Am. 1984;22:563-589.
rn the varices on chest findings
in