Diagnostic uncertainty and urinary tract infection in the emergency department: a cohort study from a UK hospital

Background Suspected urinary tract infection (UTI) syndromes are a common reason for empirical antibiotics to be prescribed in the Emergency Department (ED), but differentiating UTI from other conditions with a similar presentation is challenging. We investigated how often an ED diagnosis of UTI is confirmed clinically/microbiologically, and described conditions which present as UTI syndromes. Methods Observational study using electronic health records from patients who attended the ED with suspected UTI and had a urine sample submitted for culture. We compared the ED diagnosis to diagnosis at discharge from hospital (ICD-10 codes), and estimated the proportion of cases with clinical/microbiological evidence of UTI. Results Two hundred eighty nine patients had an ED diagnosis of UTI syndrome comprising: lower UTI (191), pyelonephritis (56) and urosepsis (42). In patients admitted to hospital with an ED diagnosis of lower UTI, pyelonephritis or urosepsis, clinical/microbiological evidence of UTI was lacking in 61/103, 33/54 and 31/42 cases respectively. The ED diagnosis was concordant with the main reason for admission in less than 40% of patients with UTI syndromes, and antibiotics were stopped within 72 h in 37/161 patients. Conclusions Clinical/microbiological evidence of UTI was lacking in 60–70% of patients, suggesting scope to revise empirical prescribing decisions for UTI syndromes in light of microbial culture and clinical progression.


Background
Urinary tract infection syndromes are a major cause of ED attendance and hospital admission [1]. Concerns over delaying antibiotic treatment for severe infection in patients with suspected UTI means that clinicians have a low threshold for initiating antibiotics in the ED, however differentiating UTI from other conditions with a similar presentation can be challenging. These difficulties are compounded by conflicting messages which promote rapid instigation of antibiotics for suspected sepsis on the one hand, and prudent antibiotic use on the other. Diagnosing UTI is particularly challenging because patients may present with atypical signs and symptoms [2], diagnostic tests based on microbial culture of urine or blood take 48-72 h, the finding of bacteriuria without associated urinary symptoms does not usually warrant antibiotic treatment [3], and rapid urine dipstick tests have a poor positive predictive value for bacteriuria [4].
Previous studies have suggested that UTIs may be both over and under-diagnosed in the ED [5,6], particularly in the elderly [7,8], with one study suggesting that > 40% of patients aged > 65 years who had been diagnosed with UTI had no evidence of this condition [9]. Although a non-infectious cause is established for many of these cases, antibiotics are often continued unnecessarily, which drives the emergence of anti-bacterial resistance (ABR).
Using electronic health records supplemented by medical note review, we set out to estimate the frequency of over-diagnosis of UTI syndromes in the ED, and to describe the conditions which present as UTI in the ED in order to estimate the potential to reduce antibiotic prescribing by stopping antibiotics early in patients with no evidence of bacterial infection.

Study design and source of data
We undertook a retrospective cohort study in patients who were investigated for suspected UTI syndromes between 1st January 2014 and 30th June 2017 at the Emergency Department (ED) of the Queen Elizabeth Hospital Birmingham (QEHB), which is part of University Hospitals Birmingham NHS Foundation Trust. QEHB is the largest single site hospital in England with over 1200 inpatient beds and a local resident population of around 1.3 million people, which is ethnically diverse with high levels of social deprivation. Individuals were eligible for inclusion in the study if they attended the ED and had a urine sample submitted for microbiological culture from the ED (defined as receipt of a urine culture in the laboratory ±2 days from ED attendance since there can be a delay between sample collection and when the urine sample is recorded in the laboratory system). From this group we used ED diagnostic codes to identify patients with suspected UTI, and used simple random sampling to select 300 patients who were discharged from the ED and 700 patients who had been admitted to hospital. ED diagnostic codes represent the main reason for the patients ED attendance, are recorded by the ED physician and distinguish between urological conditions and different UTI syndromes.
Patients with suspected UTI who were discharged from the ED We estimated the proportion of patients discharged from the ED with suspected UTI who had clinical/ microbiological evidence of this diagnosis, based on the presence of urinary symptoms and microbial culture of urine and/or blood. A single investigator (LS) evaluated whether urinary symptoms had been recorded in the ED notes, considering: dysuria, frequency, hesitancy, urinary retention, difficulty passing urine, haematuria and malodorous urine. Information was also extracted about pain localised to the back, abdomen, flank, loin and suprapubic region. Symptoms were classified as probable if patients reported at least two urinary symptoms or a urinary symptom and pain, and possible if patients reported only one urinary symptom or only pain. Patients with no record of relevant pain or urinary symptoms were classified as having no symptoms of UTI. This information was integrated with results of urinary white cell count and microbial culture of urine and/or blood to categorise patients as probable, possible or no evidence of UTI (Table 1).

Patients admitted to hospital for suspected UTI
For patients admitted to hospital with suspected UTI, we compared two approaches to estimate scope for improving antibiotic prescribing for UTI. First we estimated the proportion of patients with microbiological/clinical evidence of UTI as previously described (Table 1). Next we compared diagnoses made in the ED to the primary ICD-10 diagnostic code, which can identify patients who were admitted for treatment of a UTI syndrome, using code lists derived from the NHS outcomes framework [10]. The primary ICD-10 code represents the main reason for the patient's admission and is entered retrospectively after the patient has left hospital based on information recorded in the medical record. Consequently the primary ICD-10 code and the ED diagnosis may be discordant when test results and the patient's clinical progression in hospital make it clear that the ED diagnosis was incorrect.
Investigations, test results and antibiotic prescriptions that were ordered in the ED were not recorded electronically, consequently information on urinary symptoms, results of urinalysis, antibiotic prescriptions, presence of an indwelling urinary catheter, Standardised Early Warning Score (SEWS) [11], and temperature were transcribed from the medical notes and entered into a database which was stored securely at QEHB. Demographic data (age, gender, index of multiple deprivation-IMD), were obtained from the electronic health record in addition to microbiological culture and sensitivity results from urine and blood. Antibiotics were categorised as broad (carbapenems, cephalosporins, co-amoxiclav, colistin, piperacillin/tazobactam) or narrow spectrum (all other antibiotics) [12]. We further obtained all ICD-10 and OPCS codes for the 5-year period preceding admission in order to calculate Charlson Comorbidity Indices. Patients with no relevant code recorded in the 5 years prior to admission were classified as not having any comorbidities.
Microbiological culture of urine and/or blood was undertaken in accordance with standard clinical laboratory procedures current at the time (UK Standards for Microbiology Investigations: SMI B41, Investigation of Urine; SMI B37: investigation of blood cultures (for organisms other than Mycobacterium species) [13]. Urine cell counts were done on the UF100 Urine Analysis (Sysmex, Milton Keynes, UK) until September 2015 and subsequently on its newer version UF1000i. Cell counts were not possible for urine samples less than 4 mL or for samples too viscous to pass through the instrument. Samples for which cell counts could not be done were always cultured. When cell counts were available, only urines with white blood cell counts above a threshold value were cultured. At the start of the study the threshold value was 40/μL or bacteria counts > 4000/μL. This was adjusted to 80/μL or bacteria counts > 8000/μL following the introduction of a revised standard operating procedure in the microbiology laboratory in October 2015. Semiquantitative culture was done on Chromid® CPS® Elite agar (bioMérieux, Basingstoke, UK). Microbial samples of urine or blood were classified as multidrugresistant based on criteria agreed by international experts in a joint initiative of the European Centre for Disease Prevention and Control, and the Centers for Disease Control and Prevention [14].

Analytical approach
The primary outcome was the proportion of patients with an ED diagnosis of suspected UTI who met clinical/microbiological criteria for possible or probable UTI (Table 1), comparing patients who were and were not admitted to hospital. Secondary outcomes in patients admitted to hospital with an ED diagnosis of suspected UTI included: the proportion without a primary ICD-10 code for a UTI syndrome; the proportion with a primary ICD-10 code for a non-infectious condition; and the proportion stopping antibiotics within 72 h following admission (defined as the first calendar day with no record of antibiotic treatment). We undertook sensitivity analysis by extending comparison of the ED diagnosis and ICD-10 code to include the highest ranked secondary diagnostic code, since conditions such as urosepsis may be recorded using two codes.
For patients who had been admitted to hospital, we generated alluvial plots to illustrate the relationship between diagnoses that were made in the ED, and the patients' main reason for admission to hospital based on ICD-10 coding, comparing patients with and without microbiological evidence of UTI.

Results
Nine hundred forty three patients met the study inclusion criteria and 289 patients were diagnosed with a UTI syndrome in the ED (Fig. 1). Patients who were admitted to hospital were older compared to those discharged from the ED (mean age 72 versus 44 years), more likely to be male (39.8% versus 29.8%) and to have at least one comorbidity (44.9% versus 24.3%), (Table 2). Patients who were discharged from the ED were also more likely to have a positive urinalysis compared to patients who were admitted to hospital (72.1% versus 66.4%). Overall there were 191 patients with an ED diagnosis of lower UTI, 56 cases with pyelonephritis and 42 patients with urosepsis ( Table 3).
Comparison of the ED diagnosis to clinical/ microbiological criteria There were 191 patients with an ED diagnosis of lower UTI, 56 cases with pyelonephritis and 42 patients with urosepsis (Table 3). Approximately one half (83/176) of the patients who were admitted to hospital, and nearly

Comparison of ED diagnoses and ICD-10 codes
For patients who were admitted to hospital we compared the ED diagnosis to the main reason for the patient's admission based on ICD-10 codes. In this group, lower UTI remained the main reason for admission in 33/83 cases (39.8%) who were diagnosed with this condition in the ED (Additional file 1, Fig. 2) and 3 patients in this group were diagnosed with pyelonephritis or urosepsis. However, the main reason for admission was non-infectious in 40.9% (95% CI: 30.3-52.3%) of patients with an ED diagnosis of lower UTI.
Pyelonephritis remained the main reason for admission in about a quarter (13/51) patients who were diagnosed with this condition in the ED. The diagnosis was downgraded to lower UTI in 20 (39.2, 95% CI: 25.8-53.9%) patients with this ED diagnosis and one patient was diagnosed with urosepsis. The main reason for admission was non-infectious in 14 patients (27.5, 95% CI: 15.9-41.7%).
Forty two patients were diagnosed with urosepsis in the ED, but this remained the main reason for admission in just 2 patients, with 3 cases of pyelonephritis. The diagnosis was downgraded to lower UTI in almost half of patients (20/42) who had been diagnosed with urosepsis in the ED and the main reason for admission was non-infectious in 11 cases (26.2, 95% CI:13.9-42.0%).  Sensitivity analysis comparing the ED diagnosis against both the primary and secondary ICD-10 diagnostic code slightly reduced the proportion of patients who had a non-infectious reason for admission to 29/83 (34.9, 95% CI: 24.8-46.2%) of patients with an ED diagnosis of lower UTI, 12/51 (23.5, 95% CI: 12.8-37.5%) with an ED diagnosis of pyelonephritis, and 7/42 (16.7, 95% CI: 7.0-31.3%) with an ED diagnosis of urosepsis. The relationship between the diagnosis of UTI in the ED and the main reason for admission based on ICD-10 coding is illustrated in Fig. 2a and b.

Discussion
Clinical/microbiological evidence of infection was lacking in > 60% of patients who were treated empirically for UTI syndromes in the ED, but antibiotics were stopped within 72 h in less than one-quarter of patients. Our findings suggest there is scope to further reduce a Multiple antibiotics may be given to a single patient and multiple specimen might be identified from a single patient's specimen b Threshold for undertaking urine culture was adjusted from 40 urinary white cells/μL to 80 urinary white cells/μL in October 2015, following the introduction of a new laboratory standardised operating procedure. c Heavy mixed growth: samples in which there were multiple organisms of >1 species, where no single species predominates and colonies are so numerous as to be unquantifiable by the culture method used. These samples are most likely to represent contamination antibiotic prescribing for this common condition, and highlights the potential impact of reviewing and revising antibiotic prescribing decisions in light of microbial culture and clinical progression. There are many reasons why it is challenging to estimate how often UTI is over-diagnosed in the ED. Firstly there is a lack of consensus on diagnostic criteria for UTI which is compounded by the fact that around 20% of patients with genuine lower UTI will have negative urine cultures [15,16]. Prior antibiotic treatment also reduces the likelihood of obtaining a microbial culture from urine or blood, and this may partly explain why we failed to identify a urinary pathogen in over half of patients with suspected UTI in this study. Using the medical notes we estimated that 9.4% (89/943) patients had received antibiotics prior to their ED attendance, but this is likely to be an underestimate as this information is poorly recorded. Previous studies in patients aged > 65 years have highlighted the discrepancy between an ED diagnosis of bacterial infection (including UTI) against standardized criteria for infection [7], and this is supported by a number of studies which have estimated the prevalence of "inappropriate" antibiotic prescribing in the ED to range from 30 to 40% [17,18]. In common with these studies, we found that the reason for admission was not an infection in around 40% of patients with an ED diagnosis of lower UTI, suggesting that these patients are unlikely to have benefitted from antibiotic The diagnosis of pyelonephritis is "correct" in 11 (36.7%) cases, but it is revised to lower UTI or a non-infectious condition in 11 (36.7%) and 5 (16.7%) cases respectively. In Fig. 2b, 61 cases are diagnosed with lower UTI in the ED (but none of these individuals have clinical/microbiological evidence of UTI). The diagnosis of lower UTI is confirmed 23 (37.7%) cases, but 12 patients (19.7%) are diagnosed with a different infection and 24 (39.3%) are diagnosed with a non-infectious condition treatment. We also note that a proportion of cases diagnosed as "non-infectious" in the ED were subsequently allocated a primary ICD-10 code for UTI, suggesting that under-diagnosis of this condition also occurs. Our findings highlight the uncertainty around the diagnosis and diagnostic coding of UTI syndromes in secondary care, a problem which severely undermines the ability to implement and evaluate antimicrobial stewardship initiatives at scale.
We demonstrate a clear discrepancy between diagnoses that are made in the ED and clinical and microbiological information that accumulates during admission to hospital, which emphasises the importance of antibiotic prescribing review as a core element of hospital antibiotic stewardship programmes [19,20]. Our findings also demonstrate the limitations of rapid diagnostic tests such as urinalysis that are widely used in the ED, since more than 60% of patients who were not diagnosed with a UTI syndrome had evidence of UTI on urinalysis i.e. detection of leukocytes and/or nitrites. Overall, antibiotics were stopped in less than one quarter of patients who were diagnosed with a UTI syndromes in the ED, despite the fact that an estimated 40% of these patients may not have benefitted from antibiotic therapy. The ED diagnosis of pyelonephritis and/or urosepsis was downgraded to UTI in 40-50% of patients once they had been admitted to hospital, highlighting the opportunity to deescalate antibiotics. We were unable to estimate how often de-escalation took place in this study, due to the small number of patients with each type of UTI syndrome. However, recent national audits in patients with sepsis in England suggests that > 80% of antibiotic prescriptions are continued following antibiotic prescribing review at 48-72 h [12].

Strengths and limitations
Diagnostic criteria for UTI syndromes are notoriously difficult with laboratories, and therefore different studies each employing different protocols. As in most UK laboratories, our laboratory used the UK Standards in Microbiology Investigations (SMI) for urine current at the time [13]. In its guidance for interpreting urine cultures, this SMI reflects the variation in interpretation that might be placed on urines that grow microorganisms. In a range of growth from 10 2 to 10 5 cfu/mL, the nature of the specimen, the urine white cell count and the patient's symptoms all need to be considered in assigning the likelihood of a urinary tract infection. Even the SMI states that interpretation may need to be supplemented with local reporting policies. It is possible that the laboratory algorithm that is used to determine whether samples undergo microbial culture may have underestimated the prevalence of bacteriuria in patients included in this study. Our approach may have erroneously classified patients with asymptomatic bacteriuria as having genuine UTI or have failed to identify patients with urinary symptoms where this information was not clearly recorded. In addition it is possible that some patients with an ED diagnosis of UTI did have this condition, but the diagnosis was superseded by a different complaint which was recorded as the primary ICD-10 code. We attempted to investigate this issue by undertaking a sensitivity analysis looking at secondary ICD-10 codes. We restricted our analysis to patients who had a urine sample submitted from the ED and acknowledge that only a proportion of patients with suspected UTI have a urine sample submitted for microbial culture. Preliminary, unpublished estimates from a related, ongoing research study at this site suggest a urine sample is submitted for culture from 44% of patients who are diagnosed with suspected UTI in the ED. It seems likely that there are reasons why doctors submit urine samples for culture in some patients but not others, potentially relating to diagnostic uncertainty, so it is unclear whether our findings can be generalised to all patients with suspected UTI. A further limitation of our analysis is that a number of conditions that are included in the Charlson Comorbidity Index are unlikely to be recorded in secondary care.
A strength of our approach is that fact that ED physicians use structured codes to record the diagnosis at discharge from the ED, increasing the likelihood that this information accurately represents the doctors' decision making. We capitalised on the availability of high quality electronic health records at this site and also undertook a detailed review of the medical notes to extract information on urinary symptoms and antibiotics that were prescribed in the ED. However, a limitation of our approach is that this study was conducted at a site which was selected because of its electronic health records. Whilst there is a widely held view that clinical infection syndromes, and UTI's in particular, are likely to be overdiagnosed in the ED, more sites would be required in order to generalise these findings beyond this hospital.

Clinical implications and future research
Efforts to monitor and improve prescribing for UTI syndromes are undermined by the complexity of the diagnosis, and the difficulty in differentiating UTI from other conditions with a similar presentation. For these reasons, we emphasise the importance of reviewing antibiotic prescribing decisions in patients with suspected UTI, and the potential impact on total prescribing that could be achieved if this approach was adopted more widely. However, the diagnostic challenge of UTI could be addressed most effectively through the development of rapid diagnostics with sufficient negative predictive power to reliably rule out bacterial infection.