13C-Urea breath test threshold calculation and evaluation for the detection of Helicobacter pylori infection in children

Background The 13C-urea breath test (UBT) is performed in adults and children with epigastric pain for non-invasively diagnosing a suspected H. pylori infection. Criteria for UBT interpretation have not been generally agreed on and test reliability has not been established in children of different ages. This study aimed at identifying reliable UBT thresholds in children by using 251 UBTs in conjunction with reference histology and by analyzing 1232 UBTs. Methods At baseline and 30 and 60 minutes after the administration of 75 mg 13C-urea to children and adolescents (0.25 to 18 years of age), the differences (Δ) of 13CO2/12CO2 ratio in exhaled air (δ) were determined by mass spectrometry. UBT Δδ value thresholds were calculated in random subgroups and evaluated in complementary subgroups using logistic regressions on reference histology or bimodal distribution analyses of Δδ values from UBTs alone. Results Δδ values were higher (median, 15.4‰) in positive (133/251, 53 %) than in negative histology (2.4‰). At 30 minutes, the calculated cut-off was 5.3‰ (mean regression determination R2 = 0.91), and sensitivity (0.95), specificity (0.97), positive (0.97) and negative predictive values (0.95) were higher than at 60 minutes (threshold 6.8‰, R2 = 0.85). Similar thresholds resulted from UBTs analysis (5.8‰ and 6.2‰) when sensitivity and specificity were maximized (concordance probabilities, 0.99 and 0.99). There was no systematic age effect. Conclusions In children, 13C UBT cut-offs were obtained and specially validated, entailing high accuracy of non-invasively testing for gastric H. pylori infection.


Background
Painful chronic gastritis and duodenal ulcerations in children are mostly associated with an infection of the gastric mucosa by H. pylori (for reviews see [1,2]). Invasive tests for the detection of H. pylori are usually performed in biopsy specimens of the stomach (histological staining, bacterial culture and the rapid urease test) or in the blood (anti-HP antibodies). The use of these tests in adults and children is well established (for reviews see [3][4][5]). Since the introduction of the non-invasive 13 C-urea breath test (UBT) by Graham et al. 1987 [6], several attempts have been made to define the minimum criteria for a positive UBT in adults [4,[7][8][9][10][11][12].
For general use in children, however, criteria for the performance of the UBT and the interpretation of its results are not yet sufficiently defined. Feasibility and promising precision of the UBT had been shown in conjunction with invasive tests, for example, in studies including 50 [13], 88 [14], and, recently, 115 children [15], in whom a 98% sensitivity and specificity at a rather low cut-off (∆δ = 3.5‰)was asserted.
Another recent study in 247 Gambian children up to four years reported a cut-off of 5.47‰ for H. pylori infection of in 1532 UBTs at 30 minutes, validating with histology results in only 14 cases [16]. Their method was similar to our secondary aim's method, which was to calculate a threshold from 1232 UBTs alone in children and adolescents from 1 to 18 years of age. Primarily, our study aimed at calculating UBT thresholds from reference histology in 251 children from 5 to 16 years of age, validating with a prospective method. We additionally demonstrate that the thresholds produced by both methods are in close agreement.

C-urea breath test
The 13 C urea breath test (UBT) was performed as described previously [17]. In short, after at least 4 hours of fasting, 75 mg 13 C-urea (99% 13 C; Promochem, Wesel, Germany) was given in 50 ml water shortly after a meal, which usually consisted of half a non-sweetened roll. Breath samples were collected in duplicate before ingestion of the 13 C-urea (t 0 ) and at 30 and 60 minutes afterwards (t i ). If a child could not expire into the collection bag, breath samples were collected by applying a breath mask with two unidirectional valves connected to a breath bag. Normal free breathing activity filled the bag, from which a sample was drawn and transferred to an evacuated tube [17].
Breath samples were analyzed twice by gas isotope ratio mass spectrometry (Delta E; Finnigan MAT, Bremen, Germany). The measured 13 CO 2 / 12 CO 2 ratio was compared with the known isotope ratio of a standard gas. The 13 CO 2 values were corrected for the international PDB-standard, then averaged and expressed as relative delta per mil difference to the standard gas (δ [‰]). In the following, ∆δ [‰] is used to denote the difference between δ at time t i and at baseline.
Subjects 251 children and adolescents (135 boys and 116 girls; aged 5-16 years, median 11 years) who had not received antibiotics during the previous two months were included because of recurrent epigastric pain for at least 3 months. (Children younger than four years almost never presented with symptoms indicating esophagogastroduodenoscopy.) Esophagogastroduodenoscopy, UBT, and histological examinations were performed by three different investigators, each unaware of the others' results. The histological identification of H. pylori in four biopsy specimens from the gastric antrum and corpus was used as reference. At least one biopsy specimen was required to be positive. The pathologist performed histological identification according to established staining methods and criteria [18] and was blinded to all specimens.
Additionally, a total of 1232 UBTs were performed (between January 1st 1995 and June 1st 1998) in 1232 German children with chronic abdominal pain (1 month to 18 years of age). Informed consent was given by the parents and children, where obtainable.
The study was carried out in accordance with the Helsinki Declaration for Human Studies and the protocol was approved by the local ethics committee.

Calculations
Analyses and graphics were produced with S-Plus4.5 (MathSoft, Seattle) and SAS6.12 (SAS Institute, Cary). Robust statistics included the median and the rescaled median absolute deviation, which compare to mean and standard deviation. Natural logarithms of ∆δ values were used throughout. In histograms, the bin width was chosen by data-based kernel smoothing [19] and the bin height was scaled to a unit area under the curve [ [20], p.168].
Logistic regressions of the histological detection of H. pylori (positive or negative) on the ∆δ values were performed in randomly chosen 50% of the data. The resulting threshold ∆δ values were evaluated in the complementary 50% of the data on the basis of the observed H. pylori detection rate. Coefficients of determination (R 2 ) were obtained according to [21].
The observed binormal distributions of logarithmic ∆δ values from UBTs alone were modeled with sum functions of two normal distributions using maximum likelihood estimation [ [19], p.287ff]. Assuming the two normal distributions represent either absent or present H. pylori infection, threshold ∆δ values were calculated at the maximum of specificity plus sensitivity and of negative plus positive predictive values.
Average bias corrected estimates of the thresholds and the associated test performance characteristics were obtained by 1000-fold bootstrapping [22].
The concordance rate was the area under the receiver operating characteristic curve [23].
The Kolmogorov-Smirnov goodness-of-fit-test was used to compare distributions.

Feasibility and ∆δ values
Children from 4 months of age on were successfully studied with the aforementioned UBT procedure. The standard error of measuring delta values (∆δ) was 0.05‰. ∆δ values measured in breath samples taken 30 and 60 minutes after 13 C-urea ingestion correlated neither with body weight (Spearman's rank correlation R 2 = 0.03) nor with age (R 2 = 0.04).
∆δ values at 30 minutes were maximum 6.6‰ in the histological absence and minimum 3.4‰ in the histological presence of H. pylori. Only 18.3% of these 30 minutes UBT ∆δ values were between 3.4‰ and 6.6‰.
When positive biopsies occurred, always both antral biopsy specimens were positive. Inconsistently, the corpus was positive in addition to the positive antral specimens.
The modeled sum functions were not different from the observed ∆δ distributions (P = 0.45 and P = 0.08, respectively). The ∆δ values and the respective thresholds were similarly located in histology validated UBTs (figure 2, panel A) and in UBTs alone ( figure 2, panel B).
The portions of presumably positive UBTs were 24.4% at 30 minutes and 23.9% at 60 minutes in UBTs alone.

Threshold in UBTs in different age groups
Threshold ∆δ values at 30 minutes varied from 9.1 ± 4.5‰ in children up to 5 years of age to 5.0 ± 1.5‰ in adolescents (14-18 years of age). This seemingly negative association in the arbitrarily chosen age groups was not evident in a plot of all UBT values by age (not shown). In the youngest age group, the bootstrap procedure's variance indicates this model's instability (table 1).

Threshold evaluation
Standard performance measures (sensitivity, specificity, negative and positive predictive values, correct classification rates, and concordance rates) were evaluated at the calculated cut-offs and almost always exceeded 0.90 (table  1).

Discussion
This study newly found 13 C urea breath test (UBT) ∆δ thresholds that were robustly associated with histological H. pylori detection in children. Thresholds obtained in separate groups varied from 9‰ in small children to 5‰ in children and adolescents. The non-optimistic validation and the special evaluation of the UBT demonstrated high accuracy.
The threshold calculation of this study comprised logistic regressions and iterative modeling analyses. The analytical principle is superior to threshold selection procedures, even if the often used single value dichotomy is replaced by receiver operator characteristics including all test values (for example, [11]). For comparison, we used iterative analyses in a large number of UBTs from symptomatic children in whom gastric biopsies were not performed. These analyses returned threshold values, which were similar to the aforementioned results, apart from our youngest age group, which had a rather large error. However, a recent study of 3 months to 4 years old Gambian children agreeably complements our findings by obtaining a 5.5‰ threshold with a very similar method [16].
This study's design was based on others' findings that factors suspected to influence the UBT, such as the quantity and formulation of 13 C-urea, the fasting period, the test meal composition, and the timings of breath sample col- Here, ∆δ values were higher at 60 than at 30 minutes, which is in agreement with the 13 C-urea elimination kinetics. According to our results, breath test samples collected at 30 minutes were slightly more useful when analyzed for correspondence with the histological detection of H. pylori in gastric biopsy specimen. On the other hand, the two peaks of UBT ∆δ value distribution were slightly more separated in samples obtained 60 minutes after the test meal. In face of the small differences, either time point seems appropriate.
The many approaches for performing the UBT in adults have resulted in various threshold proposals ranging from about 3.5‰ to 6‰. In our study, only 16.3% of UBT samples at 30 minutes were in this comparatively wide range. In order to clarify the impact of H. pylori in relation to that of the respective approach for performing the UBT, multivariable models could be set up from the internationally accumulated data.
With regard to the effect of age, there were only few studies on rather small groups of children, not including infants [3,14,23]. Recently, a correlation of UBT values with the children's age was proposed, while the portion of explained variance was not given and the significance might be explicable by the large numbers (1016, of which 149 were histologically validated [25]). We could not demonstrate a systematic influence of age, notwithstanding the higher, but greatly variable thresholds in children up to 5 years. A partial shortcoming rests in (not only) our assumption that the modes of the UBT results represent negative or positive H. pylori detection. In contrast to other studies, we demonstrated that the model distributions based on this assumption are not different from the observed distributions. However, clarification of an age effect would require endoscopy in all age groups and groups of different H. pylori prevalence.
In our study, endoscopy with biopsies was prompted only by patients' symptoms, and not by UBT results. Re-sampling by bootstrapping was used to additionally reduce patient bias. Furthermore, thresholds were cross-validated, that is, thresholds obtained in one portion of the tests were evaluated prospectively in the complementary portion. Thus, the great accuracy reported in this study is not optimistically biased. Although these methods are very useful to validate conclusions and cut-offs in data analysis, they have not been reported with the UBT previously.
Additionally, thresholds other than the proposed ones could be chosen on the basis of the provided figures, which indicate the associated H. pylori detection probabilities.

Conclusions
In conclusion, this study calculated and validated urea breath test thresholds in a large pediatric patient population, and showed that this test reliably and non-invasively detects a H. pylori infection. In children, cut-offs such as 5.3‰ and 6.8‰ in UBT samples obtained at 30 and 60 minutes after ingestion of 75 mg 13 C-urea entail high accuracy (0.94 and 0.98, respectively) for gastric H. pylori detection.

Competing interests
None declared.