Article Text
Abstract
Aims: To test the reproducibility of a highly sensitive clinical decision rule proposed to predict vesicoureteral reflux (VUR) after a first febrile urinary tract infection in children. This rule combines clinical (family history of uropathology, male gender, young age), biological (raised C reactive protein), and radiological (urinary tract dilation on renal ultrasound) predictors in a score, and provides 100% sensitivity.
Methods: A retrospective hospital based cohort study included all children, 1 month to 4 years old, with a first febrile urinary tract infection. The sensitivities and specificities of the rule at the two previously proposed score thresholds (⩽0 and ⩽5) to predict respectively, all-grade or grade ⩾3 VUR, were calculated.
Results: A total of 149 children were included. VUR prevalence was 25%. The rule yielded 100% sensitivity and 3% specificity for all-grade VUR, and 93% sensitivity and 13% specificity for grade ⩾3 VUR. Some methodological weaknesses explain this lack of reproducibility.
Conclusions: The reproducibility of the previously proposed decision rule was poor and its potential contribution to clinical management of children with febrile urinary tract infection seems to be modest.
- CRP, C reactive protein
- UTI, urinary tract infection
- VCUG, voiding cystourethrogram
- VUR, vesicoureteral reflux
- epidemiology
- urinary tract infection
- validation studies
- vesicoureteral reflux
Statistics from Altmetric.com
- CRP, C reactive protein
- UTI, urinary tract infection
- VCUG, voiding cystourethrogram
- VUR, vesicoureteral reflux
Urinary tract infection (UTI) is one of the most frequent bacterial infections in children.1 It has been estimated that 7% of girls and 2% of boys will have a UTI before 6 years of age.2 Vesicoureteral reflux (VUR) is diagnosed at the first febrile UTI in 20–40% of the patients.3 VUR is a risk factor for recurrent UTI, renal scarring, hypertension, and chronic renal failure,3 and the risk is correlated to VUR grade.4 Thus, American,5 Swedish,6 Swiss,7 and French8 paediatric societies recommend a systematic voiding cystourethrogram (VCUG) after the first febrile UTI in children. This strategy assures 100% sensitivity for VUR diagnosis. However, VCUG is a posteriori normal in 60–80% of patients. Moreover, VCUG is associated with a risk of iatrogenic UTI,9 and it is irradiating,10 painful,11 and expensive.12 Therefore, being able to predict the absence of VUR, in order to avoid unnecessary VCUG, would be helpful.10
Two predictive tools have been proposed. Renal ultrasound alone, regardless of the criteria chosen, was shown to be poorly sensitive for VUR prediction.13–16 Oostenbrink et al proposed a multivariate approach based on a risk score combining clinical (family history of uropathology, male gender, young age), biological (raised C reactive protein (CRP)), and radiological (urinary tract dilation on renal ultrasound) variables.17 A clinical decision rule based on the score proposes that VCUG is not performed in patients with low scores. This rule yields 100% sensitivity for the prediction of all-grade VUR with 17% specificity, and 100% sensitivity for the prediction of grade ⩾3 VUR with 38% specificity.
Clinical decision rules attempt to formally test, simplify, and increase the accuracy of clinicians’ diagnostic assessments.18 Three steps are involved in the development of a clinical decision rule: creation of the rule, validation, and assessing its impact on clinical behaviour.18,19 Jodal stressed the potential clinical impact of the prediction rule proposed by Oostenbrink and colleagues,17 but also the need of its validation,10 as for all clinical decision rules.20 The aim of our study was to evaluate the reproducibility of the prediction rule proposed by Oostenbrink et al.
METHODS
Study design
We conducted a retrospective, hospital based cohort study in the Department of Paediatrics of Saint-Vincent-de-Paul Hospital, Paris, France, from January 2000 to September 2003. Data were extracted from medical files using a standardised data form.
Patients
All patients with UTI as a discharge code or a text word in their computerised medical records were evaluated for inclusion. All children 1 month to 4 years old admitted with a first, community acquired, febrile UTI were included. Febrile UTI was defined as rectal fever ⩾38°C associated with a positive urine monoculture (⩾105 colony forming units/ml in urine, collected in sterile bags which were changed every 30 minutes, or by midstream clean-void sample for older toilet trained children), and biological inflammatory syndrome (leucocyte count ⩾15 000/mm3 and/or CRP ⩾15 mg/l). Patients with a known, congenital, urinary tract abnormality at the time of the UTI diagnosis were not included.
Outcome measures
All patients underwent VCUG during the study period. The radiologists were blinded to the predictors (see below) and graded VUR from 0 to 5, according to the International System of Radiological Grading of Vesicoureteral Reflux.21
Predictors
We used the same predictors as Oostenbrink and colleagues:17 family history of uropathology, age, gender, serum CRP at the time of diagnosis, and urinary tract dilation on renal ultrasound performed by a senior paediatric radiologist.
Statistical analysis
Statistical analysis was performed using EPI INFO 6.0 software (Centers for Disease Control and Prevention, Atlanta, GA). The first step was a descriptive analysis of the characteristics of the population.
The second step was to calculate the VUR risk score for each patient using the grades proposed by Oostenbrink and colleagues.17 Points were assigned as follows: 0 for age <1 year, 1 for age 1–2 years, etc, maximum 4 for children 4 years old; 1 for boys and 0 for girls; first degree family history of uropathology was graded 1 for yes and 0 for no; 1 for CRP 0–9 mg/l, 2 for CRP 10–19 mg/l, etc, maximum 20 points for CRP >200 mg/l; urinary tract dilation on renal ultrasound was graded 0 when absent, 1 for a renal pelvic diameter ⩽10 mm, 2 for a diameter >10 and <25 mm, and 3 for a diameter ⩾25 mm.22 The risk score formula was: individual score = 6*male gender + 7*family history of uropathology − 1*age + 1*CRP + 14*renal ultrasound dilation. Theoretically, the score could range from −4 to 75.
The third step was to analyse the relation between a high score and VUR in the validation population using odds ratios (OR) and Fisher’s exact test.
The last step was to evaluate the degree of discrimination obtained with the decision rule by calculating its sensitivities and specificities at the two score thresholds (⩽0 and ⩽5) proposed by Oostenbrink et al who had chosen these thresholds to predict, respectively, the absence of all-grade and grade ⩾3 VUR.17
RESULTS
Patients
One hundred and fifty nine patients fulfilled the inclusion criteria. Ten (6%) of the 159 included patients were lost to follow up before VCUG could be performed. Hence the analysis was based on 149 patients. A family history of uropathology was found for 37 (26%) patients. Sixty eight (46%) were male. The median age was 9.8 months (SD 7.9). Urinary tract dilation on renal ultrasound was observed in 30 (20%) patients. VUR was diagnosed in 37 (25%) children, including 14 (9%) with grade ⩾3 VUR.
Validity of the decision rule
Three (2%) patients had a score ⩽0, the first proposed threshold, and none had VUR (table 1). We did not find a significant association between a score >0 and all-grade VUR (p = 0.6). A rule based on the ⩽0 threshold would have yielded 100% sensitivity and 3% specificity for the prediction of all-grade VUR (table 2).
Among 19 (13%) patients with score ⩽5, the second proposed threshold, four had VUR. We did not find a significant association between a score >5 and all-grade VUR (OR = 1.3, 95% CI 0.4 to 5.7, p = 0.8). A rule based on the ⩽5 threshold would have yielded 89% sensitivity and 13% specificity for the prediction of all-grade VUR.
One of the 19 patients with a score ⩽5 had grade ⩾3 VUR. We did not find a significant association between a score >5 and grade ⩾3 VUR (OR = 2.0, 95% CI 0.3 to 91.1, p = 1.0). A rule based on ⩽5 threshold would have yielded 93% sensitivity and 13% specificity for the prediction of grade ⩾3 VUR.
DISCUSSION
Main results
We report the first attempt to evaluate the reproducibility of the rule proposed by Oostenbrink and colleagues.17 In their study, they reported a significant association between a score >0 and all-grade VUR (p = 0.006), and between a score >5 and grade ⩾3 VUR (p = 0.0001). These significant associations were not found in our patients (p > 0.5). The high (100%) sensitivity of the rule for the prediction of all-grade VUR was confirmed, but the reproducibility of the specificity was poor: only 3% in our patients versus 17% for the construction population. The high sensitivity for the prediction of grade ⩾3 VUR was not reproduced either: 93% versus 100%. Applying the decision rule to our patients, we would not have performed VCUG in one patient with grade 5 VUR. This patient, a 25 month old girl, with no family history of uropathology, fever of 6 days’ duration, a CRP of 59 mg/l, and normal renal ultrasound, had a 4 point score.
What is already known on this topic
-
A clinical decision rule based on a risk score has been proposed to define a selective approach to avoid unnecessary VCUG
-
This decision rule has never been validated
The potential contribution of the decision rule to clinical management of these children with UTI seems to be modest. Indeed, if the ⩽0 threshold of the score is retained, few (3%) useless VCUG would be avoided. Choosing the ⩽5 threshold, cases of VUR, including grade ⩾3 VUR, would have been missed.
Study weaknesses
In our retrospective study, 6% of patients were lost to follow up before VCUG could be performed. This rate is close to that observed in other studies on the subject.4,16
The use of sterile bags for urine collection introduced a selection bias, as this technique is less specific than suprapubic aspiration or catheterisation and is not recommended by the American Academy of Pediatrics.5 Indeed, some patients who would not have had a diagnosis of UTI using suprapubic aspiration or catheterisation were included in our study. This bias explains the higher prevalence of male children in our study (46%) compared with other studies using suprapubic aspiration or catheterisation for the diagnosis of UTI (11%, 28%).4,23 However, sterile bags are used in the day-to-day practice of North American paediatricians24 and they were used by Oostenbrink et al in their study.17
Other issues
The external validation of a decision rule can fail either because the reproducibility of the rule is low or because the construction and validation populations differ too much.20 In the present case, the construction and validation populations were not significantly different (p > 0.05) for the main parameters: VUR prevalence (26% v 25%), positive family history of uropathology (27% v 26%), sex ratio (36% v 46% males), and prevalence of urinary tract dilation on ultrasound (29% v 20%).
Furthermore, methodological weaknesses of the construction study by Oostenbrink and colleagues17 might have limited the reproducibility of the rule.20 First, some variables used in the score were not clearly defined: the relative degree for a family history of uropathology was unknown and urinary tract dilation was not quantitatively graded. Second, some variables that were not significantly associated with VUR in univariate and multivariate analyses were retained in the model: age, gender, family history of uropathology.20 Third, continuous variables (age, CRP) were entered into the model based on an undemonstrated and unlikely hypothesis of a linear gradient.25
What this study adds
-
The reproducibility of the previously proposed decision rule was poor either because of a lack of sensitivity or because of a lack of specificity depending on the threshold chosen
-
The potential contribution of this rule to clinical management of children with febrile urinary tract infection seems to be modest
Implications
Systematic VCUG after a first febrile UTI in children assures a 100% sensitivity for the diagnosis of VUR. That is why selective approaches to avoid unnecessary examination require high sensitivity.10 The strategy proposed by Oostenbrink et al took into account this constraint. Unfortunately, its reproducibility seems poor according to our results. Perhaps new predictors, such as procalcitonin,26 and new strategies, such as recursive partitioning,27 rather than scores, might prove helpful to define selective approaches for imaging work up after the first UTI.
REFERENCES
Footnotes
-
Published Online First 12 May 2005
-
Funding: S Leroy was financially supported by a grant from the Association des Juniors en Pédiatrie and Laboratoire Gallia
-
Competing interests: none declared
Linked Articles
- Perspectives