Volume 136

Number 2

July 15,1992

American Journal ofEPIDEMIOLOGYCopyright © 1992 by The Johns Hopkins University

School of Hygiene and Public Health

Sponsored by the Society for Epidemiologic Research

REVIEWS AND COMMENTARY

Confounding in Studies of Adverse Reactions to Vaccines

Paul E. M. Fine1-2 and Robert T. Chen1

Several social and medical attributes are associated with both avoidance or delay ofvaccination and an increased risk of adverse events such as sudden infant deathsyndrome or childhood encephalopathy. Studies that fail to control adequately for suchconfounding factors are likely to underestimate the risks of adverse events attributableto vaccination. This paper reviews the literature on studies of severe adverse eventsafter the administration of pertussis antigen-containing vaccines, with particular attentionto the measures taken by different investigators to avoid this problem. Most publishedstudies have reported a deficit of sudden infant death syndrome among vaccinees,which may reflect confounding in their study designs. An expression is derived toexplore the extent of underestimation that may be Introduced in such studies, underdifferent sets of conditions. Confounding of this sort is a general problem for studies ofadverse reactions to prophylactic interventions, as they may be withheld from someindividuals precisely because they are already at high risk of the adverse event. Am JEpidemiol 1992; 136:121 -35.

confounding factors (epidemiology); immunization; pertussis vaccine; sudden infantdeath; vaccination

Immunization programs are undeniablyamong the most effective public health in-terventions. Reductions over recent decadesin the morbidity and mortality attributableto smallpox, measles, polio, diphtheria,whooping cough, and tetanus are eloquentreminders of this fact (1). However, the very

success of these programs brings new prob-lems. No intervention is entirely withoutrisk, and even very rare adverse reactions toa vaccination increase in importance as thetarget disease itself disappears.

Changes in the perception of risks attrib-utable to vaccination, compared with those

Received for publication September 3, 1991, and Infinal form January 17, 1992.

Abbreviations' DPT, diphtheria, pertussis, and tetanusvaccine; DT, diphtheria and tetanus vaccine; SIDS, suddeninfant death syndrome.

1 Division of Immunization, National Center for Preven-tion Services, Centers for Disease Control, Atlanta, GA.

2 Present address' Communicable Disease Epidemiol-ogy Unit, Department of Epidemiology and Population

Sciences, London School of Hygiene and Tropical Medi-cine, Keppel Street, London WC1E 7HT, United Kingdom.

Reprint requests to Information Services (MS-E06), Na-tional Center for Prevention Services, Centers for DiseaseControl, Atlanta, GA 30333.

The authors thank Dr Edgar Marcuse, Dr. Kim Wentz,Dr. Edward Mortimer, Dr. Walter Orensteln, and Dr. AlanHinman for their comments on drafts of the manuscript.

121

122 Fine and Chen

attributable to natural disease, are of im-mense importance to vaccination programs.Recognition of these changes within the sci-entific community led to termination ofsmallpox vaccination in many countriesprior to the global elimination of disease.Such recognition also is now the basis forreconsideration of polio vaccination strate-gies (2, 3). The public perception of suchchanges led to dramatic declines in the up-take of pertussis vaccination during the1970s in the United Kingdom and Japan (4,5). Similar concerns in the United Stateshave led to a large number of lawsuits, asubstantial rise in vaccine prices (6), andnew legislation governing reporting andcompensation of adverse events (7). Givensuch issues, one sees an obvious need forcontinued monitoring of vaccine safety toassist policymakers in assessing needs forimprovements in vaccine preparations or forchanges in vaccination strategy.

The monitoring of vaccine safety may bebased on either active or passive ascertain-ment of adverse events (8). To assesswhether such events are, in fact, attributableto vaccination, the investigator may use twosorts of approach. The first involves cohortlogic, i.e., the comparison of incidence ratesof the event in question between cohorts ofvaccinated and unvaccinated individuals. Ifthere are very few unvaccinated individualsin the population, then the comparison maybe between (age-specific) rates of events be-fore and at successive intervals after vacci-nation. The alternative approach involvesthe application of case-control logic, i.e.,comparisons of the frequency of a history ofrecent vaccination between individuals ex-periencing adverse events and appropriatecontrols.

Regardless of the approach used, suchstudies face several methodological difficul-ties (9, 10). Many potential sources of biashave been identified. Prominent amongthese is the problem of ensuring that adverseevents are ascertained independently of vac-cination history. Failure to control for thisfactor may lead to creation, or overestima-tion, of an association between a vaccineand an adverse event. Another problem is

that of confounding between the risk factor(vaccination) and outcome measure (ad-verse event) of interest. Many factors knownto be associated with either avoidance ordelay of vaccination may themselves be as-sociated with an increased risk of adverseevent-type medical outcomes. As an illustra-tion, table 1 presents reported risk factorsfor sudden infant death syndrome (SIDS) andfor childhood encephalopathy, on the onehand, and for failure to receive diphtheria-pertussis^-tetanus (DPT) vaccination on theother (11-22). The close correspondence be-tween these sets of factors, which includemedical contraindications and social corre-lates of low vaccine coverage, suggests thatindividuals predisposed to either SIDS orencephalopathy are relatively unlikely to re-ceive DPT vaccination. Studies that do notcontrol adequately for this form of "con-founding by indication" (23) will tend tounderestimate any real risks associated withvaccination.

This paper examines the influence of suchconfounding on vaccine adverse event stud-ies by reviewing the literature to illustrate itspresence and by modeling to demonstrateits impact under different sets of conditions.

REVIEW OF THE LITERATURE

Published studies that are relevant to theproblem of confounding between risk factorsfor DPT vaccination and for potential ad-verse events are summarized in table 2. Thisreview does not cover reports of cases orclusters of time-associated adverse events(24, 25), as these are not likely to be repre-sentative and they provide no means to eval-uate the confounding problem that is thefocus of this paper.

Studies of DPT and SIDS

The first published controlled investiga-tion of the relation between DPT and SIDSwas a case-control study by Taylor andEmery in 1982 (26), who reported that eight(31 percent) of 26 SIDS cases had ever re-ceived DPT or DT vaccine compared with27 (52 percent) of 52 age-and area-matched

Confounding and Adverse Reactions to Vaccines 123

TABLE 1. Known risk factors for failure or delay in receiving vaccines (left) and for sudden infant deathsyndrome (SIDS) or for encephalopathy (right), which may act as confounders in studies of adversereactions to vaccination*

Attributes reported as risk factors for fafejre or delay(D) In receiving vaccines

Attributes reported as risk factors for SIDS (S)or for encephalopathy (£)

Social factorsLow parental education (11 -13,18)tLarge family size (11,12,15,16)Non-wtiite race (17)Single parent (12)Low socioeconomic status (11,15,18, 22)No health insurance (12, 17, 18)f

Urban (15,18)

Low parental education (S) (14)High parity (S) (14)Black race (S) (14)

Young mother (12, 22)Maternal smoking (15)Low birth weight (D) (19)Preterm infant (0) (19)Lack of prenatal care (12)

Maternal and birth factorsYoung mother (S) (14)Smoking in pregnancy (S) (14)Low birth weight (S) (14)

Infant medical historyHistory of acute illness (bronchitis, pneumonia) History of vomiting (S) (14)

(15) No well-baby visits (S) (14)Febrile illness (20, 21)Evolving neurologic disorder (20, 21) Evolving neurologic disorder (E) (16)

Conditions predisposing to seizures (20, 21) Conditions predisposing to seizures

Behavioral factorsAggressive child (15)Solitary child (15)

* Factors that appear on both sides of trie table are printed in boM.t Numbers in parentheses, reference^).

controls (odds ratio = 0.41). Except for thematching of controls, no attempt was madeto overcome confounding by factors predis-posing to vaccination or to SIDS in thisinvestigation.

The following year, Baraff et al. (27) re-ported data on the time interval betweenDPT vaccination and the death of 27 SIDScases who had received DPT vaccine within28 days prior to death. A significant excessof deaths was noted within 24 hours (ob-served = 6; expected = 0.96; p < 0.005) andwithin 7 days (observed = 17; expected =7.72; p < 0.05) of vaccination. Subsequentcorrespondence discussed the potential forselection, recall, and observer bias in thisstudy and raised the possibility that the as-sociation might have been due in part to thesimilarity in age trends between SIDS inci-dence and DPT vaccination (28). The paper

also included an analysis of intervals be-tween visits to physicians and death for 40SIDS cases reported to have sought medicalcare (but not received vaccination) within28 days prior to death. There was an excessof visits within 7 days, which may havereflected prodromal symptoms associatedwith the subsequent deaths of these children.Given that some of these symptoms mayhave been interpreted as contraindicationsto vaccination, we again see evidence of theconcordance of risk factors summarized intable 1.

Results of the largest investigation of therelation between DPT and SIDS were re-ported by Hoffman et al. in 1987 (14). Thesewere based upon a multicenter case-controlstudy comparing risk factors in 757 SIDScases with those in randomly selected livingcontrols matched for birthplace and age

TAB

LE 2

. S

umm

ary

of p

ublis

hed

Inv

estig

atio

ns o

f D

PT

* va

ccin

es a

s ri

sk f

acto

rs f

or S

IDS

* an

d f

or e

nce

ph

ato

pat

hie

s/se

vere

neu

rolo

gic

dis

ord

ers

Aut

hor

Res

ults

Inve

stig

atio

n m

etho

dO

R*,

t9

5%

conf

iden

ceIn

terv

alS

tudy

in

terv

al

ro 31 CD § a. O CD

3

Stu

dies

of

SID

ST

aylo

r an

d E

mer

y (2

6)§

Bar

aff

et a

l. (2

7)

Hof

fman

eta

). (

14)

Wal

ker

et a

l. (2

9)

Grif

fin e

t al

. (30

)

Stu

dies

of

ence

phal

opat

hles

/ser

ious

neu

rolo

gic

diso

rder

sM

iUer

et

al. (

31)

Sim

ple

case

-con

trol

stu

dy c

ompa

ring

26 c

ases

with

52

age-

and

are

a-m

atch

ed c

ontr

ols

Cas

e se

ries

of in

terv

al s

ince

last

vac

-ci

natio

n am

ong

27 c

ases

Mul

ticen

ter

case

-con

trol

stu

dy c

om-

parin

g 75

7 ca

ses

with

equ

al n

um-

bers

of

age-

and

sex

- an

d of

age

-,se

x-,

and

area

-mat

ched

con

trol

sC

ase-

cont

rol

usin

g lin

ked

data

bas

e,co

mpa

ring

29 c

ases

with

262

con

-tr

ols.

Exc

lude

d ch

ildre

n w

ith l

owbi

rth

wei

ght,

"pre

disp

osin

g m

edic

alco

nditi

ons"

Coh

ort

anal

ysis

of

129,

834

vacc

i-na

ted

child

ren

in T

enne

ssee

. Rel

a-tiv

e ris

k ca

lcul

ated

with

ref

eren

cepe

riod

>3

0 da

ys p

ost

vacc

inat

ion

Cas

e-co

ntro

l st

udy

com

pann

g 90

4ca

ses

with

age

-, s

ex-,

and

are

a-m

atch

ed c

ontr

ols

0.41

0.69

0.6

0.16

[5.4

] I (

29)

0.54

, 0.

580

.19

,0.4

6

0.15

3.5

[0.1

8][0

.17]

[0.7

5][1

.0]

2.3

-13

0.0

5-0

.45

1.2

-9.9

0.0

4-0

.80

.04

-0.7

0.4

-1.5

0.6

-1.6

Eve

r<

3d

ay

s3

-28

days

>2

8 da

ys3

days

Eve

r1

day

Eve

r3

days

0-3

day

s4

-7 d

ays

8-1

4 da

ys1

5-3

0 da

ys

4.71

1.1

-28

.00

-7 d

ays

Confounding and Adverse Reactions to Vaccines 125

•D

3

tv. op

in (oCNJ CO

<D

I•a

<D

C

CD

•aOD

it

- S

-t— -H-«» — •-

(control group A) or for birthplace, age, race,and birth weight (control group B). Overall,SIDS cases were less likely to have receivedDPT (or any vaccine) than were theirmatched controls (odds ratio = 0.54 (controlgroup A) and 0.58 (control group B)). Thesignificant negative association betweenprior DPT vaccination and SIDS was main-tained in multiple logistic analysis control-ling for 11 other factors: birth weight, sex,race, parity, maternal age, maternal educa-tion, smoking during pregnancy, alcoholconsumption during pregnancy, use of pre-natal care, prepregnancy weight, and preg-nancy weight. Case children were less likelythan were controls to have had postnataloutpatient visits, but more likely to have hadsick visits; no attempt was made, however,to control for these factors in the analyses.The negative association between vaccina-tion and SIDS was strongest when analyseswere restricted to vaccination within 24hours of death (crude odds ratio = 0.19(control group A) or 0.46 (control group B)).The authors concluded, "DPT immuniza-tion does not appear to be a significant factorin the occurrence of SIDS" (14, p. 610).

A smaller case-control study based onlinked data was reported by Walker et al.(29). These authors compared 29 SIDS caseswith 262 age-matched controls drawn fromlinked vaccination and mortality records of26,500 children registered between 1972 and1983 with the Group Health Cooperative ofPuget Sound. SIDS was defined as " . . . anydeath for which no cause could be discernedamong infants of normal birth weight(>2,500 g) and without predisposing medi-cal conditions . . . " (29, p. 945). The criteriafor "predisposing medical conditions" werenot stipulated in detail, but led to the exclu-sion of two children with "life-threateningmedical conditions" (29, p. 950). Such ex-clusions represent an effort to control forconfounding in the design of this study andshould have compensated to some degreefor the concordance of risk factors illustratedin table 1. These authors found a negativeassociation between SIDS and a history ofhaving ever received DPT (odds ratio =

126 Fine and Chen

0.15). On the other hand, when nonimmun-ized children were excluded from analysis,detailed breakdown by successive intervalsbetween DPT vaccination and death sug-gested that the daily mortality risk in theperiod 0-3 days after vaccination (fourdeaths observed) was 7.5 (95 percent confi-dence interval 1.7-31) times greater thanthat during the period more than 30 daysafter vaccination (nine deaths observed).

Griffin et al. (30) linked birth, death, andimmunization records in Tennessee in orderto follow up 129,834 infants who were bornover the years 1974-1984 and recorded ashaving received at least one dose of DPTvaccine. Sudden infant death was reportedin 109 of these children between the ages of29 days and 1 year of life. Cohort logic wasused in order to calculate the relative risksof SIDS in successive intervals after receiptof DPT vaccine, compared with the risk ofSIDS occurring more than 30 days aftervaccination. A clear gradient in relative riskwas observed, from a low of 0.2 during thefirst 72 hours after vaccination to unity forthe period 2 weeks or more after vaccina-tion. The trend remained when controlledfor age, sex, race, year, birth weight, andMedicaid enrollment. The authors inter-preted the finding as follows: "The mostplausible explanation for the decreased rateof SIDS in the period immediately afterimmunization is that children may be im-munized when they are in better health andthat this healthier state is associated with alower risk of SIDS" (30, p. 621). The authorsthen attempted to evaluate the potential im-pact of such confounding on their investi-gation, noting that other studies had shownthat

nearly half of all children who die of SIDShave either no symptoms or very minorones before death. Therefore, these studiessuggest that selective immunization ofasymptomatic cohort children could atmost account for a 50% decrease in therate of SIDS after immunization in thisstudy, but that the decrease could not be ofsufficient magnitude to mask a true in-crease in the incidence of SIDS after im-munization (30, p. 622).

This statement implies two things. First,even if more than half of the children whodied of SIDS had prior symptoms that mighthave rendered them ineligible for vaccina-tion shortly before death, such selection"could" still only have reduced the observedrelative risk by a maximum of 50 percent,at least under the conditions of their study.Second, their finding of a relative risk of0.18 (see table 2) was therefore incompatiblewith a true relative risk greater than unity.We will return to the logic of this argumentbelow.

In summary, we see that all investigatorshave found that SIDS cases are less likely tohave ever been vaccinated than are livingage-matched controls. On the other hand,analyses of time intervals between DPT vac-cination and SIDS have shown a deficit ofdeaths shortly after vaccination in somestudies (14, 26, 30) and an excess of suchdeaths in others (27, 29). The two positiveshort-interval associations were based uponsmall numbers (27 and 29 total cases) andmay have been due in part to the fact thatthe peak age distribution of SIDS coincideswith the recommended onset of DPT vac-cination. This was exacerbated by the use oftime more than 30 days after vaccination asthe reference period, as this extends into agesof low background risk.

All in all, the negative associations be-tween DPT vaccination and SIDS are im-pressive. None of the investigators citedabove has suggested that these findingsmight be due to DPT's being protectiveagainst SIDS, and several have noted thatthe findings are probably attributable to thefact that risk factors for SIDS are similar tofactors known to be associated with eitheravoidance or delay of vaccination (e.g., table1). The negative associations between SIDSand having ever been vaccinated reflectavoidance of vaccination. On the otherhand, the negative associations betweenSIDS and having recently been vaccinatedcould reflect either avoidance or delay ofvaccination by those predisposed, for onereason or another, to die of SIDS.

Confounding and Adverse Reactions to Vaccines 127

Studies of DPT and encephalopathy

The British National Childhood Enceph-alopathy Study represents the largest con-trolled study of encephalopathy and DPTvaccination thus far carried out. It also in-cludes the most thoughtful discussion in theliterature on the issue of confounding be-tween factors predisposing to both avoid-ance of vaccination and the adverse reactionunder study (16, 31).

The British National Childhood Enceph-alopathy Study was designed as a case-controlstudy comparing detailed vaccination histo-ries of more than 1,000 encephalopathycases with those of controls (two per case)matched for sex, date, and area of birth.Significant associations were revealed be-tween encephalopathy and receipt of DPTvaccine less than 7 days before onset ofillness or between encephalopathy and re-ceipt of measles vaccine within 7-14 daysprior to onset of illness, but no associationwas detected with prior DT vaccination.Many aspects of this study, in particular,biases that may have been introduced by themethod of case ascertainment, have beendiscussed extensively in the literature (e.g.,10).

The authors explored the potential forconfounding in four ways. First, they re-stricted their most rigorous analyses to thosecases who had no evidence of neurologicabnormality prior to onset of the encepha-lopathy. This should have controlled formost neurologic factors (except for febrileconvulsions, which were not treated as priorneurologic abnormalities) that might haveserved as contraindications for vaccination.Second, they carried out separate analysesexcluding all cases and controls with a pre-vious history of fits (again in an effort tocontrol for factors that might have influ-enced both the risk of encephalopathy andthe propensity to be vaccinated). Third, theycarried out a separate analysis, matching forsocial class (manual vs. nonmanual occu-pation of the head of the family). The sig-nificant association remained, leading theauthors to comment, "There is, therefore,

no evidence that correcting for the effect ofsocial class eliminates or diminishes the sig-nificant association demonstrated betweenserious neurological disorder and immuni-zation against pertussis, or that social classis a significant confounding variable" (16, p.132). Finally, the authors considered

other possible confounding variables...such.. . as past family and personal medi-cal history, and other environmental con-ditions. For these, or any other factor, tocause significant bias in the calculations ofrelative risk they would need to operatepowerfully and consistently in one direc-tion, to be specific for one vaccine (DPT)and not another (DT), and to concentratetheir influence on the observed associationsover relatively short time intervals beforeonset which differed between vaccines(DPT and measles). It seems highly im-probable that all of these criteria would besatisfied by any of the confounding vari-ables postulated in this Study (16, p. 132).

The authors of the National ChildhoodEncephalopathy Study were concernedwhether confounding factors might havebeen responsible for creating the observedsignificant association between DPT vacci-nation and encephalopathy. Given that allof the factors listed in table 1 would beexpected to reduce rather than to create suchan association, the conclusion of the Studyof a significant association between recentDPT vaccination and encephalopathy doesnot appear to be threatened by any failureto control for additional factors that relateto both the propensity for (avoidance of)vaccination and the risk of encephalopathy.Indeed, as might have been predicted, con-trolling for previous neurologic status, priorhistory of fits, and social class led to in-creases in the estimated relative risks, theonly exception being in a subanalysis of onesocial class group (manual), for which theestimated relative risks associated with DPTremained virtually unchanged.

Three other investigations of the relationbetween DPT vaccination and encephalo-pathy or serious neurologic illness have nowappeared. Both Walker et al. (32) and Griffinet al. (33) have extended their studies ofDPT and SIDS to include encephalopathies.

128 Fine and Chen

TABLE 3. Summary of variable definitions Inalgebraic argument to predict the extent of biasattributable to confounding

S = Risk erf SIDS* In unvaccinated children who lack thecontraindication

fl - True relative risk of SIOS associated wfth vaccinationD - Relative risk of SIDS associated with the contraindicationC m Proportion of chidren with the contraindicationV - Proportion vaccinated among chidren without the con-

traindicationP - Proportion vaccinated among chidren with the contrain-

dication

* SIDS, sudden infant death syndrome.

Neither found any evidence of an associa-tion with DPT vaccination, but the numbersof cases were small and none had recentlyreceived DPT, which may reflect avoidanceof vaccination by children at risk. In addi-tion, a preliminary report (34) has appeared,describing a major case-control study ofacute, serious, neurologic diseases of chil-dren in Oregon and Washington states inthe United States. Matched-set analysis ofthe first 100 severe cases revealed an oddsratio of 2.5 (95 percent confidence interval0.7-9.3) with a history of DPT vaccinationwithin the previous 7 days. Adjustment forseveral factors that might be related to vac-cine avoidance (personal or family historyof seizures, prior DPT reaction, and illnesswithin 30 days) led to an increase in the oddsratio to 3.6 (95 percent confidence interval0.8-15.2), although the relation was still notstatistically significant. Once again, we seeevidence of confounding and must askwhether the adjustment actually carried outhas removed the effects entirely.

THEORETICAL ARGUMENT

The extent of bias introduced by con-founding will be a function of several vari-

ables. In order to explore the quantitativeimplications of these variables, we beginwith the following definitions (table 3), usingDPT and SIDS as an example.

• S is the risk of SIDS in unvaccinatedchildren who lack the contraindication tovaccination. (It should be noted that werefer to "contraindication" here to exem-plify any factor associated with avoidanceor delay of vaccination.)

• R is the true relative risk of SIDSassociated with vaccination.

• D is the relative risk of SIDS associatedwith the contraindication.

• C is the proportion of children withthe contraindication.

• V is the proportion vaccinated amongchildren without the contraindication.

• P is the proportion vaccinated amongchildren with the contraindication.

Using these definitions, we can calculate theexpected risk of SIDS in different segmentsof the child population, as shown in table 4.It should be noted that these predictionsassume that the risks of SIDS associated withvaccination and with the contraindicationare independent and, thus, the risk of SIDSamong children who are vaccinated despitehaving the contraindication is R * D timesthat in unvaccinated children who lack thecontraindication. Given these expressions,we can estimate what would be the observedrelative risk of SIDS associated with vacci-nation, if an investigation were to take noaccount of the potential confounding bycontraindication (i.e., no appropriatematching or stratification). With cohortlogic, the observed relative risk would bea(c + d)/c(a + b), using conventional defini-tions for the cells of table 5. In a case-control

TABLE 4. Proportional breakdown of population by contraindication status, vaccination status, and suddenInfant death syndrome, using variable definitions from table 3

Without contraindicationVaccinatedUnvaccinated

With contraindicationVaccinatedUnvaccinated

Proportion withSIDS'

V(1 - C)RS(1 - VX1 - C)S

PCRDS(1 -P)CSD

Remaining population

V(1 - CX1 - RS)(1 - VX1 - CX1 - S)

PC(1 - RDS)(1 - P)C(1 - SD)

Total

V(1 - C)(1 - VX1 - C)

PC(1-P)C

" SIDS, sudden infant death syndrome.

Confounding and Adverse Reactions to Vaccines 129

TABLE 5. A 2 x 2 table derived from table 4 to show the relation between vaccination status and suddenInfant death syndrome (SIDS), Independent of contraindication status, and analogous standard 2 x 2 tableas basis for relative risk and odds ratio expressions

VaccinatedUnvaccinated

VaccinatedUnvaccinated

Proportion with SIDS

V(1 - C)RS + PCRDS

(1 - VX1 ~C)S +(1 - P)CSD

ac

a + c

Remaining population

Derived 2x2 table

V(1 - CX1 - RS) + PC(1 - RDS)

(1 - VX1 - CX1 - S) + (1 - P)C(1 - SD)

Standard 2x2 tablebd

b + d

V(1 -( 1 - 1 /

Total

U) + ru

X1 - C) +1

a + bc + d

a + b + c

;i - PJC

+ d

study, the odds ratio (ad/be) should give aclose approximation of the relative risk,given that the adverse event is rare (i.e., aand c are small).

We explore the implications of theseexpressions under two general sets of cir-cumstances. The first relates to probabilitiesof vaccination and of SIDS such as wouldaccumulate over a year (analogous to studiesthat have used a history of having ever beenvaccinated as the risk factor). In this long-term case, the (annual) risk of SIDS may beon the order of S = 0.001 (35), and theoverall proportion vaccinated at least oncemay be on the order of V = 0.7-0.9. Thesecond uses parameter levels such as mightarise in short-term studies that examine therisk of SIDS within 1 day or 1 week ofvaccination. In this case, the risk of SIDSwill be small, on the order of S = 3 * 10"6

per day or 2 * 10~5 per week, and the prob-ability of vaccination will also be small, onthe order of V = 0.01 per day or 0.07 perweek.

Figures 1 and 2 present the ratios betweenthe observed and the "true" relative risks ofSIDS, associated with vaccination, undereach of these circumstances, and given dif-ferent sets of assumptions as to the values ofthe several parameters. Although risk factorssuch as those listed in table 1 are unlikely tobe associated with relative risks (D) greaterthan 10, D = 30 is included for sensitivityanalysis to examine the impact of extremevalues.

An interesting feature of this relation be-tween observed and true relative risks is its

independence of R (the true relative risk)and 5" (the background risk of SIDS in thepopulation). The magnitude of the bias is afunction of the degree to which the contrain-dications are observed (i.e., the ratio V/P),as this determines the proportions with con-traindications and, hence, the risks of ad-verse events in the vaccinated and unvacci-nated populations. The lower the proportion(P) vaccinated among those with "contrain-dications" (i.e., the greater the extent towhich contraindications are observed bythose responsible for vaccination), thegreater will be the bias in a study that doesnot control for these factors. Under both thelong- or short-term assumptions, we see thata substantial bias in estimating R can occur,given levels of D greater than 10 and preva-lences of the contraindication (C) greaterthan 1 percent.

Table 6 illustrates the implications of var-ious combinations of variables for the ob-served relative risks of SIDS associated withthe vaccination, as a function of the truerelative risks and the observed proportion ofSIDS cases who have the contraindication.We see that it is possible for the observedrelative risk of SIDS associated with vacci-nation to be less than half the true relativerisk, even if half the children with SIDS havecontraindications.

DISCUSSION

Review of the literature on SIDS, enceph-alopathies, and DPT suggests that a largenumber of factors are associated with both

130 Fine and Chen

1.2 r

1 0.8tr

0.6£

£0.4

0.2

0%

1.2

£ 0.8

S. 06

0.4

0.2

0%

1 % 2 % 3% 4% 5% 6% 7%% Population with Contraindication (C)

8% 9% 10%

1% 2 % 3% 4 % 5% 6% 7%% Population with Contraindication (C)

D=»1 D-6 D=10

8%

[X30

9% 10%

FIGURE 1. Ratio of observed to true relative risk of sudden Infant death syndrome (SIDS) associated withvaccination, as a function of the percentage of children with contraindication to vaccination (C) and the relative riskof SIDS associated with vaccination (D). These graphs show the impfications of different proportions ever vaccinatedamong infants without (V) and with (P) contraindications. Top, V - 0.7, P = 0.5; bottom, V - 0.9, P = 0.2.

a tendency to avoid or delay vaccinationand an increased risk of SIDS and otherserious neurologic events (table 1). That fail-ure to control for such factors may lead tospurious negative associations between vac-cination and adverse events is evident inseveral published investigations (table 2).Examination of the logic underlying this

relation reveals that failure to control forsuch factors in analyses may mask true as-sociations between vaccinations and certainadverse outcomes under certain conditions(tables 3-6; figures 1 and 2). In particular,we note that the extent of relative risk under-estimation will be related directly to theproportion of individuals with contraindi-

Confounding and Adverse Reactions to Vaccines 131

1.2

0.4

0% 1% 2% 3% 4% 5% 6% 7%

% Population with Contraindication (C)

8% 9% 10%

1.2 r

1% 2% 3% 4% 5% 6% 7%% Population wtth Contraindication (C)

8% 9% 10%

—— D-1 —•— D=S - * - 0=10 - « - D=30FIGURE 2. Ratto of observed to true relative risk of sudden infant death syndrome (SIDS) associated withvaccination, as a function of the percentage of children with contraindication to vaccination (C) and the relative riskof SIDS associated with vaccination (D). These graphs show the implications of different proportions recentlyvaccinated (e.g., within 1-3 days) among infants without (V) and with (P) contraindications. Top, V = 0.03, P -0.0075; bottom, V - 0.03, P = 0.003.

cations to vaccination that are also risk fac-tors for the adverse outcome, the relativerisk of the adverse outcome associated withthese contraindications, and the extent to

which these contraindications to vaccinationare observed (i.e., the difference in vaccina-tion coverage between individuals with andwithout the contraindications).

132 Fine and Chen

TABLE 6. Numerical examples illustrating the ratio between the observed and true relative risk of suddenInfant death syndrome (SIDS) associated with vaccination (OBS RR/R), and the proportion of SIDS caseswith contraindications (SIDS WITH/SIDS), in circumstances In which contraindications for vaccination arealso risk factors for SIDS

Variable name*

SBDCVP

OBSRR

OBS RR/R

SIDS WITH/SIDS

SRDCVP

OBSRR

OBS RR/R

SIDS WITH/SIDS

Baselinevatuest

Variations In boktf

"Long-term" or "ever-vaccinated" circumstances

0.0015

200.10.950.3

0.654

0.131

0.505

0.0025

200.10.950.3

0.654

0.131

0.505

0.0014

200.10.950.3

0.523

0.131

0.523

0.0015

100.10.950.3

1.007

0.201

0.337

0.0015

200.080.950.3

0.659

0.132

0.444

"Short-term" or "recently vaccinated" circumstances

10"6

2100.20.030.002

0.808

0.404

0.709

10"5

2100.20.030.002

0.808

0.404

0.709

icr"3

100.20.030.002

1.212

0.404

0.703

10"*2

150.20.030.002

0.636

0.318

0.785

10"*2

100.10.030.002

1.109

0.554

0.519

0.0015

200.10.900.3

0.901

0.180

0.515

10~«02

100.20.020.002

0.863

0.431

0.711

0.0015

200.10.950.4

0.780

0.156

0.546

10-*2

100.20.030.001

0.756

0.378

0.708

* Variable names are as defined In the text and table 3.t A baseline set of parameter values.t Columns 3-8 show the effect of varying each of the baseine assumptions.

The magnitude of such confounding ef-fects may be considerable. The five studiesof DPT and SIDS summarized in table 2reported relative risk estimates ranging from0.15 to 5.4 using various methods; however,most of the estimates were below 1.0, andfour of the studies have reported at leastsome relative risk measures below 0.2 (table2). It seems unlikely to us, though, on bio-logic grounds that the true relative risk inthis situation could be less than unity (asthis would imply that such vaccines providesome immediate nonspecific protectionagainst sudden infant death). Although theunderestimation may have been due in partto biased case ascertainment, inappropriatecontrol selection, or chance effects, its mostobvious source is the confounding problemdiscussed in this paper. Major reductions are

seen when the prevalence of contraindica-tions exceeds 1 percent, and the effect ap-proaches its maximum when their preva-lence reaches 5 percent (figures 1 and 2). Itmay not be unreasonable to suppose that 5percent of infants in many populations willhave at least one of the confounding riskfactors cited in table 1 (36, 37).

In contrast to the conclusion of Griffin etal. (30), our simulations demonstrate that itis at least possible for the observed relativerisk to be less than half the true value evenif more than half of the cases (e.g., of SIDS)have risk factors for avoidance of vaccina-tion (tabh 'J). On the other hand, our explo-ration of parameter values, such as mightarise in "recent vaccination history" studiesexemplified by Griffin et al., does not easilyexplain the very low relative risks of SIDS

Confounding and Adverse Reactions to Vaccines 133

associated with DPT vaccination observedby some investigators (14, 30). Samplingerrors aside, observed relative risks on theorder of 0.2 could arise even if the truerelative risk were greater than 1.0, if oneassumes that the contraindications werehighly prevalent (high C) and associatedwith a very high relative risk of the adverseoutcome (high D) (e.g., if V = 0.01, C = 0.2,D = 50, and P = 0.01, then a true relativerisk of R = 1.2 would be observed as 0.24).Such a high prevalence of so strong a con-traindication/risk factor, however, seemsimplausible. Risk factors such as those listedin table 1 are unlikely to be associated withrelative risks (Z>) greater than 10, let alone30 or 50. Thus, whether the low observedrelative risks of SIDS associated with vacci-nation reflect sampling error, interactionsamong several contraindications/risk fac-tors, or other sorts of biases, or, indeed,whether they do reflect some "protective"effect of vaccination remains unclear to usand awaits elucidation. We note, however,that reanalysis of the British National Child-hood Encephalopathy Study of all cases andcontrols with any potential contraindica-tions to vaccination has led to a fourfoldincrease, from 3.3 to 12.6 (95 percent con-fidence interval 2.8-114.7), in the estimatedrelative risk of encephalopathy subsequentto DPT vaccination (D. Miller, St. Mary'sHospital Medical School, London, personalcommunication, 1990).

In theory, it might be possible to estimatethe extent of this bias in a particuliar situa-tion, but this would require knowledge ofthe nature, frequency, and implication ofeach of the six factors that may influenceboth propensity to be vaccinated and therisk of adverse event (table 3). The difficultyof obtaining such information on all sixfactors makes it extremely hard to assesswhether an observed relative risk of, forexample, 0.2 is consistent with a true relativerisk greater than 1.0. This inference is madeeven more problematic by the fact that manyother sorts of bias, for example, relating tocase ascertainment, may influence the ob-served relative risk.

In reviewing the literature for this paper,we have been impressed that much more isknown about factors associated with a fail-ure to receive adequate vaccination in dif-ferent societies than about the nature andfrequency of factors that lead to postpone-ment of vaccination. It may be expected thata number of situations (ill health on the partof the child or other family member, do-mestic crises in the family) will lead parentsto delay taking their child to be vaccinatedand that some of these situations will them-selves be risk factors for severe neurologicepisodes or SIDS. For example, Stanton etal. found that parents reported prior symp-toms classified as "major," i.e., " . . . usuallyneeding a medical opinion on the same dayand continuing close supervision . . . " (38,p. 1,250), in 48 percent of 145 SIDS casesas compared with 12 percent of age-matchedcontrols (odds ratio = 7). It is likely thatmost parents and health care providerswould postpone vaccination of children withsuch symptoms. Given that studies of asso-ciations between vaccination and severe ad-verse reactions typically focus on narrowtime intervals between prior vaccination andonset of the "reaction," it becomes impor-tant to understand the nature and frequencyof vaccination-postponing factors in studypopulations. This is an area of research thathas attracted inadequate attention in thepast.

We have focused in this paper on just oneof many methodological problems confront-ing studies of adverse reactions to vaccina-tions. Most published discussions of the sub-ject have concentrated upon biases that actto overestimate the relative risk of adverseevents after vaccination (10). Biases thatunderestimate the risk, as discussed here, havereceived less attention. The fact that suchbiases do exist makes it difficult to demon-strate convincingly that a vaccine is not re-sponsible for rare, severe, adverse reactions.The avoidance of so many potential con-founding factors presents a difficult chal-lenge to epidemiologists who would studythe problem of rare, severe, adverse reactionsto vaccines. If such studies are to prove

134 Fine and Chen

useful, they must include strenuous effortsto control for such factors in their design,analysis, and interpretation. Whether this ispossible at all may be open to discussion.The difficulty of doing so is indisputable.

REFERENCES

1. Plotkin SA, Mortimer EA, eds. Vaccines. Philadel-phia: WB Saunders, 1988.

2. Lane JM, Millar JD. Routine childhood vaccina-tion against smallpox reconsidered. N Engl J Med1969^81:1220-4.

3. Institute of Medicine. An evaluation of poliomye-litis vaccine policy options. Washington, DC: Na-tional Academy of Medicine, 1988. (Publicationno. (IOM)88-O4).

4. Miller DL, Alderslade R, Ross EM. Whoopingcough and whooping cough vaccine: the risks andbenefits debate. Epidemiol Rev 1982,4:1-24.

5. Kimura M, Kuno-Sakai H. Epidemiology of per-tussis in Japan. Tokai J Exp Clin Med 1988; 13(suppl):l-7.

6. Hinman AR. DTP vaccine litigation. Am J DisChild 1986;14O:528-3O.

7. Smith MH. National Childhood Vaccine InjuryCompensation Act. Pediatrics 1988;82:264-9.

8. Strom BL, ed. Pharmacoepidemiology. New York:Churchill Livingstone, 1989.

9. Ray WA, Griffin MR. Use of Medicaid data forpharmacoepidemiology. Am J Epidemiol 1989;129:837-49.

10. Wentz KR, Marcuse EK. DTP and serious neuro-logic illness: an updated review of the epidemio-logic evidence. Pediatrics 1991;87:287-97.

11. Marks JS, Halpin TJ, Irvin JJ, et al. Risk factorsassociated with failure to receive vaccinations. Pe-diatrics 1979;64:304-9.

12. Hutchins SS, Escolan J, Markowitz LE, et al. Mea-sles outbreak among unvaccinated preschool-agedchildren: opportunities missed by health care pro-viders to administer measles vaccine. Pediatrics1989,83:369-74.

13. Markland RE, Durand RE. An investigation ofsotiopsychological factors affecting infant immu-nization. Am J Public Health 1976;66:168-70.

14. Hoffman HJ, Hunter JC, Damus K, et al. Diphthe-ria-tetanus-pertussis immunization and sudden in-fant death: results of the National Institute of ChildHealth and Human Development cooperative epi-demiological study of sudden infant death syn-drome risk factors. Pediatrics 1987;79:598-611.

15. Golding J. Immunizations. In: Butler NR, GoldingJ, eds. From birth to five: a study of the health andbehaviour of Britain's 5 year-olds. Oxford: Perga-mon Press, 1986:295-319.

16. Alderslade R, Bellman M, Rawson NSB, et al. TheNational Childhood Encephalopathy Study. In:Whooping cough. London: Her Majesty's Station-ery Office, 1981:79-169.

17. Woodward HB. Some factors associated with non-

immunization. In: Proceedings of the 21st immu-nization conference. Atlanta: Centers for DiseaseControl, 1987:67-9.

18. Gergen PJ, Ezzati T, Russell H. DTP immuniza-tion status and tetanus antitoxin titers of MexicanAmerican children ages six months through elevenyears. Am J Public Health 1988;78:1446-50.

19. Roper J, Day S. Uptake of immunisations in lowbirthweight infants. Arch Dis Child 1988;63:518-21.

20. Committee on Infectious Diseases, AmericanAcademy of Pediatrics. Report of the Committeeon Infectious Diseases. 21st ed. Elk Grove, IL:American Academy of Pediatrics, 1988:1-566.

21. Immunization Practices Advisory Committee.Diphtheria, tetanus, and pertussis: guidelines forvaccine prophylaxis and other preventive measures.MMWR 1985,34:405-414, 419-26.

22. Orenstein WA, Allmen KC, Eddins D, et al. Pre-school immunization—1988. In: Proceedings ofthe 22nd immunization conference. Atlanta: Cen-ters for Disease Control, 1988:121-8.

23. Greenland S, Neutra R. Control of confounding inthe assessment of medical technology. Int J Epide-miol 1980,9:361-7.

24. Bemier RH, Frank JA, Dondero TJ, et al. Diph-theria-tetanus toxoids-pertussis vaccination andsudden infant deaths in Tennessee. J Pediatr 1982;101:419-21.

25. Solberg LK. DPT vaccination, visit to child healthcenter, and sudden infant death syndrome (SIDS):evaluation of EPT vaccination. Report to the OsloHealth Council 1985. Bethesda, MD: Food andDrug Administration, 1985. (NIH laboratory trans-lation 85-152).

26. Taylor EM, Emery JL. Immunisation and cotdeaths. (Letter). Lancet 1982^2:721.

27. Baraff LJ, Ablon WJ, Weiss RC. Possible temporalassociation between diphtheria-tetanus toxoid-pertussis vaccination and sudden infant death syn-drome. Pediatr Infect Dis J 1983̂ 2:7—11.

28. Mortimer EA Jr, Jones PK, Adelson L. DTP andSIDS. (Letter). Pediatr Infect Dis J 1983;2:492-3.

29. Walker AM, Jick H, Perera DR, et al. Diphtheria-tetanus-pertussis immunization and sudden infantdeath syndrome. Am J Public Health 1987;77:945-51.

30. Griffin MR, Ray WA, livengood JR, et al. Risk ofsudden infant death syndrome after immunizationwith the diphtheria-tetanus-pertussis vaccine. NEngl J Med 1988;319:618-23.

31. Miller DL, Ross EM, Wadsworth J, et al. Riskfactors for acute neurological illness in children. In:Proceedings of the Fourth International Sympo-sium on Pertussis, Geneva, 1984. Dev Biol Stand1985,61:389-94.

32. Walker AM, Jick H, Perera DR, et al. Neurologicevents following diphtheria-tetanus-pertussis im-munization. Pediatrics 1988;81:345-9.

33. Griffin MR, Ray WA, Mortimer EA, et al. Risk ofseizures and encephalopathy after immunizationwith the diphtheria-tetanus-pertussis vaccine.JAMA 1990,263:1641-5.

34. Gale JL, Thapa PB, Bobo JK, et al. Acute neuro-logical illness and DTP: report of a case-controlstudy in Washington and Oregon. (Abstract). In:Manclark CR, ed. Proceedings of the Sixth Inter-

Confounding and Adverse Reactions to Vaccines 135

national Symposium on Pertussis. Bethesda, MD:Department of Health and Human Services,1990:228-9. (DHHS publication no. (FDA)90-1162).

35. Kraus JF, Bulterys M. The epidemiology of suddeninfant death syndrome. In: Kiely M, ed. Reproduc-tive and perinatal epidemiology. Boca Raton, FL:CRC Press, 1989.

36. Jelley DM, Nicholl AG. Pertussis: what percentage

of children can we immunize? BMJ 1984;288:1582-3.

37. Hewitt M. Incidence of contraindications to im-munisation. Arch Dis Child 1989;64:1052-3.

38. Stanton AN, Downham MAPS, Oakley JR, et al.Terminal symptoms in children dying suddenlyand unexpectedly at home: preliminary report ofthe DHSS multicentre study of postneonatal mor-tality. BMJ 1978,2:1249-51.