Added value of non-contrast CT for the diagnosis of acute pyelonephritis in older patients with suspected infection with an unknown focus: a retrospective diagnostic study

Introduction

Acute pyelonephritis (APN) is one of the most common infectious diseases in adults.1 To select appropriate antibiotic agents and treatment duration, it is important to make a correct diagnosis of APN.2 However, in older patients, the diagnosis of APN is challenging due to the high prevalence of asymptomatic bacteriuria, as well as chronic urinary symptoms.3 Furthermore, cognitive impairment often makes it difficult to evaluate history and physical examination in this population.

While some experts have suggested several potentially useful radiological findings of CT for the diagnosis of APN,4–7 very few studies have evaluated their diagnostic performance. These signs can be useful to assess the possibility of APN, especially when the focus of infection is unknown, which is common in older patients. Accordingly, we have recently assessed the diagnostic performance of five CT signs (perirenal fat stranding, pelvicalyceal wall thickening, enlargement of the kidney, thickening of Gerota’s fascia and decreased attenuation) for APN in older patients with a suspected infection with an unknown focus.8 In our previous study involving 473 patients, non-contrast CT findings were evaluated in all patients, while contrast-enhanced CT findings were assessed in 101 patients. Contrast medium was withheld mainly due to concerns about contrast-induced acute kidney injury (AKI). Even without contrast, these CT signs were especially useful for ruling in APN (positive likelihood ratio 2.5–11.3).8 However, whether those CT findings have any added value to readily available information before CT (ie, history, physical examination, urinalysis, etc) remains unknown. Since the diagnosis of APN is always made based on all available information in clinical practice, whether CT provides added value is more important than its diagnostic performance as a sole test.

Accordingly, the aim of this study was to evaluate the added value of CT findings to readily available information for the diagnosis of APN in older patients with a suspected infection with an unknown focus.

Methods

The present study was performed using the data from the aforementioned study in which we reported the diagnostic performance of CT findings for the diagnosis of APN.8 The study was designed as a retrospective, diagnostic study. Standards for Reporting of Diagnostic Accuracy 20159 was followed.

Patients

The study was conducted in the Department of General Medicine of Shirakawa Kosei General Hospital, an acute care hospital in Japan (471-bed capacity), between April 2015 and March 2017. Consecutive patients who satisfied the following inclusion criteria were included: (1) age 65 years or older; (2) admitted with suspected infection; and (3) chest and abdominal CT scan performed on admission to detect the focus of infection (ie, the focus of infection was unknown prior to CT scan according to comments on CT ordering). As in previous studies,10 11 the physician’s decision to obtain a urine culture and two or more sets of blood cultures on admission was used as an indicator of the clinical suspicion of infectious disease including urinary tract infection. Patients with only one kidney were excluded, because the assessment of CT findings needed to compare the findings between the left and right kidneys (see below in the Index test section).

Index test

All patients underwent 64-row multislice CT scan (Aquillion 64, Toshiba Medical Systems, Japan). Of the five CT findings that were assessed in the previous study,8 four non-contrast CT findings were selected for the present analyses: (1) perirenal fat stranding, (2) pelvicalyceal wall thickening, (3) enlargement of the kidney and (4) thickening of Gerota’s fascia. Decreased attenuation was not investigated because this sign can be assessed only on contrast-enhanced CT, and there was an insufficient number of patients who underwent contrast-enhanced CT due to concerns about contrast-induced AKI. Typical images of these four CT findings are shown in figure 1. Two certified radiologists (radiology panel), HH and MM, independently reviewed admission CT without any clinical information. As the CT signs are often deemed non-specific, the effect of laterality of the signs was evaluated. To this end, two types of analyses were conducted with different definitions of positive CT findings: laterality-sensitive and laterality-insensitive analyses. In the former analysis, each CT finding was judged as positive only when the sign was observed differently between the left and right kidneys. In the latter, the presence of CT findings was judged to be positive regardless of whether there was any difference between the left and right kidneys. Any conflicts between the two radiologists were resolved by discussion between them.

Figure 1
Figure 1

Findings of CT in acute pyelonephritis. (A) An 86-year-old man with acute pyelonephritis. Axial non-contrast CT scan of the abdomen showing perirenal fat stranding (white arrowheads). (B) An 86-year-old woman with acute pyelonephritis. Axial non-contrast CT scan of the abdomen showing pelvicalyceal wall thickening (white arrowheads). (C) An 84-year-old woman with acute pyelonephritis. Axial non-contrast CT scan of the abdomen showing renal enlargement (white arrowheads). (D) A 66-year-old woman with acute pyelonephritis. Axial non-contrast CT scan of the abdomen showing thickening of Gerota’s fascia (white arrowheads).

Information available before CT

Based on existing literature for the diagnosis of APN in older patients,3 12–14 the following information was a priori selected to which the added value of CT findings was evaluated: (1) age, (2) sex, (3) living in nursing home, (4) genitourinary symptoms (dysuria, urinary frequency and urgency), (5) medication for urinary incontinence; (6) urinary catheter, (7) body temperature, (8) costovertebral tenderness, (9) pyuria (white blood cells in urine ≥10 per high-power field, ×400) and (10) bacteriuria.

Reference standard

Though various diagnostic criteria for APN in younger adult patients can be found in the existing literature,15 16 it is not possible to simply apply them to older patients since signs, symptoms, pyuria and bacteriuria are not reliable in this population.17 When there are no definite diagnostic criteria, expert panel diagnosis is recommended as a reference standard.18 In the present study, the clinical expert panel consisted of two members, one a board-certified internist (TT) and the other a certified emergency physician (TY). To summarise the information to make the final diagnosis of each patient, a research assistant extracted the following information on admission from the electronic medical record system in the hospital using a data extraction form: age, sex, history, vital signs, findings on physical examination, results of laboratory studies, microbiological assessments (eg, blood, urine and sputum cultures), reports of imaging studies and clinical course during hospitalisation. To avoid information bias, the research assistant excluded any comments on the urinary tract from the reports of any imaging studies made in actual practice, while comments on the other organs (eg, lung, hepatobiliary tract, etc) were provided to the expert panel. Each physician of the clinical expert panel independently made the final diagnosis of each patient based on the information extracted on the form. Any conflicts in the final diagnoses by each panel member were resolved by discussion between them.

Statistical analysis

The number of predictors that can be included in the diagnostic model for APN

In accordance with the recent methodological recommendation,19 we calculated the maximum number of predictors that could be included in the diagnostic model for APN. Based on the prevalence of APN (14.8%) and a target mean absolute prediction error of 0.05, a total of 14 predictors could be included in the model.

Handling of missing data

There were some missing values in the dataset. To avoid biased results by excluding those with any missing values,20 they were imputed using chained equations with all available information. Ten imputed datasets were created and separately analysed. The results for each dataset were pooled using Rubin’s rules.21

Reliability of the clinical and radiology panels

The reliability of the assessment of CT findings by the radiology panel and the diagnosis of APN by the clinical expert panel was assessed using kappa coefficients. The kappa coefficient was interpreted as follows: ≤0=no agreement; 0.01–0.20=none to slight; 0.21–0.40=fair; 0.41–0.60=moderate; 0.61–0.80=substantial; and 0.81–1.0=almost perfect agreement.22

Evaluation of added value of CT findings

To evaluate the added value of CT findings, two prediction models for the diagnosis of APN were developed: one with the 10 predictors available before CT (base model), and the other in which CT findings were additionally included (extended model). Logistic regression modelling was used for deriving the models. For continuous variables such as age and body temperature, non-linearity of the association with the outcome was assessed using a restricted cubic spline function with three knots and incorporated as such if it significantly improved the model fit based on the likelihood-ratio test (ie, p<0.05). The added value of CT findings was quantified by comparing the model performance between the base and extended models. First, the model fit of those two models was compared using the likelihood-ratio test. When the likelihood-ratio test was significant, the discriminative performance was evaluated in terms of the c-index, and calibration was visually assessed by calibration plots.23 Furthermore, to illustrate the clinical utility of the CT findings, the net percentage of patients correctly reclassified by adding the CT findings (extended model) was calculated. To this end, patients were classified into low, intermediate and high risk of APN based on the risk estimated by each model. Since there has been no established guidance on what threshold is appropriate for the diagnosis of APN, the thresholds for those three categories were decided as <15%, 15–40% and ≥40% based on discussion among the coauthors from the clinical context. The CI of the net percentage of correctly reclassified patients was obtained by the percentile method with 2000 bootstrap samples. The same procedures were performed for both laterality-sensitive and laterality-insensitive analyses.

All analyses were performed with R V.3.6.3 (R Foundation for Statistical Computing, Vienna, Austria).

Patient and public involvement

No patients were involved in the development of the research question, the outcome measures, or in the design and implementation of the study.

Results

Patients’ characteristics

Of the 475 eligible patients, 2 were excluded due to a history of nephrectomy. Patients’ characteristics are shown in table 1.24 The median estimated glomerular filtration rate (eGFR) was approximately 50 mL/min/1.73 m2 with 17% of patients having less than 30 mL/min/1.73 m² and 38% less than 45 mL/min/1.73 m². The prevalence of APN was 14.8% (n=61). The most common diagnosis in patients without APN was respiratory infection including pneumonia (295 patients, 71.6%), followed by abdominal infection, bone and joint infection, and neoplasms. Patients with APN were more likely to be women, reported urinary symptoms, have pyuria and bacteriuria, and have positive CT findings in both laterality-sensitive and laterality-insensitive analyses than those without APN.

Table 1

Patients’ characteristics

Reliability of the judgement of CT findings

The kappa coefficients for the judgement of each CT finding by the radiological expert panel are shown in table 2. In both laterality-sensitive and insensitive analyses, all CT signs except thickening of Gerota’s fascia showed fair to substantial reliability.

Table 2

Reliability of the judgement of CT findings between the radiologists

Clinical diagnosis and judgement by the clinical expert panel

The reliability of the diagnosis of APN between the physicians in the clinical expert panel was almost perfect, with a kappa coefficient of 0.82. Of the 61 patients diagnosed with APN, 55 (90.2%) were diagnosed as such in actual clinical practice. On the other hand, of 412 patients not diagnosed with APN by the panel, only 5 (1.2%) were diagnosed with APN in actual practice.

Added value of CT findings in the laterality-sensitive analyses

The formulas of the base and extended models are shown in online supplemental table. Age and body temperature were incorporated linearly into the models since the model fit was not improved by taking non-linearity into account. Adding CT findings did not improve the model fit (likelihood-ratio test p=0.11).

Supplemental material

Added value of CT findings in the laterality-insensitive analyses

The formulas of the base and extended models are shown in online supplemental table. Of the CT findings, only pelvicalyceal wall thickening showed a significant contribution to predict the diagnosis of APN. The extended model showed better model fit than the base model (likelihood-ratio test p<0.001). The c-index was improved from 0.89 (95% CI 0.85 to 0.93) to 0.91 (95% CI 0.87 to 0.95) by adding CT findings (p=0.03). In calibration plots (figure 2), underprediction around the estimated risk of 0.1 was improved in the extended model. In the reclassification table (table 3), in patients with APN, 15 were correctly classified into the higher risk category, whereas 8 patients were misclassified into the lower risk category. Thus, the net proportion of patients with APN correctly classified using CT findings was 11.5% (95% CI −4.2% to 25.5%). Similarly, the corresponding value for those without APN was 1.9% (95% CI −0.4% to 4.5%).

Figure 2
Figure 2

Calibration plots of the base and extended models. Ideally, all decile groups of predicted probabilities fit close to the dashed diagonal line (perfect calibration). The base model shows underprediction around the estimated risk of 0.10, which is improved by adding CT findings in the laterality-insensitive analysis.

Table 3

Reclassification table in the laterality-insensitive analysis

Discussion

The added value of CT findings to readily available information (ie, history, physical examination and urinalysis) for the diagnosis of APN was evaluated in older patients with a suspected infection with an unknown focus. When the laterality of the CT findings was considered, the added value of the CT findings was insignificant. On the other hand, in the laterality-insensitive analysis, the model to predict the diagnosis of APN was improved in terms of model fit and statistical predictive measures (ie, likelihood-ratio test, discrimination and calibration). However, their clinical utility was limited to improve the classification of 11.5% (95% CI −4.2% to 25.5%) of patients with APN.

To the best of our knowledge, this is the first study to evaluate the added value of CT findings to readily available information for the diagnosis of APN. Diagnostic reasoning usually starts with history taking followed by physical examination. Based on the list of differential diagnoses formulated by these initial processes, physicians subsequently order necessary investigations. Thus, rather than assessing the diagnostic performance of CT as a sole test, it is more clinically relevant to assess its added value to information available beforehand.

For full appreciation of the present findings, some limitations should be noted. First, the external validity of the findings should be interpreted with caution since this study was a single-centre, retrospective study including patients with the average age of 85 years. For younger patients, the validity of the findings should be further examined. Multicentre, prospective studies are warranted to enhance the generalisability and robustness of our results. Second, in accordance with existing studies,10 11 the physician’s decision to obtain urine and blood cultures was used as an indicator of clinical suspicion of infectious disease. However, such a subjective criterion might suffer from an issue of reproducibility. In particular, the decision to order CT might be different in countries where CT is not broadly available. Thus, future studies should apply more objective inclusion criteria. Third, the added value of findings on contrast-enhanced CT could not be evaluated due to the insufficient number of patients in whom contrast-enhanced CT was performed (N=101). In the target population of the present study, older patients admitted with suspected infection, the use of contrast medium is not always feasible due to concerns about contrast-induced AKI. As shown in table 1, impaired kidney function, a primary risk for contrast-induced AKI,25 was prevalent. Additionally, it is crucial to recognise that eGFR is often overestimated in older populations.26 It should also be noted that reduced kidney perfusion, another risk factor for contrast-induced AKI,25 is common in those with infectious diseases. However, it would not be immediately reflected as changes in eGFR. While contrast-enhanced CT after volume expansion as a preventive measure for contrast-induced AKI might be considered,25 this process typically takes several hours in older patients due to concerns about congestive heart failure resulting from rapid volume expansion. In contrast, prompt initiation of antibiotic treatment is essential for improving patient prognosis.27 Therefore, physicians often start antibiotics based on presumptive diagnoses based on available information at admission, but without information from contrast-enhanced CT. However, among the 101 patients who underwent contrast-enhanced CT, the four CT signs on CT with and without contrast showed high correlations (tetrachoric correlation 0.6–1.0). Thus, it is not very likely that the added value of those findings is drastically improved with contrast. Yet, the added value of decreased attenuation, which can be assessed only with contrast, remains to be studied. Finally, the clinical expert panel was used as the reference standard for APN. Although an ideal gold standard for APN might be pathological assessment, it is neither feasible nor ethical to perform kidney biopsies in all patients. The judgement by the panel might be inherently subjective and have an issue of reproducibility. However, the judgement showed very high reliability between the panel members and was substantially consistent with the diagnoses in actual clinical practice.

In older patients with infectious diseases, atypical presentations are very common, especially in patients with frailty and cognitive impairment.28 While symptoms specific to the focus of infection (eg, cough, sputum, dyspnoea for pneumonia, urinary symptoms for APN) are commonly lacking, non-specific symptoms like confusion, lethargy, agitation, appetite loss and falls can be the sole sign in this population.28 Thus, it is sometimes difficult to make a correct diagnosis in older patients with suspected infection. In such cases, chest and abdominal CT scan can be a useful tool to identify the source of infection by detecting pneumonia, intra-abdominal abscess and so on.29 For the diagnosis of APN, several CT findings have been considered useful. However, whether the CT findings have any added value to information available before CT has not been studied. When the non-contrast CT findings were judged as positive in cases where there was a difference between the right and left kidneys, CT findings did not show any significant added value. When both unilateral and bilateral findings were judged as positive, only 11.5% of patients with APN were correctly classified into higher risk groups of APN, indicating its limited clinical value for ruling in APN given the cost and radiation exposure associated with CT scan. It should be noted that our findings do not negate the role of CT in the evaluation of older patients with suspected infections, since CT is not performed solely for the diagnosis of APN. Yet, CT scan, particularly without contrast enhancement, is not recommended if the purpose of the investigation is only to rule APN in or out.

In conclusion, despite the diagnostic usefulness of non-contrast CT findings as a sole test, their added value to history, physical examination and urinalysis was limited for the diagnosis of APN in older patients with a suspected infection with an unknown focus.

This post was originally published on https://bmjopen.bmj.com