- Email: [email protected]
A simple model to predict bacteremia in women with acute pyelonephritis
Journal of Infection (2011) 63, 124e130
www.elsevierhealth.com/journals/jinf
A simple model to predict bacteremia in women with acute pyelonephritis Kyung Su Kim a, Kyuseok Kim b,*, You Hwan Jo b, Tae Yun Kim b, Jin Hee Lee b, Se Jong Lee a, Joong Eui Rhee b, Gil Joon Suh a a
Department of Emergency Medicine, Seoul National University Hospital, Seoul, Republic of Korea Department of Emergency Medicine, Seoul National University Bundang Hospital, 166 Gumi-ro, Bundang-gu, Seongnam 463-707, Republic of Korea
b
Accepted 10 June 2011 Available online 22 June 2011
KEYWORDS Pyelonephritis; Bacteremia; Logistic models; Risk assessment; Community-acquired infections
Summary Objectives: To construct a simple model to predict bacteremia in women with uncomplicated acute pyelonephritis (APN) for the judicious use of blood cultures. Methods: A prospective database including 735 women with uncomplicated APN at an academic urban emergency department was analyzed retrospectively. Independent risk factors were determined using multivariate logistic regression in two-thirds of patients. Cutoff values representing 10% and 30% of risk were selected for the stratification. This model was internally and externally validated using a remaining one-thirds of patients and 169 independent patients, respectively. Results: Independent risk factors were as follows: age 65 years (odds ratio [OR]Z5.18, 4 points), vomiting (OR Z 2.40, 2 points), heart rate >110 beats/min (OR Z 2.35, 2 points), segmented neutrophils >90% (OR Z 3.17, 3 points), and urine WBC 50/HPF (OR Z 4.27, 4 points). Patients were stratified as low (points <4), intermediate (points, 4e6), or high risk (7 points). The areas under receiver operating characteristics curves were 0.707 and 0.792 in internal and external validation cohorts, respectively. The model stratified internal and external validation cohort into low (8.5% and 5.7%), intermediate (16.5% and 14.8%), and high risk of bacteremia (42.0% and 56.4%). Conclusions: This model provides a useful tool to predict the risk of bacteremia, which can be helpful to decide whether to perform blood cultures and whether to admit the patient for the intravenous antibiotics in women with uncomplicated APN. ª 2011 The British Infection Association. Published by Elsevier Ltd. All rights reserved.
* Corresponding author. Tel.: þ82 31 787 7572; fax: þ82 31 787 4055. E-mail address: [email protected] (K. Kim). 0163-4453/$36 ª 2011 The British Infection Association. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2011.06.007
Predicting bacteremia in women with APN
Introduction Urinary tract infections (UTIs) are the most common bacterial infections in women. Among UTIs, acute pyelonephritis (APN) is a potentially severe disease with approximately 250,000 cases and $2.14 billion costs in the United States each year.1,2 Uncomplicated APN is defined as APN without structural and functional abnormalities within the urinary tract, without relevant kidney diseases and without relevant comorbidities.3 Some authors have suggested that routine blood cultures are not necessary for patients with uncomplicated APN.4e6 However, there is no consensus whether to perform blood culture or not. Therefore, many unjustified blood cultures are being performed concerning about high rate of bacteremia and increasing prevalence of resistant strains.7,8 The aim of this study was to construct a risk stratification model for bacteremia in uncomplicated APN to guide the judicious use of blood culture.
Materials and methods Study design We analyzed retrospectively a prospective database. This database was originally designed to assess the effectiveness of an institutional admission protocol in women with APN.9 After the study, we have continued registering the patients and maintaining the database. The cohorts obtained from this database were used for the derivation and internal validation of the risk prediction model. Furthermore, we retrospectively identified the patients with uncomplicated APN in an independent academic hospital emergency department (ED) for the external validation. This study was approved by the institutional review board and was exempted from informed consent.
Study setting and population A prospective database was obtained in a single academic, urban ED with an annual census of approximately 65,000 between January 2006 and September 2009. We reviewed the database, which was collected by following criteria.9 All consecutive female patients who were 15 years old and had APN were prospectively enrolled between January 2006 and September 2009. The diagnostic criteria for APN included the following: (1) a body temperature (BT) over 38 C, (2) pyuria, and (3) lumbar tenderness. Pyuria was defined as the presence of leukocytes 10 per high-power field in a centrifuged urine specimen. If the patient had taken any antipyretic agent before visiting the ED, a BT over 38 C was not required for diagnosis. Exclusion criteria were as follows: (1) severe sepsis or septic shock; (2) an immunocompromised state (on chemotherapy or immunosuppressive medication); (3) history of chronic renal failure; (4) acute renal failure (serum creatinine >2 mg/dL); (5) obstructive APN; (6) neurogenic bladder; (7) patients with urinary catheters; and (8) previous kidney transplantation. Advanced age, liver cirrhosis, diabetes mellitus, and other chronic diseases were not considered as exclusion criteria.
125 External validation cohorts were obtained in an independent academic, urban ED with an annual census of approximately 50,000 between January 2009 and December 2009. Patients were considered eligible for the external validation cohort when they had the discharge diagnosis of APN and underwent blood culture tests. After reviewing medical records of all eligible patients, patients were enrolled using the same criteria described previously.
Measures All data were collected based on both the established protocol and the unified definition of variables. We evaluated potential risk factors usually available prior to the decision to perform the blood cultures. Those were as follows: (1) epidemiologic data including age, diabetes mellitus, and hypertension; (2) clinical findings including fever, chill, anorexia, and vomiting; (3) antibiotics therapy before ED visit; (4) initial vital signs including systolic blood pressure (BP), diastolic BP, heart rate, respiration rate, and BT; and (5) initial laboratory findings including WBC counts, percentage of segmented neutrophils, hemoglobin, platelet counts, blood urine nitrogen (BUN), creatinine, C-reactive protein, and urine WBC counts. In retrospective cohort, missing information about clinical findings was considered to indicate the absence of that finding. Patients with insufficient data were further excluded from the analysis. Blood culture was performed at the discretion of the attending physician. To perform the blood cultures, 10 mL of blood was obtained and inoculated into 2 culture bottles (5 mL into an aerobic bottle and 5 mL into an anaerobic bottle). An additional 10 ml of blood was obtained from a different site and the procedure was repeated. Blood culture bottles were processed in an automated blood culture system. The following isolates were considered to be contaminants if they were isolated at only one culture bottle and discordant from urine cultures: Streptococcus pyogenes, Bacillus species, Micrococcus species, and coagulase-negative staphylococci except Staphylococcus saprophyticus.
Data analysis After excluding the patients without blood culture from the prospective database, the patients were randomized into 2 groups to create derivation cohort with two-thirds of the patients and an internal validation cohort with the remaining one third. Then, we compared patients with and without bacteremia in derivation cohort. Continuous variables were dichotomized considering both clinical relevancy and statistical significance. Univariate logistic regression analyses using various cutoffs were performed for predicting bacteremia. The most discriminant cutoff was determined using the p value and Bayesian information criterion (BIC).10 The cutoffs selected were as follows: age 65, systolic BP < 100 mm Hg, heart rate >110 beats/min, respiratory rate >20 cycles/ min, body temperature >39 C, WBC counts <4000 or >12,000 cells/mL, segmented neutrophils >90%, hemoglobin <12 mg/dL, platelet counts <150,000 cells/mL, serum creatinine >1.5 mg/dL, C-reactive protein >10 mg/dL, and urine WBCs 50/HPF. Statistically significant (p < 0.05) variables
126 were selected as candidate variables for the model derivation. Using all the candidate variables, multivariate logistic regression model predicting bacteremia was constructed. Variables was then removed from this model in a stepwise manner until none remained. The final model with best fit was determined using BIC. BIC is a criterion for model selection that penalizes the number of variables to avoid overfitting.10 A model is better than another if it has a smaller BIC value.11 Then, the corresponding logistic regression model coefficients were divided by 0.4 and rounded to the numbers. These numbers were assigned to the independent variables to estimate the risk of bacteremia. To stratify the risk of bacteremia as low, intermediate, or high, we selected the specific cutoffs representing predicted probability of 10% and 30%. Results of multivariate analysis are reported as odds ratios (OR) with 95% confidence intervals (CIs) and p values. The performance of the model was tested with respect to discrimination and calibration. Discrimination was quantified with the areas under the receiver operating characteristics curve (AUCs) with its corresponding 95% CIs. Calibration was tested with stratification tables. Continuous variables were expressed as means with SDs, and categorical data were presented as the percent frequency of occurrence. A Student t test was used to compare continuous variables, and the c2 or Fischer exact test was used to compare binomial variables. Two-tailed p < 0.05 was considered to indicate statistical significance. All analyses were performed using Stata version 10.1 (Stata Corp, College Station, Tex).
Results A total of 862 cases with uncomplicated APN were enrolled in the prospective cohort. Among them, 127 cases did not have blood cultures. The characteristics of these cases are described in Table 1. Cases without blood culture had fever (p < 0.001), chill (p < 0.001), and anorexia (p Z 0.027) less frequently than those with blood culture. They received antibiotics before ED visit more frequently (p Z 0.004). Initial heart rate was slower (p < 0.001) and BT (p < 0.001), the percentage of segmented neutrophils in white blood cells (p < 0.001), and C-reactive protein (p < 0.001) were lower in cases without blood culture. The other characteristics were not different. Among 735 cases with blood cultures in the prospective cohorts, true bacteremia was identified in 141 (19.2%) cases and contaminants were detected in 41 (5.6%) cases. Among the true pathogens, Escherichia coli was the most frequently isolates, found in 134 (95.0%) cases, followed by Klebsiella pneumoniae, found in 3 (2.1%); Enterobacter aerogenes, found in 1 (0.7%); Enterococcus faecalis, found in 1 (0.7%); Corynebacterium species, found in 1 (0.7%); Proteus mirabilis found in 1 (0.7%); and Salmonella species, found in 1 (0.7%). There was one case with polymicrobial infections (K. pneumonia and P. mirabilis). Among 134 isolates of E. coli, 25 (18.7%) were resistant to ciprofloxacin and 4 (3.0%) were extended-spectrum b-lactamases-producing organisms. All 4 extended-spectrum b-lactamases-producing organisms were sensitive to ciprofloxacin. One of 3 isolates of K. pneumoniae was resistant to ciprofloxacin.
K.S. Kim et al. Among these 735 cases, 494 (67.2%) were randomly assigned to the derivation cohort, and 241 (32.8%) were assigned to the internal validation cohort. The characteristics of cases in derivation and internal validation cohorts are presented in Table 1. In the derivation cohort, 10 variables were significantly associated with bacteremia, including epidemiological data (age 65, diabetes mellitus, and hypertension), vomiting, tachycardia (heart rate >110 beats/min), and laboratory findings (WBC counts <4000 or >12,000 cells/mL, segmented neutrophils >90%, platelet counts <150,000 cells/mL, C-reactive protein >10 mg/dL, and urine WBCs 50/HPF) (Table 2). Variables selected in the final multivariate logistic regression model were as follows: age 65, vomiting, heart rate >110 beats/min, segmented neutrophils >90%, and urine WBCs 50/HPF (Table 3). According to the regression coefficient, points were assigned as described in Table 3. The specific cutoffs representing predicted probability of 10% and 30% were 4 and 7, respectively. The application of the model enables us to categorize derivation cohort into low (4.3%), intermediate (16.7%), and high (50.9%) risk groups of bacteremia (Table 4). The AUC of the stratification model was 0.792 (95% CI, 0.746e0.839) in the derivation cohort (Fig. 1). Internal validation cohort was categorized into low (8.5%), intermediate (16.5%), and high (42.0%) risk groups of bacteremia (Table 4). The AUC of the model was 0.707 (95% CI, 0.623e0.790) in the internal validation cohort (Fig. 1). A total of 198 cases who had the discharge diagnosis of APN and underwent blood culture tests were enrolled for the external validation cohort. After excluding 29 cases with exclusion criteria or without insufficient data, 169 cases were analyzed. Table 1 shows the characteristics of the external validation cohort. Among 43 cases with bacteremia, E. coli was found in 39 (90.7%) cases, followed by K. pneumoniae, found in 2 (4.7%); Bacteroides fragilis, found in 1 (2.3%); and Citrobacter freundii complex, found in 1 (2.3%). The model stratified external validation cohort into low (5.7%), intermediate (14.8%), and high (56.4%) risk of bacteremia (Table 4). The AUC of the model was 0.792 (95% CI, 0.720e0.865) (Fig. 1).
Discussion In this study, we developed and validated a stratification model for the risk of bacteremia in uncomplicated APN. The discrimination ability of the model was generally good (AUC > 0.700) in all cohorts and the stratification ability at predetermined probability was also reasonable. Recently, the model to predict the bacteremia in febrile urosepsis syndrome was published, but our study was different from the previous one in terms of subjects and methods to develop the model. The merit of our study was to validate the model both internally and externally with good discrimination. On the contrary, they added procalcitonin to the usual clinical variables in the prediction model, and yielded more discriminative power.12 Among 862 cases in prospective database, 127 (14.7%) did not receive blood culture. Cases without blood culture tend to have mild symptoms and relatively normal vital signs. They also have received antibiotics before ED visit
Predicting bacteremia in women with APN Table 1
127
Characteristics of all cohorts.a Prospective cohorts Excluded
b
External validation cohort
Derivation
Internal validation
Number of patients
127
494
241
169
Epidemiologic data Age Diabetes mellitus Hypertension
48.5 20.0 16 (12.6) 24 (18.9)
50.5 18.6 77 (15.6) 120 (24.3)
48.0 18.5 28 (11.6) 56 (23.2)
54.2 18.8 32 (18.9) 54 (32.0)
Clinical findings Fever Chills Anorexia Vomiting
78 69 29 29
415 376 160 101
203 (84.2) 184 (76.4) 80 (33.2) 61 (25.3)
144 (85.2) 110 (65.1) 49 (29.0) 46 (27.2)
Antibiotics before ED visit
19 (15.0)
34 (6.9)
20 (8.3)
20 (11.8)
Initial vital signs Systolic BP, mmHg Diastolic BP, mmHg Heart rate, beats/min RR, cycles/min Body temperature, C
127.6 21.3 71.5 13.6 93.3 18.2 19.9 1.4 37.4 1.2
127.4 21.5 70.1 12.8 99.1 18.2 20.1 2.0 38.1 1.2
126.8 20.0 70.4 12.0 99.5 18.5 20.1 1.7 38.0 1.1
126.0 21.5 73.0 13.3 101.1 18.5 20.3 1.7 37.9 1.3
Laboratory findings WBC counts, 103 cells/mL Seg. neutrophils, % Hemoglobin, mg/dL Platelet, 103 cells/mL BUN, mg/dL Serum creatinine, mg/dL C-reactive protein, mg/dL Urine WBCs 50/HPF
11.0 4.8 77.8 13.2 12.6 1.5 252.8 90.5 13.7 6.9 1.0 0.4 7.4 6.7 47 (37.0)
11.9 4.8 82.7 8.7 12.6 1.4 240.6 86.7 14.0 6.7 1.0 0.3 9.9 8.5 188 (38.1)
12.0 6.8 81.7 10.5 12.7 1.4 247.2 82.0 13.0 6.2 1.0 0.2 10.2 8.9 93 (38.6)
11.6 4.8 82.7 10.4 12.2 1.5 234.7 103.4 15.6 7.9 1.1 0.3 10.6 9.3 82 (48.5)
Bacteremia
e
97 (19.6)
44 (18.3)
43 (25.4)
(61.4) (54.3) (22.8) (22.8)
(84.0) (76.1) (32.4) (20.5)
ED Z emergency department; BP Z blood pressure; RR Z respiratory rate; WBC Z white blood cells; Seg. Z segmented; BUN Z blood urea nitrogen; HPF Z high-power field. a Data are presented as mean SD or frequency with percentage. b Cases were excluded because they did not have blood culture.
more frequently. These characteristics might make the physician decide not to perform blood culture. Blood culture is a simple and relatively inexpensive test with high specificity if skin contaminants are excluded. Therefore, it is a common practice to obtain blood culture at the presentation of patient suspicious of systemic infections. Treatment with antibiotics could be guided by the sensitivity test of the isolated organism. However, there is little evidence when to order a blood culture. As a result, the yields of blood culture are generally less than 10%.13,14 Significant medical resources are utilized for this low yield test. Furthermore, false positive result of blood culture can also increase unnecessary antibiotics treatment, admissions, and resource utilization.15 Nevertheless, the identification of bacteremia is critical because bacteremia is a potentially fatal condition, which may lead to severe sepsis, septic shock, and mortality. Thus, assessing the risk of bacteremia can be helpful to select the patients who benefit from blood culture.13e15 Some authors doubted the utility of blood culture in uncomplicated APN, because blood culture may only give
the information about whether bacteremia exists or not and did not change the antibiotic therapy in most cases.4e6 However, many physicians are still performing blood culture in uncomplicated APN with several concerns. First, the rate of bacteremia in APN is relatively high (about 20%).4,12,16 Second, the presence of bacteremia itself may give the information indicating that the active treatment is warranted because the patients with bacteremia have higher mortality and complication rate.2 Third, urine cultures are not always indicating the true pathogen. Urine cultures obtained by voiding urine are susceptible to contamination. Fourth, to identify the true pathogen and to obtain the susceptibility test is important because of the high prevalence of antimicrobial resistance in E. coli and the increasing trends of resistance experienced by most countries. The rate of ciprofloxacin-resistant E. coli isolated in this study was much higher than that in the previous study (18.7% vs 1%).7 In addition, 4 (3.0%) E. coli were extended-spectrum b-lactamases-producing organisms. With this background, we thought that a risk stratification model would help the clinician to perform blood culture
128 Table 2
K.S. Kim et al. Characteristics in patients with or without bacteremia in the derivation cohort.a Patients without bacteremia
Patients with bacteremia
Number of patients
397
97
Epidemiological data Age 65b Diabetes mellitus Hypertension
81 (20.4) 54 (13.6) 78 (19.7)
54 (55.7) 23 (23.7) 42 (43.3)
<0.001 0.014 <0.001
Clinical findings Fever Chills Anorexia Vomiting
334 (84.1) 298 (75.1) 121 (30.5) 66 (16.6)
81 78 39 35
0.880 0.268 0.066 <0.001
Antibiotics before ED visit
29 (7.3)
5 (5.2)
0.453
Initial vital signsb Systolic BP <100 mm Hg Heart rate >110 beats/min Respiratory rate >20 cycles/min Body temperature >39 C
20 94 57 97
5 (5.2) 38 (39.2) 20 (20.6) 27 (27.8)
0.962 0.002 0.128 0.489
Laboratory findingsb WBC counts <4000 or >12,000 cells/mL Segmented neutrophils >90% Hemoglobin <12 mg/dL Platelet counts <150,000 cells/mL Serum creatinine >1.5 mg/dL C-reactive protein >10 mg/dL
157 (39.6) 50 (12.6) 99 (24.9) 32 (8.1) 12 (3.0) 130 (32.8)
55 (56.7) 36 (37.1) 25 (25.8) 16 (16.5) 7 (7.2) 44 (45.4)
0.002 <0.001 0.865 0.012 0.054 0.020
Urine WBCs 50/HPF
124 (31.2)
64 (66.0)
<0.001
(5.0) (23.7) (14.4) (24.4)
(83.5) (80.4) (40.2) (36.1)
p value
ED Z emergency department; BP Z blood pressure; WBC Z white blood cells; HPF Z high-power field. a Data are presented as mean SD or frequency with percentage. b Continuous variables are dichotomized using the most discriminant cutoff point.
judiciously. To perform blood culture in patients with high risk of bacteremia seems reasonable. When considering the utility of blood culture in uncomplicated APN, patients with low risk of bacteremia (about 40% of all the cohorts) may not benefit from blood culture. Thus, considerable amount of resources to perform two sets of blood cultures can be saved. Nevertheless, the relevance of saving blood culture in low or intermediate risk patients is a resource-based issue. Among patients with low and intermediate risk of bacteremia, 5.7% and 16.2% still had bacteremia. In the setting with adequate resources or with the high prevalence of resistant strains, to perform blood culture in patients with low or intermediate risk of bacteremia is acceptable.
Therefore, the physician should consider the changing condition of the patient, the available resources and the rates of resistant strains in the community to decide whether to perform the blood culture or not in patients with low or intermediate risk of bacteremia. Complicated APN is a heterogeneous condition including severe sepsis, obstructive uropathy, immunocompromised state, and chronic renal disease. Although one study suggested that the utility of blood culture should be reevaluated,16 blood culture can be justified in these potentially fatal conditions. The definition of complicated APN also varies by the researchers. We have reported that old age and diabetes were not associated with hospital
Table 3 Variables selected in the final multivariate logistic regression model for predicting bacteremia and corresponding points assigned for the model in the derivation cohort. Age 65 Vomiting Heart rate >110 beats/min Segmented neutrophils >90% Urine WBCs 50/HPF
OR (95% CI)
Coefficient (95% CI)
p value
Points
5.18 2.40 2.35 3.17 4.27
1.64 0.88 0.85 1.15 1.45
<0.001 0.003 0.003 <0.001 <0.001
4 2 2 3 4
(3.03e8.85) (1.34e4.30) (1.34e4.12) (1.77e5.69) (2.53e7.22)
(1.11e2.18) (0.29e1.46) (0.29e1.42) (0.57e1.74) (0.93e1.98)
OR Z odds ratio; CI Z confidence interval; WBC Z white blood cells; HPF Z high-power field.
Predicting bacteremia in women with APN Table 4
Actual distribution of bacteremia stratified by the model.a
Cohorts
Derivation cohort Internal validation cohort External validation cohort a
129
Scores, 0e3
Scores, 4e6
Scores, 7e15
Low risk
Intermediate risk
High risk
Up to 10%
10e30%
Above 30%
9/208 (4.3) 9/106 (8.5) 3/53 (5.7)
28/168 (16.7) 14/85 (16.5) 9/61 (14.8)
60/118 (50.9) 21/50 (42.0) 31/55 (56.4)
Total
97/494 (19.6) 44/241 (18.3) 43/169 (25.4)
Data were presented as patients with bacteremia per total patients (percent).
Figure 1 ROC curves of the stratification model for predicting bacteremia (A) in derivation cohort, (B) in internal validation cohort, (C) in external validation cohort. Stratification cutoffs are at 0, 4, 7, and 15.
admissions.17 Thus, we have included these patients as uncomplicated cases. Although old age was selected as an independent risk factor of bacteremia in this study, patients with old age alone are not sufficient enough to be considered as high risk of bacteremia. This result is consistent with previous study that has evaluated 3,730 adult patients who underwent blood culture.18 Although patients with different diagnosis and severity were examined, the independent predictors of bacteremia in this study were comparable to the previous study,18 regarding vomiting and systemic inflammatory responses. This prediction rule needs the results of some blood and urine tests. One might argue that blood cultures could be obtained when routine blood tests are performed, and that this way be convenient and time-saving. In another point of view, however, the blood of patients could be discarded with no gain if blood culture seems useless with prediction rule after the results of blood test. The major component of prediction rule is also based on basic clinical survey, so higher probability of bacteremia could be predicted before blood and urine tests. In that case, blood culture could be taken at the same time with initial blood tests. This study has several limitations. First, the selection of patients might be biased because blood culture was not performed in all patients enrolled. As discussed above, patients without blood culture had mild symptoms, and the patients retrospectively enrolled had the definite diagnosis of APN. Therefore, the actual rate of bacteremia might be lower than the result of this study. Furthermore, this study was performed in the two EDs. Therefore, this model should be re-evaluated in the different population and settings for the generalizability. Second, the external validation cohort was enrolled retrospectively. Although we eliminated cases without adequate data, potential biases still remain.
In conclusion, the risk of bacteremia in patients with uncomplicated APN can be stratified by this internally and externally validated model using clinical variables. This model can be helpful to decide whether to perform blood cultures and whether to admit the patient for the intravenous antibiotics in patients with uncomplicated APN.
Financial support/conflict of interest Nothing to declare.
References 1. Hooton TM, Stamm WE. Diagnosis and treatment of uncomplicated urinary tract infection. Infect Dis Clin North Am 1997; 11:551e81. 2. Brown P, Ki M, Foxman B. Acute pyelonephritis among adults: cost of illness and considerations for the economic evaluation of therapy. Pharmacoeconomics 2005;23:1123e42. 3. Ramakrishnan K, Scheid DC. Diagnosis and management of acute pyelonephritis in adults. Am Fam Physician 2005;71: 933e42. 4. Velasco M, Martı´nez JA, Moreno-Martı´nez A, Horcajada JP, Ruiz J, Barranco M, et al. Blood cultures for women with uncomplicated acute pyelonephritis: are they necessary? Clin Infect Dis 2003;37:1127e30. 5. Thanassi M. Utility of urine and blood cultures in pyelonephritis. Acad Emerg Med 1997;4:797e800. 6. McMurray BR, Wrenn KD, Wright SW. Usefulness of blood cultures in pyelonephritis. Am J Emerg Med 1997;15:137e40. 7. Talan DA, Krishnadasan A, Abrahamian FM, Stamm WE, Moran GJ. Prevalence and risk factor analysis of trimethoprim-sulfamethoxazoleand fluoroquinoloneresistant Escherichia coli infection among emergency department patients with pyelonephritis. Clin Infect Dis 2008;47: 1150e8.
130 8. Wagenlehner FM, Weidner W, Naber KG. An update on uncomplicated urinary tract infections in women. Curr Opin Urol 2009;19:368e74. 9. Kim K, Lee CC, Rhee JE, Suh GJ, Lee HJ, Kim HB, et al. The effects of an institutional care map on the admission rates and medical costs in women with acute pyelonephritis. Acad Emerg Med 2008;15:319e23. 10. Schwarz G. Estimating the dimension of a model. Ann Stat 1978;6:461e4. 11. Kass RE, Raftery AE. Bayes factors. J Am Stat Assoc 1995;90: 773e95. 12. van Nieuwkoop C, Bonten TN, van’t Wout JW, Kuijper EJ, Groeneveld GH, Becker MJ, et al. Procalcitonin reflects bacteremia and bacterial load in urosepsis syndrome: a prospective observational study. Crit Care 2010;14:R206. 13. Bates DW, Cook EF, Goldman L, Lee TH. Predicting bacteremia in hospitalized patients. A prospectively validated model. Ann Intern Med 1990;113:495e500.
K.S. Kim et al. 14. Pfitzenmeyer P, Decrey H, Auckenthaler R, Michel JP. Predicting bacteremia in older patients. J Am Geriatr Soc 1995;43: 230e5. J, Ruiz15. Falguera M, Trujillano J, Caro S, Mene ´ndez R, Carratala lez A, et al. A prediction rule for estimating the risk of Gonza bacteremia in patients with community-acquired pneumonia. Clin Infect Dis 2009;49:409e16. 16. Chen Y, Nitzan O, Saliba W, Chazan B, Colodner R, Raz R. Are blood cultures necessary in the management of women with complicated pyelonephritis? J Infect 2006; 53:235e40. 17. Rhee JE, Kim K, Lee CC, Kang J, Lee JW, Shin JH, et al. The lack of association between age and diabetes and hospitalization in women with acute pyelonephritis. J Emerg Med; 2008. Nov 8 [Epub ahead of print]. 18. Shapiro NI, Wolfe RE, Wright SB, Moore R, Bates DW. Who needs a blood culture? A prospectively derived and validated prediction rule. J Emerg Med 2008;35:255e64.
www.elsevierhealth.com/journals/jinf
A simple model to predict bacteremia in women with acute pyelonephritis Kyung Su Kim a, Kyuseok Kim b,*, You Hwan Jo b, Tae Yun Kim b, Jin Hee Lee b, Se Jong Lee a, Joong Eui Rhee b, Gil Joon Suh a a
Department of Emergency Medicine, Seoul National University Hospital, Seoul, Republic of Korea Department of Emergency Medicine, Seoul National University Bundang Hospital, 166 Gumi-ro, Bundang-gu, Seongnam 463-707, Republic of Korea
b
Accepted 10 June 2011 Available online 22 June 2011
KEYWORDS Pyelonephritis; Bacteremia; Logistic models; Risk assessment; Community-acquired infections
Summary Objectives: To construct a simple model to predict bacteremia in women with uncomplicated acute pyelonephritis (APN) for the judicious use of blood cultures. Methods: A prospective database including 735 women with uncomplicated APN at an academic urban emergency department was analyzed retrospectively. Independent risk factors were determined using multivariate logistic regression in two-thirds of patients. Cutoff values representing 10% and 30% of risk were selected for the stratification. This model was internally and externally validated using a remaining one-thirds of patients and 169 independent patients, respectively. Results: Independent risk factors were as follows: age 65 years (odds ratio [OR]Z5.18, 4 points), vomiting (OR Z 2.40, 2 points), heart rate >110 beats/min (OR Z 2.35, 2 points), segmented neutrophils >90% (OR Z 3.17, 3 points), and urine WBC 50/HPF (OR Z 4.27, 4 points). Patients were stratified as low (points <4), intermediate (points, 4e6), or high risk (7 points). The areas under receiver operating characteristics curves were 0.707 and 0.792 in internal and external validation cohorts, respectively. The model stratified internal and external validation cohort into low (8.5% and 5.7%), intermediate (16.5% and 14.8%), and high risk of bacteremia (42.0% and 56.4%). Conclusions: This model provides a useful tool to predict the risk of bacteremia, which can be helpful to decide whether to perform blood cultures and whether to admit the patient for the intravenous antibiotics in women with uncomplicated APN. ª 2011 The British Infection Association. Published by Elsevier Ltd. All rights reserved.
* Corresponding author. Tel.: þ82 31 787 7572; fax: þ82 31 787 4055. E-mail address: [email protected] (K. Kim). 0163-4453/$36 ª 2011 The British Infection Association. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2011.06.007
Predicting bacteremia in women with APN
Introduction Urinary tract infections (UTIs) are the most common bacterial infections in women. Among UTIs, acute pyelonephritis (APN) is a potentially severe disease with approximately 250,000 cases and $2.14 billion costs in the United States each year.1,2 Uncomplicated APN is defined as APN without structural and functional abnormalities within the urinary tract, without relevant kidney diseases and without relevant comorbidities.3 Some authors have suggested that routine blood cultures are not necessary for patients with uncomplicated APN.4e6 However, there is no consensus whether to perform blood culture or not. Therefore, many unjustified blood cultures are being performed concerning about high rate of bacteremia and increasing prevalence of resistant strains.7,8 The aim of this study was to construct a risk stratification model for bacteremia in uncomplicated APN to guide the judicious use of blood culture.
Materials and methods Study design We analyzed retrospectively a prospective database. This database was originally designed to assess the effectiveness of an institutional admission protocol in women with APN.9 After the study, we have continued registering the patients and maintaining the database. The cohorts obtained from this database were used for the derivation and internal validation of the risk prediction model. Furthermore, we retrospectively identified the patients with uncomplicated APN in an independent academic hospital emergency department (ED) for the external validation. This study was approved by the institutional review board and was exempted from informed consent.
Study setting and population A prospective database was obtained in a single academic, urban ED with an annual census of approximately 65,000 between January 2006 and September 2009. We reviewed the database, which was collected by following criteria.9 All consecutive female patients who were 15 years old and had APN were prospectively enrolled between January 2006 and September 2009. The diagnostic criteria for APN included the following: (1) a body temperature (BT) over 38 C, (2) pyuria, and (3) lumbar tenderness. Pyuria was defined as the presence of leukocytes 10 per high-power field in a centrifuged urine specimen. If the patient had taken any antipyretic agent before visiting the ED, a BT over 38 C was not required for diagnosis. Exclusion criteria were as follows: (1) severe sepsis or septic shock; (2) an immunocompromised state (on chemotherapy or immunosuppressive medication); (3) history of chronic renal failure; (4) acute renal failure (serum creatinine >2 mg/dL); (5) obstructive APN; (6) neurogenic bladder; (7) patients with urinary catheters; and (8) previous kidney transplantation. Advanced age, liver cirrhosis, diabetes mellitus, and other chronic diseases were not considered as exclusion criteria.
125 External validation cohorts were obtained in an independent academic, urban ED with an annual census of approximately 50,000 between January 2009 and December 2009. Patients were considered eligible for the external validation cohort when they had the discharge diagnosis of APN and underwent blood culture tests. After reviewing medical records of all eligible patients, patients were enrolled using the same criteria described previously.
Measures All data were collected based on both the established protocol and the unified definition of variables. We evaluated potential risk factors usually available prior to the decision to perform the blood cultures. Those were as follows: (1) epidemiologic data including age, diabetes mellitus, and hypertension; (2) clinical findings including fever, chill, anorexia, and vomiting; (3) antibiotics therapy before ED visit; (4) initial vital signs including systolic blood pressure (BP), diastolic BP, heart rate, respiration rate, and BT; and (5) initial laboratory findings including WBC counts, percentage of segmented neutrophils, hemoglobin, platelet counts, blood urine nitrogen (BUN), creatinine, C-reactive protein, and urine WBC counts. In retrospective cohort, missing information about clinical findings was considered to indicate the absence of that finding. Patients with insufficient data were further excluded from the analysis. Blood culture was performed at the discretion of the attending physician. To perform the blood cultures, 10 mL of blood was obtained and inoculated into 2 culture bottles (5 mL into an aerobic bottle and 5 mL into an anaerobic bottle). An additional 10 ml of blood was obtained from a different site and the procedure was repeated. Blood culture bottles were processed in an automated blood culture system. The following isolates were considered to be contaminants if they were isolated at only one culture bottle and discordant from urine cultures: Streptococcus pyogenes, Bacillus species, Micrococcus species, and coagulase-negative staphylococci except Staphylococcus saprophyticus.
Data analysis After excluding the patients without blood culture from the prospective database, the patients were randomized into 2 groups to create derivation cohort with two-thirds of the patients and an internal validation cohort with the remaining one third. Then, we compared patients with and without bacteremia in derivation cohort. Continuous variables were dichotomized considering both clinical relevancy and statistical significance. Univariate logistic regression analyses using various cutoffs were performed for predicting bacteremia. The most discriminant cutoff was determined using the p value and Bayesian information criterion (BIC).10 The cutoffs selected were as follows: age 65, systolic BP < 100 mm Hg, heart rate >110 beats/min, respiratory rate >20 cycles/ min, body temperature >39 C, WBC counts <4000 or >12,000 cells/mL, segmented neutrophils >90%, hemoglobin <12 mg/dL, platelet counts <150,000 cells/mL, serum creatinine >1.5 mg/dL, C-reactive protein >10 mg/dL, and urine WBCs 50/HPF. Statistically significant (p < 0.05) variables
126 were selected as candidate variables for the model derivation. Using all the candidate variables, multivariate logistic regression model predicting bacteremia was constructed. Variables was then removed from this model in a stepwise manner until none remained. The final model with best fit was determined using BIC. BIC is a criterion for model selection that penalizes the number of variables to avoid overfitting.10 A model is better than another if it has a smaller BIC value.11 Then, the corresponding logistic regression model coefficients were divided by 0.4 and rounded to the numbers. These numbers were assigned to the independent variables to estimate the risk of bacteremia. To stratify the risk of bacteremia as low, intermediate, or high, we selected the specific cutoffs representing predicted probability of 10% and 30%. Results of multivariate analysis are reported as odds ratios (OR) with 95% confidence intervals (CIs) and p values. The performance of the model was tested with respect to discrimination and calibration. Discrimination was quantified with the areas under the receiver operating characteristics curve (AUCs) with its corresponding 95% CIs. Calibration was tested with stratification tables. Continuous variables were expressed as means with SDs, and categorical data were presented as the percent frequency of occurrence. A Student t test was used to compare continuous variables, and the c2 or Fischer exact test was used to compare binomial variables. Two-tailed p < 0.05 was considered to indicate statistical significance. All analyses were performed using Stata version 10.1 (Stata Corp, College Station, Tex).
Results A total of 862 cases with uncomplicated APN were enrolled in the prospective cohort. Among them, 127 cases did not have blood cultures. The characteristics of these cases are described in Table 1. Cases without blood culture had fever (p < 0.001), chill (p < 0.001), and anorexia (p Z 0.027) less frequently than those with blood culture. They received antibiotics before ED visit more frequently (p Z 0.004). Initial heart rate was slower (p < 0.001) and BT (p < 0.001), the percentage of segmented neutrophils in white blood cells (p < 0.001), and C-reactive protein (p < 0.001) were lower in cases without blood culture. The other characteristics were not different. Among 735 cases with blood cultures in the prospective cohorts, true bacteremia was identified in 141 (19.2%) cases and contaminants were detected in 41 (5.6%) cases. Among the true pathogens, Escherichia coli was the most frequently isolates, found in 134 (95.0%) cases, followed by Klebsiella pneumoniae, found in 3 (2.1%); Enterobacter aerogenes, found in 1 (0.7%); Enterococcus faecalis, found in 1 (0.7%); Corynebacterium species, found in 1 (0.7%); Proteus mirabilis found in 1 (0.7%); and Salmonella species, found in 1 (0.7%). There was one case with polymicrobial infections (K. pneumonia and P. mirabilis). Among 134 isolates of E. coli, 25 (18.7%) were resistant to ciprofloxacin and 4 (3.0%) were extended-spectrum b-lactamases-producing organisms. All 4 extended-spectrum b-lactamases-producing organisms were sensitive to ciprofloxacin. One of 3 isolates of K. pneumoniae was resistant to ciprofloxacin.
K.S. Kim et al. Among these 735 cases, 494 (67.2%) were randomly assigned to the derivation cohort, and 241 (32.8%) were assigned to the internal validation cohort. The characteristics of cases in derivation and internal validation cohorts are presented in Table 1. In the derivation cohort, 10 variables were significantly associated with bacteremia, including epidemiological data (age 65, diabetes mellitus, and hypertension), vomiting, tachycardia (heart rate >110 beats/min), and laboratory findings (WBC counts <4000 or >12,000 cells/mL, segmented neutrophils >90%, platelet counts <150,000 cells/mL, C-reactive protein >10 mg/dL, and urine WBCs 50/HPF) (Table 2). Variables selected in the final multivariate logistic regression model were as follows: age 65, vomiting, heart rate >110 beats/min, segmented neutrophils >90%, and urine WBCs 50/HPF (Table 3). According to the regression coefficient, points were assigned as described in Table 3. The specific cutoffs representing predicted probability of 10% and 30% were 4 and 7, respectively. The application of the model enables us to categorize derivation cohort into low (4.3%), intermediate (16.7%), and high (50.9%) risk groups of bacteremia (Table 4). The AUC of the stratification model was 0.792 (95% CI, 0.746e0.839) in the derivation cohort (Fig. 1). Internal validation cohort was categorized into low (8.5%), intermediate (16.5%), and high (42.0%) risk groups of bacteremia (Table 4). The AUC of the model was 0.707 (95% CI, 0.623e0.790) in the internal validation cohort (Fig. 1). A total of 198 cases who had the discharge diagnosis of APN and underwent blood culture tests were enrolled for the external validation cohort. After excluding 29 cases with exclusion criteria or without insufficient data, 169 cases were analyzed. Table 1 shows the characteristics of the external validation cohort. Among 43 cases with bacteremia, E. coli was found in 39 (90.7%) cases, followed by K. pneumoniae, found in 2 (4.7%); Bacteroides fragilis, found in 1 (2.3%); and Citrobacter freundii complex, found in 1 (2.3%). The model stratified external validation cohort into low (5.7%), intermediate (14.8%), and high (56.4%) risk of bacteremia (Table 4). The AUC of the model was 0.792 (95% CI, 0.720e0.865) (Fig. 1).
Discussion In this study, we developed and validated a stratification model for the risk of bacteremia in uncomplicated APN. The discrimination ability of the model was generally good (AUC > 0.700) in all cohorts and the stratification ability at predetermined probability was also reasonable. Recently, the model to predict the bacteremia in febrile urosepsis syndrome was published, but our study was different from the previous one in terms of subjects and methods to develop the model. The merit of our study was to validate the model both internally and externally with good discrimination. On the contrary, they added procalcitonin to the usual clinical variables in the prediction model, and yielded more discriminative power.12 Among 862 cases in prospective database, 127 (14.7%) did not receive blood culture. Cases without blood culture tend to have mild symptoms and relatively normal vital signs. They also have received antibiotics before ED visit
Predicting bacteremia in women with APN Table 1
127
Characteristics of all cohorts.a Prospective cohorts Excluded
b
External validation cohort
Derivation
Internal validation
Number of patients
127
494
241
169
Epidemiologic data Age Diabetes mellitus Hypertension
48.5 20.0 16 (12.6) 24 (18.9)
50.5 18.6 77 (15.6) 120 (24.3)
48.0 18.5 28 (11.6) 56 (23.2)
54.2 18.8 32 (18.9) 54 (32.0)
Clinical findings Fever Chills Anorexia Vomiting
78 69 29 29
415 376 160 101
203 (84.2) 184 (76.4) 80 (33.2) 61 (25.3)
144 (85.2) 110 (65.1) 49 (29.0) 46 (27.2)
Antibiotics before ED visit
19 (15.0)
34 (6.9)
20 (8.3)
20 (11.8)
Initial vital signs Systolic BP, mmHg Diastolic BP, mmHg Heart rate, beats/min RR, cycles/min Body temperature, C
127.6 21.3 71.5 13.6 93.3 18.2 19.9 1.4 37.4 1.2
127.4 21.5 70.1 12.8 99.1 18.2 20.1 2.0 38.1 1.2
126.8 20.0 70.4 12.0 99.5 18.5 20.1 1.7 38.0 1.1
126.0 21.5 73.0 13.3 101.1 18.5 20.3 1.7 37.9 1.3
Laboratory findings WBC counts, 103 cells/mL Seg. neutrophils, % Hemoglobin, mg/dL Platelet, 103 cells/mL BUN, mg/dL Serum creatinine, mg/dL C-reactive protein, mg/dL Urine WBCs 50/HPF
11.0 4.8 77.8 13.2 12.6 1.5 252.8 90.5 13.7 6.9 1.0 0.4 7.4 6.7 47 (37.0)
11.9 4.8 82.7 8.7 12.6 1.4 240.6 86.7 14.0 6.7 1.0 0.3 9.9 8.5 188 (38.1)
12.0 6.8 81.7 10.5 12.7 1.4 247.2 82.0 13.0 6.2 1.0 0.2 10.2 8.9 93 (38.6)
11.6 4.8 82.7 10.4 12.2 1.5 234.7 103.4 15.6 7.9 1.1 0.3 10.6 9.3 82 (48.5)
Bacteremia
e
97 (19.6)
44 (18.3)
43 (25.4)
(61.4) (54.3) (22.8) (22.8)
(84.0) (76.1) (32.4) (20.5)
ED Z emergency department; BP Z blood pressure; RR Z respiratory rate; WBC Z white blood cells; Seg. Z segmented; BUN Z blood urea nitrogen; HPF Z high-power field. a Data are presented as mean SD or frequency with percentage. b Cases were excluded because they did not have blood culture.
more frequently. These characteristics might make the physician decide not to perform blood culture. Blood culture is a simple and relatively inexpensive test with high specificity if skin contaminants are excluded. Therefore, it is a common practice to obtain blood culture at the presentation of patient suspicious of systemic infections. Treatment with antibiotics could be guided by the sensitivity test of the isolated organism. However, there is little evidence when to order a blood culture. As a result, the yields of blood culture are generally less than 10%.13,14 Significant medical resources are utilized for this low yield test. Furthermore, false positive result of blood culture can also increase unnecessary antibiotics treatment, admissions, and resource utilization.15 Nevertheless, the identification of bacteremia is critical because bacteremia is a potentially fatal condition, which may lead to severe sepsis, septic shock, and mortality. Thus, assessing the risk of bacteremia can be helpful to select the patients who benefit from blood culture.13e15 Some authors doubted the utility of blood culture in uncomplicated APN, because blood culture may only give
the information about whether bacteremia exists or not and did not change the antibiotic therapy in most cases.4e6 However, many physicians are still performing blood culture in uncomplicated APN with several concerns. First, the rate of bacteremia in APN is relatively high (about 20%).4,12,16 Second, the presence of bacteremia itself may give the information indicating that the active treatment is warranted because the patients with bacteremia have higher mortality and complication rate.2 Third, urine cultures are not always indicating the true pathogen. Urine cultures obtained by voiding urine are susceptible to contamination. Fourth, to identify the true pathogen and to obtain the susceptibility test is important because of the high prevalence of antimicrobial resistance in E. coli and the increasing trends of resistance experienced by most countries. The rate of ciprofloxacin-resistant E. coli isolated in this study was much higher than that in the previous study (18.7% vs 1%).7 In addition, 4 (3.0%) E. coli were extended-spectrum b-lactamases-producing organisms. With this background, we thought that a risk stratification model would help the clinician to perform blood culture
128 Table 2
K.S. Kim et al. Characteristics in patients with or without bacteremia in the derivation cohort.a Patients without bacteremia
Patients with bacteremia
Number of patients
397
97
Epidemiological data Age 65b Diabetes mellitus Hypertension
81 (20.4) 54 (13.6) 78 (19.7)
54 (55.7) 23 (23.7) 42 (43.3)
<0.001 0.014 <0.001
Clinical findings Fever Chills Anorexia Vomiting
334 (84.1) 298 (75.1) 121 (30.5) 66 (16.6)
81 78 39 35
0.880 0.268 0.066 <0.001
Antibiotics before ED visit
29 (7.3)
5 (5.2)
0.453
Initial vital signsb Systolic BP <100 mm Hg Heart rate >110 beats/min Respiratory rate >20 cycles/min Body temperature >39 C
20 94 57 97
5 (5.2) 38 (39.2) 20 (20.6) 27 (27.8)
0.962 0.002 0.128 0.489
Laboratory findingsb WBC counts <4000 or >12,000 cells/mL Segmented neutrophils >90% Hemoglobin <12 mg/dL Platelet counts <150,000 cells/mL Serum creatinine >1.5 mg/dL C-reactive protein >10 mg/dL
157 (39.6) 50 (12.6) 99 (24.9) 32 (8.1) 12 (3.0) 130 (32.8)
55 (56.7) 36 (37.1) 25 (25.8) 16 (16.5) 7 (7.2) 44 (45.4)
0.002 <0.001 0.865 0.012 0.054 0.020
Urine WBCs 50/HPF
124 (31.2)
64 (66.0)
<0.001
(5.0) (23.7) (14.4) (24.4)
(83.5) (80.4) (40.2) (36.1)
p value
ED Z emergency department; BP Z blood pressure; WBC Z white blood cells; HPF Z high-power field. a Data are presented as mean SD or frequency with percentage. b Continuous variables are dichotomized using the most discriminant cutoff point.
judiciously. To perform blood culture in patients with high risk of bacteremia seems reasonable. When considering the utility of blood culture in uncomplicated APN, patients with low risk of bacteremia (about 40% of all the cohorts) may not benefit from blood culture. Thus, considerable amount of resources to perform two sets of blood cultures can be saved. Nevertheless, the relevance of saving blood culture in low or intermediate risk patients is a resource-based issue. Among patients with low and intermediate risk of bacteremia, 5.7% and 16.2% still had bacteremia. In the setting with adequate resources or with the high prevalence of resistant strains, to perform blood culture in patients with low or intermediate risk of bacteremia is acceptable.
Therefore, the physician should consider the changing condition of the patient, the available resources and the rates of resistant strains in the community to decide whether to perform the blood culture or not in patients with low or intermediate risk of bacteremia. Complicated APN is a heterogeneous condition including severe sepsis, obstructive uropathy, immunocompromised state, and chronic renal disease. Although one study suggested that the utility of blood culture should be reevaluated,16 blood culture can be justified in these potentially fatal conditions. The definition of complicated APN also varies by the researchers. We have reported that old age and diabetes were not associated with hospital
Table 3 Variables selected in the final multivariate logistic regression model for predicting bacteremia and corresponding points assigned for the model in the derivation cohort. Age 65 Vomiting Heart rate >110 beats/min Segmented neutrophils >90% Urine WBCs 50/HPF
OR (95% CI)
Coefficient (95% CI)
p value
Points
5.18 2.40 2.35 3.17 4.27
1.64 0.88 0.85 1.15 1.45
<0.001 0.003 0.003 <0.001 <0.001
4 2 2 3 4
(3.03e8.85) (1.34e4.30) (1.34e4.12) (1.77e5.69) (2.53e7.22)
(1.11e2.18) (0.29e1.46) (0.29e1.42) (0.57e1.74) (0.93e1.98)
OR Z odds ratio; CI Z confidence interval; WBC Z white blood cells; HPF Z high-power field.
Predicting bacteremia in women with APN Table 4
Actual distribution of bacteremia stratified by the model.a
Cohorts
Derivation cohort Internal validation cohort External validation cohort a
129
Scores, 0e3
Scores, 4e6
Scores, 7e15
Low risk
Intermediate risk
High risk
Up to 10%
10e30%
Above 30%
9/208 (4.3) 9/106 (8.5) 3/53 (5.7)
28/168 (16.7) 14/85 (16.5) 9/61 (14.8)
60/118 (50.9) 21/50 (42.0) 31/55 (56.4)
Total
97/494 (19.6) 44/241 (18.3) 43/169 (25.4)
Data were presented as patients with bacteremia per total patients (percent).
Figure 1 ROC curves of the stratification model for predicting bacteremia (A) in derivation cohort, (B) in internal validation cohort, (C) in external validation cohort. Stratification cutoffs are at 0, 4, 7, and 15.
admissions.17 Thus, we have included these patients as uncomplicated cases. Although old age was selected as an independent risk factor of bacteremia in this study, patients with old age alone are not sufficient enough to be considered as high risk of bacteremia. This result is consistent with previous study that has evaluated 3,730 adult patients who underwent blood culture.18 Although patients with different diagnosis and severity were examined, the independent predictors of bacteremia in this study were comparable to the previous study,18 regarding vomiting and systemic inflammatory responses. This prediction rule needs the results of some blood and urine tests. One might argue that blood cultures could be obtained when routine blood tests are performed, and that this way be convenient and time-saving. In another point of view, however, the blood of patients could be discarded with no gain if blood culture seems useless with prediction rule after the results of blood test. The major component of prediction rule is also based on basic clinical survey, so higher probability of bacteremia could be predicted before blood and urine tests. In that case, blood culture could be taken at the same time with initial blood tests. This study has several limitations. First, the selection of patients might be biased because blood culture was not performed in all patients enrolled. As discussed above, patients without blood culture had mild symptoms, and the patients retrospectively enrolled had the definite diagnosis of APN. Therefore, the actual rate of bacteremia might be lower than the result of this study. Furthermore, this study was performed in the two EDs. Therefore, this model should be re-evaluated in the different population and settings for the generalizability. Second, the external validation cohort was enrolled retrospectively. Although we eliminated cases without adequate data, potential biases still remain.
In conclusion, the risk of bacteremia in patients with uncomplicated APN can be stratified by this internally and externally validated model using clinical variables. This model can be helpful to decide whether to perform blood cultures and whether to admit the patient for the intravenous antibiotics in patients with uncomplicated APN.
Financial support/conflict of interest Nothing to declare.
References 1. Hooton TM, Stamm WE. Diagnosis and treatment of uncomplicated urinary tract infection. Infect Dis Clin North Am 1997; 11:551e81. 2. Brown P, Ki M, Foxman B. Acute pyelonephritis among adults: cost of illness and considerations for the economic evaluation of therapy. Pharmacoeconomics 2005;23:1123e42. 3. Ramakrishnan K, Scheid DC. Diagnosis and management of acute pyelonephritis in adults. Am Fam Physician 2005;71: 933e42. 4. Velasco M, Martı´nez JA, Moreno-Martı´nez A, Horcajada JP, Ruiz J, Barranco M, et al. Blood cultures for women with uncomplicated acute pyelonephritis: are they necessary? Clin Infect Dis 2003;37:1127e30. 5. Thanassi M. Utility of urine and blood cultures in pyelonephritis. Acad Emerg Med 1997;4:797e800. 6. McMurray BR, Wrenn KD, Wright SW. Usefulness of blood cultures in pyelonephritis. Am J Emerg Med 1997;15:137e40. 7. Talan DA, Krishnadasan A, Abrahamian FM, Stamm WE, Moran GJ. Prevalence and risk factor analysis of trimethoprim-sulfamethoxazoleand fluoroquinoloneresistant Escherichia coli infection among emergency department patients with pyelonephritis. Clin Infect Dis 2008;47: 1150e8.
130 8. Wagenlehner FM, Weidner W, Naber KG. An update on uncomplicated urinary tract infections in women. Curr Opin Urol 2009;19:368e74. 9. Kim K, Lee CC, Rhee JE, Suh GJ, Lee HJ, Kim HB, et al. The effects of an institutional care map on the admission rates and medical costs in women with acute pyelonephritis. Acad Emerg Med 2008;15:319e23. 10. Schwarz G. Estimating the dimension of a model. Ann Stat 1978;6:461e4. 11. Kass RE, Raftery AE. Bayes factors. J Am Stat Assoc 1995;90: 773e95. 12. van Nieuwkoop C, Bonten TN, van’t Wout JW, Kuijper EJ, Groeneveld GH, Becker MJ, et al. Procalcitonin reflects bacteremia and bacterial load in urosepsis syndrome: a prospective observational study. Crit Care 2010;14:R206. 13. Bates DW, Cook EF, Goldman L, Lee TH. Predicting bacteremia in hospitalized patients. A prospectively validated model. Ann Intern Med 1990;113:495e500.
K.S. Kim et al. 14. Pfitzenmeyer P, Decrey H, Auckenthaler R, Michel JP. Predicting bacteremia in older patients. J Am Geriatr Soc 1995;43: 230e5. J, Ruiz15. Falguera M, Trujillano J, Caro S, Mene ´ndez R, Carratala lez A, et al. A prediction rule for estimating the risk of Gonza bacteremia in patients with community-acquired pneumonia. Clin Infect Dis 2009;49:409e16. 16. Chen Y, Nitzan O, Saliba W, Chazan B, Colodner R, Raz R. Are blood cultures necessary in the management of women with complicated pyelonephritis? J Infect 2006; 53:235e40. 17. Rhee JE, Kim K, Lee CC, Kang J, Lee JW, Shin JH, et al. The lack of association between age and diabetes and hospitalization in women with acute pyelonephritis. J Emerg Med; 2008. Nov 8 [Epub ahead of print]. 18. Shapiro NI, Wolfe RE, Wright SB, Moore R, Bates DW. Who needs a blood culture? A prospectively derived and validated prediction rule. J Emerg Med 2008;35:255e64.