Diagnostic stewardship and AI could improve UTI care
Click Here to Manage Email Alerts
Antibiotic use in asymptomatic bacteriuria (ASB) is a commonly encountered issue for a stewardship pharmacist and one of the most frustrating.
UTI is a top indication for antibiotic use in both the hospital and outpatient settings and leads to a significant amount of inappropriate antibiotic use. The inappropriate use is multifactorial and includes unnecessarily broad antibiotic coverage, prolonged duration of therapy and treatment of ASB. With few exceptions — such as in pregnant women and patients undergoing invasive urologic procedures — treatment of ASB has not been found to be beneficial.
ASB is defined as having a urine specimen with one or more bacterial organisms with quantitative counts of 105 colony-forming units/mL or more irrespective of the presence of pyuria when there are no signs or symptoms attributable to a UTI. It is a common finding among patients, including healthy women, persons with underlying urologic abnormalities, older adults, and especially those living in long-term care facilities. It is a misnomer that urine is sterile; the healthy urinary tract has its own microbiome.
The incidence of ASB among women aged between 15 and 24 years is 2.7% but increases to as high as 50% among those aged older than 80 years. ASB in men is lower but still occurs in up to 20% of those aged older than 80 years.
Clinicians often feel compelled to treat patients with antibiotics when they have a positive urine culture despite the absence of symptoms. Antibiotic treatment of ASB is frequent, especially without antibiotic stewardship program interventions, particularly in hospital and long-term care facilities. When antibiotics are prescribed for UTI, the prescribed antibiotic and duration of therapy is often not optimal. This can lead to consequences such as adverse events from antibiotics, induction of antibiotic resistance, Clostridioides difficile infection and increased costs of care.
Diagnostic stewardship
Because a positive urine culture often triggers antibiotic treatment, obtaining urine cultures when not clinically indicated encourages inappropriate antimicrobial use. To reduce antibiotic treatment of ASB, it is paramount to optimize the ordering, processing and reporting of urinalysis (UA) and urine culture results.
Urine cultures are often ordered by “reflex” based on the results of the UA. Although the criteria for triggering a urine culture are not standardized, pyuria is often used. The current cutoff values for significant pyuria generally used are in the range of five to 10 leukocytes per high-powered field. A recent study evaluated the optimal cutoff values for pyuria for UTI diagnosis in older women. They found that the commonly used 10 leukocytes/µg had a poor specificity (36%) but a sensitivity of 100%. A higher cutoff of 264 leukocytes/µg had a much better sensitivity of 88%, although the specificity dropped to 88%. Based on these data, it appears that the currently used pyuria cutoff is too low, which may contribute to overdiagnosis of UTI in older women, leading to overuse of antibiotics.
An expert panel used a Modified-Delphi Procedure to identify best practices for urine culture diagnostic stewardship practices. The recommendations that this group identified as best practices included requiring documentation of signs or symptoms of UTI to obtain a urine culture, discouraging the ordering of urine cultures in the absence of symptoms of UTI, replacing stand-alone urine culture orders with reflex urine cultures and automatically canceling repeat urine cultures within 5 days of a positive culture during the same hospital admission or 7 days for long-term care residents.
Conversely, the group identified several examples of inappropriate urine culture use: the inclusion of urine cultures in most standing order sets, and urine cultures in response to a change in urine characteristics, such as smell or look.
The group offered additional guidance regarding the reporting of urine culture results. Many of the best practice recommendations for reporting centered around encouraging clinicians not to treat asymptomatic bacteriuria and not treating cultures with mixed flora or potential contaminants. Additionally, it also encouraged using selective reporting of antibiotic susceptibility to prioritize preferred antibiotics while suppressing those with higher risk of harmful side effects such as fluoroquinolones.
Machine learning applications
Machine learning and artificial intelligence (AI) are now being used in many different areas, including in antimicrobial stewardship.
Researchers recently reported the results of a study evaluating the deployment of a machine learning decision tree algorithm to optimize UA parameters for predicting positive urine cultures. In this study, researchers developed a decision tree algorithm for predicting urine culture positivity using macroscopic and microscopic UA features with a supervised rule-based machine learning algorithm. The aim was to implement this in the electronic health record as a diagnostic stewardship initiative to increase the appropriateness of urine culture testing.
The researchers included adult patients in nonmaternity inpatient and outpatient units at five hospitals in an academic health care system. The algorithm training used results from 19,511 paired UA and urine culture cases from patients with an average age of 57.4 years, with 70% being from female patients. The model identified urine white blood cells, leukocyte esterase and bacteria as the best predictors of urine culture positivity. Overall, the algorithm met the target negative predictive value (NPV) of greater than 90%. However, the NPV for samples from women and patients in the ED failed to meet the 90% prespecified NPV target, likely because these patient groups tend to have an increased prevalence of positive urine cultures.
Although machine learning may increase laboratory efficiency in predicting who has the highest likelihood of positive urine culture, the impact of this on antibiotic use is not well studied. A study from Denmark developed machine learning models and proposed prescription policies based on model predictions for patents with suspected UTI. They found that the implementation of their model and polices reduced antibiotic prescribing by 7.42% without reducing the number of treated patients who had a bacterial infection. Although this reduction is quite low, it is possible that a greater reduction in antibiotic use could be seen in populations with higher antibiotic utilization, since Denmark has lower antibiotic utilization than other parts of the world, such as the United States.
The bottom line is that urine cultures should only be obtained when there is a significant suspicion for a UTI based on patient symptoms. Management of ASB can be challenging because it requires clinical judgment, particularly when your patient has altered mental status or cognitive impairment. Newer strategies such as AI or machine learning may help reduce the number of urine cultures performed, reducing overuse of antibiotics for ASB.
References:
- Bilsen MP, et al. Clin Infect Dis. 2023;doi:10.1093/cid/ciad099.
- Claeys KC, et al. Clin Infect Dis. 2022;doi:1010.1093/cid/ciab987.
- Luu T, Albarillo FS. Am J Med. 2022;doi:10.1016/j.amjmed.2022.03.015.
- Nicolle LE, et al. Clin Infect Dis. 2019;doi:10.1093/cid/ciy1121.
- Ribers MA, Ullrich H. arXiv. 2019;doi: 10.48550/arXiv.1906.03044.
- Seheult JN, et al. J Clin Micro. 2023;doi:10.1128/jcm.00291-23.
For more information:
Jeff Brock, PharmD, MBA, BCIDP, is a Healio | Infectious Disease News Editorial Board Member and infectious disease pharmacy specialist at Mercy Medical Center in Des Moines, Iowa. He can be reached at jeff.brock@mercyoneiowa.org.