It has long been assumed that the risk of a first-time childhood UTI progressing to long-term kidney damage is significant. In investigating the relationship between UTI and long-term damage, we are primarily concerned with established renal failure (ERF), as the relationship between UTI and other potential morbidities is ambiguous and, in most cases, not measurable. Whether kidney damage results from VUR alone or in combination with UTI remains uncertain.

Stark^{217} has argued that the number of patients who have a single UTI in childhood who then go on to have ERF is small, and that the risk of ERF following a UTI is low. From this position, he argues that the investigations undertaken to diagnose VUR and kidney scarring in children who have experienced a first-time UTI are unwarranted. He estimates that between 10 000 and 15 000 girls would need to be investigated to prevent a single case of ERF.

## Estimating risk

The assumptions used in the model developed by Stark result in a much lower risk that a first-time UTI in childhood will lead to ERF than was previously assumed. Accepting the risk presented by Stark would lead to a significant change in clinical practice in the NHS and it is important that such a change in practice be supported by robust evidence. The question that must be addressed is whether we can identify with confidence the true level of risk that a patient with a first-time UTI will develop ERF as a direct result of that infection.

In order to examine whether this is the case, a model was developed using the assumptions made by Stark to assess the risk of a first UTI in childhood leading to ERF at any time. The model is represented graphically by the Venn diagram in .

Venn diagram showing associations between UTI and ERF.

There is a population of patients with UTI (b+d+e+g), a population of patients with ERF (a+b+c+d) and a population of patients with pyelonephritic scarring/reflux nephropathy (b+c+e+f). The proportion of patients in which we are interested is (b+d) / (b+d+e+g) − that is, the risk of a patient developing ERF given that they have had a UTI. Using the figures given by Stark, the risk is calculated as about 1/10 000, where:

Lifetime risk of UTI (b+d+e+g) = 80 000 per million population (pmp)

Incidence of ERF (a+b+c+d) = 87 pmp

ERF attributable to pyelonephritic scarring/reflux nephropathy (b+c) = 9% (0.09)

To calculate the risk, Stark makes two key assumptions. Firstly, that in all cases of ERF attributable to pyelonephritic scarring/reflux nephropathy the patient has had a UTI (c = 0), and secondly, that pyelonephritic scarring/reflux nephropathy is the only mechanism by which UTI can lead to ERF (d = 0). The proportion of patients with pyelonephritic scarring/reflux nephropathy that do not go on to develop ERF (e+f) is not of importance. The proportion of ERF attributable to pyelonephritic scarring/reflux nephropathy, where c = 0, is

(b) x (a+b+c+d) = 0.09 x 87 pmp = 8 pmp

Therefore the risk of UTI leading to ERF, where d = 0 is:

(b) / (b+e+g) = 8 pmp / 0.08 = 100 pmp = 1/10 000

During the development of the model, questions were raised about a number of the crucial assumptions made by Stark, primarily the estimate of ERF incidence and the links between pyelonephritic scarring/reflux nephropathy and ERF and between UTI and pyelonephritic scarring/reflux nephropathy. These assumptions and their implications for the results of the model are examined below.

## Estimating the incidence of ERF

Stark assumes that the statistical risk of a person developing ERF in their lifetime will be very close to the mean incidence of ERF during that person’s lifetime. Rate of acceptance for renal replacement therapy (RRT) is used as proxy for the rate of ERF in the absence of accurate data on the number of people that develop ERF. In the initial analysis undertaken for the guideline, this assumption was not challenged. However, the estimate of 87 pmp used by Stark for incidence was an estimate of annual incidence and reflects not the likelihood of an individual developing ERF during their lifetime but the likelihood of them developing ERF in a given year. Annual incidence has been used where cumulative incidence was the appropriate measure.

In the absence of a reliable published estimate of the true lifetime risk of developing ERF, a table was constructed to model a cohort of 1 000 000 patients to determine the number that would develop ERF in their lifetime. Estimates of risk shown in are reported as annual incidence per million population for females and applied to the proportion of the cohort at risk (those who were alive and who had not already developed ERF). It is worth noting that males have a greater lifetime risk of ERF than females. The lifetime risk for males younger than 60 years is nearly 3800 pmp compared with about 2500 pmp in females. Data for females is used in this analysis to allow comparison with the previous estimate by Stark, but simply substituting the data for males, or for the total population, into the above model will alter the estimate of risk accordingly.

Cumulative incidence of ERF: females (source: EDTA).

shows how the lifetime estimate of developing ERF was calculated, using data for females from the European Dialysis and Transplant Association (EDTA). The estimated lifetime risk of developing ERF from this calculation is nearly 6000 pmp. This represents a much greater risk of developing ERF than that presented by Stark, with significant implications for the model. Substituting the whole lifetime estimate based on the data from the EDTA into the formula (a+b+c+d), lifetime risk for females developing ERF as a result of having had a childhood UTI is estimated at about 1/155. The age group of interest is represented by those patients younger than 60 years, as it is believed that ERF that occurs after this age is unlikely to be attributable to a childhood UTI. In this group, the risk of developing ERF is approximately 2500 pmp. For that group of patients younger than 60 years, the risk of UTI leading to ERF is about 1/355. Estimates of risk then range from 1/155 to 1/10 000, and the considerable uncertainty in other model parameters must also be explored to illustrate why no reliable estimate of risk can be achieved based on the available data.

## The causal relationship between UTI and ERF

Uncertainty in two other key assumptions in the analyses by Stark and the GDG call into question the strength of the relationship between UTI and ERF and need to be addressed. These are the relationship between pyelonephritic scarring/reflux nephropathy and UTI, and the proportion of ERF that can be attributed to pyelonephritic scarring/reflux nephropathy. These are addressed below.

### UTI and pyelonephritic scarring/reflux nephropathy

In all of the analyses presented to date it is assumed, in those cases of ERF where pyelonephritic scarring/reflux nephropathy is believed to be the cause, that all patients have also had a UTI. However, no evidence has been presented in support of this assumption. In the analysis by Stark, this assumption is not explicitly stated, although it is evident from the results. In making this assumption, the risk of a UTI leading to ERF is overestimated – in fact, if the converse is true and no patients with ERF caused by pyelonephritic scarring/reflux nephropathy had a UTI in childhood (unlikely though it is), then the risk of UTI leading to ERF is non-existent. It is not possible, based on current evidence, to estimate the true proportion of patients in whom ERF is attributed to pyelonephritic scarring/reflux nephropathy and who have had a UTI in childhood.

In the absence of a reliable estimate of this relationship, it is not possible to make a reliable estimate of overall risk. This can be illustrated using the lifetime risk data in the above table. If the proportion of pyelonephritic scarring/reflux nephropathy that is associated with UTI is assumed to be 0 the risk of UTI leading to ERF is equal to 0, although when the proportion of pyelonephritic scarring/reflux nephropathy that is associated with UTI is assumed to be 1 the risk that UTI will lead to ERF in females is 1/155. The range of risk estimates generated is so great, that in the absence of accurate data on the link between pyelonephritic scarring/reflux nephropathy and UTI, no conclusions can be drawn about the true risk of UTI leading to ERF.

### Pyelonephritic scarring/reflux nephropathy and ERF

In addition to the uncertainty around the link between UTI and pyelonephritic scarring/reflux nephropathy, there is also uncertainty over the proportion of ERF that can be attributed to pyelonephritic scarring/reflux nephropathy. Stark assumes this rate is 9%, using an approximate average of published estimates that are based on data from various renal registers, including European and the US ones. In many cases renal registry data is not classified in such a way that a reliable estimate of those cases of ERF attributable to pyelonephritic scarring/reflux nephropathy can be made. In contrast to the estimate of 9% assumed by Stark, the current proportion of ERF that is attributed to pyelonephritic scarring/reflux nephritis in the United States Renal Data System 2003 report is 0.46 per cent, or roughly one in every 200 cases of ERF. The EDTA estimates that 8% of all cases of ERF in England and Wales can be attributed to a more generic classification of pyelonephritis. It is not clear what proportion of this is pyelonephritic scarring/reflux nephropathy. The wide range of estimates for the likelihood of pyelonephritic scarring/reflux nephropathy being attributable as the cause of ERF introduces further uncertainty in the model.

This uncertainty further decreases the reliability of any estimate of the risk of UTI leading to ERF. Again, using lifetime risk data in the above table, if it is assumed that 8% of ERF cases are attributable to pyelonephritic scarring/reflux nephropathy, the risk of UTI leading to ERF is 1/155. When 0.50% of cases of ERF are attributed to pyelonephritic scarring/reflux nephropathy, then the risk is approximately 1/2800. Once again, the range is sufficiently wide to prevent a reliable estimate from being made based solely on the currently available data.

## Implications

Given the degree of uncertainty around the key assumptions and data used by Stark, and in turn by the GDG, no reliable estimate of the risk of UTI leading to ERF can be calculated. It is not clear what the true rate of ERF caused by pyelonephritic scarring/reflux nephropathy is, nor is it clear what proportion of these cases have had a UTI in childhood. Without reliable estimates of these figures, as well as of lifetime risk, the level of uncertainty in the model is such that no reliable conclusions can be drawn based on the data alone.