Little evidence for genetic susceptibility to influenza A (H5N1) from family clustering data.

Abstract

The apparent clustering of human cases of influenza A (H5N1) among blood relatives has been considered as evidence of genetic variation in susceptibility. We show that, by chance alone, a high proportion of clusters are expected to be limited to blood relatives when infection is a rare event.

Full text

Page 1

DISPATCHES
1074 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 13, No. 7, July 2007
Little Evidenc e
for Genetic
S usc eptibility
to Infl uenza A
(H5N1) from Family
Clustering Data
Virginia E. Pitzer,* Sonja J. Olsen,†
Carl T. Bergstrom,‡ Scott F. Dowell,†
and Marc Lipsitch*
The apparent clustering of human cases of infl uenza A
(H5N1) among blood relatives has been considered as evi-
dence of genetic variation in susceptibility. We show that, by
chance alone, a high proportion of clusters are expected to
be limited to blood relatives when infection is a rare event.
S
been documented (1,2). These clusters range in size from
2 to 8 infected persons; in only 4 clusters were 2 unrelated
family members (e.g., husband and wife) infected. This pat-
tern has been considered by the World Health Organization
as evidence of genetic variation in susceptibility (3–5), but
we show this observation provides little grounds for this in-
ference. We describe a null model in which nuclear families
experience a common exposure to an avian infl uenza virus.
The observed degree of clustering in blood relatives is con-
sistent with that expected by chance alone in the absence of
genetic variation in susceptibility; other features of the data
are also consistent with the null model.
Our model assumes all persons are equally susceptible,
such that they have the same probability of infection, τ, and
ignores possible human-to-human transmission (see online
Technical Appendix, available from www.cdc.gov/EID/
content/13/7/1074-Techapp.htm). The number of infected
family members follows a binomial distribution with mean
nτ, where n is the number of exposed persons in each fam-
ily. A cluster is defi ned as a family in which >1 person is
infected; clusters are limited to blood relatives unless both
parents are infected.
We compare our model to the observation that 32 of
36 clusters that occurred from December 2003 to Decem-
ber 2006 consisted only of blood relatives (pB = 0.89, 95%
ince December 2003, 36 family clusters among 261
confi rmed human cases of infl uenza A (H5N1) have
confi dence interval 0.74–0.97; Table in online Techni-
cal Appendix). When the probability of infection is low,
most clusters consist of 2 infected family members, and by
simple combinatorics, these 2 are usually blood relatives,
which is consistent with the observed date (Figure 1).
For a given a nuclear family size, the null model also
predicts the proportion of all cases that are part of a clus-
ter and the average number of cases per cluster. Neither of
these measures follows a simple distribution; we therefore
use simulated data to determine what ranges of our param-
eters (τ and n) are consistent with the observed degree of
clustering both in families and among blood relatives. We
estimate the mean and 95% prediction intervals for the pro-
portion of cases occurring in clusters when there are 261
cases, and for the average number of cases per cluster when
there are 36 clusters. The expected proportion of cases oc-
curring in clusters is similar to the observed data when the
probability of infection is low (τ<0.15) (Figure 2). The
observed average number of cases per cluster, however,
is consistent with slightly higher probabilities of infection,
larger family sizes, or both (Figure 2).
The discrepancy between the number of cases per clus-
ter and the proportion of cases in clusters may be due to be-
tween-family variation in τ. If the probability of infection
is low for members of most exposed families and higher
for members of a few exposed families, then most cases
may come from families in which τ is low, but most of the
clusters will occur among families for which τ was higher.
This will lead to a lower proportion of cases occurring in
clusters and a higher average number of cases per cluster,
as is observed. Although it is possible that such variation
may be genetic, it could also result from between-house-
hold heterogeneity in intensity of exposure to infected birds
*Harvard School of Public Health, Boston, Massachusetts, USA;
†Centers for Disease Control and Prevention, Atlanta, Georgia,
USA; and ‡University of Washington, Seattle, Washington, USA
Figure 1. Proportion of clusters limited to blood relatives versus the
probability of infection (τ) under the null hypothesis (no variation in
susceptibility). Point estimate of the observed data is represented
by the solid black line; the shaded region represents the 95%
confi dence interval.

Page 2

Little Evidence for Genetic Susceptibility to Infl uenza
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 13, No. 7, July 2007 1075
(or intensity of shedding in birds to which different house-
holds are exposed), household hygiene, living conditions,
and the like. Human-to-human transmission of the virus
could also lead to larger than expected cluster sizes because
having >1 case(s) within a family would increase the risk of
subsequent cases occurring, and it could not be ruled out in
several clusters (6,7).
Qualitatively, the data suggest the existence of nonge-
netic, between-household variation in risk. If such nonge-
netic variation were absent, then in any given village, nearly
all pairs of cases occurring among unrelated persons in the
same village would be in different households. Roughly,
the chance that a pair of cases in unrelated persons in a vil-
lage would be from the same household as opposed to dif-
ferent households would be 1/H, where H is the number of
households in a village. With 4 pairs of cases in unrelated
persons in the same household, ≈4H pairs of cases would be
expected within a village, mostly in different households. If
the average village size of ≈138 households estimated for
an area of Thailand (8) is typical, then if members of all
households in a village were at equal risk, we would expect
to see far more pairs of unrelated cases within a village
than have actually been observed (4H ≈550 pairs of cases
in unrelated persons, which greatly exceeds the observed
261 total cases). Clearly, this argument is only heuristic,
but when this argument is combined with the likelihood of
biologic and behavioral differences between households, it
seems likely that τ would vary considerably from 1 house-
hold to another.
Furthermore, the model does not account for addition-
al individual variability in susceptibility possibly related
to age, level of exposure, or other risk factors. If younger
persons have a higher risk for infection or likelihood of
exposure, clustering would be promoted, primarily with-
in blood relatives, because siblings would be more likely
than either parent to become infected. Approximately half
of all cases have occurred in those <20 years of age (9).
Similarly, if female persons (for example) were at higher
risk for exposure, infection, or both, then clusters including
non–blood relatives (e.g., spouses) would tend to include
the low-risk sex and thus be less probable. Female persons
of ages 10–29 years were slightly overrepresented among
laboratory-confi rmed case-patients, but the difference was
not statistically signifi cant (9).
The null model presented here is not designed to cap-
ture all of the heterogeneities in exposure and complexity
of real families exposed to infl uenza subtype H5N1. Rather,
it simply illustrates that a large proportion of family clus-
ters limited to blood relatives may occur by chance in the
absence of genetic variation in susceptibility, particularly
when the probability of infection is low and family sizes
are large. Although genetic heterogeneity may possibly
contribute to the clustering of avian infl uenza cases within
blood relatives, it is neither a necessary nor the most likely
explanation for the data currently available.
This work was supported by US National Institutes of
Health grants T32 AI07535 (V.E.P.) and cooperative agreement
5U01GM076497 (Models of Infectious Disease Agent Study
[M.L.]).
Ms Pitzer is a Doctor of Science candidate in the Depart-
ment of Epidemiology at Harvard School of Public Health. Her
research interests include epidemiologic methods for investigat-
ing emerging infectious diseases.
References
1. Olsen SJ, Ungchusak K, Sovann L, Uyeki TM, Dowell SF, Cox NJ,
et al. Family clustering of avian infl uenza A (H5N1). Emerg Infect
Dis. 2005;11:1799–801.
2. World Health Organization. Cumulative number of confi rmed hu-
man cases of avian infl uenza A/(H5N1) reported to WHO. 2006.
[cited 2007 Apr 12]. Available from http://www.who.int/csr/disease/
avian_infl uenza/country/cases_table_2006_12_27/en/index.html
Figure 2. Relationship between
data simulated under the null model
and the observed pattern of family
clustering for A) the proportion of
cases occurring in clusters (given
261 total cases) and B) the average
number of cases per cluster (given
36 clusters). Estimates of the mean
are represented by solid lines;
the shaded regions between the
dotted lines show 95% prediction
intervals for 1,000 simulations. The
observed data are represented by
the solid black lines.

Page 3

DISPATCHES
1076 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 13, No. 7, July 2007
3. World Health Organization. Infl uenza research at the human and ani-
mal interface. 2006. [cited 2006 Nov 3]. Available from http://www.
who.int/csr/resources/publications/influenza/who_cds_epr_gip_
2006_3c.pdf
4. Perdue ML, Swayne DE. Public health risk from avian infl uenza
viruses. Avian Dis. 2005;49:317–27.
5. Beigel JH, Farrar J, Han AM, Hayden FG, Hyer R, de Jong MD, et
al. Avian infl uenza A (H5N1) infection in humans. N Engl J Med.
2005;353:1374–85.
6. Ungchusak K, Auewarakul P, Dowell SF, Kitphati R, Auwanit W,
Puthavathana P, et al. Probable person-to-person transmission of
avian infl uenza A (H5N1). N Engl J Med. 2005;352:333–40.
7. Butler D. Family tragedy spotlights fl u mutations. Nature.
2006;442:114–5.
8. Longini IM Jr, Nizam A, Xu S, Ungchusak K, Hanshaoworakul W,
Cummings DA, et al. Containing pandemic infl uenza at the source.
Science. 2005;309:1083–7.
9. Epidemiology of WHO-confi rmed human cases of avian infl uenza A
(H5N1) infection. Wkly Epidemiol Rec. 2006;81:249–57.
Address for correspondence: Virginia E. Pitzer, Department of
Epidemiology, Harvard School of Public Health, 677 Huntington Ave,
Boston, MA 02115, USA; email: vpitzer@hsph.harvard.edu