Richard J.Kryscio, PhD, Associate Editor, Biostatistics, Neurologykpieper@neurology.org
Submitted March 10, 2009
Editor’s Note –
Richard J. Kryscio, PhD, Associate Editor, Biostatistics
The case control study reported by Mikaeloff et al. is
negative since its primary analysis failed to demonstrate a relationship
between use of HB vaccine and the risk of CNS inflammatory demyelination.
The authors carefully examined several factors that could influence the
risk including the time interval between the last delivery of the vaccine
and the index case report, the most recently used brand of the vaccine,
and the interaction between the time interval and the last brand. None of
these and other factors produced a positive result.
Then an unspecified number of subgroups were examined. The subgroup analysis reported in Table 3 relied on one of
several possible definitions of a vaccine schedule compliant with the
guidelines for optimal immunization. For that subgroup, an interaction
was found. The most widely used vaccine, Energix, delivered at least 3
years prior to the index case was associated with a significant elevated
risk for the demyelination (1.74; 95% C.I.: 1.03-2.95). Unfortunately, the
press took advantage of this report and presented it as a definitive
finding prompting a Correspondence challenging this finding from Michel
Lievre and several of his colleagues in the Epidemiology Working Group of
the French pharmacovigilance commission.
Lievre et al. argue that this subgroup analysis could simply be an
artifact produced from multiple comparisons and it could be biased since
it based on vaccine compliers. These are legitimate concerns but Lievre et al. fail
to acknowledge that Mikaeloff et al.--in the Discussion section--admit that the
P value associated with this result could be affected by multiple looks at
the data. While an analysis based on a specific definition of a completer
could be biased, this interaction remains significant under various
definitions of a compliant subject.
The Working Group argues that the database is not
large enough to rule out a difference among the vaccine brands. The
authors likely contributed to this reaction since they state that
differential HB antigen targets and varying production processes could
explain why one vaccine is involved in a significant interaction while the
other is not. Both parties appear to be ignoring the fact that the second
vaccine examined (GenHevac B) is associated with an non-significant but
elevated risk for the interaction with the three year window (risk 1.5;
95% C.I. 0.71-3.17). The latter could be attributed to a lack of power
since the second vaccine is not nearly as popular as the first.
Finally, Lievre et al argue that the choice of a three-year
window is not justified and that the interaction between specific vaccines
and shorter windows are associated with relative risks less than one.
However, these risks do increase as the length of the window increases.
In conclusion, I believe the authors generated a hypothesis
worthy of investigation in future studies. However, this is just a
hypothesis, a yellow light at best but not a a definitive finding.
Editor’s Note – Richard J. Kryscio, PhD, Associate Editor, Biostatistics
The case control study reported by Mikaeloff et al. is negative since its primary analysis failed to demonstrate a relationship between use of HB vaccine and the risk of CNS inflammatory demyelination.
The authors carefully examined several factors that could influence the risk including the time interval between the last delivery of the vaccine and the index case report, the most recently used brand of the vaccine, and the interaction between the time interval and the last brand. None of these and other factors produced a positive result. Then an unspecified number of subgroups were examined. The subgroup analysis reported in Table 3 relied on one of several possible definitions of a vaccine schedule compliant with the guidelines for optimal immunization. For that subgroup, an interaction was found. The most widely used vaccine, Energix, delivered at least 3 years prior to the index case was associated with a significant elevated risk for the demyelination (1.74; 95% C.I.: 1.03-2.95). Unfortunately, the press took advantage of this report and presented it as a definitive finding prompting a Correspondence challenging this finding from Michel Lievre and several of his colleagues in the Epidemiology Working Group of the French pharmacovigilance commission.
Lievre et al. argue that this subgroup analysis could simply be an artifact produced from multiple comparisons and it could be biased since it based on vaccine compliers. These are legitimate concerns but Lievre et al. fail to acknowledge that Mikaeloff et al.--in the Discussion section--admit that the P value associated with this result could be affected by multiple looks at the data. While an analysis based on a specific definition of a completer could be biased, this interaction remains significant under various definitions of a compliant subject.
The Working Group argues that the database is not large enough to rule out a difference among the vaccine brands. The authors likely contributed to this reaction since they state that differential HB antigen targets and varying production processes could explain why one vaccine is involved in a significant interaction while the other is not. Both parties appear to be ignoring the fact that the second vaccine examined (GenHevac B) is associated with an non-significant but elevated risk for the interaction with the three year window (risk 1.5; 95% C.I. 0.71-3.17). The latter could be attributed to a lack of power since the second vaccine is not nearly as popular as the first.
Finally, Lievre et al argue that the choice of a three-year window is not justified and that the interaction between specific vaccines and shorter windows are associated with relative risks less than one. However, these risks do increase as the length of the window increases.
In conclusion, I believe the authors generated a hypothesis worthy of investigation in future studies. However, this is just a hypothesis, a yellow light at best but not a a definitive finding.
Disclosure: The author reports no disclosures.