Main

The 13C-urea breath test is a well-established method for the diagnosis of gastric Helicobacter pylori colonization(1), and its noninvasive nature makes it ideal for use in all populations at all ages. It has been used in epidemiological studies of infants and children in developing countries(2,3), to suggest that H. pylori colonization may occur very early in life. We have recently described in a prospective study in The Gambia that colonization is predominantly acquired in infancy, and by 36 mo the majority of children are colonized with H. pylori(4). A concern about these studies is that the urea breath test has not been validated in young children from a developing country, and their results must therefore be interpreted with caution.

The 13C-urea breath test detects gastric urease activity caused by H. pylori, by measuring 13 enrichment in expired breath after oral ingestion of 13C labeled urea. To distinguish positive from negative results, a cut-off value for 13 enrichment must be defined at a specified time point after ingestion of substrate. This cut-off value is generally determined by empirical comparison with endoscopic diagnosis of H. pylori colonization (usually regarded as the gold standard diagnostic test) in a relevant population. Although validation studies have shown that the 13C-urea breath test has a high sensitivity and specificity among children from the developed world(5–9), there is an absence of such information from developing countries, where H. pylori colonization is much more common.

Recently it has been appreciated that the distribution of results for any set of urea breath tests contains two sub-populations, corresponding to the presence or absence of H. pylori in gastric biopsies(10,11). The aim of this study was to propose a method for analyzing a large population of breath tests that identifies and describes these two sub-populations and thereby to determine an appropriate cut-off value to distinguish H. pylori colonized from noncolonized children, and to illustrate the appropriateness of this value in young children from a developing country by comparing urea breath test results with gastric biopsy findings.

EXPERIMENTAL METHODS AND SUBJECTS

13C-urea breath tests were performed over a 2-year period, at intervals of at least 3 mo, on 1532 occasions in 247 infants and children aged from 3 to 48 mo, from a community of rural subsistence farmers in The Gambia, as part of a epidemiological study of the prevalence of H. pylori in early life and its influence upon growth(4,12). Full details of subject selection and other aspects of the study are described elsewhere(4,12). Study protocols were approved by the Joint MRC and Gambian Government Ethical Committee, and by a meeting of village elders. Children were only entered into the study after informed parental consent had been obtained by fieldworkers fluent in the local language.

A baseline breath sample was obtained from children shortly after awakening, using a face mask and expired breath reservoir(13). Immediately thereafter, a drink containing 50 mg 13C-urea (99 atom % excess, Cambridge Isotopes, Cambridge, MA), 50 mg naturally abundant urea, 0.5 mg/kg lactulose elixir (B.P.), and a test meal of 5 g of glucose polymer (Polycose, Abbott Laboratories, Queensborough, Kent, UK) dissolved in an appropriate amount of water was given. Further breath samples were collected 5, 15, 30, and 60 min later.

Breath samples were stored in 20-mL vacutainers (Becton Dickinson, Cowley, Oxford, UK) and transported to Cambridge, UK. Carbon dioxide was cryogenically separated from the samples using an automated system, and analyzed by gas isotope-ratio mass spectroscopy (SIRA 10, VG Isotech, UK). Isotopic enrichment was expressed as Craig corrected(14) delta (‰) relative to the international standard PDB, using the formula (Equation 1), where R is the isotopic ratio (13C:12C) of the sample or standard. The baseline-corrected 13C enrichment was defined as the difference in isotopic enrichment between the baseline breath sample and those at time points thereafter. Breath hydrogen concentrations were measured electrochemically.

Initial assessment of 13C-urea breath test data was undertaken before and after log transformation using Kernel estimates of the density function(15) to produce smoothed histograms, which were used to obtain a visual impression of the existence of separate populations within the data set. After estimation of the number of such sub-populations, the means and variances of the separate sub-populations in the cohort at each age and within the overall dataset were estimated by a Genstat procedure using the expectation maximization algorithm(16). These values were then used to define the optimum cut-off value that discriminated among sub-populations, and to calculate the proportion of results that would have been correctly assigned to each sub-population by using this calculated cut-off value within the entire dataset.

To illustrate the appropriateness of the cut-off value, another group of Gambian children was studied. Upper gastrointestinal endoscopy was performed on 14 children aged 6 to 28 mo (mean 16 mo) with persistent diarrhea and malnutrition, to obtain a diagnostic small bowel biopsy. None of the children was undergoing treatment with antimicrobial agents at the time of or immediately before the procedure. Four gastric biopsies were also obtained, two from separate sites in the antrum, one from the body, and one from the incisura. They were fixed in buffered formalin, and transported to Oxford, UK, where they were examined for the presence of H. pylori. Sections from each biopsy were stained with hematoxylin and eosin, toluidine blue, Alcian blue, and periodic acid-Schiff. H. pylori were identified by morphology, and a specific antibody stain (DAKO kit, Glostrup, Denmark) was used to verify the presence or absence of H. pylori in 10 cases. 13C-urea breath tests were performed using the method described above on each of these 14 children on the day of endoscopy.

RESULTS

The baseline-corrected enrichment of 13C in expired breath at each time point (5, 15, 30, and 60 min after substrate) was transformed (natural logarithm) and smoothed histograms summarizing all 1532 breath tests were produced, to allow visual inspection of the distribution of results. This demonstrated that at each time point, two sub-populations could be clearly detected. The time point that produced the most discrete sub-populations of baseline-corrected 13C enrichment in expired breath was 30 min after ingestion of substrate. A smoothed histogram of results at this time point is shown in Figure 1, illustrating the two sub-populations.

Figure 1
figure 1

Smoothed histogram produced by Kernel estimate of the density function of baseline 13C enrichment in expired breath 30 min after ingesting substrate for 13C urea breath test results among Gambian infants, revealing the existence of two sub-populations of results. True values of baseline-corrected 13C enrichment relative to PDB are given on an ln-x axis. Total number of tests included = 1532.

Full size image

Means (SD) of baseline-corrected 13C enrichment at 30 min for both sub-populations were 2.77 (1.67) and 24.12 (16.86) before transformation, and 1.05 (0.767) and 3.119 (0.573) after natural log transformation. These mean and SD values remained consistent throughout the cohort at each sampling age (Table 1), and were used to calculate a cut-off value of 5.47 δ ‰ relative to PDB baseline-corrected 13C enrichment at 30 min, that identified 95% of the normally distributed negative sub-population and 99.4% of the log normal distribution for the positive sub-population. This cut-off value remained appropriate at all ages studied, from 3 to 48 mo (Table 1 and Figure 2). Figure 2 also allows a visual impression of the relative proportion of children at each age with either positive or negative test results. Although children remained within one of the two sub-populations thus defined at each age division, the proportion of children within the positive sub-population increased with age. Only 2.6% of all results fell between 5 and 6 δ ‰ relative to PDB baseline-corrected 13C enrichment at 30 min.

Table 1 Mean values for the negative and positive sub-populations among the cohort of children at successive ages throughout the study
Full size table
Figure 2
figure 2

Sequential histograms of 13C urea breath test results among Gambian infants and children. The y axis shows the overall average values followed by values for successive age categories running from age 3 mo to age 45 mo in 3 monthly increments, the x axis shows baseline-corrected enrichment of 13C in expired breath (δ ‰ relative to PDB) 30 min after ingestion of substrate. The height of the bars shows proportion of children of each age that fell into each category of baseline-corrected enrichment of 13C in expired breath (δ ‰ relative to PDB) 30 min after ingestion of substrate. The unfilled bars show average data over all ages. All 1532 breath test results are included.

Full size image

Many subjects showed a transient rise in baseline-corrected 13C enrichment in expired breath five minutes after ingestion of substrate that was not maintained at other time points. Simultaneous measurement of breath hydrogen concentration showed no significant elevation of breath hydrogen levels in the first 30 min, but by 60 min, elevation of breath hydrogen greater than 20 ppm above baseline levels was detected in 133/1532 tests (8.7%).

Table 2 summarizes the characteristics of the 14 children who underwent both upper endoscopy and 13C urea breath tests. In 7/14 cases H. pylori was identified upon gastric mucosal biopsies. All seven children without H. pylori on their mucosal biopsies had negative urea breath tests, and 6/7 children with H. pylori in their stomachs had positive breath tests (Table 2). As only 14 children underwent endoscopy, meaningful values for the sensitivity and specificity of the urea breath test could not be calculated.

Table 2 13C-urea breath test results and histology from 14 Gambian children aged 6 to 28 mo who underwent gastroscopy
Full size table

DISCUSSION

We have described a method for analyzing 13C-urea breath test data that identifies two sub-populations within the overall data set, and have demonstrated how this technique can be used to define an appropriate cut-off value to distinguish positive from negative results. We have illustrated the appropriateness of our cut-off value in this population, by comparing the 13C-urea breath test with demonstration of H. pylori on gastric biopsies in a group of children of similar age from the same population.

Mion et al. described a similar method of analyzing 13C-urea breath test data among adult subjects, but their approach did not acknowledge the normal distribution of the negative sub-population before log transformation(10). While this does not significantly affect results among most series of adult subjects, in which the two sub-populations are effectively discrete, failing to analyze the negative sub-population before transformation in a data set such as ours makes it almost impossible to define a suitable cut-off value. This appears to be because our negative sub-population of Gambian children had a considerably broader distribution than that usually measured for European adults.

In many stable isotope breath tests, calculating the percentage of ingested dose of isotope that is recovered [percentage dose recovered (PDR)] can provide additional valuable information. This requires either the exact measurement of CO2 production or its estimation from weight and height(17). Calculation of PDR does not improve the discriminatory information provided by measurement of the baseline-corrected 13C enrichment in expired breath 30 min after substrate for the detection of H. pylori colonization in early childhood. In any case, when the substrate is given in sufficient excess to saturate the enzyme activity, PDR is inappropriate because it will apparently decrease with increasing substrate excess.

Examining smoothed histograms of our data (Figure 1) clearly reveals the existence of two sub-populations of results. These represent negative and positive populations with respect to urease activity. The normal distribution of values in the negative sub-population may be accounted for by baseline drift in 13C enrichment of expired breath over time. The positive sub-population, which is the result of urease activity superimposed upon the negative population, is positively skewed, but log transformation renders it adequately normal for the purposes of data analysis. The most appropriate time point for sample collection was 30 min after administration of substrate. Samples collected earlier frequently showed evidence of an early rise in 13C breath enrichment that was not sustained, and was probably associated with oral bacterial urease activity. Samples from 60 min were associated with a concomitant rise in breath hydrogen levels in 8.7% of cases, suggesting that the substrate had reached fermentative bacteria in the contaminated small or normal large bowel. Such bacterial populations could also be a source of urease activity.

The value of 5.47 δ ‰ relative to PDB above baseline at 30 min, which we found to be the best discriminator between positive and negative populations, was similar to published empirically-derived cut-off values (range 3.5 to 6 δ ‰). All are associated with high predictive values for the diagnosis of H. pylori associated gastritis(1,6–9,18) Within our data set, only 2.6% of baseline enrichment results fell between 5 and 6 δ ‰, so that in practice the selection of any arbitrary cut-off value within this range would not have significantly affected the results. The cut-off of 5.47 ‰ relative to PDB above baseline at 30 min is relatively high compared with other series that have used a detailed data set analysis approach(5,10,11). When we compared the urea breath test results with evidence of H. pylori colonization made at upper endoscopy, we found no false-positive results, and one false-negative result. Although the small number of children who underwent endoscopy does mean that we can only suggest, rather than prove, that our cut-off is appropriate, lowering the cut-off value to one used in developed countries(5) would have produced false-positive test results.

Among children in the longitudinal cohort study, our cut-off value remained appropriate at all ages. Although more children became breath test positive with increasing age, Figure 2 shows that this was brought about by children moving from the negative to the positive sub-populations, and that results that lay close to the cut-off were infrequent at all ages.

Most studies suggest that H. pylori colonization begins in childhood. In The Gambia, most children are colonized in infancy(4), and our previous work has shown that H. pylori colonization in early infancy is associated with growth faltering(12,19). Defining a cut-off for the urea breath test appropriate for the study of young children in the developing world is therefore vital if this noinvasive technique is to be used to study naturally occurring. H. pylori colonization and assess its potential impact on nutrition and growth. In addition, our method for deriving a binary value from continuous data by identifying two sub-populations within a dataset is applicable to other similar measurements.