Real-Time PCR for Diagnosing Helicobacter pylori Infection in Patients with Upper Gastrointestinal Bleeding: Comparison with Other Classical Diagnostic Methods

Abstract

The aim of this study was to determine the diagnostic usefulness of quantification of the H. pylori genome in detection of infection in patients with upper gastrointestinal bleeding (UGB). A total of 158 consecutive patients with digestive disorders, 80 of whom had clinical presentation of UGB, were studied. The number of microorganisms was quantified using a real-time PCR system which amplifies the urease gene with an internal control for eliminating the false negatives. A biopsy sample from the antrum and corpus of each patient was processed. The rapid urease test, culture, histological study, stool antigen test, and breath test were done. The gold standard was a positive culture or positive results in at least two of the other techniques. When a positive result was defined as any number of microorganisms/human cell, the sensitivity of real-time PCR was greater in bleeding patients, especially in the gastric corpus: 68.4% (95% confidence interval [CI], 52.3 to 84.5%) in non-UGB patients versus 91.5% (95% CI, 79.6 to 97.6%) in UGB patients. When a positive result was defined as a number of microorganisms/human cell above the optimal value that maximizes the Youden index (>3.56 microorganisms/human cell in the antrum and >2.69 in the corpus), the sensitivity and specificity in UGB patients were over 80% in both antrum and corpus. Our findings suggest that some bleeding patients with infection caused by H. pylori may not be correctly diagnosed by classical methods, and such patients could benefit from the improved diagnosis provided by real-time PCR. However, the clinical significance of a small number of microorganisms in patients with negative results in classical tests should be evaluated.

INTRODUCTION

Peptic gastroduodenal ulcers are responsible for a high percentage of cases of upper gastrointestinal bleeding (21), and H. pylori is present in over 90% of cases of duodenal ulcer and 70% of cases of gastric ulcer (13, 31). Various studies have shown that eradication of H. pylori leads to a significant reduction in the incidence of recurring bleeding in patients with gastrointestinal bleeding of peptic origin (8, 14, 25). Therefore, the H. pylori status of patients with bleeding peptic ulcers should be documented.

Some studies indicate that many classical diagnostic methods show a significant decrease in their sensitivity in patients with bleeding peptic ulcers (2, 20). However, various studies have shown that in cases of upper gastrointestinal bleeding due to peptic ulcers, PCR techniques for detecting the genome of the microorganism are more reliable than invasive techniques (19, 20, 24).

We performed this prospective study to evaluate the usefulness of real-time PCR for detecting H. pylori infection in patients with bleeding peptic ulcers and to compare its diagnostic efficacy with that of other conventional tests (5).

MATERIALS AND METHODS

Patients and samples.

We undertook a prospective study including 158 consecutive symptomatic patients who attended the Endoscopy Unit of Elche University General Hospital for an upper digestive tract endoscopy, in whom investigation of H. pylori status was indicated: duodenal ulcer (n = 68), duodenitis (n = 21), gastric ulcer (n = 36), gastritis (n = 5), dyspepsia (n = 11), family history of gastric cancer (n = 3), and failure to eradicate H. pylori (n = 14). For 154 of the 158 patients (97.4%), biopsy specimens of both the gastric antrum and corpus were obtained, whereas in four patients only samples from the antrum were taken.

The patients' demographic and clinical data were collected: age, sex, weight, height, smoking habit, alcoholism (consumption of more than 80 g/day in men and 60 g/day in women), comorbidity (high blood pressure, diabetes, cardiopathy, chronic obstructive pulmonary disease, kidney failure), history of complicated ulcer, consumption of proton pump inhibitors, nonsteroidal anti-inflammatory drugs, or antibiotics in the 7 days prior to the endoscopy, and clinical presentation with upper gastrointestinal hemorrhage.

All subjects were informed of the study objectives and gave their informed consent prior to their inclusion in the study. The research protocol was approved by the local ethics and research committees of the participating hospital.

Sampling and classical diagnostic methods.

Two samples of the antrum and two of the corpus were taken and processed in under an hour. One of each was used for Gram staining and culture [Columbia agar base with H. pylori selective supplement and defibrinated horse blood (Difco)] and the other for DNA extraction. Likewise, an antral sample was taken for the rapid urease test (CLO test; Ballard Medical Products) and another was taken for anatomopathological study. In addition, the presence of H. pylori antigen in feces was detected (Letitest; Leti), and in 36 patients the breath test was done to detect H. pylori (TAU kit; Isomed Pharma). The gold standard for diagnosing infection due to H. pylori was a positive culture or a positive result with two or more of the conventional techniques used.

Molecular methods. (i) Extraction of DNA from biopsy specimens.

DNA was extracted using the EZ1 DNA system (Qiagen, Germany) and BioRobot EZ1 workstation (Qiagen, Germany). It was quantified and its purity studied using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific).

(ii) Quantification of microorganisms.

A system was designed in our laboratory, using the software package Primer Express 3.0, to amplify a fragment of the urease gene of H. pylori with the following primers: forward, 5′-GCTCTCACTTCCATAGGCTATAATGTG-3′; reverse, 5′-GCGCATGTCTTCGGTTAAAAA-3′; FAM-labeled probe, 5′-TAGGGCCTATGCCTACCCCTGCGA-3′. This system has an internal control, which was also designed in our laboratory from a 166-bp fragment of Chlamydia trachomatis, replacing the end of the fragment with sequences complementary to the primers used to amplify the urease gene. The microorganisms were detected using a specific control probe marked with a different fluorochrome (VIC) (see Fig. S1 in the supplemental material) (1). The reaction took place in a real-time PCR 7500 apparatus (Applied Biosystems). The sensitivity of the technique was 40 microorganisms per sample. This was determined by a standard curve made from an H. pylori strain (ATCC 43504), using serial dilutions of the strain, quantification by quantitative culture, and extraction and quantification of total DNA (4). The specificity was evaluated with 60 samples of leukocytes from patients with no gastrointestinal symptomatology and with different microorganisms frequently found in the human gastrointestinal microbiome (Corynebacterium urealyticum, Klebsiella pneumoniae, Enterococcus faecalis, Pseudomonas aeruginosa, and Campylobacter jejuni). It was also confirmed, using the program CLC Sequence Viewer (CLC Bio), that the probe used was 100% homologous to the H. pylori gene but less so to other species of the genus.

(iii) Normalization of results according to the number of human cells in each biopsy specimen.

In order to prevent the size of the gastric biopsy specimen (which varied from one sample to another) from affecting DNA quantification, the number of human cells in the biopsy sample was quantified by amplifying the human albumin gene with the primers 5′-CTGCATTGCCGAAGTGGAA-3′ and 5′-CAAACATCCTTACTTTCAACAAAATCA-3′ plus the FAM-labeled probe 5′-TGCCTGCTGACTTGCCTTCATTAGCTG-3′. The standard curve for human cells was made from leukocytes extracted from peripheral blood using Histopaque (Sigma) and quantified by Celdyn 3600 (Abbott). The results are expressed as the quotient between the number of microorganisms and number of human cells in each gastric sample.

Statistical analysis.

Categorical variables were compared using the chi-square test. In order to compare the number of microorganisms present in the biopsy specimens of bleeding and nonbleeding patients, nonparametric tests were used (Mann-Whitney U test for independent samples). The level of statistical significance was set at 0.05, and all tests were two-tailed.

For the 154 patients from whom samples of both antrum and corpus were taken (79 patients with UGB and 75 without clinical presentation of UGB), the area under the receiver-operating characteristic (ROC) curve (AUC) was obtained for both antrum and corpus. The optimal cutoff points were calculated in two ways: (i) by defining a positive result as any number of microorganisms (real-time PCR > 0; bacteria present or absent), that is, maximizing the cost of false negatives; (ii) by defining a positive result as the number of microorganisms that maximized the weighted combination of sensitivity and specificity (i.e., that maximized the Youden index) for the same cost of false negatives and false positives and for the prevalence of the sample.

Based on these cutoff points, the main parameters of diagnostic validity were estimated: sensitivity, specificity, positive predictive value, and negative predictive value.

In addition, in order to estimate the association between the possibility of obtaining a positive result in each specific diagnostic test in bleeding and nonbleeding patients, crude and adjusted odds ratios (OR) and 95% confidence intervals (95% CI) were calculated using unconditional logistic regression.

The following potential confounders were included in the models: sex, age (≤50 years, >50 years), and consumption of proton pump inhibitors (PPI) in the days prior to endoscopy (yes/no).

All analyses were performed with SPSS v.19.0.

RESULTS

Eighty of the 158 patients (50.6%) had clinical presentation of upper gastrointestinal bleeding (UGB). In 13 of these (16.3%) blood was seen during endoscopy. Table 1 shows the patients' ages, sexes, clinical characteristics, and lifestyle in relation to UGB clinical presentation. Bleeding patients were on average 5.9 years older than nonbleeding ones (P = 0.03). In 61% of the women there was clinical presentation of UGB, whereas only 38% of the men had bleeding (P = 0.04). The group of bleeding patients presented with a statistically significant higher comorbidity and consumption of nonsteroid anti-inflammatory drugs (NSAID) in the days prior to bleeding.

Table 1.

Baseline characteristics of patients as a function of upper gastrointestinal bleeding (UGB) status

Characteristic	Result (%) for patient group		P
	UGB negative (n = 78)	UGB positive (n = 80)
Age (mean [SD])	52.92 [16.0]	58.78 [18.0]	0.03
Male	44 (62)	27 (38)	0.04
Female	34 (39)	53 (61)
Smoker	13 (45)	16 (55)	0.93
Drinker	1 (13)	7 (88)	0.07
Comorbidity	20 (34)	38 (66)	0.04
PPI consumption in previous days	21 (46)	25 (54)	0.8
NSAID consumption in previous days	7 (18)	32 (82)	<0.01
Antibiotic consumption in previous days	2 (40)	3 (60)	0.85
cagA positivity	17 (31)	37 (69)	0.07
Bacterial load (median) [interquartile range]a	69.9 [435.9]	244.1 [1,129.6]	0.38

^aNumber of copies of H. pylori per cell in samples from antrum according to real-time PCR.

The distribution of the number of microorganisms in the biopsy specimens of both antrum and corpus was very asymmetric. Bacterial load was higher in patients with UGB, although these differences were not statistically significant (P = 0.38) due to the high variability observed in the bacterial load of both bleeding and nonbleeding patients.

The area under the ROC curve (AUC) for patients with UGB was 0.89 for antrum (95% CI, 0.81 to 0.97) and 0.85 for corpus (95% CI, 0.75 to 0.94).

When a positive result was defined as any number of microorganisms (real-time PCR > 0 microorganisms/human cell; bacteria present or absent), the sensitivity in the gastric antrum in bleeding patients increased slightly compared with that in nonbleeding patients (94.7% in nonbleeding patients and 97.9% in bleeding patients). The increase was greater for the gastric corpus (68.4% in nonbleeding versus 91.5% in bleeding patients) (Table 2).

Table 2.

Validity indexes and predictive values of the real-time PCR test^a

Source of sample	UGB	AUC (95% CI)	% (95% CI)
			Se	FN	Sp	FP	PV+	PV−
Antrum	No	0.93 (0.86–0.99)	94.7 (82.3–99.4)	5.3	73.0 (57.3–88.6)	27.0	78.3	93.1
	Yes	0.89 (0.81–0.97)	97.9 (88.7–99.9)	2.1	53.1 (34.3–72.0)	46.9	75.4	94.4
Corpus	No	0.80 (0.69–0.90)	68.4 (52.3–84.5)	31.6	78.4 (63.8–93.0)	21.6	76.5	70.7
	Yes	0.85 (0.75–0.94)	91.5 (79.6–97.6)	8.5	56.3 (37.5–75.0)	43.8	75.4	81.8

^aValues were calculated as a function of upper gastrointestinal bleeding (UGB) status, considering as positive any number of microorganisms (real-time PCR > 0 microorganisms/human cell). The patients were considered to be infected by H. pylori if the culture was positive or the results of two of the other techniques were positive, either in antrum or in corpus. Se, sensitivity; Sp, specificity; FN and FP, false negative and false positive, respectively; PV, predictive value for sample prevalence.

In Table S1 in the supplemental material, real-time PCR results in antrum (bacteria present or absent) are compared to the results of each of the classical tests individually. Sensitivity in bleeding patients was 31.1%, 56.1%, 54.2%, 55.9%, and 90.9% for culture, stool antigen test, histology, rapid urease test, and breath test, respectively, compared with real-time PCR as the gold standard.

In this respect, the odds ratio for obtaining a positive result with real-time PCR (more than 0 copies) in bleeding patients was greater (OR = 2.1 and 3.1 for antrum and corpus, respectively) than with the classical gold standard tests as a whole (OR 1.4) and each of them individually, with the exception of the breath test. However, these differences tended to decrease after adjustments were made for age, sex, and consumption of protein pump inhibitors: the adjusted OR for real-time PCR was 2.5 for antrum and corpus, and the adjusted OR for the classical gold standard tests as a whole was 2.2 (Table 3).

Table 3.

Odds ratios for positive results in the conventional test and real-time PCR, in bleeding patients compared to nonbleeding patients

Diagnostic test [cutoff] (na)	Result	No. according to UGB status		Crude ORb	95% CI	Adjusted ORb	95% CI
		No	Yes
Real-time PCR, antrum [>0 copies] (154)	Neg	29	18	1		1
	Pos	46	61	2.1	1.1–4.3	2.5	1.1–5.5
Real-time PCR, antrum [>3.56 copies] (154)	Neg	41	36	1		1
	Pos	34	43	1.4	0.8–2.7	1.9	0.9–4.0
Real-time PCR, corpus [>0 copies] (154)	Neg	41	22	1		1
	Pos	34	57	3.1	1.6–6.1	2.5	1.2–5.3
Real-time PCR, corpus [>2.69 copies] (154)	Neg	51	34	1		1
	Pos	24	45	2.8	1.5–5.4	2.9	1.4–6.1
Classical gold standard (158)	Neg	39	33	1		1
	Pos	39	47	1.4	0.8–2.7	2.2	1.1–4.6
Culture (148)	Neg	42	58	1		1
	Pos	28	20	0.5	0.3–1.0	0.6	0.3–1.3
Stool antigen test (98)	Neg	22	24	1		1
	Pos	23	29	1.2	0.5–2.6	2.0	0.8–5.1
Histology (136)	Neg	43	44	1		1
	Pos	17	32	1.8	0.9–3.8	2.0	0.9–4.2
Rapid urease test (CLO test) (135)	Neg	38	43	1		1
	Pos	21	33	1.4	0.7–2.8	1.1	0.5–2.4
Breath test (36)	Neg	11	1	1		1
	Pos	13	11	9.3	1.0–83.9	6.6	0.6–78.1

^aNumbers in parentheses are numbers of valid cases (cases in which the diagnostic test was done).

^bOdds ratios (OR) and 95% confidence intervals (95% CI) were adjusted for gender, age (≤50 years versus >50 years) and consumption of PPI (yes/no). Negative results in PCR and classical tests were taken as the reference category.

The specificity in both antrum and corpus was less in bleeding than in nonbleeding patients, and the percentage of false positives was higher in bleeding than in nonbleeding patients: 46.9% false positives were found for antrum (n = 25) and 43.8% for corpus (n = 22). The quantitative values in these discrepant patients (false positives) varied greatly both in antrum and in corpus, although the bacterial load was greater in bleeding (median antrum, 1.5 microorganisms/cell [interquartile range, 62.2]) than in nonbleeding patients (median antrum, 0.6 [interquartile range, 41]). Only in bleeding patients were extreme real-time PCR values found (47,722.8 in antrum and 62,991.9 in corpus), and in these patients, conventional tests gave negative results (Table 4 and Fig. 1).

Table 4.

Real-time PCR false-positive values (FP) for antrum and corpus as a function of UGB status

Calculation	No. of copies of H. pylori per human cell according to real-time PCR
	FP in antrum (n = 25)		FP in corpus (n = 22)
	No UGB (n = 10)	UGB (n = 15)	No UGB (n = 8)	UGB (n = 14)
Mean	36.7	3,209.5	4.4	4,742.0
SD	77.0	12,314.4	10.7	16,777.7
Mean trimmed at 5%	29.4	914.9	3.2	1,769.3
50th percentile (Median)	0.6	1.5	0.2	1.2
75th percentile	41.0	62.2	1231.7	2,310.5
90th percentile	200.6	1,9251.1	6,012.5	26,839.4
Mínimum	0.002	0.010	0.002	0.011
Maximum	205.5	47,722.8	30.9	62,991.9
Interquartile range	41.0	62.2	2.1	308.1

Fig 1.

Standard box plot of bacterial load in RT-PCR false-positive cases (on a logarithmic scale) as a function of upper gastrointestinal bleeding status for antrum and corpus.

Using the cutoff point that maximizes the Youden index, the optimal cutoff point was 3.56 microorganisms/human cell for the antrum and 2.69 microorganisms/human cell for the corpus. When a number of microorganisms/human cell above these cutoff points was considered a positive result in real-time PCR, the sensitivity and specificity were over 80% in bleeding patients in both antrum and in corpus (see Table S2 in the supplemental material).

DISCUSSION

Helicobacter pylori has been associated with a large number of pathological conditions, such as chronic gastritis, duodenal ulcers, gastric cancer, and mucosa-associated lymphoid tissue (MALT) lymphoma (3). The true prevalence of infection caused by this microorganism in patients with bleeding peptic lesions is still a source of controversy. While some authors report a prevalence of over 90% (19), other authors have described a lower prevalence, between 20 and 70% (17, 30). These discrepancies may be due to the different diagnostic methods used, since the techniques employed to diagnose this infection in this type of patients have often been reported to be less sensitive (17, 18, 32).

Techniques based on detection of the bacterial genome using different molecular methods have proved to be of great use in patients with nonbleeding peptic ulcers (12). However, to date, very few studies have addressed the usefulness of these techniques, especially very sensitive and specific techniques, such as real-time PCR with an internal control, in bleeding patients (7, 19, 20, 24).

Our results suggest that our method has a good diagnostic capacity compared with the classical gold standard in bleeding patients, and is slightly greater in antrum than in corpus. (AUC, 0.89 for antrum versus 0.85 for corpus). These values indicate that in a gold standard-positive bleeding patient, chosen at random from the population studied, there is a 89% and 85% probability of bacterial quantification being higher in antrum and corpus, respectively, than the PCR levels of a gold standard-negative bleeding subject chosen at random from the same population.

In contrast to the study by Lin et al. (19), in which a decrease in sensitivity of PCR in bleeding patients was described, our results show an increase in sensitivity in bleeding compared with nonbleeding patients, probably due to the improved methods of extracting DNA from biopsy specimens, greater capacity to eliminate substances that inhibit the reaction, and amplification of the different regions of the genome. This difference is small in the gastric antrum samples (94.7% in nonbleeding and 97.9% in bleeding patients); however, it is more marked in the gastric corpus samples (68.4% in nonbleeding versus 91.5% in bleeding patients). This supports taking a biopsy sample from the antrum when only a single biopsy sample is taken in bleeding patients.

A global analysis of all the techniques used shows that the test with the greatest odds of giving a positive result in bleeding patients is the breath test, followed by real-time PCR, which confirms previously published findings (6, 23, 28). In this respect, the odds of a positive result with real-time PCR are greater than the odds with the classical gold standard tests as a whole and with the other invasive tests in which a gastric biopsy is necessary. These data support the view that the breath test and real-time PCR are the most useful techniques for diagnosing this pathology in this type of patient.

When a study to validate a diagnostic test is being designed, the reference test should be appropriate to measure the disease or condition under study (33). The limitations of traditional methods used with H. pylori, in particular the problem of false negatives, have been described by various authors (9, 11, 16, 27, 28).

The low sensitivity of classical methods in these patients makes it difficult to correctly evaluate the new methods. Therefore, it may be that results identified as false positives obtained with real-time PCR are not actually false positives but rather that the results obtained with classical methods were false negatives, especially when only one gastric sample was processed, since an increase in the number of biopsy specimens taken has been reported to improve the sensitivity of these techniques (7, 18, 29).

Our study shows that patients with UGB have more copies of the H. pylori genome, although the differences were not statistically significant. These data do not support the theory that the blood serum in the stomachs of bleeding patients has bactericidal activity against this microorganism and thus decreases the bacterial load (10). However, this theoretical decrease could also be attributed to the limited sensitivity of classical tests together with the fact that gastric lavage is carried out in these patients or that albumin is present in the stomach (27, 28).

Likewise, our results suggest that using real-time PCR in the diagnosis of H. pylori infection in bleeding patients, in addition to being an important diagnostic advance, since it enhances the sensitivity of diagnostic methods, makes it possible to determine the number of microorganisms present in the gastric mucosa (26). This information has been associated with disease eradication rate, ulcer healing rate, and antral inflammation (15). In addition, our previously reported results showed that upper gastrointestinal bleeding was more closely associated with the bacterial load than the presence of the cagA gene, suggesting the importance of quantifying the bacterial load in order to gain a better understanding of the complex interaction between the microorganism and gastric mucosa (1).

A quantitative analysis of these discrepant patients shows that in some cases the number of microorganisms is very high, which confirms the diagnostic limitations of classical methods in these patients. It also suggests that using a combination of the breath test and real-time PCR would increase the probability of a correct diagnosis, which is important given the disease severity in these patients and the need to prevent recurrences. However, further studies should be done to determine the clinical significance of a small number of microorganisms in patients in whom classical tests give negative results, since transitory colonizations with no clinical importance have been described (22).