Author: Evelyn Schraft, MD
This paper by Shaikh et al published in JAMA Pediatrics in April 2018 examines the utility of a prediction tool created to evaluate the probability of a urinary tract infection (UTI) in febrile children aged less than two.
I found this paper useful because I still find urinalysis to be a bit of a mystery. In adult patients, it is a benign and easy test to order, so the difficulty for me is interpreting an indeterminate result. In children, the question gets shifted even sooner in the process as to whether or not to obtain urine by bag or catheter. As a provider, I am motivated to avoid unnecessary catheterizations or prolonged stays waiting for a bag sample. Of note, bag urine is only useful if it is negative (meaning no leukocyte esterase, nitrites, bacteria on Gram stain, or less than five white blood cells) given its high false positive rate.
The background of research in this area has shown it is difficult to evaluate preverbal children, especially for symptoms like dysuria. In the presentation of febrile young children, the fever source is UTI in 7% of the cases. We have calculators for other diseases including acute coronary syndrome, pulmonary embolism, alcohol withdrawal, and upper GI bleeding, but no widely utilized calculator for pediatric UTI prior to this study. Furthermore, a calculator would be useful considering culture is the gold standard for diagnosis, but is not timely enough to guide antibiotic therapy. The American Academy of Pediatrics (AAP) algorithm for UTI published in 2011 was the previous decision pathway intermittently accepted into practice prior to this study.
Previous studies have evaluated who needs further lab work based on clinical features and who needs antibiotic therapy based on labs prior to culture. This study goes on to question what COMBINATION of clinical and laboratory findings best predicts the risk of urinary tract infection.
The study was designed as a consecutive retrospective case control of children aged two months to two years with a fever of 38°C or greater who presented to the Children’s Hospital of Pittsburgh between January 2007 and April 2013. To be included, the patients must have had a urinalysis and culture within three hours of each other and a documented fever. Patients were excluded if they had genitourinary abnormalities, were taking antimicrobial drugs, immunosuppressants, or who had urine collected using a bag instead of by straight catheterization. Patients were considered to be positive for UTI if they had either pyuria defined as WBC ≥5/HPF or ≥10/μL or the presence of any leukocyte esterase AND uropathogen growth on culture of at least 50,000 CFU/ml.
There were two parts to the study. First, the study sought to develop a clinical and biochemical prediction tool for assessing the risk of UTI in preverbal febrile children. Second, it validated the newly developed prediction tool in a separate database.
In the first part of the study, a database of 1,686 patients was compiled with a disease to control ratio of 1:2 to test the development of the prediction tool via a nested case control study. Potential risk factors were based on literature review and analyses were run, removing factors if they did not meaningfully change the area under the curve (AUC). The AUC is a graph of sensitivity versus specificity, where it assesses the sensitivity and specificity at various values. The larger the “area” under this curve, the bigger the number, and the closer it gets to one (which is perfect sensitivity and specificity).
Five iterations of the prediction calculator were tested in the training database. The clinical model included risk factors of age less than twelve months, temperature greater than or equal to 39°C, nonblack race, female or uncircumcised male, and no other fever source. The remaining four models (laboratory models) included variations of lab tests including leukocyte esterase, nitrites, Gram stain, and urine white blood cell count.
The sensitivity and specificity of the prediction tool was examined in a second database that had a more realistic prevalence of UTI of 7% versus the 33% ratio of the first database. Similar results were found in both.
The research team chose cutoffs of 2% for the clinical model and 5% for the laboratory model with the thought to maintain a minimum sensitivity of 95% in the developed tool. As the sensitivity and specificity of the five iterations of the calculator were tested, it was discovered those including both clinical features and all the laboratory results were the most accurate. Of note, the specificity of using only clinical characteristics to predict UTI was relatively poor.
In addition, a hypothetical population of 1,000 febrile patients with 70 confirmed UTI based on the 7% prevalence was used to compare UTICalc to the AAP guidelines on UTI from 2011. The research team used this hypothetical population to elaborate the value of their calculator. UTICalc decreased the number of missed UTI, unnecessary antibiotic use, and delay in antibiotics.
With this information UTICalc was created, which is available at uticalc.pitt.edu. This clinical calculator can be used to input clinical characteristics of a child to recommend catheterization of urine based on the pretest probability. If the pretest probability is greater than two percent, it recommends urine catheterization. Furthermore, the results of urinalysis can be included to calculate the posttest probability of UTI to guide decision for antibiotic therapy. If the posttest probability is greater than five percent, it recommends empiric antibiotics.
UTICalc provides the next iteration of a calculator coming closer to a clinician’s clinical gestalt by using both the pretest probability AND the posttest probability to determine the need for antibiotic therapy.
The strengths of the study include:
- the large population size in the database used to create the prediction tool
- a separate validation cohort
- good retrospective chart review methodology
- the clinical importance of decreasing unnecessary catheterizations and treatment delays
- the creation of an app easily used in the emergency department given the complexity of the algorithm
- the app provides specific numbers to assist in shared decision-making versus a dichotomous yes/no
The weaknesses of the study include:
- it was a retrospective study and internally validated with potential unmeasured confounders
- the validation cohort was fairly small with only 30 total UTI cases included in the 7% prevalence
- the calculator should not be used indiscriminately due to the risk of false positives and over-testing if it is used in a case with low suspicion for UTI
- it excluded patients with genitourinary abnormalities, those taking antimicrobial drugs, immunosuppressants, or bagged specimens. Therefore, it should not be utilized in these populations.
UTICalc is a tool to risk stratify UTI. By quantifying the likelihood of UTI, the provider can better align with the risk threshold and engage in shared decision-making with the patient’s parents. The tool allows the provider to often avoid unnecessary urine samples. Incorporating the posttest probability in addition to the pretest probability helps quantify risk more accurately and can shed some of the ambiguity around indeterminate samples. However, as with any decision-support tool, it should be treated as a tool and not a rule.
Special thanks to Dr. Michael Gottlieb for his significant contributions to this article.
Citations:
Shaikh N, Hoberman A, Hum SW, Alberty A, Muniz G, Kurs-Lasky M, Landsittel D, Shope T. Development and Validation of a Calculator for Estimating the Probability of Urinary Tract Infection in Young Febrile Children. JAMA Pediatr. 2018 Jun 1;172(6):550-556. doi: 10.1001/jamapediatrics.2018.0217. PMID: 29710324; PMCID: PMC6137527.