IJE Advance Access originally published online on November 22, 2005
International Journal of Epidemiology 2005 34(6):1356-1358; doi:10.1093/ije/dyi243
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Commentary |
Commentary: H. pylori infection in early life and the problem of imperfect tests
Stanford University School of Medicine, Stanford, CA, USA
* Corresponding author. E-mail: shnperry{at}stanford.edu
Helicobacter pylori infection is associated with significant disease outcomes, including peptic ulcers and gastric cancers. While persistent infection appears to be established during childhood, little is known about infection patterns in early life. Goodman et al.1 report a first time infection rate of 27% among 468 children of the Pasitos cohort followed at 6-month intervals from birth to 24 months using the 13C urea breath test. They also report that 77% of the children lost infection during follow-up and 19% of those who did so were reinfected at a later time. Since infection in childhood may well determine the outcome of disease over a lifetime, this well-designed study—including a bi-national structure and systematic testing targets—can add immensely to our knowledge of early childhood infection as well as to our experience with non-invasive testing in infants and toddlers.
A major finding of the study is that transient infection may be a common characteristic of early childhood infection and should be considered in the design of prospective studies as well as in therapeutic interventions targeted at this age group. While a growing number of studies have reported high rates of transient infection in the very young,2–4 uncertainty about the performance of non-invasive tests in infants and small children5 continues to haunt our understanding of this phenomenon. Serology, for instance, is considered unreliable in children under 2 years of age.6 Although stool antigen7 and the C-urea breath test8,9 are thought to be more accurate, studies are still required to establish operating characteristics in this age group. Serial testing in children of this age is even more problematic, because any test misclassification is inherited from one test generation to the next.10 For this reason, even a test with reasonably good operating characteristics might produce a not insignificant cumulative error.
Assume that we study a population of 468 infants with a ‘true’ prevalence of infection of 10%. Given a test with 95% sensitivity and 95% specificity, we would expect to observe 66 positive tests, rather than 47 (Figure 1, Cycle 1). In other words, 33% of observed infections will be false positives. Now let us assume that after some interval we test the infants again. Suppose, too, that no new infections have actually occurred and that the test delivers exactly the same performance (95% sensitivity and specificity). Given these assumptions, we would, by chance alone, observe another 22 new infections among prior negatives and another 22 reversions among prior positives (Cycle 2). Over four testing cycles, if no new incident cases occurred, the cumulative first time infection rate would appear to be 27%, which is approximately what Goodman et al. observed, with a cumulative reversion rate of about 61% (vs Goodman's report of 77%). In addition, we would observe seven (15%) recurrences of infection among children who appeared to lose infection at a prior visit (vs 5 reported). Given these operating characteristics and four testing cycles, almost all first time conversions and reversions could have been observed by chance.
|
Suppose that the test is virtually 100% specific and 95% sensitive. In this case, the only error incorporated throughout our testing cycles is due to the false negative error rate of the test. Assuming constant prevalence of 10%, the first time conversions will be 10% and only 14% will appear to revert. Conversely, if the sensitivity were 100% and the specificity 95%, the only error is due to the false positive rate of the test. In this instance, the rate of first time infection will be 27% as in the first example, and the reversion rate
55%. In fact, as we vary the specificity and hold the sensitivity constant, the conversion rate is simply 1-specificity of the test, and if we vary the sensitivity the reversion rate is simply 1-sensitivity of the test. In general, it can be appreciated that at lower prior probabilities the specificity of the test is driving the results (Table 1).
|
Of course, tests do not behave exactly as we describe. Repeated tests within individuals are not independent, i.e. a person whose specimen yields a false positive first test may be predisposed to have a false positive on all subsequent tests. As we have learned in serology, antibody production depends on the development of humoral immunity. Similarly, for stool and breath tests, we do not know how changes in intestinal flora during the first few years of life might affect the response of different individuals to the same test or of the same individuals to different diagnostic agents. The fact that, in Goodman's study, incidence was similar on both sides of the border—despite expected differences in exposure—and that point prevalence was also fairly stable after 6 months tends to support a concern that test misclassification rates, as well as any changes in the true force of infection accounts for an unknown, but potentially large, portion of the incidence, reversion, and transience of H. pylori infection observed.
Incidence studies are notoriously complex to conduct, and results are often difficult to analyse. The authors of this study are to be commended for their thorough analysis of common pitfalls, including losses to follow-up and selection bias. While it may not be possible to exclude test error as an explanation for serial testing results, there are measures that could be undertaken to minimize misclassification and increase the investigator's confidence in results. First, if a test has an equivocal zone, removing these results from calculations provides a more conservative estimate of incidence. Second, use of a second test to confirm positive results can improve confidence significantly. For instance, in the examples above, if another test with similar operating characteristics had been used to confirm positives, the predictive value positive (PV+) of the first visit result would have jumped by 45%. This can be appreciated by pretending that the Cycle 2 test in Figure 1 is a corroborative test conducted at Cycle 1. The confirming test need not be as competitive—for instance, a confirming test with a sensitivity of 80 and a specificity of 80 would still boost predictive value positive by 32%. Conversely, adding a second test that will exclude individuals who are positive on the corroborative test might be a useful tactic for studying reversions. In either case, as it may be prohibitively expensive to use a second test universally, corroborative testing on a random sample can still provide a basis for quantifying the investigator's confidence in results. Finally, using a test such as stool antigen that provides material for DNA sequencing can help support results. In one series, we found that 56% of children <2 years of age with transient stool antigen results could be sequenced for the Helicobacter species at the first visit, only one of which was sequenced at the second.11
An important finding of this study is that the point prevalence of infection was 15% at 12 months of age, and was largely reproducible thereafter. This is not inconsequential from a public health standpoint. Other attributes of this cohort, including size and density of households, years of education, and number of siblings, are well-known risk factors for H. pylori infection. Thus, regardless of whether transient infection is an important feature of early childhood infection or an artefact of the tests we use, H. pylori infection is common in this cohort early in life, and these children have palpable, continuing risks for exposure and infection. Ongoing work from this important cohort should help clarify important questions regarding modifiable risk factors in early childhood.
 |
References |
|---|
 Top References |
|---|
1 Goodman KJ, O'Rourke K, Day RS et al. Dynamics of Helicobacter pylori infection in a US-Mexico cohort during the first two years of life. Int J Epidemiol 2005; 34:1348–55.
2 Klein PD, Gilman RH, Leon-Barua R, Diaz F, Smith EO, Graham DY. The epidemiology of Helicobacter pylori in Peruvian children between 6 and 30 months of age. Am J Gastroenterol 1994;89:2196–200.[Web of Science][Medline]
3 Rothenbacher D, Bode G, Brenner H. Dynamics of Helicobacter pylori infection in early childhood in a high-risk group living in Germany: loss of infection higher than acquisition. Aliment Pharmacol Ther 2002;16:1663–68.[CrossRef][Web of Science][Medline]
4 Perez-Perez GI, Sack RB, Reid R, Santosham M, Croll J, Blaser MJ. Transient and persistent Helicobacter pylori colonization in Native American children. J Clin Microbiol 2003;41:2401–07.
5 Koletzko S, Feydt-Schmidt A. Infants differ from teenagers: use of non-invasive tests for detection of Helicobacter pylori infection in children. Eur J Gastroenterol Hepatol 2001;13:1047–52.[CrossRef][Web of Science][Medline]
6 Khanna B, Cutler A, Israel NR et al. Use caution with serologic testing for Helicobacter pylori infection in children. J Infect Dis 1998;178:460–65.[Web of Science][Medline]
7 Gisbert JP, Pajares JM. Stool antigen test for the diagnosis of Helicobacter pylori infection: a systematic review. Helicobacter 2004;9:347–68.[CrossRef][Medline]
8 Gisbert JP, Pajares JM. C-urea breath test in diagnosis of Helicobacter pylori infection: a critical review. Aliment Pharmacol Ther 2004;20:1001–17.[CrossRef][Web of Science][Medline]
9 Konstantopoulos N, Russmann H, Tasch C et al. Evaluation of the Helicobacter pylori stool antigen test (HpSA) for detection of Helicobacter pylori infection in children. Am J Gastroenterol 2001;96:677–83.[CrossRef][Medline]
10 Rosenstock S, Jorgensen T, Andersen L, Bonnevie O. Seroconversion and seroreversion in IgG antibodies to Helicobacter pylori: a serology based prospective cohort study. J Epidemiol Community Health 2000;54:444–50.
11 Haggerty TD, Perry S, Sanchez L, Perez-Perez G, Parsonnet J. Significance of transiently positive enzyme-linked immunosorbent assay results in detection of Helicobacter pylori in stool samples from children. J Clin Microbiol 2005;43:2220–23.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  C. V Phillips and K. J Goodman Interpreting data in the face of competing explanations: assessing the hypothesis that observed spontaneous clearance of Helicobacter pylori was all measurement error Int. J. Epidemiol., March 23, 2009; (2009) dyp006v1. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Perry, L. Sanchez, S. Yang, Z. Agarwal, P. Hurst, and J. Parsonnet Reproducibility of QuantiFERON-TB Gold In-Tube Assay Clin. Vaccine Immunol., March 1, 2008; 15(3): 425 - 432. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. V Phillips Commentary: Lack of scientific influences on epidemiology Int. J. Epidemiol., February 1, 2008; 37(1): 59 - 64. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||