Partial protection of seasonal trivalent inactivated vaccine against novel pandemic influenza A/H1N1 2009: case-control study in Mexico City

RE: Partial protection of seasonal trivalent inactivated vaccine
against novel pandemic influenza A/H1N1 2009: case-control study in Mexico
City BMJ 2009;339:b3928

The effectiveness reported by Garcia-Garcia et al. for 2008-09
trivalent influenza vaccine (TIV) against pandemic A/H1N1 (pH1N1) virus
(73%; 95%CI 34%-89%) is remarkable. It is not much different from vaccine
effectiveness (VE) estimates of ~80% during seasons when TIV is closely
matched to circulating strains and far exceeds VE estimates of ~50% during
seasons of minor TIV mismatch to closely-related drift variants[1,2]. How
then to reconcile such dramatic cross-protection for TIV against pH1N1
virus when it shares so much less in common antigenically with TIV
components? The clue to explaining these results may be found in the
second set of paradoxical results reported by the same study: participants
with chronic conditions appear at 80% (95% CI 55%-91%) reduced risk of
pH1N1 hospitalization compared to healthy individuals –a finding that
lacks coherence with known risk factors for severe influenza outcomes and
signals methodological concerns.

This case-control study was conducted at the National Institute of
Respiratory Diseases in Mexico City, a specialty clinic that accepts
referrals and also provides primary and hospital care to the local
community. Although the 60 cases and 180 controls came from the same city
and Institute, there is evidence that they emerged from different source
populations. Cases more often presented directly and with severe disease
requiring hospitalization (98% vs. 34%). Controls were more often seen
following specialty referral. As such, hospitalized cases more closely
reflect immunization rates in the community whereas controls reflect
referrals with higher health care contact and greater immunization
opportunities.

A significantly higher proportion of controls (65%) compared to cases
(25%) had chronic conditions. Individuals with comorbidity are a priori
more likely to receive TIV under the kind of high-risk influenza
immunization program administered in Mexico. The types of chronic
conditions also differed between cases and controls: 28% of controls had
asthma compared to 3% of cases; whereas 3% of controls compared to 12% of
cases had diabetes. Asthmatics are more likely to be referred to a
specialty respiratory clinic and may also have higher likelihood of TIV
immunization than diabetics. The proportion of young adult controls with
chronic conditions in this study (57%) was 10-fold higher than population
estimates of ~5% for Mexico City[3]. The proportion immunized among
controls (29%) was also >3-fold higher than expected based on TIV doses
distributed per capita in Mexico (94/1000 or 9%)[4]. Selection bias
driving high rates of chronic conditions among controls would have
resulted in the paradoxical impression that both underlying comorbidity
and prior TIV immunization provide pH1N1 protection. Comparison of crude
immunization rates between cases (13%) and the general population (9%)
gives a different impression of negative rather than protective vaccine
effect and adjusted case-cohort analyses based on community profiles may
be worth further examination.

Authors do not present the timing of recruitment in relation to
immunization efforts. Within a few weeks of recognizing the epidemic in
Mexico, the Mexican Health Minister and other health authorities strongly
recommended seasonal vaccine for those not yet immunized[5,6]. This led to
highly publicized queues for influenza vaccine in Mexico City. This
introduces further potential bias related to non-contemporaneous controls.
If recruited cases had developed their severe illness earlier in the
outbreak they would have lost subsequent opportunities for immunization
that controls then accrued following public pronouncements. Proxy
respondents were also used in this study to ascertain immunization status
for some participants. The distribution of proxy-elicited TIV status was
not reported, but since cases more often than controls were ventilated
(43% vs. 2%) and died (30% vs. 1%), it is likely that ascertainment of
immunization status for cases relied more on proxies with imperfect recall
compared to self-report by controls. This could have contributed to
higher immunization rates among controls relative to cases and further
exaggerated the impression of vaccine benefit.

Finally, authors report study design based on frequency matching
using broad categories of age and socio-economic status yet also report
data analysis based on conditional logistic regression (CLR). CLR is the
wrong analytical approach for adjustment when frequency (rather than
paired) matching has been used[7]. Age exerts a strong influence on both
likelihood of immunization and risk of pH1N1 illness. Failure to
appropriately adjust for this confounder overall or in sub-analyses of
hospitalized or healthy participants would also be critical to consider
had not other biases already superseded this concern. Ultimately, no
analytical technique –not restriction, stratification or adjustment -can
overcome selection bias. Based on conspicuous evidence in the data
presented, we conclude that this study’s statistically significant but
implausible findings of >70% TIV protection against pH1N1 illness and 5
-fold reduced risk of hospitalization in people with chronic conditions
are best explained by selection bias exacerbated by a combination of
recall bias and inadequate adjustment for confounding.

Submitted by:

Naveed Z. Janjua MD, MSc, DrPH

Danuta M. Skowronski MD, MHSc, FRCPC

Travis S. Hottes MSc

BC Centre for Disease Control
Vancouver, British Columbia, Canada

Gaston De Serres MD, PhD

Institut national de santé publique du Québec

Natasha S. Crowcroft MA MSc MD(Cantab) MRCP FFPH

Laura C. Rosella PhD, MHSc

Ontario Agency for Health Protection and Promotion
Toronto, Ontario, Canada

References:

1) Jefferson T, Rivetti A, Harnden A, Di Pietrantonj C, Demicheli V.
Vaccines for preventing influenza in healthy children. Cochrane Database
Syst Rev. 2008 Apr 16;(2):CD004879.

2) Jefferson TO, Rivetti D, Di Pietrantonj C, Rivetti A, Demicheli V.
Vaccines for preventing influenza in healthy adults. Cochrane Database
Syst Rev. 2007 Apr 18;(2):CD001269.

3) Fedson D. Increasing the overall epidemic vaccination coverage:
The macroepidemiology of Influenza vaccination. Third European Influenza
Conference Vilamoura 14-17 September 2008.
www.flucentre.org/files/fedson.ppt Accessed: October7, 2009.

4) Kuri-Morales P, Emberson J, Alegre-Díaz J, Tapia-Conyer R, Collins
R, Peto R, Whitlock G. The prevalence of chronic diseases and major
disease risk factors at different ages among 150,000 men and women living
in Mexico City: cross-sectional analyses of a prospective study. BMC
Public Health. 2009 Jan 9;9:9. http://www.biomedcentral.com/1471-2458/9/9

5) Lacey M, McNeil Jr. DG. Fighting Deadly Flu, Mexico Shuts Schools.
New York Times. April 24, 2009.
http://www.nytimes.com/2009/04/25/world/americas/25mexico.html . Accessed
October 7, 2009.

6) Randewich N, Tovar A. Mexico closes schools, begins emergency
vaccination program after swine flu outbreak. Agence France-Presse and
Canwest News Service. Ottawa Citizen. April 24, 2009. Accessed October 7,
2009.
http://www.ottawacitizen.com/health/Mexico+closes+schools+begins+emergen...

7) Greenland S, Schwartzbaum JA, Finkle WD. Problems due to Small
Samples and Sparse Data in Conditional Logistic Regression Analysis Am. J.
Epidemiol. 2000;151:531 - 539.

Competing interests:
Danuta M. Skowronski has previously (>3 years ago) received research grant funding from GSK and Sanofi-Pasteur. Gaston De Serres has received research grant funding from GSK and Sanofi-Pasteur. No other authors have conflicts of interest to declare.