The Diagnosis of Urinary Tract infection in Young children (DUTY): a diagnostic prospective observational study to derive and validate a clinical algorithm for the diagnosis of urinary tract infection in children presenting to primary care with an acute illness

Other risk factors

Summary of symptoms and signs

In our review we found that none of the risk factors, individually or in combination, was sufficient to rule in a diagnosis of UTI. Some combinations of symptoms, signs and proposed clinical prediction rules did reduce the probability of UTI below 2% and may be considered low enough to rule out UTI (see Appendix 2, Table 84). However, the studies on which these findings are based did not necessarily systematically sample urine, and were based on populations with a different ethnic mix and rate of circumcision from the UK and, therefore, may not be generalisable to UK general practice.

In the absence of accurate predictive symptoms and signs, a broad urine sampling strategy in children is advocated by NICE. 2 Others have also advocated urine ‘screening’ in some groups of children, for example in all febrile infants or broader urine sampling strategies in ill children. 70,73,74

A survey of 200 paediatricians in 1983 concerning the management of febrile infants found that all of the respondents felt that a UTI prevalence of 5% would warrant urine sampling in all; more than 80% felt that a prevalence of more than 3% would warrant urine sampling in all, and about half felt that a prevalence of between 1% and 3% would warrant sampling urine from all febrile children. 74

A recent retrospective study concluded that urine analysis should be added to the NICE ‘traffic light’ system for the detection of serious illness in febrile children < 5 years old. 73 This would require a large increase in urine sampling from acutely ill children.

Bacteriuria threshold

Laboratory diagnosis of UTI is based on colony counts following culture, which reflect the concentration of organisms in urine and hence the likelihood that the bacteria grown arise from a UTI rather than contamination. The standard threshold for diagnosing UTI from culture of ≥ 10⁵ CFU/ml was established by Kass over 50 years ago. 75 This was based on studies of adult women with acute pyelonephritis and asymptomatic women. Some have questioned if this threshold is appropriate for children,76,77 with most suggesting a lower threshold76–79 but one advocating an increase. 80 UTI is typically thought to be caused by a single organism present in a high concentration, usually ≥ 10⁵ CFU/ml. 81 However, laboratory guidelines differ regarding the urine sampling method, nature and extent of bacterial growth required to confirm UTI. 82,83 For a clean-catch specimen in children, > 10³ CFU/ml of a single species ‘may be diagnostic of UTI’; and a pure growth of between 10⁴–10⁵ CFU/ml is ‘indicative of UTI’. 82 Although usually a pure or predominant growth is required for the diagnosis of UTI, the growth of two organisms, each with a growth of ≥ 10⁴ CFU/ml, would also be considered as positive by these guidelines. 82 NICE guidelines do not provide a definitive threshold for diagnosing UTI on culture but provide advice about the level of bacterial growth in relation to symptoms and signs. 2 Although NHS laboratories in the UK follow the UK Standards for Microbiological Investigation for the examination of urine, application of the method varies between laboratories.

Some secondary care studies have required two consecutive urine samples with significant bacteriuria to diagnose UTI. 84,85 It has been suggested that obtaining two samples from children in primary care would reduce the risk of false-positive results;80 however, it is unlikely that this would prove successful in primary care given the challenges and current low levels of urine sampling. It is also unclear if two samples would improve validity as there may be a greater number of false-negative results with this technique. Current guidelines advocate one sample. 2

Chapter 4 reports the DUTY study analyses used to determine the reference standard and bacteriuria threshold to be used for the development of the DUTY study diagnostic algorithm.

Culture techniques and uropathogenic organisms

Standard techniques for culturing urine vary but usually rely on either cystine-lactose-electrolyte-deficient (CLED) agar or, more recently, chromogenic agar. 83 No single medium is likely to support the growth of (and detection of) all possibly significant organisms; however, most organisms commonly associated with UTI are supported. Organisms less commonly associated with UTI may not grow at all or may grow at an insufficient rate to be detected, for example Haemophilus influenzae, Streptococcus pneumoniae and the coagulase-negative staphylococci. 86 There is no definitive list of uropathogenic organisms and the distinction between uropathogenic and non-uropathogenic is not always clear. 82 Staphylococaus saprophyticus is commonly associated with UTI in females and growth is supported on the normal media used. Most studies report that most UTI is caused by Escherichia coli, both in adult and paediatric populations. 14 Enterobacteriaceae are usually considered to be uropathogens. 30,34–36,82,87 Non-enterobacteriaceae such as enterococci (Lancefield Streptococcus group B), coagulase-negative staphylococci, Staphylococcus aureus, S. saprophyticus and pseudomonads are also considered to be potentially significant isolates. 88 Colonies are counted following culture (for ‘between 18–24 hours’). However, reporting varies from laboratory to laboratory and often depends on the clinical information given on the clinical request form.

Contamination

It is recognised that difficulty in specimen collection and the interpretation of specimens potentially contaminated prior to culture, from skin, faeces and other sources may contribute to the misdiagnosis of UTI. 89–98 Overdiagnosing UTI can lead to unnecessary investigations and treatment, which entail risks of complications and psychological stress to the child and family. Discriminating between contamination with faecal organisms and potential UTI pathogens in the laboratory is difficult as the most common faecal organism and the most common pathogen causing UTI is E. coli, and high colony counts may be found in children without UTI.

Contamination rates in collection methods vary, as indeed do the definitions of contamination. Contamination rates range from 0% in clean catch to 48% of bag urines. 98 In a retrospective observational cohort study, contamination in clean catch, catheter specimen of urine and bag was 1%, 12% and 26%, respectively. 98 Definitions of contamination vary from single organism growth < 10⁵ CFU/ml OR ≥ 2 organisms to ≥ 2 organisms present at ≥ 10⁵ CFU/ml of urine. 89,91

In a US study, no institutional factors such as access to refrigeration were found to be associated with either low or high contamination rates. 99 Gender and diarrhoeal symptoms also had no association with higher contamination rates in children. 100 Perineal cleansing in female adults had no association with contamination rates, while urine contamination rates were higher in midstream urine collected from toilet-trained children when obtained without perineal/genital cleaning. 89,94 One of the only factors affecting contamination rates which has been published is changing nappy pads every 30 minutes. 91

Although contamination is generally considered to increase the probability of false-positive results, there is some suggestion that contaminated samples or samples with mixed growth may hide a true UTI, that is lead to false-negative results. 101,102

Summary of the research brief

The background given to the NIHR HTA research brief for improving the recognition of UTI in children in primary care (see Appendix 1) was summarised indicating the nature of the problem, that young children with UTI may present with non-specific symptoms such as poor feeding, vomiting, irritability, jaundice (in newborns) or fever alone, suggesting that a broader approach to testing may be appropriate. The research question that the commissioning brief proposed should be answered was: ‘Which clinical features of potential infection are useful in making a preliminary diagnosis of UTI in children < 2 years of age and indicate the need for a urine specimen to be taken?’

Summary of the justification for the DUTY study design

DUTY was a diagnostic cohort study, set up to systematically sample urine from acutely unwell children, with constitutional and/or urinary symptoms, in UK primary care in order to determine the diagnostic value of symptoms and signs for UTI. We proposed to develop and validate an algorithm in two steps: first, to use symptoms and signs to help clinicians to efficiently identify which young children should have their urine tested; and second, to determine the added value of dipstick testing. We included a health economic evaluation to determine the cost-effectiveness of the clinical algorithm compared with existing practice.

We recognised that the age at which children can verbalise their symptoms and achieve bladder control varies between children and that no single age < 5 years would adequately reflect this. We also knew that collecting urine samples from children under the age of 2 years would be challenging and that information from older pre-school children might still be valuable for younger pre-school children. We therefore proposed a study that recruited children up to but not including the age of 5 years.

Research objectives

The DUTY study’s ‘Detailed Project Description’ (the final protocol submitted with the final funding application) contained the following five research objectives:

1. To develop an algorithm that accurately identifies children presenting in primary care with an acute illness in whom a urine sample should be obtained, based on socio-demographic factors, medical history, symptoms and signs (see Chapter 5).

2. To assess whether dipstick urinalysis for nitrite, LE, protein, blood and glucose gives additional diagnostic information to objective (1) in the identification (ID) of urine samples that should be sent to the laboratory (see Chapter 5).

3. To model the cost-effectiveness (cost per correct diagnosis of UTI, cost per symptomatic day avoided and lifetime cost per quality-adjusted life-year (QALY), including potential long-term complications of UTI) from NHS and societal perspectives of one or more diagnostic algorithm guided strategies (see Chapter 6).

4. To compare contamination rates for nappy pad versus clean-catch urine sampling methods (see Chapter 7).

5. To explore the significance of two categories of positive urine culture (laboratory diagnosed UTI vs. asymptomatic bacteriuria/contamination) by (a) comparing the number of NHS contacts at 3 months in children with positive urine culture stratified at presentation by the responsible clinician’s assessment of possible UTI versus plausible alternative diagnosis, and (b) investigating for persisting bacteriuria in a second urine sample.

Planned change to age inclusion criterion

In response to reviewers’ comments (suggesting eligibility to < 4 years) to our detailed project description, and in communication with the HTA, we decided to increase eligibility to < 5 years. This was a pragmatic decision informed by early findings from the EURICA study (a similar study being conducted by the Cardiff members of the DUTY group) that showed higher recruitment, urine return rates and UTI prevalence in the ≥ 4 and < 5 years group and a judgement by the DUTY investigators that children ≥ 4 and < 5 years were often similar, with regard to development of language and bladder training skills, to children aged ≥ 3 and < 4 years.

Planned change to research objective 5

The original funding proposal submitted to the HTA on 12 February 2009 contained five research objectives. The outcomes of the first four objectives are reported in this publication but the fifth was changed prior to publication of the final protocol105 (see Appendix 3). This was because the study eligibility criteria meant all children were unwell (or had urinary symptoms) thus making it difficult to determine asymptomatic bacteriuria as no children were asymptomatic. We therefore removed this as an objective and, instead, have undertaken exploratory analysis on the children who had positive or contaminated urine, investigating the impact of their clinician’s working diagnosis on symptom duration and number of NHS contacts at 43 months. These latter findings are reported in Chapter 7. We also decided that taking a second urine sample in this population would have been too large a logistic undertaking and that persistent bacteriuria was not the focus of the study.

Unanticipated changes to the validation process and reference standard

Our original intention had been to derive and externally validate the diagnostic algorithm, using an approximate 66/33 data split for derivation/validation and to analyse the clean-catch and nappy pad samples together. However, analyses reported in Chapter 4 suggested that the urine collection method was having a greater than anticipated effect on laboratory culture results. Therefore, and in consultation with a number of international experts, we decided to stratify algorithm development by the clean-catch and nappy pad collection method (approximately a 50/50 division of data) and to use bootstrapping to validate the algorithms. This decision was made for two reasons. First, the smaller than anticipated numbers of outcome events in the stratified analyses meant that analyses restricted to a development sample would have been underpowered and estimated coefficients imprecise. Second, our statistical advice was that the properties of a random split into development and validation samples could be mimicked by the bootstrap procedure that we adopted, and statistical overoptimism quantified using calibration slopes.

As it would better reflect day-to-day clinical practice, we originally intended to use ≥ 10⁵ CFU/ml of a known uropathogen from the NHS laboratories as our reference standard. However, as the analyses presented in Chapter 4 show, there was greater than expected disagreement between local and research laboratories, with evidence that the research laboratory was both more reliable and more accurate than NHS laboratories. We therefore decided to use the research laboratory result as the reference standard.

Other changes from the detailed project description

Recruitment period and locations

The study was open to participant recruitment from 7 April 2010 until 30 April 2012. Recruitment was implemented from four research centres at the universities of Bristol, Cardiff, Southampton and King’s College London. Each centre recruited children from primary care, defined as any NHS facility providing first-point-of-contact, face-to-face advice for parents of unwell children.

Selection of primary care sites

The majority of UK acute paediatric care is provided by GPs in primary care practices. To ensure we recruited to target, we estimated that we required between 80 and 100 primary care sites per centre, with each centre needing to contribute around 1500 children to achieve the study sample of 6000.

General practitioner practices were supplemented by NHS walk-in centres (WICs) and children’s emergency departments (CEDs). WICs manage large numbers of young children with acute, undifferentiated illnesses. For example, the South Bristol WIC estimated 2500 contacts for acute, non-traumatic, illnesses in children < 5 years old annually. Parents also use CEDs as a first point of contact for large numbers of acutely unwell children, particularly outside normal working hours. For example, > 5% of all attendances at UK CEDs are due to fever,106 which represents approximately 1400 children annually in each of Bristol, Cardiff and Southampton, and more in London. The majority of these are < 5 years old.

Site recruitment approach

In England and Wales, practice recruitment was supported by research staff from the primary care research network (PCRN) and the National Institute for Social Care and Health Research – Clinical Research Centre (NISCHR-CRC) Network.

General practitioner practices were invited to take part in the study either by letter or via a PCRN newsletter. Interested practices were contacted initially by e-mail to provide further information about the study and were followed up by telephone or face-to-face visit with the practice manager or lead research GP to discuss the study in more detail.

Working with the local research networks in both Wales and England provided the study team with the knowledge of which practices had experience of recruiting to studies within primary care and also highlighted practices to approach. The role of the local research networks is described below.

The role of the local research networks

Local research network support, primarily the PCRN and comprehensive local research networks (CLRNs) in England and, in Wales, NISCHR-CRC, was a decisive factor in enabling this study to recruit ahead of target. Collaboration with the local research networks provided the following benefits.

Recruitment models

Primary care sites were offered two models of recruitment. The first was known as option 1, in which the majority of the recruitment procedures were undertaken by a RN or clinical studies officer (CSO) external to the primary care site, and funded by the DUTY study or the local PCRN, CLRN or NISCHR-CRC, to work with and recruit for the primary care site’s clinical team.

In option 2, recruitment was undertaken entirely ‘in-house’ by the primary care site’s practice team. The acute environment of CED and WIC recruitment meant that, for these settings, the option 1 model was the only viable model.

Site and staff training

The DUTY study team from each centre would meet with individual sites to explain the study and identify their working requirements and match this to the appropriate model of recruitment in order to meet both the study’s needs and those of each individual primary care site. The geographical area of the site was also taken into consideration. Rural practices were encouraged to adopt the option 2 approach, as it was impractical to use DUTY recruiters (RNs/CSOs).

DUTY recruiters were provided with study-specific training including paediatric informed consent, data confidentiality, data collection [using paper-based and electronic case report forms (CRFs)], collection of urine samples and safety reporting. Staff were encouraged to access PCRN or NISCHR-CRC’s informed consent and good clinical practice training days.

Sites were also offered practice-based training adjusted to meet their needs in order to tailor recruitment methods to the individual site. Reception staff were advised on how to introduce DUTY to parents of potential participants. Practice staff were trained on the method of processing the urine sample to send to local NHS and research laboratories and clinicians were provided guidance on informed paediatric consent. All staff involved with the study were provided training on data collection and CRF completion.

Eligibility criteria

The study inclusion criteria were designed to be as broad as possible. Children were eligible if they were aged before their fifth birthday and presented to primary care with a new acute illness episode of ≤ 28 days’ duration.

This illness needed to be associated with (1) at least one ‘constitutional’ symptom or sign identified by NICE2 as a potential marker for UTI – that is, fever, vomiting, lethargy/malaise, irritability, poor feeding and failure to thrive; and/or (2) at least one urinary symptom identified by NICE2 as a potential marker of UTI – that is, abdominal pain, jaundice (children < 3 months only), haematuria, offensive urine, cloudy urine, loin pain, frequency, apparent pain on passing urine and changes to continence.

Therefore, children consulting with other apparently obvious causes for their symptoms such as otitis media or bronchiolitis, as well as those with a history of previous UTI and known abnormalities of the urinary tract, learning difficulties, or reconsulting for an existing illness, were all included, as long as none of the exclusion criteria applied. Where possible, data were included for children recruited and immediately referred from primary care to secondary care. The eligibility criteria are summarised in Table 1. The same inclusion and exclusion criteria also made up the CRF section 1: screening form (see Appendix 4).

Widening participation

The study included parents who spoke non-English languages. Parent information sheets and consent forms were translated into other languages as required by participating GP practices (e.g. Welsh, Polish and Brazilian Portuguese). For languages less commonly spoken in the UK, particularly for those in which oral translation was more useful than written translation (e.g. Somali), translational services were accessed. Where possible, interpreters employed by recruiting primary care sites were used to support patient-clinician communications. When these services were not available, translational services were provided via Language Line (www.languageline.co.uk). In primary care sites in urban areas, the clinical care team often included a translator whose services were used when available.

Registration and informed consent

All recruiting primary care sites displayed posters detailing the study in waiting areas. Parents and children were invited into the study in a number of ways:

1. Where possible, the study was mentioned to the parents of children < 5 years old by reception staff when they telephoned for an appointment or during telephone triage. In this instance, the parent was invited to come to the surgery 15 minutes early to receive further information from the RN/CSO.

2. Where the study was not raised at the time of making the appointment, parents of children already booked in were telephoned and advised about the study and invited to attend a little earlier.

3. If contact prior to attendance at the site was not possible, the parent was approached on arrival, given study information sheets and asked if they were happy to discuss participation with the RN/CSO.

4. Once the parent indicated that they were happy to discuss the study, the RN/CSO explained the study, answered any questions, ensuring that they fully understood the implications of participation, and checked the child’s eligibility.

Where possible, the RN/CSO recruited the participant while they were waiting to see the GP, in order that the participant was not delayed. However, if the participant’s appointment was at risk of being delayed, or it was more convenient for them, the RN/CSO offered to see them after their appointment or visit the parent and child later the same day at their home.

If the parent agreed for their child to participate, written informed consent was obtained from the parent. If the parent was not interested in hearing more about the study, no further approach was made.

In addition to the child being seen by the RN/CSO, they were seen by the child’s responsible clinician who managed the child according to their normal practice. The study CRF was then completed for all consented children by the RN/CSO and with clinical examination, diagnosis and management sections completed by the treating clinician (see Appendix 4).

Non-registration

Sites were asked to complete a screening log of all children whose parents were approached by the recruiting clinician and invited to participate in the study. Details were recorded such as their eligibility, whether consent was given or declined, and reason for declining to participate.

Rationale for urine collection methods

Suprapubic aspiration and ‘in-out’ catheterisation carry the lowest risk of contamination56 but are invasive and unacceptable to parents, and so are uncommon in UK primary care. They were, therefore, not appropriate or feasible for widespread use in the DUTY study. The risk of contamination is low with clean-voided midstream urine or ‘clean-catch’ samples compared with SPA samples. 2 However, obtaining clean-catch samples can be difficult for children in nappies, and parents find nappy pads or urine bags preferable. 63 Many parents prefer pads to bags as the bags have an adhesive strip that attaches to, and can irritate, the child’s skin. These may be suitable alternatives to SPA,2 but there are few data comparing ‘clean-catch’ with nappy pad samples. Thus, the preferred primary care method is the ‘clean catch’, and NICE recommend pad or bag when ‘clean catch’ is not possible. 2 Therefore, we followed the NICE recommendation for collecting urines, with clean catch being preferred to pads, and pads being preferred to bag.

Maximising urine retrieval rates

Obtaining urine samples can be challenging, which is one of the reasons why urine samples are not currently routinely collected in primary care at a rate sufficient to avoid missed diagnoses. As a suboptimal rate of return of samples to the surgery by parents would have diminished power and increased risk of bias, a number of strategies were implemented to maximise retrieval rates. In addition to the role of the RN/CSO in following up parents whose child was not able to provide a sample during the recruitment visit, the data recorded in the study database were used to monitor the retrieval of urine specimens, allowing the research team to identify children for whom urine samples had not been provided, and therefore followed up with the recruiter. Additionally, primary care sites’ urine sample return rates were linked to the level of reimbursement via service support costs.

Urine dipstick methods

The RN/CSO tested the urine sample with a urine dipstick (Siemens Multistix^® 8 SG, Siemens Healthcare Diagnostics, Surrey, UK) provided by the study. The dipstick point-of-care test used tested for presence of blood, protein, glucose, ketones, nitrite and LE, and measured pH and specific gravity. A dipstick was placed in the urine as instructed, and results were interpreted according to the manufacturer’s instructions.

Urine collection information was collected in section 5 of the CRF (see Appendix 4). The time, the urine collection method and the dipstick test results (using the study-supplied Siemens Multistix^® 8 SG) were also recorded in this section.

Dispatching urines to the laboratories

All urine samples, if sufficient quantity of urine was available, were divided into two fractions at the recruiting sites. The priority fraction was sent to the local NHS laboratory for routine diagnostic processing, and the second ‘research’ fraction to a research laboratory – the SACU, Public Health Wales Microbiology Laboratory, Cardiff – for more in-depth analysis. As only small volumes of urine (minimum 1 ml) were required for each laboratory, it was thought possible that most urine samples would be split into the two fractions.

The NHS ‘clinical’ fraction was placed in the urine specimen container stipulated by the site’s local NHS laboratory and labelled with the child’s unique DUTY study ID number on DUTY specific labels as provided in patient recruitment packs. Similar DUTY labels were adhered to specific DUTY microbiology requisition form and the sample sent to the local NHS laboratory using the site’s normal method of transport. Any samples returned to the site and not collected within 4 hours were refrigerated on site and processed within 36 hours. Clinicians received and acted on reports from their local laboratory as per usual clinical care. Research laboratory urine results were not routinely fed back to clinicians; this occurred only if a discrepancy was identified between local and research laboratory results.

The remaining portion of urine was decanted into a sterile container (Monovette^®, Sarstedt AG & Co., Nümbrecht, Germany) containing boric acid. This was labelled with the child’s study ID number and sent by first-class Royal Mail using Post Office-approved Safeboxes^™ to the research laboratory.

The urine sampling method, time of sample, time of dipstick test, results of dipstick and quantity of urine were recorded in section 5 of the CRF (see Appendix 4).

Clinical case report form

The purpose of the CRF was to capture children’s medical histories, examination findings and any known risk factors for UTI. It balanced the need to include as many of the known and potential features associated with UTI with the maximisation of speed and simplicity of completion.

Telephone follow-up at day 14

At 14 days from the recruitment interview, study centre staff contacted parents of all children selected for follow-up according to the proportional selection rules (described in Table 2) to record symptom duration and health-care resource use during the 14-day period after recruitment. In older children (> 9 months), the interview also included a parent-completed questionnaire measuring child health-related quality of life (see Appendix 5). Owing to the young age of the participants, standard methods [e.g. European Quality of Life-5 Dimensions (EQ-5D)] for measuring health utilities were thought to be invalid. Instead, we used the TNO-AZL (Netherlands Organisation for Applied Scientific Research Academic Medical Centre) Preschool children Quality of Life (TAPQOL) questionnaire, completed at 14 days, to describe the health profiles of children with and without UTI. 110 TAPQOL measures parents’ perceptions of health across 12 domains (e.g. sleep, social functioning) and has been shown to be a reliable instrument for both infants and toddlers. 111 No mapping from TAPQOL responses to utility values exists, hence we used a proxy value from a condition (rotavirus) thought to have a similar health-related quality of life impact to UTI. The TAPQOL responses allowed us to validate this choice and make robust comparisons between the health-related quality of life of children with and without UTI.

A booklet version was posted to the parent when telephone contact failed. Details of data collected in the DUTY questionnaire are described below.

1. Symptom duration: we asked if children responded to treatment < 48 hours as NICE has identified failure to respond within 48 hours as a marker of ‘atypical UTI’ and recommends dimercaptosuccinic acid (DMSA) scan and micturating cystourethrogram (MCUG) in such children.

2. Primary care resource use: practice-based contacts in hours, out-of-hours contacts and community-based contacts (e.g. WIC, NHS Direct or health visitor) and associated expenses for the child’s family (fares or mileage, car parking).

3. Hospital resource use [visits to accident and emergency (A&E), attendance at hospital clinics, overnight stays in hospital, use of ambulance services, hospital tests] and associated expenses for the child’s family.

4. Consumption of prescribed and over-the-counter medicines.

5. Other out-of-pocket expenses impinging on the child’s family, for example time off work, loss of earnings, additional childcare costs.

6. Quality of life (TAPQOL or TNO-AZL Preschool children Quality of Life, after Fekkes et al. 110). The parent was asked to rate the child’s quality of life against specific measures of child health including symptoms, sleeping, feeding, behaviour and well-being. The TAPQOL questionnaire is designed for children aged 9 months to 6 years. However, a subset of questions on social, mobility and communications skills are used only in children aged > 18 months.

Three-month notes review process

The 3-month notes review (see Appendix 6) collected the following data: (1) number and type of consultations (not including routine immunisations, screening checks or NHS Direct contacts); (2) results of any further urine samples; (3) secondary care utilisation [including A&E, hospital clinics, outpatient attendances, admissions and investigations/tests (e.g. ultrasound, MCUG or DMSA scans)]; and (4) the dates, types and doses of any prescribed medications.

The majority of these reviews were completed by the study centre RNs and administrators; however, where recruiting sites were geographically remote from the study centre, the assistance of RNs employed by the PCRNs, NISCHR-CRC and practice staff in completing these reviews was sought.

Day-14 and 3-month selection

Selection of DUTY participants to complete the follow-up interview at day 14 and notes was carried out proportionally according to the growth result of the urine sample provided. All children with a pure predominant positive culture (as defined by the NHS or research laboratory at ≥ 10⁵ CFU/ml) and a representative number of children with other growth results were selected as described in Table 2.

An automatic algorithm within the database was created in order for those children to be selected from growth results posted to the database from the day of recruitment until 28 days post recruitment. Results for any child posted after that time would be deemed too late for them to be selected for a follow-up interview.

Serious adverse events

We did not expect to see any related serious adverse events (SAEs) happening as a result of this research, as the study involved only a non-invasive urine test which is often part of the routine clinical care of patients. However, we did anticipate a relatively high number of unrelated and expected SAEs as a consequence of hospital admissions of acutely unwell children, especially (though not exclusively) for children presenting to recruiting CED sites.

Specimen collection

Specimens were collected in sterile containers (some with boric acid, depending on local NHS laboratory procedures) and, where possible, filled to the line indicated on the tube. All samples were sent to the laboratory as soon as possible after collection, but refrigerated if transport to the laboratory was delayed for more than 4 hours. Specimens in boric acid were refrigerated for no longer than 48 hours.

Sample processing

On arrival at the laboratory, all DUTY urines and requisition forms were processed as routine specimens using standard laboratory procedures. The patient ID information provided by the requesting primary care site was entered onto the Laboratory Information Management Systems (LIMS). In addition, laboratories were asked to record unique DUTY patient ID, date of birth, date and time of sample collection, sample receipt, and urine processing in laboratory on the DUTY web-based database.

Samples were not processed if (1) there was too little sample volume (< 0.5 ml); (2) a boric acid container was under- or overfilled; (3) the sample was leaking; (4) the data on the form and sample did not match; or (5) if the container was unlabelled.

Urine microscopy

Microscopy was performed on all specimens by either manual or automated methods. Manual methods vary by laboratory but are based upon the methods set out in Health Protection Agency (HPA) guideline for the ‘Investigation of urine’. 83 Validated automated urine analysers were used in some laboratories, in accordance with manufacturers’ instructions.

Numbers of white blood cells (WBCs), red blood cells (RBCs), squamous epithelial cells and bacteria were recorded according to local ranges (e.g. WBC: < 10; 10–30; > 30–100; or > 100). The ranges varied by laboratory and were based around the method used.

The presence of other organisms (e.g. Trichomonas, Schistosoma or Amoeba), casts [stating type and quantity (scanty, moderate or numerous)], crystals and yeasts was also recorded.

Urine culture

Urine culture was performed using local techniques as described in Table 4. Standard media, usually UTI chromogenic and CLED media, were inoculated, typically using 1 µl to 2 µl, and plates incubated at 35 °C to 37 °C for 16 to 24 hours or according to local SOPs. Further specialised plates for detection of yeasts, for example, were added if required. Plates were read after incubation by a qualified BioMedical Scientist and results recorded.

Bacterial identification

The semiquantitative calibrated loop method of culture is only sensitive for screening down to 10³ CFU/ml if a 5-µl or 10-µl loop is used (e.g. 5 or 10 colonies), or 10⁴ CFU/ml if a 1-µl or 2-µl loop is used (e.g. 10 or 20 colonies).

Bacterial ID of significant isolates was performed according to local SOPs but reported according to Table 5.

Susceptibility testing

Susceptibility testing was performed by a variety of standard methods including disc testing and breakpoint testing. Susceptibility testing was performed on all significant isolates using local SOPs based upon susceptibility testing guidelines [for e.g. British Society for Antimicrobial Chemotherapy (BSAC) standardise disc susceptibility testing method (version 11)]. 112

Disc susceptibility testing can be performed either from isolates recovered from urine or direct from urine. Briefly, a standard inoculum was plated onto susceptibility testing media and antimicrobial discs applied. Plates were incubated for 18 to 20 hours in air at 35 °C to 37 °C. Zone sizes were read by a qualified BioMedical Scientist and interpreted using BSAC or other guidelines such as the European Committee on Antimicrobial Susceptibility Testing (EUCAST) and Clinical & Laboratory Standards Institute (CLSI).

The breakpoint method utilises agar plates containing antimicrobials at concentrations around the breakpoint. Significant isolates were emulsified in sterile saline then spot inoculated onto the plates.

Typical antimicrobials tested were firstline antimicrobials: amoxycillin/ampicillin, nitrofurantoin, trimethoprim, ciprofloxacin, co-amoxiclav, cefpodoxime and cephalexin. Cefpodoxime-resistant coliforms were tested for production of extended spectrum beta-lactamase (ESBL) enzymes in some local laboratories.

Growth on the plates by either method indicates susceptibility/resistance to the antimicrobials tested.

Reporting procedure and storage of organisms

NHS laboratories reported results as per local SOP to the requesting clinician and were required to transcribe the results onto the DUTY web-based database within 1 week in order to allow for patient sampling for follow-up. This was also the trigger to activate laboratory payment (see Appendix 8 for details of laboratory CRF).

Isolates from pure or predominant culture at ≥ 10³ CFU/ml were stored on cryogenic beads at –20 °C or –70 °C.

Quality control

All stock reagents, media, antimicrobials and equipment were monitored for quality assurance at locally specified times. Quality assurance records were completed for lot numbers and expiry dates. All laboratories perform internal quality control and participate in internal quality assurance and external quality assurance schemes.

Data processing

Urines were sent overnight by Royal Mail SafeBoxes^™ by the participating sites. Monovette containers containing boric acid were used to stabilise bacterial counts.

On arrival in the laboratory all DUTY urines and forms were checked for matching identifiers. The following data from each sample were recorded: centre ID, patient ID, date of birth, date and time of sample collection, sample receipt and urine processing in laboratory on the DUTY web-based database. If samples were received out of hours then urines were stored at 4 °C until processed and the date received recorded.

Samples were not processed if (1) there was too little sample volume (< 0.5 ml); (2) the sample was leaking; (3) the sample was received in a non-sterile container; (4) the data on the form and sample did not match; or (5) the container was unlabelled.

Before processing, each Monovette was weighed using a fine-scale balance and the total weight recorded.

Urine microscopy

Urine microscopy was performed on all urine samples with sufficient volume. If 1 to 2 ml only were received then culture only was performed. Microscopy was performed using the iQ200SPRINT analyser (Beckman Coulter Ltd, High Wycombe, UK). Samples that exceed the pre-set thresholds for a given category were manually reviewed by the BioMedical Scientist.

Microscopy results including WBC counts, RBC counts, bacteria counts, squamous epithelial cell counts, presence of yeasts and presence of casts were recorded.

Urine culture

Urines were diluted prior to spiral plating in accordance with standard dilution factors and 50 µl of urine dilution was inoculated onto UTI chromogenic agar and Columbia Blood Agar (CBA) and allowed to dry. Spiral plating was repeated for other dilutions. The stylus was washed between urine samples using 5% hypochlorite (× 1) then sterile water (× 2). Agar plates were incubated at 35 °C to 37 °C for 18 to 24 hours in aerobic conditions. If yeasts were seen on microscopy, these were also spiral plated onto two Sabouraud agar plates and incubated at 35 °C to 37 °C and 30 °C for 5 days.

Colony counting

Total colony counts were performed for all sample CBA plates following the spiral plater manufacturers’ instructions. Colonies were counted in sectors of a grid which overlays the agar plate and the number of colonies in each sector translated into CFUs per ml using the manufacturers’ tables and counts were recorded. Species-specific counts were performed on all isolates present at > 10³ CFU/ml in pure or predominant growth cultures on the UTI chromogenic agar.

Identification of isolates

A preliminary identity of all counted isolates was ascribed using the UTI chromogenic agar. Significant isolates were identified further by basic microbiological tests (e.g. colony morphology, biochemical tests such as indole for E. coli). Potentially significant isolates were considered to be Enterobacteriaceae (E. coli, Klebsiella spp., Citrobacter spp., Enterobacter spp., Serratia spp., Proteus spp., Morganella spp.) (see Appendix 8).

For samples with mixed ≥ 3 organisms with colony counts of 10³ CFU/ml and/or no significant isolates, the ID was performed by UTI chromogenic agar alone.

Susceptibility testing

Antimicrobial susceptibility was determined for all pure or predominant significant organisms according to BSAC disc susceptibility testing guidelines. Isolates were cultured onto CBA plates for single colonies and inspected for purity. A few colonies were suspended in 3 ml sterile water to turbidity 0.5 McFarland and inoculated onto Iso-Sensitest Agar (Oxoid Thermofisher, Basingstoke, UK). The surface was allowed to dry for a few minutes and then the antimicrobial discs applied and plates were aerobically incubated for 18 to 20 hours at 35 °C to 37 °C.

Antimicrobials tested for Gram-positive isolates were trimethoprim, nitrofurantoin, amoxicillin, cefoxitin, vancomycin and novobiocin.

Antimicrobials tested for Gram-negative isolates were trimethoprim, nitrofurantoin, amoxicillin, co-amoxiclav, cephalexin, ciprofloxacin and cefpodoxime.

All antimicrobials except vancomycin were purchased from Sigma-Aldrich Ltd, Poole, Dorset, UK. Disc zone sizes were recorded and interpreted using BSAC guidelines.

Detection of antimicrobial substance in urine

All urines were tested for the presence of any antibacterial substance in the urine. A 0.5 McFarland solution of Bacillus subtilis (NCTC 10400) was inoculated onto an Iso-Sensitest Agar plate. After 5–10 minutes’ drying time, 10 µl of urine was spotted onto the plate, and the plate incubated for 18 to 24 hours’ aerobic incubation at 35 °C to 37 °C. This detects any antimicrobial substances that are in the urine.

Storage of urine and cultured organisms

For samples exhibiting either pure or predominant culture at > 10³ CFU/ml, each isolate was stored on cryogenic beads at –70 °C. For samples exhibiting mixed ≥ 3 organisms at 10³ CFU/ml, a sweep of the growth was saved on cryogenic beads at –70 °C. Any significant isolates were stored on cryogenic beads at –70 °C. Samples with no growth or growth < 10³ CFU/ml were not stored.

Two aliquots of all urine samples were stored in cryogenic vials at –70 °C, one containing 5% sterile glycerol for bacterial preservation.

Data entry to website

All data recorded were transcribed to the DUTY web-based database on a weekly basis (see Appendix 8 for details of laboratory CRF). Lists of those patients considered to have a positive UTI result based on research laboratory data were collated and reported to the study centres.

Quality control

All stock reagents, media, antimicrobials and equipment were monitored for quality assurance at locally specified times. Quality assurance records were completed for lot numbers and expiry dates. See Appendix 8 for research laboratory quality control methods.

Returning positive isolates to the research laboratory after the study

After closure of recruitment for the DUTY study, all isolates retained by the local NHS laboratories were transported to the research laboratory for long-term storage. All isolates were labelled pseudo-anonymously with the study patient ID only. The transfer was made using an Amies charcoal transport swab (Technical Service Consultants Ltd, Heywood, UK) and using Hayes DX couriers (www.dxdelivery.com) (Royal Mail was used as an alternative). Hayes DX ensured that specimen delivery services were fully compliant and International Air Transport Association-approved to meet the current HPA legislation requiring that infectious samples have to be moved in accordance with strict guidelines.

Data entry in primary care sites

The web-based database system was presented as the preferred method of data collection, and DUTY recruiting staff were encouraged and supported to enter CRF data onto the database directly or, if using paper-based CRFs during the recruitment interviews, to retrospectively enter the data in a timely manner (consent and registration within 24 hours, and full eCRF data within 5 working days).

Data entry in the NHS and research laboratories

All laboratory data generated from microscopy and culture for each urine sample received from the NHS local laboratory and research laboratory were entered into the DUTY web-based database. Local NHS and research laboratories staff were only able to access the microbiology data collection pages to log the samples on receipt and enter the results within 1 week of data being available.

Follow-up data entry in research centres

Follow-up data collected at day 14 and 3 months from telephone interviews, postal questionnaires and patient records were entered either directly at time of collection, or retrospectively from paper CRFs, onto the web-based follow-up data collection forms by research staff at each centre.

Data quality

Sites where concerns regarding the possibility of protocol deviations and/or data entry issues not being concordant with electronic data recorded were subject to a data quality audit. The audit questions to be addressed were developed after discussion with both the sponsor and University Hospitals Bristol NHS Foundation Trust. Audits were followed by updated training and agreed reaudits. Following the reaudit of one particular site, by the University Hospital NHS Bristol R&D team on behalf of the sponsor, a study-wide audit was initiated with a random selection of participating primary care sites.

Data cleaning

Where data between paper CRFs and web-based database conflicted, the value on the paper CRF was deemed the true value (making the assumption that the database value was the result of a keying error), unless the paper CRF had already been appropriately annotated with a correction.

All missing data that could not be resolved, for example where no paper record was available, were recorded.

Some ‘self-evident correction’ rules were developed for areas where it was evident that a data entry mis-key had occurred. Self-evident corrections did not involve any changes to actual data or result values; they were used only for process-orientated variables such as dates.

Other cleaning issues were considered, as shown below in order of importance:

1. duplicate patients/NHS numbers

2. missing urine samples and illogical microbiology data

3. time periods – overlong and negative times (e.g. date of birth and date of recruitment)

4. overall missing data – large numbers of missing data for individual recruits

5. outliers – for example in examination findings (temperature, pulse rate, etc.)

6. relationship of consenter to child – to confirm that consent was given by the parent.

All CRF data queries were provided to each centre for verification either against the paper CRFs or by contacting the recruiting site for clarification.

In order to maximise the urine sample laboratory data, all entries on the CRF/database indicating that a urine sample had been obtained, but the laboratory had not entered the results, were followed up directly with the laboratories.

An inbuilt SQL quality control facility provided a full audit of all amendments made and was available as a CSV download. All amendments made to the database were also manually audited and recorded separately.

Data storage and retention

All data will be archived until the youngest participant reaches the age of 21 years, in line with each centre’s governance framework regulations for clinical research. The archive will consist of the protocol and all of the amendments, patient information sheets, CRFs and the follow-up data, and analysis records. Additionally, all of the legally required documentation relating to ethical approval, sponsorship and indemnity will be archived.

Research objective algorithm

The overall objective of the study was to derive and validate a clinical algorithm for the diagnosis of UTI, in children aged before their fifth birthday presenting to primary care with an acute illness. The algorithm was constructed in two stages: ID of children at risk (in whom a urine sample should be obtained) and determination of the added value of point-of-care urine dipstick testing. The determination of when to obtain a urine sample was based on sociodemographic factors, medical history, symptoms and signs. The purpose of the dipstick analysis was to assess whether dipstick urinalysis for nitrite, LE, protein, blood and glucose gives additional diagnostic information to assist in the ID of children who should receive immediate antibiotic treatment and urine samples that should be sent to the laboratory. The findings were combined to produce an overall algorithm.

Sample size calculation

Our sample size calculation was targeted at our main objective, namely the development of the clinical algorithm. We assumed a candidate predictor with 10% prevalence and UTI prevalence of 2%. With 80% power and a two-sided alpha of 5%, 3000 urine sample results are required to detect an odds ratio (OR) of 2.4, while 3100 results give a 95% CI with width 10% for an algorithm with 80% sensitivity. We originally proposed to recruit 4000 children with a target of recovering urines from at least 77.5% for algorithm derivation and a further 2000 children for external validation. However, we did not originally anticipate the need to stratify analyses by urine collection method. We therefore decided to use all available results to derive the models, with internal bootstrap validation instead of external validation as originally planned.

Determining prevalence rates

We calculated the prevalence in all samples and also by the collection method using methods appropriate for small proportions. 113 We also looked at the degree of variation in prevalence stratified by collection method. First, we considered the degree of variation in prevalence between practices using a two-level random-effect logistic regression model with practice/site as a random effect and no fixed effects. Second, we used a LR test between this multilevel model and logistic regression to show if there was any evidence of clustering, using a conservative p-value threshold of 0.1. Finally, using the chosen model type, multilevel logistic or logistic regression, we then added a number of variables to the model: area, type of recruitment site, age, sex and an Index of Multiple Deprivation (IMD) deprivation score for each child. We then assessed the effect these variables had on prevalence by listing the crude and adjusted ORs (with 95% CIs) for each of these variables.

Other analyses

We planned to use methods appropriate for small proportions113 to estimate the prevalence (with 95% CI) of culture positive urines in acutely unwell children under 5 years. This was undertaken on all children with cultured urines. The degree of variation in prevalence between practices and geographical areas was explored. This analysis was also explored by looking at difference by recruitment site type (general practice, WICs, out-of-hours providers and CEDs).

Recruited children for whom urine samples were obtained were compared with those for whom no urine sample was obtained in terms of clinical presentation and demographics.

We compared the probability of contamination in samples that were retrieved via a ‘clean-catch’ method with those using nappy pads and investigated the factors associated with contamination. We examined the impact of the time between obtaining the urine, transportation (including day of the week) and laboratory analysis on the rates of positive and contaminated urine samples (e.g. exploring if delayed samples such as those taken after daily laboratory collection have an impact on contamination rates) using recommended methods. 114

Types of primary care site

The majority of primary care sites were GP surgeries (n = 226, 96.6%), with four WICs and four CEDs. The Cardiff centre was the only centre to recruit exclusively from GP surgeries (n = 44). The final recruitment figures were 6797 (94.9%) in GP surgeries, 284 (4.0%) in CEDs and 82 (1.1%) in WICs. Three WICs were co-ordinated by the Bristol centre and one by Southampton. The Bristol centre also co-ordinated two CEDs, and London and Southampton each co-ordinated one CED site.

Recruitment models used by primary care sites

The two models of recruitment (option 1 and option 2) offered to primary care sites are described in Chapter 2. Owing to the practicalities of recruiting in busy clinical environments, the WICs and CEDs employed only the option 2 recruitment model. Table 7 shows the distribution of recruitment methods by centre.

Over half of the primary care sites (n = 124) employed an option 2 recruitment model, while 99 (42.5%) employed an option 1 model. There were 12 (5.1%) sites that used a mixed recruitment model, where both the site staff and the DUTY RN/CSOs would recruit at the site.

Eligibility assessment

A total of 14,724 children were assessed for eligibility across the four study centres according to completed screening logs. However, this figure is an underestimation of the total number of children assessed, as not all sites completed screening logs and some were incomplete.

Of the 234 primary care sites taking part, 198 (85%) completed and returned at least one screening log to the study centres. These show that 7350 children were screened but not recruited (see Figure 4), as they declined (1276) or were not eligible (4390), or for other reasons (1684) including that they left the primary care site prior to invitation (811), they did not give consent (214) or there was a language barrier (112) and an appropriate translator was not available at the time of recruitment. Table 8 shows the distribution of these children by centre.

Comparison of children recruited to DUTY with those not recruited

From the screening log information, which recorded the sex and date of birth of the children visiting the primary care sites, we were able to compare the age and sex of the children who were recruited to DUTY (n = 7163) with that of the children whose parents declined to participate (n = 1276) (Table 9).

Comparing the proportion of males in the two samples with a two-sample proportion test shows strong evidence that the proportion of males is higher in the declined sample, with a mean difference of 5.2% (95% CI 2.2% to 8.2%; p < 0.001). The mean age in the declined sample was 24.06 months and in the recruited sample it was 26.88 months. Comparing the mean age in the two samples with an independent sample t-test also shows strong evidence that the mean age is higher in the recruited sample, with a mean difference of 2.04 months (95% CI 1.08 to 3 months; p < 0.001).

Differences in sex and age are not particularly large and the small p-value and narrow 95% CIs are due to the large sample size. Figure 5 shows that the distribution of age in the two samples is similar, with both skewed towards the younger age.

FIGURE 5.

Histograms of the ages of the children recruited and those who declined to participate. Histogram of age for (a) recruits and (b) decliners.

Recruitment accruals

A total of 7374 children were enrolled into the study (50.1% of those assessed for eligibility); of these 196 were excluded and 15 were withdrawals (see Exclusions and Withdrawals, below).

Of the 7163 total recruited children, 6797 (94.9%) were recruited in GP surgeries, 284 (4.0%) in CEDs and 82 (1.1%) in WICs. The Bristol centre recruited 2947 (41.1%), the Cardiff centre recruited 1768 (24.7%), the London centre recruited 1435 (20.0%) and the Southampton centre recruited 1013 (14.1%) participants to the study during the recruitment period.

When comparing the total number of children recruited to the DUTY study with the original recruitment targets set by the HTA for study viability assessment, it can be seen that that the study consistently exceeded its projected study target (Figure 6). The sample size requirement (6000) was reached in January 2012, when we sought approval (protocol amendment 9) to continue recruiting until the end of the scheduled period (30 April 2012) in order to maximise the statistical power of the sample and provide more precise estimates of association of index tests with UTI in the final algorithm, and to take account of missing data.

FIGURE 6.

DUTY recruitment accruals against HTA targets.

Exclusions

A total of 196 children were retrospectively excluded, including 44 children who were retrospectively ineligible because they had been recruited in a CED after being referred by their GP. These participants had not been recruited on their first point of contact in primary care, and were likely to be selected as more unwell. The study protocol was subsequently amended (protocol amendment 6) to ensure that only children consulting the CED for the first time, without prior GP or other primary care contact, were included. An additional 55 children were excluded from a site co-ordinated by the London centre due to poor data quality issues. This site received intensive training and recruited a further 10 participants; however, the data quality was still poor, and therefore the site was closed to further recruitment. During data cleaning a further 97 children were also found retrospectively to be ineligible owing to absence of urinary and/or constitutional symptoms, previous (duplicate) recruitment or invalid consent, or because someone other than the parent had (inappropriately) consented.

Withdrawals

There were 20 withdrawals during the course of the study, of which nine consented to use the data already collected. However, in four cases data were inadvertently removed from the database instead of retaining data collected up to the point of withdrawal. Therefore, all data were withdrawn for 15 children, whereas five withdrawals were made with the retention of data already collected. The most common reason for withdrawal was the parent changing his or her mind about the child participating in the study.

Positive unintended consequences of the DUTY study

One child was diagnosed with type I diabetes as a direct consequence of taking part in the DUTY study. This was as a result of the dipstick showing high glucose levels in combination with their medical history. Type I diabetes can be missed as the symptoms may often be missed in primary care.

Urine collection methods

Of the 6390 urine samples collected, 3721 (58.2%) were collected in the surgery and 2632 (41.2%) were collected in the child’s home and returned to the surgery or collected by the RN/CSOs. In 37 cases, no data were recorded on where the sample was collected.

Figure 7 shows the method of urine collection, which included the DUTY-preferred method of clean catch which accounts for just under half (47.5%) of the samples collected (n = 3036), while 50.2% were collected using our alternative method of Newcastle nappy pads (n = 3205). A small proportion of samples were collected by urine bag (n = 100, 1.6%), with a further 49 cases where the method of collection was not recorded (0.8%).

FIGURE 7.

Urine collection methods.

Proportional selection for follow-up

The algorithm for proportional selection of participants for follow-up was set up to ensure that children with a positive culture as defined by the NHS laboratory or research laboratory at ≥ 10⁵ CFU/ml were followed up (see Chapter 2, Patient follow-up, Day-14 and 3-month selection). For a short period, the algorithm was not working as intended, in that if the results from the laboratories had different levels of growth, the lower category was chosen, meaning that the higher (≥ 10⁵ CFU/ml) result category was under-represented. This error was detected too late to follow up 277 children at 14 days, but 3-month notes review data were collected for these children.

Table 11 shows the proportion of participants that were selected and the number of 14-day interviews that were completed for each of the result categories (omitting the 277 urine samples that failed to be selected by the algorithm). Of the 13 urine sample result categories, nine were near the target proportion set, or on, or above it. There were 6390 children for whom a urine sample had been taken; 6314 had results entered into the DUTY database during the course of the study and 1276 (20%) children were successfully selected for follow-up.

All of the pure/predominant samples of ≥ 10⁵ CFU/ml should have been selected by the algorithm for follow-up; however, only 579 were actually selected, giving a selection rate of 67.7%. There were 413 interviews completed, giving a completion rate of 71.3% for this category.

Fourteen-day follow-up

There were 1276 participants who were selected for 14-day interviews, for whom 918 (71.9%) were successfully completed. Table 12 shows the completion rates across the four study centres.

Of the 1276 participants successfully selected for follow-up, interviews could not be completed for 358 (28.1%). The majority of cases (n = 164, 45.8%) were missed because the parent could not be contacted, while in 129 cases (30.6%) no reason was recorded, and a further 13 interviews (3.6%) were not completed for other reasons (e.g. parent refused to complete the interview or the participant was withdrawn).

The majority of the 14-day follow-ups were completed by telephone interview. However, postal questionnaires were sent if, after three or four attempts at contacting the parents by telephone were made, no successful contact resulted. Postal questionnaires were implemented from September 2011 onwards. In total, 69 14-day follow-ups (7.5%) were completed by postal questionnaire. We can only estimate the response rate of the postal questionnaires, as only three of the centres recorded the number of postal questionnaires sent out. With the information we do have from those centres, the collective response rate for the whole of the study was 45.9%. The Cardiff centre had a response rate of 44.2% (19 out of 43 questionnaires returned), London had a response rate of 31.3% (10 out of 32 questionnaires returned) and Southampton had a 100% response rate (10 out of 10 questionnaires returned). The Bristol centre had 32 questionnaires returned. Fifty-two postal questionnaires were not returned (14.5%).

Three-month notes review

Of the 1553 children selected for follow-up, 3-month note reviews were completed for 1542 (99.3%) (Table 13). Only 11 3-month note reviews were not completed, the main reason being that the child had left the surgery. In the London centre, reviews could not be completed in instances where the child was recruited outside their usual GP practice (e.g. WIC or CED) and no PCT R&D approval was in place for the recruited GP practice.

Description of study population

Table 14 shows the demographic characteristics of the 7163 recruited children. Approximately half of the children were aged < 2 years with equal numbers of boys and girls (49.2% and 50.8%, respectively). The sample was predominantly white (82.3%), with one-quarter of the parents educated to degree level or equivalent, and nearly another quarter educated to GCSE level or equivalent. The financial status of parents was also assessed by asking a question regarding the cost of living, where 40.4% reported they had to be careful about money, and almost one-quarter said that they were able to manage without much difficulty.

The comparison of DUTY recruits with the national census data for age, sex and ethnicity can be seen in Table 15. This shows that the children recruited to DUTY were broadly similar in age to the national profile, bearing in mind that children recruited to DUTY were presenting unwell to primary care.

Description of presenting symptoms and signs

Table 16 demonstrates the frequencies of the symptoms and signs categories (see Appendix 4, section 3 of the case report form) for all 7163 children who were recruited to the DUTY study. The children had been unwell for a median of 4 days (including the day of recruitment) before consulting at the primary care site with their acute illness. In 5261 (73%) children, it was the first time parents had consulted their doctor or nurse with the illness episode. The most prevalent sign and symptom was that the ‘child was not themselves’, with this being reported as a moderate/severe problem in 4682 (65.4%) participants. The second most prevalent sign/symptom was confusion/disorientation, which was reported as a moderate/severe problem in 4628 (64.6%) participants. Fever at any time during the illness was a moderate/severe problem in 3803 (53.1%) participants. Other most prevalent presenting symptoms and signs, reported as a moderate/severe problem, included cough (n = 3771, 52.6%), blocked or runny nose (n = 3695, 51.6%), refused feeds/eating less than normal (n = 3670, 51.2%), and fever at any time during the past 24 hours (n = 3057, 42.7%).

Description of clinical examination findings

Clinical examinations were performed by either a doctor or a nurse, and the reported findings recorded on the CRF (see Appendix 4, section 4 of the case report form) for all recruited patients are presented in Table 17.

For the 7163 children recruited, the responsible clinician’s global impression score was recorded as between 0 and 4 in over 90% of cases. Hydration and consciousness level were reported as normal in 94.6% and 96.8% of cases, respectively. From a general examination, 68.0% of cases were recorded as normal. Normal findings were recorded by the responsible clinician on examination of the throat, ear, chest and abdomen in 60.5%, 63.0%, 79.6% and 81.2% for recruited children, respectively (see Table 17).

For the 2213 (30.9%) cases with an abnormal general examination the following findings were recorded: pallor (27.8%); flushed (35.1%); jaundice (0.4%); lymphadenopathy (15.6%); distressed (11.5%); rash (7.6%); and other (17.3%). For the 1707 cases with an abnormal throat examination the following findings were recorded: red or inflamed (84.7%); swollen (17.3%); quinsy (10.0%); lymph nodes (1.1%); tonsillitis (1.55); mouth ulcer (1.2%); oral thrush (1.1%); and other (4.2%). For the 1751 with an abnormal ear examination an acute abnormality was recorded in 1576 cases (90.0%) and a chronic abnormality in 4.7%. The main findings recorded were pink (43.9%); red or bulging (42.6%); fluid (1.8%); acute perforation (0.1%); chronic perforation (2.2%); glue ear (0.6%); grommets (0.3%); otitis externa (1.3%); otorhoea (2.2%); otitis media (3.2%); wax (6.45); and other (3.0%). For the 919 cases with an abnormal chest examination the following findings were recorded: bronchial breathing with distribution (7.95); wheeze (44.5%); crackles (57.8%); recession (8.1%); grunting (1.5%); nasal flaring (0.5%); transmitted sounds with distribution (2.6%); and other (4.4%). For the 179 cases with an abnormal abdomen examination the following findings were recorded: mass or organomegaly (3.4%); loin tenderness (5.6%); suprapubic tenderness (22.9%); other abdominal tenderness (25.7%); genital inflammation or nappy rash (10.1%); and other (33.0%). These data are not included in the report, but are available from the authors on request. Chapter 5 provides a description of the inclusion of symptoms and signs in the prediction rule statistical analysis (see Methods, Statistical analysis) and the results of the rule (see Results, Clean-catch models and Nappy pad models).

Urine collection and sample processing

Urine samples were obtained by clean catch, where possible, for children who were toilet trained or for whom the parent was happy to attempt such collection. A full description of the urine collection methods is given in Chapter 2.

NHS laboratories processed urine samples using their local SOPs and recorded data according to local reporting procedures, including, where possible, quantifying bacterial growth (as < 10³; 10³ to < 10⁵; or ≥ 10⁵ CFU/ml), purity of growth (pure/predominant; mixed growth two species; mixed growth > 2 species), organism speciation for up to two species, and microscopy for white and red cells. The research laboratory quantified absolute colony counts (range 10¹–10¹⁰ CFU/ml) for all organisms present and established species ID for organisms present at ≥ 10³ CFU/ml.

Statistical analysis

Results reported by the NHS and research laboratories were initially classified based on extent and purity of growth and whether or not the species grown was a uropathogen, defined as a member of the Enterobacteriaceae group. For NHS laboratory results, samples for which the laboratory reported pure/predominant growth of a uropathogen at ≥ 10⁵ CFU/ml were considered UTI positive. For research laboratory results, samples were considered UTI positive if there was growth of ≥ 10⁵ CFU/ml of a single uropathogen (‘pure growth’) or growth of ≥ 10⁵ CFU/ml of a uropathogen with ≥ 3 log₁₀ difference between the growth of this and the next species (‘predominant growth’). Agreement between laboratories was assessed using kappa statistics, with analyses additionally stratified by urine collection method (clean catch or nappy pad) and by age (0 to < 2, 2 to < 3 and 3 to < 5 years).

Analyses were restricted to samples with a result from both the NHS and research laboratories. Samples classified as UTI positive by both NHS and research laboratories results were denoted ‘Agree UTI positive (step 1)’. Where there was disagreement between the NHS and research laboratories, we considered whether or not the combined evidence was consistent with a UTI. When the result classified as negative in one of the laboratories met either of the two conditions (1) growth of ≥ 10⁵ CFU/ml of a uropathogen with lesser growth of at most one other species, or (2) pure/predominant growth of 10³ to 10⁵ CFU/ml of a uropathogen, samples were denoted ‘Agree UTI positive (step 2)’.

A priori (before examining their associations with different definitions of microbiological positivity from different laboratories), we selected a small number of variables (symptoms, signs and dipstick test results) reported in the literature to be clearly related to presence of a UTI. 116 Those thought suitable for all ages and collection methods were urinary symptoms (pain/crying when passing urine, passing urine more often, changes in urine appearance); temperature ≥ 39 °C, and nitrite- or leucocyte-positive results from urine dipstick tests. Additional symptoms and signs thought to be relevant mainly for older children were daytime or bed wetting when previously dry, and a history of UTI. Most parent-reported symptoms were recorded using categories ‘no’, ‘slight’, ‘moderate’ or ‘severe problem’; we decided a priori (based only on inspection of symptom frequencies) to dichotomise these into ‘no or slight problem’ and ‘moderate or severe problem’. Responses of ‘not known’ were coded with ‘no or slight problem’. A small number of samples for which there were missing data on most or all urine symptoms (five samples), or for which urine dipstick tests were not available (12 samples), or there was missing information on prior infection (three samples) were excluded. Missing data on temperature (204 children) were coded as < 39 °C. The remaining, sporadic missing values (on five children) were coded as ‘no or slight problem’.

We used separate logistic regression models to quantify associations of the selected variables with UTI positivity in the NHS and research laboratories, and for the different outcomes ‘agree UTI’, ‘disagree (NHS positive, research negative)’ and ‘disagree (NHS negative, research positive)’, all compared with ‘agree UTI negative’. We plotted receiver operating characteristic (ROC) curves and AUROCs to quantify diagnostic utility. The maximum value of the AUROC is 1 (perfect prediction) while a value of 0.5 corresponds to no association with any predictor. The symptoms ‘day or bed wetting when previously dry’ and ‘history of UTI’ were recorded in too few of the children who provided nappy pad samples to permit their associations with microbiology results to be examined. For these children, therefore, we examined associations of the remaining six signs, symptoms and dipstick results with microbiology results. For children who provided a clean-catch sample, we fitted two sets of logistic regression models, one including all eight symptoms and dipstick results (the ‘eight variable model’) and the other, for comparability, including the same six symptoms and results as for children providing nappy pad samples (the ‘six variable model’).

In sensitivity analyses we (1) stratified by age (< 3 and ≥ 3 years); (2) allowed for ‘not known’ categories in the questionnaire responses for variables for which such responses occurred sufficiently frequently; (3) stratified according to whether samples were coded as ‘agree UTI positive’ at step 1 or step 2; (4) stratified by whether the sample was collected at the surgery or at home; (5) stratified by time between recruitment and laboratory sample receipt (< 24 hours and ≥ 24 hours); (6) stratified NHS laboratory results according to extent of pure/predominant growth (≥ 10⁵, ≥ 10³ to < 10⁵ CFU/ml); (7) stratified research laboratory results according to extent of pure/predominant growth (≥ 10⁷; ≥ 10⁶ to < 10⁷; ≥ 10⁵ to < 10⁶; ≥ 10⁴ to < 10⁵; and ≥ 10³ to < 10⁴ CFU/ml); (8) stratified NHS and research laboratory results according to whether the WBC count was < 30 or ≥ 30/mm³; (9) stratified research laboratory results according to whether growth was pure or predominant, and (10) as the research laboratory received only the urine available after the priority fraction was placed in the NHS collection sample, we stratified by research laboratory urine volume (using the cut-point of median urine Monovette weight for clean-catch and nappy pad samples). All analyses were carried out using Stata version 12 (StataCorp LP, College Station, TX, USA).

Summary of findings

Based on a large, unselected cohort of children presenting to primary care in England and Wales with acute illness, the reliability of microbiological diagnosis of UTI in routine NHS laboratories and a research laboratory was lower than expected and worse for urine samples collected using nappy pads than for clean-catch samples. Associations of microbiological positivity with pre-specified symptoms, signs and urine dipstick test results were lower for NHS laboratories than the research laboratory, and for nappy pad samples than clean-catch samples. Urines giving a ‘positive’ result in a NHS laboratory but not in the research laboratory had only modest associations with the preselected symptoms, signs and dipstick test results. These findings did not appear to be attributable to the younger age of the children providing nappy pad samples. Discrimination improved with increasing bacteriuria concentration and with the presence of WBCs in the urine sample (pyuria). Results of urine microbiology should, therefore, not be regarded as dichotomous result, but rather as a continuous phenomenon to be interpreted in the clinical context, with UTI possible even when bacterial concentrations are between 10³ and 10⁴ CFU/ml, and becoming increasingly probable with higher concentrations of pure or predominant bacterial growth in the presence of pyuria.

Results in context with other studies

To our knowledge, this is the largest and most generalisable primary care-based study comparing the diagnostic performance of NHS laboratories with a research laboratory, and using nappy pad and clean-catch collection methods. However, the number of UTI positive samples was relatively small, especially for clean-catch samples in younger children and for the research laboratory. We do not know which samples were sent to NHS laboratories in containers with boric acid: all samples were sent via the routine mechanisms for that laboratory. All research laboratory samples were sent in containers with boric acid by Royal Mail post, introducing longer delays (which were not associated with UTI status) before processing than with NHS laboratories. In current UK primary care, nappy pad and bag samples are often the only feasible methods for obtaining urine samples from young children: there is usually insufficient space or staffing for parents and children to wait to provide clean-catch specimens. Moreover, most primary care clinicians (other than those with specialist paediatric or emergency department training) are not trained in SPA or catheter techniques, and neither of these are acceptable to parents. Nappy pads have been shown to be acceptable to parents63 and endorsed by NICE. 2

In the early development of microbiology, laboratory diagnosis of UTIs required isolation of the same organism from repeated urine samples. It was recognised in the 1940s that high urine bacterial counts were related to UTI, and subsequently proposed that a single sample with a high count could support a laboratory diagnosis of UTI and allow early treatment. 117 Early proposed thresholds to define positivity were derived from detailed investigation of fresh urine samples and ranged from ≥ 10³ CFU/ml118 to ≥ 3 × 10³ CFU/ml119 and ≥ 10⁵ CFU/ml. 120 Recently, it was suggested that a threshold of ≥ 10⁶ CFU/ml would be more appropriate. 80 Current laboratory guidelines differ with respect to recommended thresholds. The UK Standards for Microbiological Investigations do not have specific paediatric guidance and suggest a threshold of a ‘single organism ≥ 10⁴ CFU/ml indicating UTI’, but other thresholds are also discussed. 82 European paediatric guidelines suggest a threshold of ≥ 10⁴ CFU/ml if symptoms are present and ≥ 10⁵ CFU/ml if no symptoms are present for midstream specimens, and lower thresholds for specimens collected by bladder catheterisation or SPA. 83 US guidance suggests that clinicians require both urinalysis evidence of infection and a threshold of ≥ 5 × 10⁴ CFU/ml. 121

Microbiological examination of urine requires quantification of bacteria and the ability to differentiate mixed from pure growths. The pour-plate method proved too labour-intensive given the large numbers of urine samples submitted to routine microbiology laboratories in the UK:82 in 2012, 663,355 samples (12,689 from children aged < 5 years) were submitted in Wales alone, equating to some 12.1 million samples annually (250,000 from children aged < 5 years) in England and Wales. More rapid methods using calibrated loops, filter paper strips or multipoint methods to deliver a standard inoculum were developed in response to the need for rapid throughput. 122–124 All were validated against viable counts performed by pour plates or the method of Miles and Misra. 125 The Standards for Microbiological Investigation followed by most UK laboratories provide options for urine culture using these methods to inoculate CLED or chromogenic agar, but they have not previously been calibrated against clinical symptoms. Difficulties in defining mixed growths and achieving accurate bacterial counts may be due to the small volumes of urine inoculated onto small areas of agar. 80,122 Spiral plating, which was used by the research laboratory in this study and involves a much larger inoculum (50 µl) over an entire 9-cm agar plate, is a more accurate method of quantifying bacterial counts and allows easy differentiation of mixed cultures. 126 Enhanced diagnostic performance might also be achieved through improvements in procedures for sample collection and transport.

NHS laboratory UTI positivity was consistently less strongly associated with urinary symptoms and dipstick results than research laboratory UTI positivity. A substantial number of UTI positive NHS laboratory samples [128 (5.8%) nappy pad and 59 (2.3%) clean catch] were negative in the research laboratory, and associations of the NHS laboratory positive results with symptoms, signs and dipstick results were modest. These findings suggest that the diagnostic performance of current routine NHS laboratory testing is suboptimal and may lead to overtreatment and unnecessary investigations.

Clinical and microbiological implications

Even for samples processed in the research laboratory, the diagnostic utility of microbiology based on nappy pad samples was less than for clean-catch samples. The prevalence of UTI positivity diagnosed in the research laboratory was lower for nappy pad (1.3%) than for clean-catch (2.3%) samples, suggesting that UTIs are missed in nappy pad samples because of contamination. Therefore, primary care clinicians should try to obtain clean-catch samples in even very young children in whom they suspect a UTI,127 for example by providing time and space to support urine collection. If an algorithm based on parent-reported symptoms can provide earlier ID of the children at greatest risk of UTI, then parents could be advised to obtain a urine sample prior to attending primary care.

In adult medicine, results from urine microbiology can be interpreted in the clinical context of the patient’s presentation. However, in young children the significant difficulties in obtaining uncontaminated samples, together with the non-specific nature of the presenting symptoms, mean that there is greater reliance on the laboratory result. More detailed routine microbiological examination of paediatric urine samples would have resource implications that could better be justified if urines were selected for testing through an algorithm that increased the prior probability of positivity. Our results suggest that NHS laboratories should distinguish primary care paediatric (age < 5 years) samples from adult samples and consider reporting these in more detail, and that national procedures should be correspondingly updated.

Adoption of the more accurate but labour-intensive research laboratory methods would not be appropriate to use for all urines processed in NHS laboratories but, in many laboratories, extra or augmented methods are used for ‘special urines’, usually received in small numbers, to enhance reporting. It seems reasonable that research laboratory methods be implemented in NHS laboratories for the processing of paediatric urines if associated with enhanced diagnostic performance.

Implications for the DUTY study

For the purposes of the DUTY study, samples will be considered UTI positive (called ‘the reference standard’ in Chapter 5) if there is growth in the research laboratory of ≥ 10⁵ CFU/ml of a single uropathogen (‘pure growth’) or growth of ≥ 10⁵ CFU/ml of a uropathogen with ≥ 3 log₁₀ difference between the growth of this and the next species (‘predominant growth’).

Participants

The methods of recruitment are described in detail in Chapter 2 and our study protocol (see Appendix 3). 106

Index tests and urine collection

All participating NHS staff received training in study procedures and only clinically qualified staff (GPs, nurse practitioners or emergency department doctors/nurses) performed and recorded clinical examination findings. Following consent, index tests (symptoms, signs or dipstick results) were recorded on a CRF (see Appendix 4) using a paper- or a web-based data collection system, with the latter requiring a unique log-in account for each staff member. Symptoms included the child’s medical history and parent-reported symptoms (graded as absent, mild, moderate or severe when at their worst during the illness). Signs, from a full clinical examination, included ‘clinicians’ global impression of the child’s illness severity’, rated on a scale of 0 (child displaying no constitutional upset) to 10 (child displaying life-threatening signs requiring immediate hospitalisation) and ‘abdominal tenderness’ (any of suprapubic, loin or other abdominal tenderness). In total, 107 symptoms and signs were recorded and, preceding urine dipstick testing, clinicians recorded their working diagnosis, including a rating of the likelihood of UTI (‘clinical diagnosis’) as ‘UTI fairly to very certain’, ‘no UTI fairly certain or uncertain’ or ‘no UTI certain to very certain’.

Our preferred urine collection method was ‘clean catch’. NICE-recommended ‘Newcastle nappy pads’2 were used for children still using nappies and those for whom the parent/guardian did not think clean catch would be successful. Urine collection bags were used in a few instances, but data from these samples were excluded from the present study. At the primary care site, urine samples were dipstick tested (using Siemens Multistix 8 SG) for blood, protein, glucose, ketones, nitrite, LE, pH and specific gravity (eight dipstick index tests). For full details of urine collection methods, see Chapter 2, Urine sample collection, and our protocol paper (see Appendix 3). 106 All 107 index tests and the clinicians’ working diagnosis were measured blind to the reference standard.

Reference standard

Research laboratory samples were processed by two staff members using a single, standardised procedure. We used a microbiological definition of UTI of ≥ 10⁵ CFU/ml of a single uropathogen (‘pure growth’) or ≥ 10⁵ CFU/ml of a uropathogen with ≥ 3 log₁₀ (1000-fold) difference between the growth of this and the next species (‘predominant growth’) (see Chapter 4). We defined uropathogens as members of the Enterobacteriaceae family. Aside from knowing children’s dates of birth (used as part of the urine ID process), research laboratory staff were blind to children’s index test characteristics. We compared the characteristics of children for whom research laboratory culture results were and were not available.

Statistical analysis

We examined the frequency of symptom and sign categories, blind to their associations with urine culture results, and merged the least frequent categories prior to analyses. We used logistic regression to estimate associations of index tests with urine culture positivity. We fitted separate models for clean-catch and nappy pad samples because of the markedly different age distributions of children providing samples using these methods and concerns about contamination reducing the sensitivity of urine culture in nappy pad samples. The p-values were derived using LR tests. For ordinal variables, both heterogeneity and trend p-values were derived. Continuous variables were divided into quintiles and trend p-values were derived using the median within categories. We examined plots of the log-odds of culture positivity against the median within quintiles for evidence of non-linearity.

We used two methods for dealing with missing data, including the response ‘don’t know’ to questions about the presence of symptoms such as pain or crying when passing urine. In both univariable and multivariable analyses, missing data were coded as the modal value, usually absence of the symptom. We repeated multivariable analyses using the chained equations approach to multiple imputation: estimates and Wald p-values130 based on 50 imputed data sets derived using Rubin’s rules. 131

We sequentially evaluated selected index tests in two groups: parent-reported symptoms and clinician-reported signs, and urine dipstick results. This reflects the clinical process, where symptoms and signs are used to identify children from whom urine should be collected and dipstick testing may additionally help clinicians to decide which children should be treated. First, we selected those variables with either trend or heterogeneity univariable p-value < 0.01 for either collection method or when all samples were analysed together. Second, we derived models from among all of the selected symptoms and signs, separately for nappy pad and clean-catch samples, using backwards stepwise selection and an exclusion criterion of heterogeneity p-value > 0.1. Third, we used backwards stepwise selection with the same exclusion p-value for models in which dipstick results were added to the previously selected symptoms and signs, to give models including symptoms, signs and dipstick results. We investigated the effect of using more liberal p-value thresholds of 0.1 and 0.2 at each stage, and found no important differences in the final models (results available from the authors on request).

For each model, we quantified diagnostic accuracy as the AUROC (also known as the c-statistic). We also estimated AUROC values for ‘clinician diagnoses’ of UTI. To formally test the additional value of dipsticks we calculated the difference in AUROC, along with 95% CI and p-value, between the symptoms and signs model and the symptom, sign and dipstick model. We internally validated the models, calculating the AUROC and calibration slope, using the bootstrap procedure described by Steyerberg. 132 The calibration slope is a shrinkage factor that compares predicted with observed probabilities, where 1 is perfect calibration and 0 reflects no agreement between predicted and observed probabilities. Multiplying the estimated coefficients (log-odds ratios) in a predictive model by the calibration slope would correct the estimated positive predictive values for statistical overoptimism due to the model fitting. For each model, we selected cut-points corresponding to a range of values for sensitivity, and then calculated the corresponding specificity, negative and positive predictive values, and the proportion of children classified positive. This was completed using the linear predictors that were combined using Rubin’s rules. 131 CIs were calculated for these parameters by transforming to the log-odds sale, and then using the standard error formula for log-odds, and then transforming back to a proportion. As log-odds are undefined for a proportion of 0 or 1, we were unable to calculate CIs when these parameters were equal to either 0% or 100%.

We also investigated the robustness of our clean-catch models in younger children by fitting the clean-catch models to those with clean-catch samples who were under 3 years of age. Finally, to investigate the robustness of our nappy pad models, we fitted the nappy pad models to those with clean-catch samples who were under 3 years of age.

Fever of unknown origin

In order to investigate if ‘fever of unknown origin’ was associated with UTI, we identified children with fever and no symptoms or signs suggestive of an alternative source. We did three analyses using the three fever variables – two parent-reported (‘fever now or in the past 24 hours’ and ‘fever at any time during this illness’) and one from the physical examination (temperature of ≥ 38 °C) – for children without any of the following symptoms and signs: new generalised rash; vomiting; diarrhoea (any time and last 24 hours); blocked/runny nose; cough; wheeze; shortness of breath; chest pains; earache; sore throat; oxygen saturations < 94%; any throat abnormality; any ear abnormality; and any chest abnormality. The categories ‘slight’, ‘moderate’ and ‘severe problem’ for the parent-reported fever variables were merged owing to the small number of UTI positives within these categories. We conducted two analyses for each of the clean-catch and nappy pad samples: one used a complete case approach and the other assumed ‘modal/normal’ imputation for each of the 14 variables above.

Points-based models

As the above models are relatively complex and would need a computer or calculator to estimate risk scores, we also generated points-based models where the risk of UTI could be more easily calculated. To do this, we first dichotomised all variables in our original models, including dipsticks. To make the resulting rule as user-friendly as possible, we dichotomised all variables at the ‘present/absent’ threshold. We removed clinical examination variables, as these variables contributed the least to the models and would take a disproportionate amount of time to collect in relation to their value. After deriving these dichotomised variable models using the multiple imputed data, we calculated a corresponding points-based model. Using methods similar to previous studies,133,134 points were calculated for each variable by dividing that coefficient’s variable by the smallest coefficient in the model and then rounding to the nearest integer. We quantified the diagnostic accuracy of the resulting points-based models using the AUROC, and compared these with the ‘full models’. To formally test the additional value of dipsticks, we calculated the difference in AUROC, along with its 95% CI and p-value, between the symptoms and signs model and the symptoms, signs and dipstick model. This comparison was estimated for both the coefficient and the integer points-based models. The points-based models cannot be validated using bootstrapping as this method relies on the coefficients being estimable in different (bootstrap) samples and, in the ‘points-model’, these are fixed. However, to assess possible optimism of the points-based models, we internally validated the dichotomised models without the points’ scores. For the points-based models, we selected cut-points corresponding to a range of integer risk scores, and then calculated the corresponding sensitivity, specificity, negative and positive predictive values, and the proportion of children classified positive. This was completed with the points risk scores that were combined using Rubin’s rules. 131

Additionally, for the points-based models we selected cut-points corresponding to the complete range of risk scores for each possible combination of symptoms and signs, and then calculated the corresponding sensitivity, specificity, negative and positive predictive values, and the proportion of children classified positive.

Added value of dipstick testing

In addition to looking at the difference in the AUROC between the models with and without the dipstick test results, we fitted multinomial logit models for the association of the selected symptoms and signs with the dipstick test results included in the symptoms, signs and dipstick model. We then quantified the additional value of dipstick testing in diagnosis of UTI using a three-step simulation approach based on the points-based models. The steps were (1) sample coefficient values from the multivariate normal distribution of the multinomial logit parameter estimates; (2) based on the sampled coefficients, randomly generate a set of dipstick results; and (3) compute the probability of UTI based on the symptoms, signs and dipstick points-based model. For each combination of symptoms and signs we generated 10,000 samples and calculated the probability of UTI, with and without the dipstick results, and the change in probability of UTI after accounting for the dipstick results. One of the dipstick combinations was dropped as it was observed in only 3 out of 2740 individuals and led to numerical instabilities. We grouped the distribution of these probabilities and changes in probabilities according to whether or not primary care clinicians considered it reasonable to recommend urine sampling based on symptoms and signs and the cost-effectiveness analyses; results were weighted according to the relative frequencies of different combinations of symptoms and signs within the groups.

Participants

The recruitment of participants is described in Chapter 3. In summary, 14,724 children were screened for eligibility, of whom 4390 were immediately ineligible, 1276 declined, 1684 could not be recruited for other reasons, 196 were subsequently excluded and 15 withdrew (see Figure 4). Urine was collected using clean-catch or nappy pad methods from 6241 (87%) children and sent to the NHS (priority) laboratory in 6192 (86%) and the research laboratory in 5170 (72%); these figures and where they flow from can be seen in Figure 10. Reference standard results were available from the research laboratory (our final analytic sample) in 5017 (70%), of which 2740 (55%) were collected via clean catch and 2277 (45%) via nappy pad (see Figures 4 and 10). The number in Figure 10 in the box ‘urine sent to research laboratory’ does not include those urines collected by bag or those for which we did not know the collection method.

FIGURE 10.

DUTY flow diagram for diagnostic algorithm. a, Includes left prior to invitation (n = 811), no consent (n = 214), language barrier (n = 112); and b, excluded as retrospectively ineligible/duplicate recruitment/invalid consent (n = 141) and poor data quality (n = 55).

Comparing recruited children with (n = 5017) and without (n = 2146) a research laboratory culture result, we found that children with a culture result were older (59% vs. 33% were ≥ 2 years); more likely to have a mild illness (83% vs. 80% clinician global impression < 3); more likely to report pain/crying on passing urine (12% vs. 8%); more likely to report darker urine (14% vs. 7%); more likely to have UTI suspected clinically (7% vs. 4%); less likely to report nappy rash (16% vs. 20%); and less likely to be nitrite positive (9% vs. 14%). However, they were similar in terms of sex (female 51% vs. 50%); ethnicity (non-white 17% vs. 18%); parental highest qualification (diploma or degree 49% vs. 50%); deprivation (21% vs. 21% for most deprived quintile score); parental impression of overall illness severity (19% vs. 15% < 3); days unwell prior to recruitment (both median 4 days); and the prevalence of all the other symptoms, signs and dipstick index tests found to be independently associated with UTI. Further details of these comparisons are given in Chapter 3.

The most common non-UTI clinical diagnoses in the clean-catch and nappy pad groups, respectively, were ‘upper respiratory tract infection’ (28% and 35%), ‘viral illness’ (15% and 18%) and otitis media (10% and 9%). Gastroenteritis was thought to be present in 3.6% and 5.7%, respectively. Antimicrobial substances, which can arise from the use of both systemic antibiotics and locally applied cleaning agents, were found in 4.5% and 6.6% of clean-catch and nappy pad samples, respectively, and for both collection methods samples were more likely to be present in children with than without UTI (Table 27, and see Table 38). A total of 79 children were reported hospitalised, none as a result of study participation. We are not aware of any adverse events resulting from the measurement of index or reference tests.

Clean-catch models

Table 27 shows that 2.2% of the 2740 children providing clean-catch samples met the criteria for a microbiological diagnosis of UTI, and that 94% were aged ≥ 2 years, 54% were female, and 96% of samples were provided within 24 hours of index test measurement. Clinicians correctly identified 47% of the 60 UTIs. Missing data or ‘not known’ responses were generally infrequent other than in the youngest children (nappy pad samples), in whom symptoms, such as pain/crying on passing urine, cannot be determined (see Tables 27 and 38).

‘Full’ coefficient-based models

Tables 27 and 28, respectively, show the crude and adjusted ORs for the index tests (marked with footnote marker d in Table 28) selected for the clean-catch model, for children predominantly ≥ 2 and < 5 years. Table 28 shows that parent-reported pain/crying while passing urine, smelly urine, history of UTI, and absence of severe cough were independently associated with UTI and, for the first two of these, evidence of increasing strength of association for increasing severity of symptoms with UTI. Clinician-reported global impression of illness severity, abdominal tenderness on examination and the absence of acute ear abnormality on examination were independently associated with UTI, again with the former showing a gradation of association with increasing illness severity. Leucocytes, nitrites and blood on dipstick testing were strongly and independently associated with UTI. The increase in the AUROC while using the symptoms, signs and dipstick model was around 0.06 for both the modal and the multiple imputation, and the low p-value means that there is very strong evidence that the dipsticks increase the AUROC when added to the symptoms and signs model. The validated multiple imputation AUROC for the symptoms and signs model (0.876) demonstrated very good diagnostic accuracy (see Table 28). Adding dipstick findings increased the validated AUROC to 0.903. For comparison, the AUROC for ‘clinician diagnosis’ was 0.774 (0.714 to 0.833). Figure 11 shows the multiple imputed ROC curves for the models with and without dipstick urinalysis. Model calibration slopes were 0.832 (> 0.8 indicates a model with good calibration).

TABLE 28 - Clean-catch samples: ‘full’ coefficient models based on symptoms and signs and on symptoms, signs and dipstick results, including results based on MI

MI, multiple imputation; ref., reference.

Missing values coded to modal category.

Internal validation using the bootstrap procedure.

The difference in ROC between symptoms and signs model and symptoms, signs and dipstick model.

Index tests	Symptoms and signs model		Symptoms, signs and dipstick model
	Adjusted OR (95% CI)a	MI adjusted OR (95% CI)	Adjusted OR (95% CI)a	MI adjusted OR (95% CI)
Pain/crying when passing urine
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight problem	1.68 (0.64 to 4.41)	1.87 (0.70 to 4.99)	1.01 (0.30 to 3.45)	1.19 (0.34 to 4.11)
Moderate problem	5.81 (2.58 to 13.10)	6.06 (2.60 to 14.12)	3.26 (1.19 to 8.92)	3.55 (1.25 to 10.04)
Severe problem	21.69 (9.19 to 51.23)	23.79 (9.91 to 57.10)	15.23 (5.17 to 44.89)	16.55 (5.46 to 50.11)
Smelly urine
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight problem	5.08 (2.23 to 11.60)	5.30 (2.24 to 12.55)	3.70 (1.35 to 10.11)	3.99 (1.39 to 11.42)
Moderate problem	6.43 (3.07 to 13.46)	6.56 (2.99 to 14.37)	5.55 (2.33 to 13.23)	5.83 (2.30 to 14.77)
Severe problem	11.85 (4.60 to 30.57)	12.09 (4.58 to 31.90)	5.51 (1.64 to 18.46)	6.00 (1.72 to 20.90)
History of UTI
No	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Yes	3.16 (1.46 to 6.83)	3.07 (1.40 to 6.72)	2.86 (1.15 to 7.11)	2.80 (1.13 to 6.96)
Cough
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight problem	1.33 (0.63 to 2.82)	1.37 (0.64 to 2.93)	1.33 (0.53 to 3.32)	1.37 (0.54 to 3.46)
Moderate problem	1.36 (0.65 to 2.86)	1.45 (0.68 to 3.10)	2.23 (0.94 to 5.29)	2.35 (0.98 to 5.64)
Severe problem	0.23 (0.06 to 0.92)	0.24 (0.06 to 0.97)	0.29 (0.05 to 1.60)	0.29 (0.05 to 1.64)
Clinician global impression of illness severity (0–10)
0–1	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
2	2.19 (0.94 to 5.14)	2.19 (0.92 to 5.17)	2.50 (0.92 to 6.77)	2.49 (0.90 to 6.87)
3	3.09 (1.32 to 7.23)	3.16 (1.32 to 7.55)	3.22 (1.19 to 8.74)	3.19 (1.15 to 8.85)
4–5	4.35 (1.73 to 10.92)	4.73 (1.86 to 12.04)	3.68 (1.22 to 11.12)	4.12 (1.34 to 12.62)
≥ 6	9.23 (2.91 to 29.28)	9.71 (2.98 to 31.61)	8.27 (2.04 to 33.57)	9.11 (2.18 to 37.97)
Abdominal exam: any abdominal pain
No	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Yes	2.76 (1.04 to 7.34)	2.52 (0.95 to 6.72)	1.42 (0.34 to 6.01)	1.22 (0.29 to 5.19)
Ear exam: any acute abnormality
No	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Yes	0.25 (0.08 to 0.81)	0.22 (0.07 to 0.71)	0.40 (0.12 to 1.27)	0.34 (0.10 to 1.10)
Dipstick: leucocytes
Negative			1 (ref.)	1 (ref.)
Trace			2.04 (0.63 to 6.60)	1.99 (0.61 to 6.56)
+			0.66 (0.11 to 3.94)	0.61 (0.10 to 3.92)
++			7.38 (3.04 to 17.92)	7.24 (2.94 to 17.81)
+++			16.78 (5.47 to 51.44)	16.62 (5.20 to 53.13)
Dipstick: nitrites
Negative			1 (ref.)	1 (ref.)
Positive			7.33 (3.09 to 17.40)	7.54 (3.12 to 18.23)
Dipstick: blood
Negative			1 (ref.)	1 (ref.)
Non-haem			0.86 (0.29 to 2.54)	0.86 (0.29 to 2.60)
Haem trace			5.54 (1.43 to 21.53)	5.42 (1.34 to 21.93)
Haem +			3.22 (0.85 to 12.23)	3.51 (0.90 to 13.63)
Haem ++ or +++			1.90 (0.59 to 6.11)	1.94 (0.58 to 6.47)
AUROC (95% CI)	0.892 (0.84 to 0.94)	0.899 (0.85 to 0.95)	0.926 (0.89 to 0.96)	0.933 (0.90 to 0.97)
Validated AUROCb	0.871	0.876	0.904	0.903
Δ ROCc (95% CI)			0.034 (0.01 to 0.06)	0.034 (0.01 to 0.06)
Δ ROCc p-value			0.007	0.009

FIGURE 11.

Clean-catch ROC curve for multiple imputation models for symptoms and signs only (solid line) and symptoms, signs and dipstick (dotted line).

The upper section of Table 29 shows the diagnostic test characteristics corresponding to various choices of cut-point for positivity using the coefficient-based symptoms and signs model, which could be used to identify the children in whom urine collection is warranted. The risk scores for these cut-points can be calculated using the parameters in Table 30. For a sensitivity of 60%, 70% or 80%, respective urine sampling of approximately 5%, 7% or 13% of unwell children aged ≥ 2 years presenting to primary care would be required. At these cut-points, urine samples would be positive for UTI in 27%, 23% and 13% of children, with corresponding specificities of 96.3%, 94.6% and 88.3%. The lower section of Table 29 shows the diagnostic test characteristics for the model based on the symptoms, signs and dipstick test results. If a decision to treat presumptively with antibiotics was made using the 80% sensitivity cut-point, this model would improve the specificity from 88% to 94% and would reduce the overall number of children treated from 13% to 8%, compared with the symptoms and signs only model.

TABLE 29 - Clean-catch diagnostic test characteristics (95% CI) for a range of cut-points, using both the symptoms and signs model (upper part of table) and the symptoms, signs and dipstick model (lower part of table)

NPV, negative predictive value; PPV, positive predictive value.

Results based on models using multiple imputation to deal with missing values.

Risk score (≥)	Sensitivity, %	Specificity, %	PPV, %	NPV, %	Children positive, %
Symptoms and signs model
–0.195	20.0 (11.7 to 32.0)	99.8 (99.5 to 99.9)	66.7 (42.9 to 84.2)	98.2 (97.7 to 98.7)	0.7 (0.4 to 1.0)
–0.87	30.0 (19.8 to 42.7)	99.5 (99.1 to 99.7)	56.3 (39.0 to 72.1)	98.4 (97.9 to 98.9)	1.2 (0.8 to 1.6)
–1.698	40.0 (28.5 to 52.8)	98.2 (97.6 to 98.6)	32.9 (23.1 to 44.4)	98.7 (98.1 to 99.0)	2.7 (2.1 to 3.3)
–1.98	50.0 (37.6 to 62.4)	97.5 (96.9 to 98.1)	31.3 (22.8 to 41.2)	98.9 (98.4 to 99.2)	3.5 (2.9 to 4.3)
–2.34	60.0 (47.2 to 71.5)	96.3 (95.5 to 97.0)	26.7 (19.9 to 34.7)	99.1 (98.6 to 99.4)	4.9 (4.2 to 5.8)
–2.75	70.0 (57.3 to 80.2)	94.6 (93.7 to 95.4)	22.6 (17.1 to 29.1)	99.3 (98.9 to 99.6)	6.8 (5.9 to 7.8)
–3.515	80.0 (68.0 to 88.3)	88.3 (87.0 to 89.4)	13.3 (10.1 to 17.2)	99.5 (99.1 to 99.7)	13.2 (12.0 to 14.5)
–3.884	85.0 (73.6 to 92.0)	83.9 (82.4 to 85.2)	10.6 (8.1 to 13.6)	99.6 (99.2 to 99.8)	17.6 (16.2 to 19.1)
–4.86	93.3 (83.5 to 97.5)	61.0 (59.1 to 62.8)	5.1 (3.9 to 6.6)	99.8 (99.4 to 99.9)	40.2 (38.4 to 42.0)
–5.7	96.7 (87.6 to 99.2)	37.8 (35.9 to 39.6)	3.4 (2.6 to 4.3)	99.8 (99.2 to 100.0)	63.0 (61.2 to 64.8)
–6.664	100	15.7 (14.4 to 17.1)	2.6 (2.0 to 3.3)	100	84.6 (83.2 to 85.9)
Symptoms, signs and dipstick model
1.43	20.0 (11.7 to 32.0)	99.9 (99.7 to 100.0)	85.7 (57.3 to 96.4)	98.2 (97.7 to 98.7)	0.5 (0.3 to 0.9)
0.321	40.0 (28.5 to 52.8)	99.9 (99.7 to 100.0)	88.9 (70.7 to 96.4)	98.7 (98.2 to 99.0)	1.0 (0.7 to 1.4)
–1.15	60.0 (47.2 to 71.5)	99.3 (98.8 to 99.5)	64.3 (51.0 to 75.7)	99.1 (98.7 to 99.4)	2.0 (1.6 to 2.6)
–3.275	80.0 (68.0 to 88.3)	93.8 (92.9 to 94.7)	22.5 (17.4 to 28.6)	99.5 (99.2 to 99.7)	7.8 (6.8 to 8.8)
–3.98	83.3 (71.7 to 90.8)	88.3 (87.0 to 89.5)	13.8 (10.6 to 17.7)	99.6 (99.2 to 99.8)	13.2 (12.0 to 14.6)
–5.237	96.7 (87.6 to 99.2)	66.3 (64.5 to 68.1)	6.0 (4.7 to 7.7)	99.9 (99.6 to 100.0)	35.0 (33.3 to 36.8)
–5.825	98.3 (89.1 to 99.8)	53.1 (51.2 to 54.9)	4.5 (3.5 to 5.7)	99.9 (99.5 to 100.0)	48.1 (46.2 to 49.9)
–6.69	100	29.5 (27.8 to 31.2)	3.1 (2.4 to 3.9)	100	71.2 (69.4 to 72.8)

TABLE 30 - Parameters to calculate risk score for the clean-catch ‘full’ coefficient-based models

0 = completely well, 10 = extremely unwell.

Variables	Coefficients
	Symptoms and signs model	Symptoms, signs and dipstick model
Intercept	–6.0152	–7.0011
Pain/crying when passing urine
No problem	0	0
Slight problem	0.6263	0.1734
Moderate problem	1.8009	1.2658
Severe problem	3.1693	2.8062
Smelly urine
No problem	0	0
Slight problem	1.6683	1.3838
Moderate problem	1.8805	1.7631
Severe problem	2.4923	1.7924
History of UTI
No	0	0
Yes	1.1227	1.0299
Cough
No problem	0	0
Slight problem	0.3156	0.3118
Moderate problem	0.3736	0.8541
Severe problem	–1.4324	–1.2293
Clinician global impression of illness severity (0–10)a
0–1	0	0
2	0.7818	0.9103
3	1.1509	1.1613
4–5	1.5543	1.4148
≥ 6	2.2732	2.2090
Abdominal exam: any abdominal pain
No	0	0
Yes	0.9244	0.1987
Ear exam: any acute abnormality
No	0	0
Yes	–1.5122	–1.0919
Dipstick: leucocytes
Negative		0
Trace		0.6904
+		–0.4912
++		1.9791
+++		2.8103
Dipstick: nitrites
Negative		0
Positive		2.0204
Dipstick: blood
Negative		0
Non-haem		–0.1465
Haem trace		1.6902
Haem +		1.2556
Haem ++ or +++		0.6638

Clean-catch samples were available for 88, 91 and 612 children aged < 12, 12–23 and 24–35 months, with UTI diagnosed in 4, 2 and 16 of these children, respectively. We acknowledge that the small number of UTIs may have led to results from multiple imputation analyses being unstable and estimated AUROC values being overoptimistic.

Nappy pad models

Table 38 shows that 1.3% of children providing nappy pad samples met the criteria for a microbiological diagnosis of UTI, of whom 82% were < 2 years and 48% were female, and that 92% of samples were provided within 24 hours of index test measurement. Clinicians correctly identified 13% of the 30 UTIs.

‘Full’ coefficient-based model

Table 39 shows adjusted ORs for the index tests (footnote marker d in Table 38) selected for the nappy pad model, for children predominantly < 2 years. Parent-reported smelly urine, darker urine, female sex and the absence of a nappy rash were independently associated with UTI; for the first two of these there was evidence of graded associations. No clinical examination findings were independently associated with UTI in this age group. The presence of leucocytes and nitrites from dipstick urine testing were independently associated with UTI, though more modestly than in clean-catch samples. The increase in AUROC between the symptoms and signs model and the symptoms, signs and dipstick model was about 0.065 in the multiple imputation analysis. However, with a p-value of 0.035 there was only slight evidence to suggest an increase in the AUROC when adding the dipsticks to the symptoms and signs model. The symptoms and signs model had reasonable diagnostic accuracy (validated AUROC for the multiple imputed model was 0.778) and diagnostic accuracy increased (validated AUROC 0.821) with addition of dipstick findings. For comparison, the AUROC for ‘clinician diagnosis’ was 0.626 (95% CI 0.532 to 0.719).

TABLE 39 - Nappy pad samples: models based on symptoms and signs, and on symptoms, signs and dipstick results, including results based on multiple imputation

MI, multiple imputation; ref., reference.

Missing values coded to modal category.

Internal validation using the bootstrap procedure.

The difference in ROC between the symptoms and signs model and the symptoms, signs and dipstick model.

Index tests	Symptoms and signs model		Symptoms, signs and dipstick model
	Adjusted OR (95% CI)a	MI adjusted OR (95% CI)	Adjusted OR (95% CI)a	MI adjusted OR (95% CI)
Gender
Male	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Female	2.61 (1.16 to 5.83)	2.45 (1.08 to 5.55)	1.71 (0.71 to 4.08)	1.70 (0.70 to 4.11)
Smelly urine
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight problem	1.99 (0.64 to 6.17)	2.47 (0.77 to 7.91)	1.76 (0.54 to 5.78)	2.28 (0.68 to 7.68)
Moderate problem	3.75 (1.21 to 11.65)	5.11 (1.66 to 15.76)	3.26 (0.98 to 10.87)	4.64 (1.39 to 15.53)
Severe problem	17.81 (4.82 to 65.86)	22.04 (6.05 to 80.25)	8.42 (2.03 to 34.88)	12.94 (3.20 to 52.30)
Darker urine
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight problem	2.51 (0.55 to 11.49)	2.50 (0.56 to 11.22)	2.50 (0.51 to 12.30)	2.52 (0.52 to 12.21)
Moderate or severe problem	3.66 (0.97 to 13.82)	2.98 (0.81 to 10.96)	3.59 (0.90 to 14.42)	3.19 (0.82 to 12.37)
Nappy rash
No problem	1 (ref.)	1 (ref.)	1 (ref.)	1 (ref.)
Slight to severe problem	0.07 (0.01 to 0.55)	0.07 (0.01 to 0.52)	0.08 (0.01 to 0.59)	0.07 (0.01 to 0.56)
Dipstick: leucocytes
Negative			1 (ref.)	1 (ref.)
Trace			0.80 (0.10 to 6.37)	0.74 (0.09 to 6.13)
+			3.06 (0.88 to 10.68)	3.01 (0.83 to 10.91)
++			2.13 (0.62 to 7.30)	2.25 (0.65 to 7.84)
+++			6.23 (2.20 to 17.67)	5.53 (1.90 to 16.10)
Dipstick: nitrites
Negative			1 (ref.)	1 (ref.)
Positive			5.93 (2.72 to 12.93)	6.35 (2.86 to14.12)
ROC (95% CI)	0.769 (0.68 to 0.85)	0.805 (0.72 to 0.89)	0.858 (0.79 to 0.93)	0.870 (0.80 to 0.94)
Validated ROCb	0.744	0.778	0.799	0.821
Δ ROCc (95% CI)			0.089 (0.02 to 0.16)	0.065 (0.00 to 0.13)
Δ ROCc p-value			0.012	0.036
Calibration slopec	0.695	0.749	0.647	0.708

Figure 12 shows the multiple imputed ROC curves for the models with and without dipstick urinalysis. The calibration slopes being around 0.7 shows that we have a large degree of overoptimism in our nappy pad models.

FIGURE 12.

Nappy pad ROC curve from multiple imputation for symptoms and signs only (solid line) and symptoms, signs and dipstick (dotted line).

The upper section of Table 36 shows the diagnostic test characteristics corresponding to various choices of cut-point for positivity using the symptoms and signs model, which could be used to identify the children in whom urine collection is warranted. The risk scores for these cut-points can be calculated using the parameters in Table 40. For a sensitivity of 50%, 63% or 90%, respective urine sampling of approximately 9%, 20% or 43% of unwell children aged < 2 years presenting to primary care would be required. At these cut-points, urine samples would be positive for UTI in 8%, 4% and 3% of children, with corresponding specificities of 92%, 81% and 58%. The lower section of Table 36 shows the diagnostic test characteristics for the model based on the symptoms, signs and dipstick test results. If a decision to treat presumptively with antibiotics was made using the (only comparable) 63% sensitivity cut-point, this model would improve the specificity from 81% to 91% and would reduce the overall number of children treated from 20% to 10%, compared with the symptoms and signs only model. However, as discussed in Chapter 6, the cost-effectiveness of this strategy is not clear.

TABLE 40 - Parameters to calculate risk score for the nappy pad ‘full’ coefficient-based models

Index tests	Coefficients
	Symptoms and signs model	Symptoms, signs and dipstick model
Intercept	–5.2709	–6.0041
Gender
Male	0	0
Female	0.8972	0.5293
Smelly urine
No problem	0	0
Slight problem	0.9042	0.8251
Moderate problem	1.6310	1.5343
Severe problem	3.0927	2.5601
Darker urine
No problem	0	0
Slight problem	0.9166	0.9243
Moderate or severe problem	1.0913	1.1598
Nappy rash
No problem	0	0
Slight to severe problem	–2.6792	–2.6098
Dipstick: leucocytes
Negative		0
Trace		–0.2988
+		1.1029
++		0.8105
+++		1.7095
Dipstick: nitrites
Negative		0
Positive		1.8488