Rapid Responses to:

PAPERS: Marjatta Karvonen, Zygimantas Cepaitis, and Jaakko Tuomilehto Association between type 1 diabetes and Haemophilus influenzae type b vaccination: birth cohort study BMJ 1999; 318: 1169-1172 [Abstract] [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

Hemophilus vaccine and increased IDDM, causal relationship likely.

J Bart Classen   (7 May 1999)

Two vaccination studies offer contrast in interpretability.

Steve Simon   (9 May 1999)

Hib vaccine & increased incidence of IDDM

Heidi White   (14 May 1999)

A unique opportunity

Stephen Katona   (29 October 1999)

Hemophilus vaccine and increased IDDM, causal relationship likely. 7 May 1999
J Bart Classen Classen Immunotherapies Send response to journal: Re: Hemophilus vaccine and increased IDDM, causal relationship likely.	Hemophilus vaccine and increased IDDM, causal relationship likely. Authors: John Barthelow Classen M.D., M.B.A. (contact person) President and Chief Executive Officer Classen Immunotherapies, Inc. 6517 Montrose Avenue Baltimore, MD 21212 U.S.A. E-mail: Classen@vaccines.net http://vaccines.net Tel: (410) 377-4549 Fax: (410) 377-8526 David C. Classen, M.D., M.S. Division of Infectious Diseases LDS Hospital and University of Utah School of Medicine Salt Lake City, Utah We initiated and funded a collaborative study with Tuomilehto on the effect of the Hemophilus vaccine on insulin dependent diabetes (IDDM) and found the data supports a causal relationship between the vaccine and the development of IDDM (paper submitted). Furthermore the potential risk of the vaccine exceeds the potential benefit. We therefore disagree with our collaborator. The study involved groups receiving 4 doses, 1 dose and 0 doses of hemophilus vaccine. The cumulative incidence of IDDM/100,000 in the 3 groups were 261, 237, 207 at 7 years and 398, 376, 340 at 10 years of age respectively. Tuomilehto's analysis is not rational and his conclusion is not supported by our data. Tuomilehto compared groups receiving 4 doses to 1 dose and groups receiving 1 dose to 0 doses. His analysis minimizes the difference and misleads the reader. His calculations of relative risk are also misleadingly low and we urge the reader to check these. Most objective researchers would compare the group receiving 4 doses to the group receiving 0 doses. Alternatively they would compare the combined vaccinated groups to the group receiving 0 doses. Both reach statistical significance. The cumulative difference in cases IDDM/100,000 between those receiving 4 doses and those receiving 0 doses is 54 cases (P=0.013) at 7 years and 58 cases at 10 years (P=0.029) using a single tail Fisher test. The relative risk equals 1.26 at 7 years. The cumulative difference between those receiving 4 or 1 doses and those receiving 0 doses is 42 cases (P=0.016) at 7 years and 47 cases at 10 years (P=0.028). The results are disturbing. The rise in IDDM, just one potential adverse effect exceeds the potential benefit of the vaccine, which has been estimated to prevent 7 deaths and 7 to 26 cases of severe disability per 100,000 children immunized (1). Even the difference between the groups receiving 4 and 1 dose exceeds the mean expected benefit. Temporal changes in the incidence of IDDM do not explain the differences since there were an extra 31 cases of IDDM/100,000 children in the age 5 to 10 and the incidence of IDDM in this group had been stable for about 10 years prior to this (2). Furthermore sharp rises in IDDM have been recorded in the US (3) and the UK (4) following the introduction of the Hemophilus vaccine. It is understandable that public health officials want to avoid a scaring the public however they risk depriving damaged children of compensation. Their denial of safety concerns are likely to lead to an eroding of public confidence especially since they may be able to avoid the problem of vaccine induced diabetes by starting vaccination a few weeks earlier (5). For additional information please see http://vaccines.net Competing interest: Methods used in our research is covered by patents which are owned by Classen Immunotherapies. These patents includes methods of testing vaccines for the induction of diabetes and methods of administering vaccines without inducing diabetes. Dr. John B. Classen holds stock in Classen Immunotherapies. Dr. David Classen owns no stock in Classen Immunotherapies, receives no funding from Classen Immunotherapies and has no financial ties to Classen Immunotherapies or this research. Bibliography 1. Peltola H, Kayhty H, Sivonen A, Makela H. Hemophilus influenza type B capsular polysacharide vaccine in children: a double blind field study of 100,000 vaccinees 3 months to 5 years of age in Finland. Pediatrics 1977;60:730-7. 2. Tuomilehto J, Virtala E, Karvonen M, et al. Increase in incidence of insulin-dependent diabetes mellitus among children in Finland. International Journal of Epidemiology 1995;24:984-92. 3. Dokheel TM. An epidemic of childhood diabetes in the United States. Diabetes Care 1993;16:1606-11. 4. Gardner S, Bingley PJ, Sawtell PA, Weeks S, Gale EA. Rising incidence of insulin dependent diabetes in children under 5 years in Oxford region: time trend analysis. BMJ 1997;315:713-6. 5. Classen DC, Classen JB. The timing of pediatric immunization and the risk of insulin-dependent diabetes mellitus. Infectious Diseases in Clinical Practice 1997;6:449-54.
Two vaccination studies offer contrast in interpretability. 9 May 1999
Steve Simon, Children's Mercy Hospital Kanas City, MO U.S.A Send response to journal: Re: Two vaccination studies offer contrast in interpretability.	Two very similar studies published back to back in the May 1, 1999 issue of the British Medical Journal offer an interesting contrast in interpretability. Both studies examine an adverse event that might be associated with vaccination: type 1 diabetes with Haemophilus influenzae type B vaccine (1) and wheezing illnesses with pertussis vaccine (2). Both studies use a cohort design. Both studies summarize the association using risk ratios. Both studies even include identical risk ratios, 1.06, among the risk ratios studied. But one study uses p-values to summarize statistical significance and the other uses confidence intervals. The study using confidence interval is easier to interpret because you can assess the practical significance of the findings. You assess practical significance by examining whether the results of a study, even after allowing for sampling error, fail to exceed a threshold of clinical relevance. The threshold is a change of sufficient size to convince you to change your practice. Confidence intervals are valuable for studies that achieve statistical significance. With large sample sizes, it is possible to achieve statistical significance, even though the results of the study lie well within a range of clinical indifference. Confidence intervals are even more valuable for studies that fail to demonstrate statistical significance. If the confidence interval is sufficiently narrow, then absence of a statistically significant effect also implies absence of a clinically relevant effect. The narrow interval assures you that this finding could not be caused by an inadequate sample size. In the pertussis vaccination article, it is easy to judge practical significance because the authors present the statistical summary using confidence intervals. For example, the relative risk for intermittent wheezing is 1.06 and the 95% confidence interval is 0.81 to 1.37. This interval clearly excludes the possibility that vaccination is associated with a large relative change in either direction. As a statistician, I am in a poor position to judge how large a risk would have to be in order to achieve clinical relevance. Such a decision would require medical judgement about the frequency and severity of the adverse event, the cost of treating the adverse event, and would have to balance this against the benefits of vaccination. The decision should examine both relative and absolute changes in risk. But suppose an expert decided in this situation that a change in clinical practice should only happen for a relative risk smaller than 0.67 or larger than 1.5. With this guidance, you have evidence that no change in practice is warranted, because the confidence interval in the pertussis vaccination article lies entirely inside a range of clinical indifference. In the H. influenzae type B vaccination article, it is much harder to judge practical significance because the authors use p-values. The relative risk of type 1 diabetes comparing children first vaccinated at 3 months (cohort 2) to children first vaccinated at 24 months (cohort 3) is also 1.06, but the authors only tell us that the p-value is 0.545. This p- value does not inform us how large the relative risk might plausibly be. It only takes a little effort to calculate a confidence interval for this risk ratio and presenting this information adds very little to the length of the article. Using SPSS version 8 software (SPSS Inc. Chicago IL), I found the 95% confidence interval to be 0.88 to 1.27. The authors were fortunate that lack of statistical significance also happened to coincide with lack of practical significance. This is not always the case. Suppose that the adverse event studied was more common, so that we saw 4,700 events in cohort 2 and 4,280 events in cohort 3 and that the sizes of the two cohorts remained the same (59,238 and 57,114 respectively). You have the same relative risk, but now this value is statistically significant (p=0.0049). A confidence interval, however, informs us about the lack of practical significance of this finding. The narrow interval (1.02 to 1.10) would allow us to rule out the possibility of anything except a very small increase in relative risk. Consider another scenario: a very rare adverse event. If there were 12 events in cohort 2 and 11 in cohort 3, and the sample sizes remained the same, you would see a relative risk of 1.05. The p-value would be non- significant (p=0.90). Yet you could take little comfort from this finding. In this case, the confidence interval is extremely wide, 0.46 to 2.4. The data is consistent with a finding of no effect, but it is also consistent with a finding that one cohort has half the risk of the other. It is even consistent with a finding that the cohort has twice the risk. These are extreme and artificial examples, but they illustrate an important point: you can't judge clinical relevance by looking at the p- value. The study with a small p-value gives you more evidence of no practical effect than the study with a large p-value. It is unfortunate that the authors and reviewers and editors involved with the H. influenzae type B vaccination article have failed to heed advice about confidence intervals that is well known, especially to readers of this journal. The BMJ web pages link to a current article stressing the importance of confidence intervals (3) in its collection of articles about randomized control trials. The conclusions of this article apply equally well to cohort studies. This idea is not new either, having been discussed in BMJ thirteen years earlier (4). In summary, two articles examine the association between an adverse event and vaccination and present relative risks of 1.06. Both papers claim lack of statistical significance, but only in the paper that uses confidence intervals can you also assess the practical significance of these findings. Steve Simon, Research Biostatistician Children's Mercy Hospital, Kansas City MO, USA Competing interests: none declared. (1) Karvonen M, Cepaitis Z, Tuomilehto J. Association between type 1 diabetes and Haemophilus influenzae type b vaccination: birth cohort study. BMJ 1999; 318: 1169-1172. (2) Henderson J, North K, Griffiths M, Harvey I, Golding J. Pertussis vaccination and wheezing illnesses in young children: prospective cohort study. BMJ 1999; 318: 1173-1176. (3) Tarnow-Mordi WO, Healy MJR. Distinguishing between "no evidence of effect" and "evidence of no effect" in randomised controlled trials and other comparisons. Arch. Dis. Child. 1999; 80: 210-211. (4) Confidence intervals rather than P values: estimation rather than hypothesis testing Martin J Gardner and Douglas G Altman BMJ 1986; 292: 746-750.
Hib vaccine & increased incidence of IDDM 14 May 1999
Heidi White Send response to journal: Re: Hib vaccine & increased incidence of IDDM	In reply to Karvonen et al1, Classen2 has questioned the way the data in this paper was analysed and presented. He highlighted the fact that in Table 2. the relative risk of IDDM was only compared between cohorts 1 & 3 (ie. those receiving no Hib vaccine vs. those receiving Hib vaccine at 24 months only), and cohorts 2 & 3 (ie. those receiving 4 doses of Hib vaccine from 3 months vs. those receiving Hib vaccine at 24 months only). Why did Karvonen et al not bring a comparison between cohorts 1 & 2 (ie. those receiving no Hib vaccine vs. those receiving 4 doses of Hib vaccine from 3 months)? Furthermore, in Figure 1 of this paper (Cumulative incidence of type 1 diabetes per 100 000 person years in Finnish children aged 10 years or under), only the data for cohorts 2 & 3 was plotted on the graph. Why was the data for cohort 1 excluded from this graph? Could it be possible that the inclusion of the data for cohort 1 on the graph would have allowed for a more direct visual comparison between cohorts 1 & 2 to have been made? And would this have then make it more difficult for Karvonen et al to convince the casual observer that there is no link between the introduction of Hib vaccine and an increase in the incidence of IDDM? The greatest increase in IDDM has occurred in children aged under 4 years (Figure 2)1, which coincides with the period when the Hib vaccine was introduced in the mid 1980's. This should raise our suspicions as to whether the Hib vaccine could be responsible for this increase. While Karvonen et al has dismissed the data as not being statistically significant, the impact on the lives of a further 58 cases per 100 000 children at the age of 10 years, who will have to learn how to deal with a life-long chronic disease such as IDDM, should not be trivialised. Further research would do well to focus on the incidence of IDDM before and after the introduction of Hib vaccination programs in other countries such as Australia, the US and the UK. Heidi White Hospital Pharmacist Lyell McEwin Health Service, Elizabeth Vale, South Australia Competing interests: none declared REFERENCES: 1 Karvonen M, Cepaitis Z, Tuomilehto J. Association between type 1 diabetes and Haemophilus influenzae type b vaccination: birth cohort study. BMJ 1999; 318: 1169-1172. 2 Classen JB. Haemophilus vaccine and increased IDDM, causal relationship likely. Electronic letter: eBMJ (7th May 1999)
A unique opportunity 29 October 1999
Stephen Katona Send response to journal: Re: A unique opportunity	EDITOR-Karvonen et al ‘s cohort study1 looking at the association between Haemophilus influenzae type b vaccination and type 1 diabetes has offered a unique opportunity. Many studies are plagued by confounding variables such as factors influencing non-acceptance of vaccination . It was pleasing to see 98% vaccination in those participating in the efficacy trial . P.Brambleby2 looked at reasons for non-immunisation in 977 children recorded as not having received the measles vaccine. Over half of their parents responded to a questionnaire , in which 27% claimed their children had already been immunised elsewhere. Perhaps these were families most fit to travel . The main reasons for non-immunisation were : a history of measles infection , personal or family fits or allergy , and a mistrust of the vaccine. Would these or similar factors have an independent association with reducing risk of developing Type 1 diabetes ? The environment that non- responders live in , may be having an enormous influence on whether we are likely to suffer from autoimmune or allergic disorders. A squeaky clean world3-4 is a new , and unnatural environment for immune systems that need constant education. Classen and Classen rightly call for Karvonen et al to make a comparison between Cohort 2 that received four doses and Cohort 1 that had none. There are other analyses that could be made. The data on births was obtained from the Finnish medical birth registry. Demographic , social and environmental information could be added to the analysis , to get a clearer idea of which factors are most important and how they may interact. Studies looking at immunological mechanisms5-7 involved will inevitably miss the full picture unless these too can be analysed with other important factors. 1 Association between type 1 diabetes and Haemophilus influenzae type b vaccination:birth cohort study. Marjatta Karvonen, Zygimantas Cepaitis,Jaakko Tuomilehto BMJ Vol318 1 May 1999 p1169-72 2 Measles immunisation non-acceptance: validation of computer- held records and raising the vaccine uptake at early school age; the Maidstone experience. Brambleby P. Hanrahan J. Public Health. 103(4):289 -94, 1989 Jul. 3 Let them eat dirt. Gary Hamilton. News Scientist 18 July 1998 4 Give us this day our daily germs. Graham A.W.Rook , John L.Stanford Immunology Today March 1998 , Vol19 , No3 , 113-6 5 Nasal immunization induces Haemophilus influenzae-specific Th1 and Th2 responses with mucosal IgA and systemic IgG antibodies for protective immunity. Kurono Y. Yamamoto M. Fujihashi K. Kodama S. Suzuki M. Mogi G. McGhee JR. Kiyono H. Journal of Infectious Diseases. 180(1):122-32, 1999 Jul. 6 Haemophilus influenzae and Streptococcus pyogenes group A challenge induce a Th1 type of cytokine response in cells obtained from tonsillar hypertrophy and recurrent tonsillitis. Agren K. Brauner A. Andersson J. Journal of Oto-Rhino-Laryngology & its Related Specialties. 60(1):35-41, 1998 Jan-Feb. 7 NKT cell cytokine imbalance in murine diabetes mellitus. Frey AB. Rao TD. Autoimmunity. 29(3):201-14, 1999. Dr Stephen Katona MRCP Immunology SpR Level 7 Department of Immunology Derriford Hospital Plymouth PL6 8DH