Heriot-Watt University, Edinburgh: Angus Macdonald and Delme Pritchard

A MATHEMATICAL MODEL OF ALZHEIMER' S DISEASE AND
THE APOE GENE

B
ANGUS MACDONALD AND DELME PRI TCHARD
Heriot-Watt University, Edinburgh
ABSTRACT
Al zhei mer' s disease ( AD) account s for a significant pr opor t i on of l ong-t erm
care costs. The recent di scovery t hat the e4 allele of the ApoE gene indicates
a predi sposi t i on to earlier onset of AD raises quest i ons about the pot ent i al
for adverse selection in l ong-t erm care insurance, about l ong-t erm care costs
in general, and about the pot ent i al effects on costs of gene t herapy, or bet t er
t arget t ed t reat ment s for AD. This paper describes a simple Mar kov model
for AD, and the estinaation of the transition intensities from the medical and
epi demi ol ogi cal literature.
KEYWORDS
Al zhei mer' s Disease, Apol i popr ot ei n E Gene, Genetics, Mar kov Model s.
1. I NTRODUCTI ON
Mol ecul ar genetics has far-reaching implications for all aspect s of health
economi cs, including the effectiveness, or even practicability, of insurance-
based fundi ng of all forms of care. Ill particular, if genetic test results are
known by the individual but not by the insurer, there may be a t endency for
t hose most at risk to buy mor e insurance, known as adverse selection. This is
a nat ural subj ect for quant i t at i ve modelling.
Such discussion as has taken place (in the Uni t ed Ki ngdom) has t ended to
be confined to life insurance, in which cont ext it has been suggested that the
overall cost s of adverse selection might be limited ( Macdonal d, 1997, 1999;
Pri t chard, 1997). However , these papers acknowl edge:
(a) that different concl usi ons might hold in respect of ot her forms of
insurance; and
(b) the lack of sound epi demi ol ogi cal dat a in respect of any but a few genetic
condi t i ons.
ASTIN BULLETIN, Vol. 30, No. I, 2000, pp. 69-110
70 ANGUS MACDONALD AND DELME PRITCHARD
Thus, the model pr oposed by Macdonal d (1997, 1999) was based on
par amet er est i mat es intended to be extreme, and to suggest an upper bound
on the cost of adverse selection, rat her than on dat a- based statistical
estimates. Tan (1997) applied a similar model to annui t y business, with
results that suggested higher costs.
The U. K. Gover nment has so far avoi ded legislating on the very sensitive
issue of the use of genetic test i nformat i on by insurers, but instead has set up
the Genet i cs and Insurance Commi t t ee (GA1C), charged with the assessment
of the likely relevance and reliability of genetic test i nformat i on as it relates
to different kinds of insurance. Such consi derat i ons require actuarial model s
of the insurance process, allowing for the effects of specific genes on
mort al i t y and morbi di t y. To date, the onl y such st udy is Lemai re e t a l .
(1999) and Subr amani an & Lemaire (1999), treating breast and ovari an
cancer and the BRCA1 and BRCA2 genes.
In this paper we pr opose a simple Mar kov model for Al zhei mer' s disease
( AD) and est i mat e its transition intensities from medical and epi demi ol o-
gical studies. Genet i c vari at i on arises because the e4 allele of the
Apol i popr ot ei n E ( ApoE) gene is known to indicate a predi sposi t i on to
earlier onset of AD.
In Section 2 we give a bri ef out l i ne of genetics t ermi nol ogy, then describe
AD and the evidence for a genetic component of AD. In Section 3 we specify
the model , and in Sections 4 and 5 we est i mat e the transition intensities. We
show some results in Section 6, and our concl usi ons are in Section 7.
Appl i cat i ons of the model include the st udy of long-term care insurance,
( Macdonal d & Pritchard, 1999) and the st udy of long-term care costs
(Warren e t a l . , 1999). The processes leading to LTC insurance claims are
compl ex, when compar ed with ot her forms of insurance, and there are no
insurance dat a to speak of; t herefore we (and insurance compani es
themselves) have to rely on dat a collected and published for a variety of
reasons, most l y in the medical literature. As a result, our model is far from
definitive, but we found the process of ext ract i ng i nformat i on from the
medical literature and put t i ng it to act uari al use very instructive, and we
suggest t hat any short comi ngs of our model shed useful light on the
pr obl ems t hat might be faced in the future.
2. THE GENETI CS OF ALZHEI MER' S DI SEASE
For a recent survey of the genetic epi demi ol ogy of AD, see Sl oot er &
van Dui j n (1997); Breteler e t a l . (1992), reviews the posi t i on before much was
known about the genetic component of AD.
2. 1. A Bri ef Pri mer on Huma n Genet i cs
The fol l owi ng discussion gives onl y the briefest definitions of genetic terms,
and omi t s all compl i cat i ons. For a pr oper account , see a text such as
St rachan & Read (1999).
A MAT HE MAT I C AL MODE L OF AL Z HE I ME R ' S DI SEASE AND T HE AP OE GE NE 7 1
(a) Each cell of the human body (except sperm and egg cells) has a nucleus
cont ai ni ng 23 pairs of chr omosomes. Each cell carries an identical set of
chr omosomes at birth, unless some have been damaged.
(b) A chr omos ome is a very long doubl e st rand (the f amous doubl e helix) or
linear molecule, made up of a sequence of base pairs of DNA.
(c) A gene is a sequence of base pairs at a fixed l ocat i on on a given
chr omos ome t hat acts as a functional unit, codi ng for the pr oduct i on of
a prot ei n (usually). The term ' gene' strictly refers to the locus on the
chr omosome.
(d) The gene at a given locus might have several variants (rat her as chemical
elements might have several isotopes); these are known as alleles.
Different alleles are simply different sequences of base pairs. Differences
bet ween alleles are the result of mut at i ons, namel y al t erat i ons to a gene
caused by (for example) errors in replication of the DNA when cells
divide. If the mut at i on occurs in an adul t (somat i c) cell, it can onl y
spread t hrough cell division (which is what happens in cancers). If the
mut at i on occurs in a gene carried by a sperm or egg cell, it can be passed
to offspring.
(e) Because the chr omosomes are pai red (except the X and Y chr omosomes
that det ermi ne sex) each cell has t wo alleles of every gene, one inherited
from each parent. They may be the same ( homozygosi t y) or different
(heterozygosity).
(f) Mut ant genes will encode variants of the prot ei n t hat the gene produces.
These may have a beneficial, neutral or adverse effect on the cell or
organi sm. If the altered prot ei n is radically different, the cell will
pr obabl y die, or the organi sm will fail to devel op, or it will not survive to
reproduct i ve age. Less radical vari at i ons may manifest themselves as
susceptibility to disease, or may be harmless but non-funct i onal .
(g) The great range of genetic diseases arises from the range of effects of the
prot ei n pr oduct s of different alleles, and the simple combi nat or i cs of
inheritance.
(I) If an allele encodes a harmless but non-funct i onal prot ei n pr oduct ,
the disease will appear onl y in homozygous individuals ( aut osomal
recessive inheritance, such as cystic fibrosis).
(2) Het er ozygous cells will pr oduce a mi xt ure of vari ant s of the prot ei n
product ; if j ust one of these is lethal it will cause disease ( aut osomal
domi nant inheritance, such as Hunt i ngt on' s disease).
(3) In bet ween these extremes, alleles encode prot ei n pr oduct s that are
more or less danger ous or fully-functional, the effect oft en dependi ng
on the envi ronment . Then different levels of susceptibility to disease
will appear, and homozygous individuals will oft en be mor e
susceptible than het erozygous individuals.
(h) Lifestyle and envi ronment can affect the pot ency of susceptibility genes.
For example, the activity of the prot ei n pr oduced by a danger ous allele
may be enhanced in certain envi ronment s, or a prot ect i ve allele may be
put at great er risk of being knocked out by a mut at i on.
72
( i )
ANGUS MAC DONAL D AND DELME P R I T C HAR D
Al t hough the outline above ment i ons only single genes, most processes
in the human body are compl i cat ed and involve very many proteins, each
encoded by its own gene with its various alleles. Most genetic disease will
result from combi nat i ons of several alleles, lifestyle and envi ronment ;
the term for this is mul t i fact ori al disorder.
2. 2. Cogni t i ve Impai rment
The term ' cognitive i mpai rment ' covers AD, which account s for by far
the great er number of cases, and ot her forms of ment al det eri orat i on,
chiefly vascular in origin (for example, arising from strokes). Assessment
is liable to be imprecise, maki ng it difficult to decide on an exact dat e
of inception of cognitive i mpai rment , if such a thing exists. Moreover,
al t hough AD is the commonest form of cognitive i mpai rment , it is hard
to di agnose with cert ai nt y except by post -mort em exami nat i on. These
factors i nt roduce consi derabl e uncert ai nt y into epidemiological studies of
AD. Breteler et al. (1992) not ed that:
(a) AD itself can have a significant vascular component ;
(b) some of the neuropat hol ogi cal sympt oms of AD (see Section 2.4) can
also be sympt oms of vascular dement i a; and
(c) studies by Ti erney et al. (1988) found that post - mor t em exami nat i on
confi rmed only 64-86% of diagnoses of AD.
2. 3. The Pat hol ogy o f AIzheimer's Di s eas e
The pat hol ogy of AD includes:
(a) senile plaques (deposits on the outside of neurones (brain cells),
consisting largely of the prot ei n/ 3-amyl oi d);
(b) neurofibrillary tangles (connect i ons between neurones);
(c) amyl oi d angi opat hy (deposits of amyl oi d protein in the arteries of
the brain);
(d) loss of neurones; and
(e) decreased activity of choline acet yl t ransferase (an enzyme).
Therefore, any gene whose expression leads to the product i on, or over-
product i on, of substances associated with these changes is potentially a
genetic mar ker for AD.
2. 4. General Evi dence o f a Geneti c Contri buti on to Al zhei mer' s Di s eas e
AD is a disease of old age; it is rare below ages 60-70. These rare cases are
called ' earl y-onset ' AD, which should not be confused with early onset of
AD within the usual age range. We are concerned only with the latter.
Families with a history of AD are somet i mes observed, but AD also
occurs sporadically (t hat is, in the absence of a family history of AD) and it
A MAT HE MAT I C AL MODE L OF AL Z HE I ME R ' S DI SEASE AND THE AP OE GE NE 73
is al ways possible t hat a case of AD in an affected family is, in fact, sporadi c.
The differences bet ween familial and sporadi c AD are not clear, al t hough the
former may be mar ked by earlier onset and more rapid progressi on.
A very few fami l i es, have several cases of earl y-onset AD in
several generations, consi st ent wi t h aut osomal domi nant transmission
( Levy- Lahad & Bird, 1996), and three genes have been found. First was
the gene encodi ng for amyl oi d precursor prot ei n (APP), involved in the
pr oduct i on of fl-amyloid. It resides on chr omos ome 21, which is the
chronaosome affected in cases of Down syndrome, sufferers of which oft en
devel op AD in mi ddl e age. Several mut at i ons have been found, but t hey are
rare. Later, mut at i ons in two genes labelled presenilin-I (PS-I) and
presenilin-2 (PS-2) were identified, which appear ed to be associ at ed with
AD, t hough the mechani sms remain unclear.
Familial AD is not restricted to earl y-onset cases and family history
remains an i mpor t ant risk fact or for late-onset AD (Jarvi k e t a l . , 1996).
Susceptibility genes have been identified, of which the most st udi ed is t hat
which codes for apol i popr ot ei n E.
2. 5. The Apofi poprotei n E Gene
This summar y is not comprehensi ve, but we hope t hat it is detailed enough
to give an actuarial reader an impression of the progress made in
underst andi ng genetic fact ors of AD, as well as some of the pr obl ems and
(perhaps most i mport ant ) tile great speed at which human genetics is
advanci ng.
Apol i popr ot ei n E ( ApoE) is found in senile pl aques and neurofi bri l l ary
tangles in AD patients. It has also been studied because of its r61e in lipid
met abol i sm. The gene t hat encodes it is on chr omosome 19, which was
linked to families with late-onset AD by Peri cak-Vance e t a l . (1991), maki ng
it a clear candi dat e gene for familial AD. St ri t t mat t er e t a l . (1993) confi rmed
this hypot hesi s, which was rapidly suppor t ed by many ot her studies. The
basic facts are as follows:
(a) The ApoE gene has three common alleles - E2, e3 and ~4 - whose
frequencies are roughl y 0.09, 0.77 and 0.14 respectively.
(b) Since each offspring receives one allele from each parent , there are six
possible genot ypes (~2/e2, e2/ e3, e2/ e4, e3/ e3, ~3/ e4 and e4/ e4).
Offspring with two copi es of the same allele are called homozygot es,
while t hose with t wo different alleles are called het erozygot es.
(c) The ApoE e4 allele increases the risk of AD in a dose related fashion,
such t hat e4 homozygot es (e4/ e4) are at great er risk than
e4 het erozygot es (e2/ e4, e3/ e4), who in turn are at great er risk than
those wi t hout the e4 allele (Bickeb611er e t a l . , 1997; Cor der e t a l . , 1994;
van Dui j n e t a l . , 1995; Far t er e t a l . , 1997; Jarvi k e t a l . , 1996;
Kuusi st o e t a l . , 1994; Leht ovi rt a e t a l . , 1995; Mayeux e t a l . , 1993;
Myer s e t a l . , 1996; Poirier e t a l . , 1993; Tsai e t a l . , 1994). See Section 5.2
74 ANGUS MAC DONAL D AND DELME P R I T C HAR D
for risk estimates. The risk depends on age, being highest at ages 60-70,
t aperi ng off at ol der ages (Bickeb611er e t a l . , 1997; Cor der e t a l . , 1994;
Far r er e t a l . , 1997).
(d) It is also possible t hat the e4 allele is associ at ed with earlier onset of AD
(not to be confused with earl y-onset AD). The effect may be dose dependent
(Farrer e t a l . , 1997; Frisoni e t a l . , 1995; Gomez-Isl a e t a l . , 1996); or not
(Leht ovi rt a e t a l . , 1995; Stern e t a l . , 1997; Cor der e t a l . , 1995); or it may not
exist at all (Liddell e t a l . , 1994; Masul l o e t a l . , 1998; Nor r man e t a l . , 1995).
(e) Invest i gat i ons into the rate of mental decline of AD patients by genot ype
found no evidence for any difference (Basun e t a l . , 1995;
Gomez- l sl a e t a l . , 1996; Masul l o e t a l . , 1998; Nor r man e t a l . , 1995).
There is conflicting evidence about mortality. It is possible that younger
age at onset shoul d imply longer survival times, because of the usual age-
related mort al i t y differentials, and t herefore t hat the e4 allele shoul d be
associ at ed with longer life after onset of AD. While some studies suppor t
this ( Cor der e t a l . , 1995; Gomez- I sl a e t a l . , 1996; Nor r man e t a l . , 1995),
ot hers have found no difference (Basun e t a l . , 1995; Stern e t a l . , 1997). If
e4 is associ at ed with lighter mort al i t y in AD patients then risk est i mat es
from cross-sectional studies (the vast maj ori t y to dat e) shoul d be
i nt erpret ed with caut i on. An incidence st udy (Evans e t a l . , 1997)
confi rmed e4 to be a significant risk factor, but the est i mat ed increased
risk of onset was at the lower end of the report ed range.
(f) In cont rast , the e2 allele has been found to have a prot ect i ve
effect agai nst late-onset AD ( Cor der e t a l . , 1994; Farrer e t a l . , 1997;
Gomez- l sl a e t a l . , 1996; Jarvi k e t a l . , 1996; Lamber t e t a l . , 1998;
Masul l o e t a l . , 1998). However, a st udy of earl y-onset AD patients
(van Duijn e t a l . , 1995), found a higher frequency of the e2 allele, and an
associ at i on of e2 with a mor e aggressive form of AD, suggesting different
r61es of ApoE in earl y-onset and late-onset AD. Findings relating to the
e2 allele are based on the ~2/e3 genot ype, as e2 homozygot es are rare.
The risk at t ached to the e2/ e4 genot ype is not clear, possibly because e2
and e4 have opposi t e effects (Jarvi k e t a l . , 1996; Levy-Lahad e t a l . , 1996).
ApoE e4 is the most i mpor t ant genetic risk fact or for AD identified yet.
Though it is neither necessary nor sufficient to cause AD it does increase
susceptibility. Appr oxi mat el y 26% of Caucasi ans carry at least one e4 allele
and it has been est i mat ed t hat bet ween 42% and 79% of AD cases are
at t r i but abl e to the associ at ed excess risk ( Nal bant ogl u e t a l . , 1994).
2.6. Other Genetic Factors of Alzheimer's Disease
In 1997, a gene for the K-vari ant of but yryl chol i nest erase ( BCHE K), not a
risk fact or by itself, was found to act in synergy with ApoE ~4, such that
carriers of bot h (an est i mat ed 6% of Caucasi ans) were at over 30 times the
risk of AD as a person with neither ( Lehmann e t a l . , 1997). Subsequent
A MATHEMATICAL MODEL OF ALZHEIMER'S DISEASE AND THE APOE GENE 75
studies (Brindle e t a l . , 1998; Singleton e t a l . , 1998) failed to reproduce the
result. Al t hough some expl anat i ons have been advanced, caut i on is
advi sabl e in using BCHE K as a risk fact or for AD.
Payami e t a l . (1997) report ed an associ at i on bet ween AD and the A2 allele
of the human l eukocyt e antigen ( HLA) ; the HLA- A2 phenot ype and
ApoE E4/c4 genot ype had similar and addi t i ve effects on reducing age at
onset of AD, at ages bel ow 60 and above 75. Fur t her studies woul d be
needed to confi rm these findings.
Podusl o e t al . (1998) found the apol i popr ot ei n CI (apo CI) gene to be a
risk fact or for earl y-onset and late-onset AD, whet her sporadi c or familial.
Apo CI A homozygot es had 4 to 5 times the odds of devel opi ng AD,
het erozygot es about twice the risk. This was not unexpect ed, since Apo CI is
closely linked to ApoE and in linkage di sequi l i bri um with ApoE and AD.
Linkage disequilibrium is the non- r andom assort ment , in a popul at i on, of
t wo genes on the same chr omos ome (the strength of the linkage is inversely
pr opor t i onal to the di st ance bet ween them). It was t hought that the
associ at i on of AD with ApoE may be more significant.
3. A MODEL FOR ALZHEI MER' S DISEASE
3.1. Model Specification
We use a cont i nuous- t i me multiple state model. For general comment s on
these models, see Macdonal d (1996a) or Wat ers (1984). Lives with each
ApoE genot ype are assumed to form a homogeneous popul at i on, suffering
the different risks of AD discussed in Sections 2.5 and 5.2.
An i mpor t ant reason for using these model s is t hat they al l ow the most
compl et e represent at i on of the underl yi ng process. It is then necessary to
est i mat e a large number of transition intensities, for which adequat e dat a do
not al ways exist, but it is preferabl e to obt ai n a clear picture of the dat a
needed than to sweep the issue under the carpet by wor ki ng with some less
adequat e model in the first place. In particular:
(a) if some simpler model is event ual l y r ecommended for use, because of
missing dat a or for comput at i onal conveni ence, it is i mpor t ant to be abl e
to assess its soundness in practice; and
(b) if missing dat a become avai l abl e later, for example, as the insured lives
experience develops, it is a hi ndrance if t oo much has been invested in a
model that cannot i ncorporat e it.
Moder n comput i ng power is such t hat the comput at i onal demands of
multiple state model s (numerical i nt egrat i on of differential equat i ons) can
qui t e reasonabl y be met, for arbi t rari l y compl ex Mar kov model s
(Norberg, 1995) and for many semi - Mar kov model s (Wat ers & Wilkie, 1987;
Wat ers, 1991). The techniques can all be found in st andar d texts on
numerical analysis, and no act uar y shoul d be prevent ed from choosi ng an
adequat e model by the need to use them.
76 ANGUS MACDONALD AND DELME PRI TCHARD
Fi gure 1 shows a simple model of AD. Each genot ype is represented by
such a model; the transition intensities in each model will differ, representing
the different genetic risks, x denot es the age at out set (for example, when
insurance is purchased) and t the elapsed durat i on. The choice of states is
di ct at ed entirely by the events t hat have been studied in the medical and
epi demi ol ogi cal literature. For certain purposes, it woul d be desi rabl e to
model ot her events, such as the start of a l ong-t erm i nsurance claim. No dat a
about that event are available; however a maj or event t hat has been studied
is institutionalisation. Al t hough becomi ng institutionalised need not
coincide with the start of an i nsurance claim, it is the best available proxy.
Macdonal d (1999) consi dered frailty model s as an alternative to
Mar kov model s, for genetics and i nsurance applications. They offer the
advant age of a simple model of the genetic variability, if that is justified
by the ci rcumst ances. They may be especially useful for modelling
mul t i fact ori al disorders, or genes with very many alleles or mut at i ons, but
for a single gene with j ust a few alleles it seems reasonabl e t o model each
separately. Ot her possible model s (such as Cox- t ype model s) might be
useful for model l i ng individual t ransi t i ons but do not lend themselves to
the inclusion of payment s cont i ngent upon compl i cat ed life histories.
St a t e i1: No Alzheimer' s
disease.
i1 4 x-t-t
i1 2 x + t
St a t e i2: Onset of
Alzheimer' s disease.
i2 4 x- I - t
St a t e i4: Dead
i2 3 xd-L
St a t e i3: Institutionalised
from Alzheimer' s disease.
j i3 4 x+t
FIGURE I: A simple model of Alzheimer' s disease in the ith of M subgroups, each representing a different
ApoE genotype, x is the age at outset, and t the elapsed duration.
A MATHEMATICAL MODEL OF ALZHEIMER' S DISEASE AND THE APOE GENE 77
3.2. An Expanded Long-Term Care Model
AD al one does not account for all l ong-t erm care costs. Br oadl y speaking,
the need for care arises because of cognitive di sorder (including AD) or loss
of ability to per f or m Activities of Dai l y Living ( ADLs) such as dressing,
washi ng and feeding. A comprehensi ve model of l ong-t erm care cost s can be
specified in terms of these causes, with AD included as a component , and the
i mpact of the ApoE gene on overall care costs can t hereby be studied.
However , i ncorporat i ng AD explicitly in an expanded model will require
dat a t hat describe, at the level of individual lives, the progress of AD and the
loss of ADLs. Until such dat a are available, we can onl y est i mat e overall
care costs by crude approxi mat i ons: see Macdonal d & Pri t chard (1999).
4. ESTIMATION OF TRANSITION |NTENSITIES NOT DEPENDING
ON APOE GENOTYPE
In this section we estimate the transition intensities for the events: onset of
AD; institutionalisation; and deat h. All of these must be ' est i mat ed' from
results report ed in the medical and epi demi ol ogi cal literature. Ideally, we
woul d wor k with the original dat a, but these are al most never available.
Repor t ed results are not al ways ideal for the ext ract i on of paramet ers for an
actuarial model; often the age gr oups used are very wide, and not the same in
different surveys; somet i mes onl y graphs (such as Kapl an- Mei er survival
curves) are given.
4. 1. Prospective and Case-Based Studies
A most i mpor t ant distinction must be drawn when est i mat i ng transition
intensities from epi demi ol ogi cal studies (see, for exampl e, Kahn & Sempos
(1989), Cl ayt on & Hills (1993), Lilienfeld & Hills (1993), Selvin (1996)).
(a) Prospective studies, based on sampl es of the general popul at i on, ought t o
yield the most reliable est i mat es of popul at i on risk, but are expensive
and t i me-consumi ng. Mor eover , they are rarely even begun until
subst ant i al evidence of an effect has been accumul at ed from ot her
studies.
(b) Case-based studies, based on affected persons (and cont rol s) oft en yield
relative risks greatly in excess of the true popul at i on risks, precisely
because the subjects are affected or at risk. However , early studies into
any medical condi t i on are al most inevitably of this type.
Our current knowl edge of most genetic di sorders is derived from case-based
studies; this is certainly true of ApoE and AD (see Section 2.5). It is very
likely t hat est i mat es of risk conferred by ApoE genot ype will fall as more
prospect i ve studies are carried out (see the comment on Evans et al. (1997) at
the end of Section 2.5(e)), but this will t ake time.
78 ANGUS MAC DONAL D AND DELME P RI T CHARD
The appr oach we adopt is as follows:
(a) in Section 4.2, we state assumpt i ons about the general level of mortality;
(b) in Section 4.3, we est i mat e the aggregat e incidence of AD, denot ed IL,-+ADt,
which has been investigated extensively;
(c) in Section 4.4, we est i mat e the intensity of institutionalisation, following
the onset of AD (that is, ..,~3 ~ and the force of mort al i t y following the
bL~.+t]
onset of AD (t hat is,/~.,-+i),"~4 ~:
(d) in Section 4.5, we est i mat e the force of mort al i t y for lives institutiona-
lised with AD (t hat is,/~x+t), i 3 4 ", .
(e) in Section 5.1, we estimate the popul at i on frequencies o f the ApoE
alleles; and
(f) in Section 5.2, we estimate the incidence of AD for each genot ype using
odds ratios from the genetic studies: this gives estimates ~_af #.,-+i.n2
4. 2. Basel i ne Mort al i t y Tabl es
For convenience, we choose paramet ri c approxi mat i ons to the AM80 and
AF80 Ul t i mat e mort al i t y tables as bases for mort al i t y assumptions; for use
in the model they are adj ust ed in a variety of ways. Gomper t z curves were
fitted to/_L.~.+I at ages 65-120, using log-linear least squares (see equat i on (1)):
AMS0p,x+l = 0.0000941160 . o 8 4 5 5 4 ( . ` . + 0 ( 1 )
"~rS01Lx+ t = 0.000025934e0093605(x+0.
Experi ment s with the AMS0 and AF80 tables themselves showed that the
Gomper t z approxi mat i ons had a negligible effect in l ong-t erm care
applications; we use them because they are somet i mes useful in numeri cal
work. For i nsurance use, some al l owance must be made for fut ure
i mprovement s in mortality. No experience is available to help, but following
discussion with some actuaries experienced in pricing long-term care
insurance, we have chosen 65% of these baseline tables as the aggregat e
mort al i t y assumpt i ons.
4. 3. The Ons et o f Al zhei mer' s Di s eas e in the Popul at i on
AD has been the subject of some large-scale epidemiological studies, many
of t hem pre-dat i ng the discovery of the ApoE gene. Some of these report
incidence rates, or ' occurrence/ exposure' rates, which are exactly the
estimates we need for transition intensities.
There is general agreement , in this literature, on the shape of the intensity
AD
/~.,+1 in the age range 60-85 years; it is very low at ages 60-64 (about 0 to
0.002) and increases rapidly with age, approxi mat el y doubl i ng every 5 years.
Sayet t a (1986) and Hebert e t a l . (1995) found t hat a Gomper t z curve gave
the best fit, despite trying a number of mor e complex models.
A MATHEMATICAL MODEL OF ALZHEIMER' S DISEASE AND THE APOE GENE 79
A n u mb e r o f s t udi es r e por t t he i nci dence o f AD ( t ha t is, t he i nt ens i t y #.,-+t)AO
by age but not by ge not ype , i nc l udi ng Co p e l a n d e t a l . (1992), Hagnel l e t a l .
(1992), Ko k me n e t a l . (1993), Le t e n n e u r e t a l . (1994), Ni l s s on (1984),
Ot t e t al . (1998), Roc c a e t a l . (1998) a nd Ro r s ma n e t a l . (1986). Of pa r t i c ul a r
i nt er est , however , is t he r ecent me t a - a na l ys i s o f t he i nci dence o f AD by
J o r m & Jol l ey (1998):
(a) it dr a ws on 23 s t udi es wor l d- wi de, i nc l udi ng 13 Eu r o p e a n st udi es;
(b) t he anal ysi s is car r i ed o u t s epar at el y f or Eur ope , t he U. S. A. a nd
Eas t Asi a;
(c) t he i nci dence o f AD is e s t i ma t e d by severi t y, cat egor i s ed as Mi l d + a n d
Mo d e r a t e + AD, wher e Mi l d + i ncl udes all cases cl assi fi ed as mi l d or
wor se; a nd
(d) p o i n t es t i mat es o f AD /~.,.+t wer e o b t a i n e d f or 5-year age g r o u p s f r o m 65 t o
95, and no a p r i o r i s ha pe o f y o .,-+t was a s s ume d.
We estimated/-~.,-+1'41) f r om J or l n & Jol l ey (1998) us i ng t he fi gures f r om t he
Eu r o p e a n s t udi es a n d f or Mi l d + AD. Th e es t i mat es , 95% c onf i de nc e l i mi t s
a n d t he l og- l i near l east s quar es Go mp e r t z fit:
#,4/).~.+r = 1.31275 x 1 0 - 7 e 0.145961(x+t) ( 2 )
are s h o wn in Fi gur e 2. I t is cl ear t ha t a Go mp e r t z cur ve is a very g o o d fit.
r"3 o
< c 5
O
O
d
[ ] Estimated incidence of AD, aggregated by sex and genotype
Fitted incidence of AD, aggregated by sex and genotype
........... 95% Confidence limits
70 7'5 80 85 90
Age (years)
FIGURE 2: Aggregate incidence of Alzheimer's disease: point estimates and 95% confidence intervals.
Source: Jorm & Jolley (1998).
Dat a on the incidence of AD among the very elderly (> 90 years) are
sparse, so est i mat es at these ages have wide confidence intervals and the trend
is uncertain. The met a-anal ysi s by Ga o e t a l . (1998) found that the rate of
increase in A o # x + t slowed down with age, but ot her studies found no evidence of
this ( Heber t e t a l . , 1995; Jor m & Jolley, 1998; Let enneur e t a l . , 1994). We have
simply ext r apol at ed the Gomper t z formul a above to all ages; the effect of this
assumpt i on will depend on the part i cul ar application, or t ype of insurance,
and this shoul d be investigated when the model is used (see, for example,
Macdonal d & Pri t chard (1999)).
Many studies have found men and women to be at the same risk of AD
( Kokmen e t a l . , 1993; Nilsson, 1984; Ot t e t a l . , 1998; Rocca e t a l . , 1998) and,
when differences have been report ed ( Gao e t a l . , 1998; Jor m & Jolley, 1998;
Let enneur e t a l . , 1994), women were found to be at great er risk onl y at very
old ages. Some experi ment s (not descri bed here) in appl yi ng the rnodel to
AD-rel at ed long-term care insurance costs using different rates of AD for
men and women showed that it made little difference, and here we have used
the aggregat e rate (equat i on (2)).
4 . 4 . T i me f r o m On s e t o f Al z h e i me r ' s Di s e a s e t o ~ ns t i t ut i o na l i s a t i o n o r De a t h
t23 ] ~ t ' 2 4 .i34
The avai l abl e dat a do not allow us to anal yse /.L.,.+, .,-+t or P'.,-+I by
genot ype.
Tabl e I summari ses the literature on time to the first of institutionalisa-
tion or deat h (' first event ' ) for AD patients, Some studies give times from
ent ry to the st udy rat her than from onset, which is usually not observed
directly. A striking feat ure is that few lives die before becomi ng
institutionalised. Thi s may seem surprising as AD patients have generally
been report ed to suffer higher mort al i t y than healthy lives (see Section 4.5).
However , AD' s debilitating effects are not sudden, and we may expect
patients to be in receipt of informal care bet ween onset and institutionalisa-
tion, which might lead .,24 l~.,-+t to be relatively light.
We used the dat a from the st udy by Jost & Gr ossber g (1995). Al t hough it
is not the largest st udy, it does have advant ages:
(a) it is a brain bank st udy, so all AD cases were confi rmed by aut opsy (the
onl y reliable met hod of diagnosis);
(b) there were no censored cases; and
(c) the time from onset to i nst i t ut i onal i sat i on is estimated.
A MATHEMATI CAL MODEL OF ALZHEI MER' S DISEASE AND THE APOE GENE
TABLE I
MEAN AND MEDIAN TIMES TO INSTITUTIONALISATION (INST'N) OR FIRST EVENT FOR AD PATIENTS
81
A g e a t T i me ( y e a r s ) t o % f o r wh i c h
R e f e r e n c e i s t e v e n t i s
On s e t E n t r y I n s t 'n 1s t E v e n t d e a t h
Be r g e t al , ( 1 9 8 8 ) 7 1 . 4 ( I ) 7 . 1 %
H e y m a n e t al. ( 1 9 9 7 ) 72. 0 ( 2) 3.1 ( 2) 1 3 . 1 %
J o s t e t al . ( 1 9 9 5 ) 75.1 ( 3) 4. 3 ( I ) 1 5 . 0 %
S e v e r s o n e t al . ( 1 9 9 4 ) 79. 4 ( I ) 2. 5 ( 2) ( 4) 1 0 . 0 %
( I ) Me a n .
( 2) Me d i a n .
( 3) M e a n a g e a t o n s e t o f A D , i f i n s t i t u t i o n a l i s e d , e s t i m a t e d a s ( m e a n a g e a t i n s t i t u t i o n a l i s a t i o n
- m e a n t i me t o i n s t i t u t i o n a l i s a t i o n ) .
( 4) M e d i a n t i me f r o m o n s e t e s t i m a t e d a s 5. 6 y e a r s .
Since we c a nnot di st i ngui sh genot ypes here, we will j ust write 23 ll, x+, and
24 t23 /24
lq.+, i nst ead of P,,-+t and #,.+,, respect i vel y. Gui ded by these dat a we deri ve
" - " ~ 3 . . . . p24 (t he
mome nt estmaates of P'2-+~ (t he force of l ns t l t ut | onahs at t on) and 'x+,
force of mor t al i t y of an AD pat i ent pri or t o i nst i t ut i onal i sat i on) . We defi ne
bel ow t he usual i ndi cat or f unct i ons (/j) and sampl e pat h f unct i ons (Nj~) in
respect of a single life (see Ma c dona l d (1996b)):
1 i f life is in s t a t e j at t i me t
l j ( t ) = 0 ot her wi se
I i f life t r ansf er s f r om s t a t e j t o st at e k at t i me t
d N j ~ ( t ) = 0 ot her wi se
/ 0
N j ~ ( T ) = d N j k ( t ) = No. of t ransfers f r om s t a t e j t o st at e k
Al so let P/ J, be the pr obabi l i t y t hat a life in s t a t e j at age x is in s t a t e j at age y.
Then equat i on (3) bel ow is the mean age at onset of AD, given t hat the life
was event ual l y i nst i t ut i onal i sed wi t h AD (as in Jost & Gr os s ber g (1995)):
[ / , ;
E x + I i ( l ) d l N 2 3 ( c o - x ) = 1 and It (x) = 1 =
e , , { . L , 2 X - ~ J ' " ( I - - " 12 II 2 3 p . ~ ; d s } d l " ( 3 )
f . ~ , 1 2 n l l H , t ".x' t { J ' ~ [)',23p~,2d'~"} d l '
8 2 ANGUS MAC DONAL D AND DELME P RI T CHARD
equat i on (4) is the mean time from onset of AD to institutionalisation:
E 1 2 ( t ) d t N2 3 ( ~ - x ) = 1 and I t ( x ) = 1 =
7 2 e ' , ; ( , - , > 2 3 J s } a , .
. w, , 1 2 Di i w 23 22
J . , . , , , , . , . , { f , P , . , . d s } d t
and equat i on (5) is the probabi l i t y t hat an AD patient dies before becomi ng
institutionalised. The upper age bound, denot ed co, is taken to be 120 years:
P[ N24( c o- x) = 1 NI 2 ( W' - x ) = 1 and / i t x ) - - 1 ] :
w 12 II w 24 22 ,
f.,. #, e, . , {y, I,~ P , s d s } d l (5)
co 12 II
f x # t P.vt d t
Setting equat i ons (3), (4) and (5) equal to their est i mat ed values from Tabl e 1,
we obt ai n 3 equat i ons, which can be solved for at most 3 unknown
paramet ers. The paramet ri c forms we chose were as follows:
(a) /,12x+, = A + t.L.~.+tat), where &,.+,,4o is given by equat i on (2). This Makeham
term adj ust s the incidence of AD to a level t hat gives the same mean age
at onset (for AD patients who become institutionalised).
23
(b) &,.+, = D. We felt that the dat a did not suppor t anyt hi ng more el aborat e
than a const ant intensity.
( C) / ~24 14
x +t = P & , . + t . That is, the mort al i t y of an AD patient before becomi ng
institutionalised is a pr opor t i on of baseline mortality.
(d) ~t,,.+t,14 baseline mort al i t y, was taken as AM80 mortality, using the
Gomper t z appr oxi mat i on given by equat i on (I). Al t hough it is
appr opr i at e to allow for fut ure i mprovement s in mort al i t y in applica-
tions, it is not appr opr i at e to do so in est i mat i on based on past data. The
values of D and P do not depend strongly on the baseline mortality.
Solving these equat i ons numeri cal l y yields the solutions:
A = 0.02025038 D = 0.18895779 P = 0.33502488.
The transition intensities .f13 and 24 .,.+, ~t,.,.+, are summari sed in Tabl e 3.
The Makeham term, A, is a nuisance paramet er used to adj ust the
incidence of AD so that the mean age at onset in the model is the same as
t hat in the data. Its onl y pur pose here is to i mprove the est i mat i on of the
ot her terms, as the survival of a cohor t of AD patients is st rongl y related to
their mean age at onset. It does not furnish an est i mat e of the incidence of
AD in the whol e popul at i on, which was descri bed in Section 4.3.
4. 5. Mort al i t y o f Li ves with AIzheimer's Di sease
AD patients have been found to suffer higher mort al i t y than the general
popul at i on (Barclay e t a l . , 1985(b); Bonai ut o e t a l . , 1995; Bracco e t a l . , 1994;
A MATHEMATI CAL MODEl_ OF ALZHEI MER' S DISEASE AND THE APOE GENE 83
Burns e t al . , 1991; van Di j k e t al . , 1991; Evans e t al . , 1991;
Heyman e t al . , 1996; M61sfi e t al . , 1986; Treves et al . , 1986). However ,
there is little agreement on the magni t ude of the increase, or its dependence
on age at onset, dur at i on since onset, sex, race, level of educat i on, marital
status, level of cognitive i mpai rment , familial/non-familial AD and level of
behavi oural i mpai rment . The main fact ors we need to consi der are:
(a) Th e ma g n i t u d e o f t he i nc r e as e in mo r t a l i t y f o r A D l i ves. The mort al i t y of
lives with AD has been investigated using different met hodol ogi es. For
example, Evans et al . (1991), est i mat ed the relative risk of deat h for
AD patients as 1.44 (95% confi dence interval 1.05-1.96) times t hat of the
unaffected. Ot hers have suggested t hat AD has onl y a small i mpact on
mortality: Barclay e t al . (1985a) claimed t hat well-tended individuals
may have life expect ancy close to normal , and Sayet t a e t al. (1986) found
t hat survival did not depend on disease acquisition.
(b) Th e e f f e c t o f age at ons e t on r e l at i v e mo r t a l i t y . The mort al i t y of patients
with AD increases with age ( Bonai ut o e t al. 1995; Burns e t al . , 1991).
Most studies into survival times have found no relation bet ween
age at ent ry into the st udy and relative survival (Barclay e t al . , 1985b;
Bracco et al . , 1994; Heyman e t al . , 1996; Stern et al . , 1995;
M61sfi e t al . , 1986), except that Barclay e t al . (1985b) found that
younger lives had short er relative survival times. Diesfeldt e t al . (1986),
investigating survival from onset of AD, found t hat AD pat i ent s with
onset before age 76 had reduced survival times, but not those with later
onset. Compar i ng the t wo met hods of investigation, Walsh et al . (1990)
f ound that ol der age at onset affected survival adversely, whereas ol der
age at ent ry into the st udy did not; a possible expl anat i on was that ol der
pat i ent s have sympt oms for a short er time before present at i on. Al t hough
no definitive relationship bet ween age at onset and relative survival
emerges, it is clear that:
(I) survival with AD depends on age; and
(2) if age at onset affects relative mort al i t y, the relationship is onl y weak,
but possi bl y st ronger at younger ages.
In terms of the model in Fi gure 1, this suggests t hat mort al i t y in state i3
(institutionalised from AD) coul d be model l ed by the addi t i on of a
Makeham term to the normal force of mort al i t y; the latter is age
dependent , and the Makeham term will be less significant at ol der ages.
(c) Th e e f f e c t o f t he d u r a t i o n o f A D on r e l at i v e s ur v i v al . Perhaps surprisingly,
the dur at i on of AD has not been found to be associ at ed with increased
mort al i t y (Barclay e t al . , 1985a; Bracco e t al . , 1994; Burns e t al . , 1991;
Diesfeldt e t al . , 1986; Heyman e t al . , 1996; Sayet t a e t al . , 1986;
Wal sh e t al . , 1990). That is, AD patients with long dur at i on of s ympt oms
do not suffer higher mort al i t y than patients, of the same age, with short
dur at i on of sympt oms. In terms of the model, this means that the
mort al i t y of lives in states 2 and 3 (onset of AD and institutionalised
from AD) does not depend on the time spent in these states. This is
especially conveni ent , as it allows us to wor k in a Mar kov framework.
(d) The ef f ect o f gender on rel at i ve survi val wi t h AD. Ma n y researchers have
f ound that the differences in survival bet ween men and wome n wi th AD
can be expl ai ned by the usual mortal i ty differential between men and
wome n (Beard et al . , 1994; Bonai ut o et al . , 1995; Bracco et al . , 1994;
Burns et al . , 1991; Heyman et al . , 1996; Walsh et al . , 1990), t hough
Barclay et al . (1985a), did find greater differences. In terms o f model l i ng,
al l owi ng for the normal differences in mortal i ty bet ween genders shoul d
be sufficient.
Tabl e 2 summari ses the literature on survival with AD. Since we cannot
di sti ngui sh genot ypes here, we will just write/.t?,.+,t4 instead o f ] " z i 3 4 x + , ' As in the
previ ous secti on, we can write down the mean age at onset (see equat i on (6))
and the mean survival time (see equat i on (7)) in the model o f Figure 1:
E / ,
E x + I i ( t ) d t N i 2 ( c o - x ) : 1 a n d l l ( X) = 1 :
x + f ~ (t - x ) #12P ' ' at
- " t (6)
f ~ o , i z p l i d t
, c I ' l l . v t
E I 2 ( t ) +1 3 ( t ) d t Nj z ( w- x ) = 1 and I i ( x ) = 1 =
, f . ~ l - t l 2 p l l { f ~ ( S - - l ) ( I t ~ . - J r - I - t s 2 4 ) e t s 2 2 d s + jtcw/Z:"~3 ,',23r, s JsrW ( r _ s ) l _ z ~ 4 p ~ ) d r d s } d t ( 7 )
~v w 12 1 I
t t t P x t d t
TABLE 2
SUMMARY STATISTICS ON SURVIVAl. TIMES OF AD PATIENTS
Mean (Medi an) Mean ( M e d i a n )
Reference Age at Onset S u r v i v a l T i m e Addi ti on to #x34+t
Barclay et a l . (1985a) (73.3) yrs (8.1) yrs 0.15829
Bracco et a l . (1994) (72.4) yrs 7.3 yrs 0.25259
Diesfeldt et a l . (1986) 75.6 yrs 7.2 yrs 0.21056
Heyman et a l . (1996) (69.2) yrs (9.7) yrs 0.10993
J o s t e t a / . (1995) 75.1 yrs 8.11 yrs 0.13345
Kokmen e t a l . (1988) 80.4 yrs 6.2 yrs 0.26420
Treves e t a l . (1986) 73.9 yrs (9.3) yrs 0.08135
Average 0.17291
Setting equations (6) and (7) equal to their estimated values in Table 2, and
noting that we have estimates of 23 and 24 /1,.,.+, #x+, from the previous section, we
obtain 2 equations, which can be solved for at most 2 unknown parameters.
The parametric forms we used are as follows:
AD ~L AD is given by equation (2). This is just the (a) /z 12 = A +/Zx+,, where .,-+t
x+t
addition of a Makeham term to the force of incidence of AD, shifting the
latter to a level that gives the estimated age at onset.
(b) ~4 = K +AM80
It,~.+t l.t,,-+t. This is a Makeham term as discussed in (d) above.
The estimated values of K for each of the references cited are given in the last
column of Table 2. They range from about 0.08 to 0.27, with an average of
0.173. The Makeham term A is, again, only included to improve the
estimation of the other terms (see the end of the previous section).
4.6. Summary of the Transition Intensities for the AD Model
For clarity, we summarise the transition intensities estimated here. They all
have the form:
IZ!,.~+, = A + DB e c o ' +' ) (8)
and the calculated values are given in Table 3. Three values are given for
34
tZx+,, an upper bound, mean value and lower bound to enable a check of how
sensitive the results are, in any particular investigation, to this term.
TABLE 3
SUMMARY OF TRANSITION INTENSITIES FOR TIl E AD MODEL WITH BASELINE MORTALITY 1 0 0 % ( 6 5 %) OF AM80
AND AF80
Parameter Values
Transition B ( x 10 -5) C ( 10 -2)
Intensity A D
Mal e Female Male Female
,u 24 0 0.33502 9.4116 2.5934 8.4554 9.3605
XWt
(0.21776)
23 0.18896 0.00
# x + t
~34 (Lower bound) 0.08 1.00 (0.65) 9.4116 2.5934 8.4554 9.3605
x +t
/z 34 (Mean) 0.17291 1. 00 (0.65) 9.4116 2.5934 8.4554 9.3605
X+I
34 (Upper bound) 0.27 1.00 (0.65) 9.4116 2.5934 8.4554 9.3605
#x+t
5. ESTIMATION OF TRANSITION INTENSITIES DEPENDING
ON APOE GENOTYPE
5.1. Popul ati on Frequencies of the ApoE Genotypes
Tabl e 4 shows the popul at i on frequencies of the ApoE genot ypes estimated in
several studies. Ot hers are Cor der e t al . (1994, 1995), Gomez-l sl a e t al . (1996),
Leht ovi rt a e t al . (1996), Liddell e t al . (1994), Lopez e t al. (1998), Nal bant ogl u
e t al. (1994), Poirier e t al . (1993), Roses (1995) and Tsai e t al . (1994).
Some features are clear: the e 3 / s 4 genot ype is not uncommon (about
21%) while the e2/ e4 and e4/ e4 genot ypes are quite uncommon (about 3%
and 1% respectively). We mi ght expect to find lower proport i ons of
' danger ous' genot ypes at ol der ages, because these lives suffer a higher rate
of AD onset, but the two age-related studies (Bickeb611er e t al . (1997) and
Cor der e t al . (1995)) gave conflicting results. However, there is reasonable
agr eement on the gene frequencies at ar ound ages 60-70, which is what we
need for our modelling.
Farrer e t al . (1997) is a meta-analysis, combi ni ng the results of 40 ot her
studies, including 6,264 Caucasi an subjects. As it is the largest study, and
differentiates by ethnic group, gender and ascert ai nment met hods, and as the
ApoE s4 allele was found with the same frequency in respect of AD
di agnosed at aut opsy and clinically di agnosed probabl e AD, we use its
est i mat ed gene frequencies, namely: e2/ e2 0.008; e2/~3 0.127; s 2 / e 4 0.026;
e3/ e3 0.609; e3/E4 0.213; e4/ e4 0.018. The sharp-eyed will notice t hat these
sum to 1.001, because of roundi ngs used in Far t er e t al . (1997), but we have
left this small di screpancy unadj ust ed.
5.2. Genetic Risk of AIzheimer's Disease
When we have a het erogeneous popul at i on, it is often conveni ent to think of
a given intensity in each sub-popul at i on as a multiple (not necessarily
const ant ) of a 'baseline" intensity, either in one of the sub-popul at i ons or in
an aggregat ed ' average' popul at i on. Similarly, if Pt and P2 are the
probabilities of an event in popul at i ons 1 and 2 respectively, the relative
risk in popul at i on 2 (with respect to popul at i on 1) is P 2 / P l . A related
quant i t y is the odds ratio: the odds in popul at i ons i and 2, respectively, are
p l / ( i - P l ) and p2/ ( l - P2 ) , and the odds ratio is:
p2(l - p, )
p, (1 - P2) " (9)
When intensities are small, the odds ratio is a good approxi mat i on to the
relative risk. In many studies, the published results are either relative risks or
odds ratios.
TABLE 4
ESTIMATED POPULATION FREQUENCY OF AI oE GENOTYPES
>
Country / No. o f Age
Reference Sex
Ethnicity Lives Group
Farrer et al. (1997) Caucasian 6,262 M & F All
Afr-Amer 240 M & F All
Hispanic 267 M & F All
Japanese 1,977 M & F . All
Bickeb611er et al. (1997) France 1,030 M & F All
316 M All
40 M < 60
93 M 60-69
80 M 70-79
103 M > 80
714 F All
47 F < 60
75 F 60-69
143 F 70-79
449 F _>80
van Duijn et al. (1995) Netherlands 532 M & F < 65
228 M < 65
304 F < 65
Evans et al. (1997) E. Boston 490 M & F > 65
Jarvik et al. (1996) not given 310 M & F 48-98
I 17 M 48-98
193 F 48-98
Lambert et al. (1998) not given 308 M & F not given
Levy-Lahad et al. (1996) not given 304 M & F not given
Lucotte et al. (1997) France 248 M & F > 65
>
Allele Frequency Genotype Frequency
m
e2 e3 e4 e2/ e2 e2/e3 e2/ e4 e3/ e3 e3/ e4 e4/ e4 ~:
>
0.084 0.779 0.137 0.01 0.13 0.03 0.61 0.21 0.02 ~,
0.083 0.727 0.190 0.01 0.13 0.02 0.50 0.32 0.02 >
t "
0.067 0.823 0.110 0.00 0.12 0.01 0.67 0.18 0.02 K
0.042 0.869 0.089 0.00 0.07 0.01 0.76 0.16 0.01 O
0.085 0.770 0.145 0.01 0.12 0.03 0.59 0.24 0.01 r~ r'-
0.070 0.815 0.105 0.01 0.10 0.02 0.67 0.19 0.00 O
"r]
0.050 0.800 0.150 0.00 0.08 0.02 0.62 0.28 0.00 >
0.070 0.845 0.085 0.01 0.12 0.00 0.70 0.17 0.00 t-
N
0.060 0.840 0.090 0.00 0.10 0.02 0.71 0.16 0.00 :
m
0.081 0.795 0.125 0.01 0.11 0.03 0.64 0.20 0.01 ~-
0.090 0.750 0.170 0.01 0.13 0.03 0.55 0.27 0.02 m
7~
0.075 0.735 0.190 0.00 0.15 0.00 0.51 0.30 0.04 u5
0.075 0.730 0.195 0.01 0.11 0.02 0.52 0.31 0.03
0.065 0.765 0.160 0.00 0.12 0.01 0.56 0.29 0.01
>
0.095 0.750 0.165 0.01 0.13 0.04 0.56 0.25 0.02
0.103 0.731 0.165 0.01 0.17 0.02 0.51 0.27 0.02 >
0.105 0.705 0.190 0.01 0.16 0.03 0.48 0.29 0.03 Z
0.100 0.745 0.155 0.01 0.17 0.01 0.53 0.26 0.02 .--I
.-r
0.062 0.854 0.084 0.01 0.09 0.01 0.74 0.14 0.01
0.098 0.750 0.132 0.01 0.12 0.05 0.59 0.20 0.01 >
0.068 0.791 0.124 0.02 0.09 0.02 0.64 0.21 0.01 O
m
0.117 0.725 0.137 0.01 0.15 0.07 0.55 0.20 0.01
m
0.081 0.805 0.114 0.00 0.13 0.02 0.65 0.18 0.01 Z
0.100 0.765 0.135 0.01 0.13 0.05 0.60 0.21 0.01 m
0.069 0.804 0.127 0.02 0.09 0.01 0.65 0.22 0.01
O 0
-...O
Few studies report prospectively the incidence of AD by genot ype. Two
t hat do are Evans e t a l . , (1997) and Slooter e t a l . , (1998). Both have quite
small st udy popul at i ons, and neither provides age specific estimates of AD
risk, so they are not appropri at e for our purposes.
Tabl e 5 gives the Odds Ratios (ORs) of AD and 95% confidence intervals
from a number of genetic studies. The ' reference' popul at i ons (also shown in
the table) were either the ~3/~3 genot ype or the three non-e4 genot ypes
combi ned. The est i mat ed ORs vary consi derabl y across studies. For
example, estimates of the OR for the e3/ e4 genot ype (relative to the
~3/e3 genot ype) range from 1.8% to 3.7%, and for the e4/ 4 genot ype, from
6.2% to 30.7%. Some of the variation may be explained by the differences
between the studies themselves. In particular, differences may arise from: the
met hod of ascert ai nment of patients, the count ri es of study, the met hod of
diagnosis of AD, the age st ruct ure of the samples, the reference/risk
genotypes, and whet her they are cross-sectional or prospective studies. We
make the following observations:
(a) The st udy by Lopez e t a l . (1998) suggests t hat the risk of AD associated
with the ApoE e4 allele may be different in different countries.
(b) In support of the above, Mayeux e t a l . (1998) found that the association
between ApoE and AD may depend on et hni c group and, in particular,
may not be present in black populations.
(c) Despite the differences between studies: the presence of one or two
e4 alleles is consistently report ed to be significantly associated with AD;
and homozygot es are generally report ed to have higher risk of onset of
AD t han het erozygot es.
(d) The weakest associations between ApoE and AD were report ed in the
two prospective studies, those by Evans e t a l . (1997) and Slooter e t al .
(1998). This is as expected for the reasons given in Section 4.1.
For our purposes, the genetic risk of AD at different ages is i mport ant .
Few studies have considered this; the odds ratios from two t hat have are
given in Table 6. Bickeb611er e t a l . (1997) is based on hospital admissions,
and Cor der e t a l . (1994) on aut opsy cases; both use e3/E3 as the reference
popul at i on. Al t hough some ORs are missing, because of small sample sizes,
the trends are fairly clear:
(a) The odds of AD among the higher risk genot ypes (e3/ e4 and e4/ e4) fall
with age. This may be expected as higher risk genot ypes will succumb to
AD mor e rapidly, reduci ng the proport i on of such genot ypes within the
popul at i on.
(b) Conversel y, the prot ect i on conferred by the lower risk genot ype (e2/ e3)
seems to weaken with age, possibly as tiffs genot ype becomes mor e
common in the remai ni ng popul at i on.
A MATHEMATICAL MODEL OF ALZHEIMER'S DISEASE AND THE APOE GENE
TABI.E 5
A G G R E G A T E D ODDS RATI OS OF AD FOR T HE APoE e4 ALLELE
89
As c e r t ai nme nt Re f e r e nc e Odds Rat i o
Re f e r e nc e GenoO, pe ( 2 )
S c h e me ( I ) Ge not y pe ( 2 ) Me a n 9 5 % CI
Evans et al. (1997) P
Frisoni et al. (1995) C
Jarvik et al. (1996) C
Kuusisto et al. 0994) P
Lambert et al. (1998) C
Lehtovirta et al. (1995) C
Liddell et al. (1994) C
Lopez et al. (1998) C
Mayeux et al. (1993) P
Myers et al, (1996) P
Nal bant ogl u et al. (1994) A
Slooter et al. (1998) P
Tsai et al. (1994) C
~3/e3 e3/ e4 & e4/ s4 2.27 I. I-4.9
-/- e4/ - 6.6 2.2.-19.5
e4/ e4 17.9 4.5-70.5
~3/~3 ~2/~3 0.4 0.2-0.96
~2/~4 1.4 0.6-3
e3/ s4 3.1 2-4.7
~4/s4 30.7 7-13 I
-/- ~4/- 2.7 1.4-5.2
~4/~4 9.1 3.5-23.4
-/- E4/- & c4/ s4 4.66 3.14-6.93
-/- ~4/- 5.1 2.4-1 I.I
~4/~4 21.4 2.8-166.3
-/- e4/ - 2.2 I. 1-4.7
~4/e4 10.7 2.3-48.8
-/- e4/ - & e4/ e4 2.34 (3) 1.03-5.55
E4/- & e4/~4 3.64 (4) 1.78-7.69
-/- c4/ - 4.2 (5) 1.8-9.5
s4/E4 17.9 (5) 4.6-69.8
e4/ - & ~4/e4 15. 3 (6) 3.0-78.1
e4/ - & E4/e4 0.7 (7) 0.1-6.4
E4/- & e4/E4 4.5 (8) 0.7-27.7
s3/ e3 s3/ e4 3.7 1.9-7.5
E4/E4 30. I 10.7-84.4
-/- e4/ - & e4/ e4 15. 5 6.2-38.5
~3/e3 e2/ e3 0.4 0.1-1.0
E2/e4 1.3 0.2-8.5
e3/~4 1.8 1.0-3. I
e4/ e4 6.2 1.4-28.2
-/- e4/ - & e4/ e4 4.6 1.9-12.3
E4/- 3.6 1.5-9.8
(I) Ascert ai nment Scheme: C indicates clinic/hospital;
A, aut opsy/ br ai nbank.
(2) Dash (-) represents ~2 or E3 alleles.
(3) Study popul at i on - Gerona, Spain.
(4) Study popul at i on - Pittsburgh, USA.
(5) Mixture of White, Black and Hispanic ethnic groups.
(6) White ethnic group only.
(7) Black ethnic group only.
(8) Hispanic ethnic group only.
P, popul at i on/ communi t y; and
90 ANGUS MACDONALD AND DELME PRITCI-IARD
TABLE 6
ODDS RATI OS OF AD BY GENOTYPE AND AGE
Bickebb'ller et al. (1997) Corder et al. (1994)
Age Odds Ratio Age Odds Ratio
Group Genotype Mean 95% CI Group Genotype Mean 95% CI
60-69 e2/ ~3 60-66
70-79
8 0 +
6 0 +
E 2 / c 4
e 3 / e 4
~4/ e4
E2/ e3
e 2 / e 4
e3 I s4
e 4 / e 4
E2/E3
e 2 / e 4
e 3 / e 4
s 4/ ~4
e 2/ e 3
e 2 / e 4
c 3 / e 4
e4 I~4
0.3 0. 0-2. 3 ~3/ e3 0.1
- - E2/E4 1.2
3. I 1.4-6.9 e 3 / e 4 I I. l
29.1 3. 6-239. 5 E4/ e4 123.8
0. 4 0. 1-2. 3 67-74 e 2/ e 3 0.3
- - ~2/ e4 I. I
3.2 1.5-6.6 e 3 / c 4 4. 6
- - e 4 / c 4 20.8
0.3 0. 0-2. 6 75-92 e 2/ e 3 0.5
- - ~2/ e4 1.6
1.3 0. 5-3. 4 e 3 / e 4 3.2
- - e 4 / e 4 I 0. 0
0. 4 0. 1-0.9 6 0 + ~2/ e3 0.3
1.6 0. 5-5. 5 e2/ ~4 I. I
2.2 1.5-3.5 ~3/ e4 4. 4
11.2 4. 0-31. 6 e4/ ~4 19.3
w
These t rends are suppor t ed by the met a-anal ysi s by Farrer e t al . (1997), and
as it is from this st udy t hat we t ake our est i mat es of the ApoE genot ype
risks, we cite some relevant details:
(a) The aggregat e relative odds from Far t er e t al . (1997) (relative to the
e3/~3 genot ype) are shown in Fi gure 3.
(b) The genot ype risks of AD were not significantly different in respect of
Caucasi an males and females, except in the case of the e3/ e4 genot ype,
for which women had a significantly higher risk of AD. The relative odds
of AD by ApoE for Caucasi an men and women are shown in Figures 4
and 5. (The aut hor s kindly provi ded us with the numerical values of the
odds ratios; confidence intervals were not available. )
(c) The genot ypes e2/ e2 and e2/ e3 were combi ned as there were very few
e2/ e2 genot ypes, and the risks associ at ed with the t wo genot ypes
appear ed to be similar.
(d) Not e t hat the risks associ at ed with the ApoE E4 allele were consi derabl y
higher than t hose found in the t wo popul at i on- based studies by
Evans e t al . , (1997) and Sl oot er e t a l . , (1998).
A MATHEMATICAL MODEL OF ALZHEIMER'S DISEASE AND THE APOE GENE 9!
D
e3/ e4
+ e4/e4
/ x ~ e2/ e4
e2/ e3
0
or-
O 0 ~ ~ 0 0 0 0 0 0 0
4 0 4 5 5 0 5 5 6 0 6 5 7 0 7 ' 5 8 0 8 ' 5 9 0
Age (years)
FIGURE 3: Odds ratios (ORs) of AD relalive to E3/~3 genotype for males and Females combined.
Source: Farter et al . (1997).
0
0
v
0
r r
0
0
Men e3/ e4
+ Women e3/ e4
~ Men e4/ e4
4 0 4 ' 5 s o s . 5 6 0 6 . 5 7 0 7 . 5 8 0 8 . 5 9 0
Age (years)
FIGURE 4: Odds ratios (ORs) of AD relative to e3/e3 genotype for e3/e4 and e4/e4 genotypes.
Source: Farrer e t al , (1997).
2.
9.
0 o .
o
r r
-8o
'1o
0
,,
o -
Men 02,/04
+ Women e2./e4
v Men e2Je3
o Women e2__/e3
4 ' 0 4 ' 5 s ' o s ' s o ' o o ' s i o " / s 8 ' o 8 ' s 9 ' 0
Ag e ( year s)
FIGURE 5: Odds ratios (ORs) of AD relative to E3/E3 genotype for s2/e2 or ~2/e3 and s2/e4 genotypes.
Source: Farrer et al. (1997).
For use in our model, these odds ratios have to be converted into relative
risks. More precisely, we have to find a plausible set of age- and genotype-
dependent transition intensities that are consistent with the odds ratios and
together are consistent with the aggregate incidence of AD. There is no
unique solution to this problem. The method we used was to model the
incidence of AD for the ith genotype as:
i ~i l 2 : i A ~ ( 1 0 )
x+t ----- r l J j , + t l z ~ + t
where:
(a) iz.,.+rA is the aggregate incidence rate of AD (from Section 4.3);
(b)f,.+1 is a parametric function representing the risk relative to the
aggregate incidence rate, where we take f~+t = 1 in the case of the
e3/e3 genotype; and
(c) rl is a constant chosen so that the aggregate incidence of AD based
on the modelled intensities is consistent with the aggregate incidence
A D
I-l~.x-+t.
We did this for males and females separately and combined. We confined
our attention to ages 60 and over, in order to get a better fit in the age range
of interest in applications. The form of the ORs, either rising to a peak and
then falling, or gently declining, suggested a similar pattern of relative risks,
A MATHEMATI CAL MODEL OF ALZHEI MER' S DISEASE AND THE APOE GENE 93
and we found the fol l owi ng family of funct i ons sat i sfact ory (not e that
const ant relative risks, or pr opor t i onal hazards, result in odds rat i os with
exponent i al growth):
f ~ + t = E e - r ( ( x +t ) - k ' ) 2 - G( ( x +t ) - k 2 ) 4- H . (11)
Act uari es will recognize this as a GM( 1, 3) funct i on, familiar in the
graduat i on of life tables (Forfar, McCut cheon & Wilkie, 1988), al t hough
as descri bed bel ow either F o r G is set to zero. We found this flexible enough
to give a good appr oxi mat i on to the ORs, and also suitable for ext rapol at i ng
beyond age 90. The fitting pr ocedur e was as follow.s:
(a) by consi deri ng the form of the OR, we set either F = 0 (giving an
exponent i al funct i on) or G = 0 (giving a bell-curve function), and set H
equal to 0 or 1;
(b) the best value Ofkl o r k2 was found, to the nearest integer, by inspection;
(c) the resulting ORs were calculated from the model in Fi gure I using the
intensities from previ ous sections; and
(d) the remaining coefficients (either E and F, or E and G) were fitted by least
squares.
For the
( a ) / ~ 1 4
x +l
and
( b ) ~ 2 3
x+/
(c) u Z ,
( d ) ~34
. r+t
cal cul at i ons in (c) above we used the following paramet ers:
0.65 x AMS #x+~ for males and 14 xAb~0
= ,ux+ t = 0.65 #.~.+~ for females
in aggregate, where AMS01~.,.+t and A~'~0/.Lx+ t are given by equat i on 1.
= 0.189, cal cul at ed in Section 4.4.
= 0.335 i~1~.4 t calculated in Section 4.4.
14
= 0.173 +/-!,.+t, the mean value calculated in Section 4.5.
The fitted paramet ers are given in Tabl e 7. Fi gure 6 shows t hat the model l ed
ORs closely reproduce the estimates from Far r er e t a l . (1997) (see Fi gure 3;
onl y the aggregat e ORs are shown, and the model l ed ORs start at age 61
because we start with unaffected lives at age 60).
To det ermi ne the par amet er rl, we cal cul at ed the aggregat e incidence of
AD in the whol e model, and fitted this to #Ao by least squares, l f i p II is the
A'q-t --xt
probabi l i t y t hat a life with genot ype i, heal t hy at age x, is unaffect ed by AD
at age x + t, and if p(,. is the popul at i on frequency of the ith genot ype at age x
then the aggregat e incidence of AD is:
Aggregat e incidence of AD at age ( x + t ) = r l p.i. -_,-tJ.,-+~ /~.,-+~
i=1
94 A N G U S MA C D O N A L D A N D D E L ME P R I T C H A R D
TABLE 7
PARAMETERS FOR THE RELATIVE RISK OF A D FOR MALES, FEMALES AND IN AGGREGATE, BY GENOTYPE
G e n d e r G e n o O , p e
Parameter Values
E F G H kl kz r,
Bot h
F e ma l e
Ma l e
~4/ E4 13.5 0. 00529 0 I 60 - 0. 93
e 3 / e 4 2. 98 0. 00312 0 I 62 -
e 2 / e 4 2. 87 0. 00938 0 1 68 -
E 2 / e 2 &e 2 / e 3 0. 754 0 0. 00859 0 - 60
E4/ ~4 10.4 0. 00504 0 1 60 - 0. 88
E3/ E4 3. 68 0. 00319 0 I 62 --
e 2 / e 4 4.21 0. 01020 0 1 68 -
e 2 / e 2 & e 2 / e 3 0. 675 0 0. 00692 0 - 60
c 4 / e 4 8. 94 0. 00656 0 1 60 - 1.27
e 3 / ~4 1.92 0. 00103 0 0 51 -
e 2 / e 4 1.42 0. 00506 0 0 67 -
e 2 / e 2 & e 2 / E 3 0. 434 0 0. 01600 0 - 60
O , , I
v -
o
O
CO
n "
/ ' }
" O
C q
O
+ e 4 / e 4
. . . . . . . . . . . + ~ e 3 / e 4
,.......
. . . . . . . . . . ~ e 2 / e 4
. . . . . O e 2 / e 2 or e 2 / e 3
" . . .
' . . . .
. . . . . . . . . . . . . . . . M o d e l Ap p r o x i ma t i o n
- . . .
. . ~ , . .
" - .....
" '. . , .
' " . . . . . .
- . . .
" ' " . . . % . .
~ v ~T~.::: ::=,~6 . . ~. ..~
""::B ...... "" ""...
. . . . . . . . : : " : : : : " " " " ~ L ~ . i i L i l l i L i l i [ [ i i : " . . ~
" " ' " ' ~ 7
. . . . . . . . . . . 0 . . . . . . . . . . . . . . . . . . . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . 0 . . . . . . . . . . . . . . . . . . . . . . . . . ........................... ...........................
6 0 6 5 70 7 5 8 0 8 5 9 0
Age ( years)
FI GURE 6: O d d s r a t i o s ( OR s ) o f A D r e l a t i v e t o c 3 / ~ 3 g e n o t y p e f r o m F a r r e r e t a l . ( I 997) , c o mp a r e d wi t h O R s
c o mp u t e d u s i n g mo d e l l e d r e l a t i v e r i s k f u n c t i o n s .
A M A T H E M A T I C A L M O D E L O F A L Z H E I M E R ' S D I S E A S E A N D T H E A P O E G E N E 95
We t o o k x = 60, and for the p!,. we used the gene frequenci es o f the
Cauc as i an cont rol po pul a t i o ns in Farrer et al . ( 1997) , whi ch were gi ven in
Sect i on 5.1 The i nci dence o f AD, AD #x+t, was taken as that es t i mat ed in
e quat i on 2 and the oc c upanc y probabi l i t i es, i pl l were cal cul at ed by
- - A ' I '
s ol vi ng Ko l mo g o r o v ' s f orward e quat i ons numeri cal l y usi ng a Runge - Kut t a
al gori t hm wi th step size 0. 0005 years ( Cont e & De Boor, 1972).
The val ues o f r~ are gi ven in Tabl e 7. The adj us t ment to the overal l level
onl y had a margi nal effect on the model l ed ORs for the i ndi vi dual genot ypes .
The relative risk f unct i ons for mal es and f emal es are gi ven in Fi gures 7
and 8. For f emal es, the e 4 / e 4 , e 3 / e 4 and e 2 / e 4 ge not ype s are unambi gu-
ous l y high-risk; the relative risks exceed 1.0 at all ages. For mal es, onl y the
~ 4 / e 4 ge not ype conf ers hi gher risks at all ages. The e2 allele appears to be
protecti ve, so the e 2 / e 2 and e 2 / e 3 ge not ype s are l ow- ri sk, whi l e the e 3 / e 4
and e 2 / e 4 ge not ype s are initially at hi gher risk but are at l ower risk f rom
about age 75. The prot ect i on apparent l y gi ven by the e2 allele in mal es
means that the e 3 / e 3 ge not ype conf ers sl i ghtl y hi gher than average risk; thi s
is why, in Tabl e 7, r] > 1 for mal es. It mus t be remembered that dat a in
respect o f mal es are rel ati vel y sparse, and dat a in respect o f very ol d mal es
even sparser, s o these effects s houl d be treated wi t h caut i on; we can have
more conf i dence in the relative risks in respect o f f emal es.
CM.
O
O' ) -
CO-
t r
O r . . -
O
r r
t / I
"1~tO-
8
, ~- .
NI -
O -
\
\
\
\ \
\
\ . \
N x
" \ \ \
" \ \ \
\ \
\
\ ",,
r e . -
m -
e 4 / e 4 f e m a l e s
e 4 / e 4 m a l e s
e 3 / e 4 f e m a l e s
e 3 J e 4 m a l e s
"---.-.......
6 ' 0 6 . 5 7 ' 0 7 ' 5 8 ' 0 8 ' 5 9 ' 0
A g e ( y e a r s )
FIGURE 7: Mo d e l l e d r i s k o f A D , r e l a t i v e t o t he e 3 / e 3 g e n o t y p e , f o r e 4 / e 4 a n d e 3 / e 4 g e n o t y p e s . Ba s e d o n
o d d s r a t i o s f r o m F a r r e r et al. ( 1 9 9 7 ) .
96 A N G U S M A C D O N A L D A N D D E L M E P R I T C I - t A R D
0
0 3 .
0 D .
Of , . .
0
~ C O ,
n -
u )
"0 kt3.
"0
o
qq.
O"
- - - e 2 / e 4 f e m a l e s
---- e 2 / e 4 m a l e s
. . . . e 2 / e 2 & 0 2 / 0 3 f e m a l e s
- - - e 2 / e 2 & e 2 / e 3 m a l e s
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . : ~ . . . . . ~ " - : - ~ : ~ 2 4 Z - . = _ _ V 2
6 ' 0 6 ' 5 7 ' 0 7 ' 5 8 ' 0 8 ' 5 9 ' 0
A g e ( y e a r s )
F I G U R E 8: Modelled risk of AD, relative to the E3/ e3 genotype, for E2/E4 and ~ 2 / ~ 2 & e2/ ~3 genotypes.
Based on odds ratios from Farrer et al . (1997).
Th e s e ri sk e s t i ma t e s p r o b a b l y o v e r s t a t e t he t rue p o p u l a t i o n ri sks, p e r h a p s
q u i t e s u b s t a n t i a l l y , as t he y are f r om c l i ni c - a n d a u t o p s y - b a s e d s t udi e s , wh i c h
i n v e s t i g a t e pr e c i s e l y t he s ubj e c t s t ha t are af f e c t e d o r a l r e a d y k n o w n t o be at
ri sk. T o a l l o w f or t hi s p o s s i b i l i t y we wi l l a l s o c o n s i d e r mo d e l s a s s u mi n g t ha t
t he t rue rel at i ve ri sks are a p r o p o r t i o n m < 1 o f t h o s e e s t i ma t e d a b o v e . We
d o t hi s by a d j u s t i n g e q u a t i o n ( 10) s o t ha t f or g e n o t y p e i:
~ i l 2 i
.,.+, = r,,,{(f.~.+t - 1 ) m + I }/z~+ D ( 13)
whe r e f ~+t is as a b o v e , a n d rm is c h o s e n as a b o v e s o t ha t t he a g g r e g a t e
AD Va l u e s of r 0. 5 a nd r025 i n c i d e n c e o f A D i n t he mo d e l is c o n s i s t e n t wi t h iLx+ t.
are s h o w n i n Ta b l e 8.
T A B L E 8
r m FOR m = l , 0 . 5 AND 0.25
Gender r I to.5 r o. 2f i
Both 0.93 0.96 0.97
Female 0.88 0.94 0.97
Male 1.27 1.11 1.05
Figure 9 shows that the aggregat e incidence of AD in the genetic model
AD It also shows that increasing for bot h sexes combi ned is quite close to #x+~'
the level of relative risk tends to overst at e the incidence of AD at younger
ages, and to underst at e it at ol der ages; the reason is t hat higher relative risks
depl et e the high-risk groups more quickly, leaving a relatively healthier
popul at i on at ol der ages.
O
d
(D
o.
O
O
O
a
<~
mo
-
e- O
o.
O
O
d
O
O
Estimated Incidence A D
/
Incidence A D ( m = 0 . 2 5 ) / / / A
AggregatedAggregated Incidence AD (m = 0.5) ~.....~/
........... A g g r e g a t e d I n c i d e n c e A D ( m = 1 . 0 0 ) / . . ' /
...~'~ 2~
.... ~
...... ~.~
.............. ::.~5 ~
6 5 7 0 7 5 8 0 8 5 9 0
A g e ( y e a r s )
FIGURE 9: Compari son of estimaled population incidence of AD it.,A+~ with the aggregated incidence of AD
for different levels of relative risk, males and females combined.
Decreasi ng the level of relative risks for high-risk genot ypes means
increasing the relative risks for low-risk genot ypes. Usi ng a lower value of
rm will diminish any effects of the (possi bl y anomal ous) feature, not ed above,
t hat the e3/E4 genot ype is low-risk for males.
5. 3. C o m m e n t o n Mo d e l S e l e c t i o n
We chose a simple model for the relative risks (equat i ons (10) and (I 1)). We
did consi der alternatives, in part i cul ar cubi c pol ynomi al s and Ga mma
functions, but these gave poor er fits, and were less suitable for ext rapol at i on
(cubics to ol der ages and Ga mma funct i ons to younger ages). Also, it is
easily seen t hat if an OR is specified as a funct i on of time, and the transition
98 A N G U S M A C D O N A L D A N D D E L M E P R I T C H A R D
intensity in the reference popul at i on is given, the transition intensity in the
second popul at i on is det ermi ned (as the solution to an ODE); Figure 6,
therefore, gives good support for our choice of model. Fur t her refinement
seemed somewhat spurious, given the dat a we were using, and in view of the
maj or sensitivity analysis needed in respect of the domi nant par amet er m.
6. RESULTS
6. 1. Occupancy Probabi l i t i es
Figures 10 and 11 show occupancy probabilities up to age 90 for females
heal t hy at age 60, with high (m = I) and low (m = 0.25) relative risks,
respectively. Each shows:
(a) Occupancy probabilities in respect of each genot ype (with c2/ c2 and
c2/~3 combi ned).
(b) Occupancy probabilities calculated by aggregating all the genot ypes in
the model. In the not at i on of equat i on (12) the probabi l i t y of being in
i = 5 i i I j
state j (j = I, 2, 3, 4) at age 60 + t Is ~-~'~i-I P60 P60 t, where the sum is
over all genotypes. These are labelled ' Aggregat ed ~enot ypes' .
(c) Occupancy probabilities based on the aggregate incidence of AD, #Ao
A ' + I "
These are labelled ' Aggregat e Model'
With high relative risks (m = 1), the effect of the e4/ e4 allele is clear; AD
cases rise to a peak in the early 70s, by which time over 10% of the original
cohor t are in one of the AD states, and then fall away. A similar but smaller
effect can be seen for the e3/ e4 genot ype. With low relative risks (m = 0.25)
these features are all diminished; in part i cul ar the peaks in the early 70s
disappear.
We omi t the cor r espondi ng figures for males; the differences are as we
woul d expect, given the model l ed relative risks.
For males and females with low relative risks (Figure 11) the aggregat ed
results from the genetic model are very close to the results from the aggregate
(popul at i on) model For females with high relative risks, the rate of onset of
AD seems to be t oo low at younger ages and t oo high at ol der ages.
6. 2. Preval ence Rat es
Also of interest are prevalence rates, namel y the proport i on of those alive at
every age who are in each of the three live states. Figures 12 and 13 show
these, for females, including, for convenience, the two AD states combi ned.
We omi t the correspondi ng figures for males
The most striking feature is the prevalence of AD in respect of the
e4/ c4 genot ype under high relative risks (Figure 12); it increases al most
linearly. Again, for males and females the aggregat ed results from the genetic
model are quite close to those from the aggregat e model Moreover, t hey fall
within the range of prevalence rates actually observed. Breteler et al. (1992)
cite the following rates: 47. 2% at ages 85 and over (Evans e t a l . , (1989));
31. 7% at ages 85 and over (Pfeffer e t a l . , 1987); and 28. 0% at ages 90 and
over ( O' Connor e t a l . , 1989); some ot her studies gave lower figures.
6. 3. Ge ne Fr e que nc i e s at Hi g he r Ag e s
We have assumed t hat the gene frequencies given by Far r er e t a l . (1997) are
appr opr i at e for age 60. They will change with age, as higher-risk genot ypes
die mor e quickly. We must est i mat e these if we wish to consi der ent rant s (to
a st udy, or into insurance) at ages over 60. Tabl e 9 shows est i mat es of the
gene frequencies in respect of lives unaffected by AD at ages 65, 70 and 75.
Usi ng the not at i on of equat i on (12), these are given by:
i / ~l I
P o + , = P , % o -6o,,
~-,j=5 _ j j o t l " (14)
Z. ~j =I 1760 r 6 0 , t
These are not the gene frequencies in respect of the whol e popul at i on; lives
alive but who have AD are omi t t ed (as is appr opr i at e for insurance
applications). Nor are they the gene frequencies in respect of the heal t hy
popul at i on; lives with disabilities ot her t han AD are included.
Gene frequencies in the whol e popul at i on at ol der ages can also be
est i mat ed, as:
( i p l l . . l - i D l 2 i p 1 3 "~
P60 ~, - - 60, t " - - 60, t -~- - - 6 0 , t ]
1
13
~"~j=5 _ j ( j p I I _ t . . j p l 2 q - J P 6 0 , t ) ( 5 )
z....~j=l 1' 60 ~ 60,t - - - - 60, t
but these are not so relevant for i nsurance applications.
100 ANGUS MACDONAL D AND DELME PRI TCHARD
G e n o t y p e e4/e4
o . " " ' ..
~ "..
~ "'..,..
o
o
o
6O 6 5 "t0 7 5 8 0 8 5 ~O
.,~g* (ywm=)
G e n o t y p e e 2 / e 2 & e 2 / e 3
o
i ...... . . . . . . . . . . . . . . . .
6O 6O ~l l ~'5 6O 6O ~ 1
Genotype e3/e4 Genotype e3/e3 Genotype e2/e4
o
m
o
o
o
o
6 O 6 5 7 0 7 5 8 . O 8 5 9 O
A g g r e g a t e d G e n o l y p e s
~ , o "" " . . .
6O 6O 7o ~'5 6O 8 5 ~
o j . . . . .
: 1 .
=1 ......
6O 6 5 7 0 7 5 6O 6O gO
(year=)
A g g r e g a t e M o d e l
o
" - - . . . .
~ "......
o
6 O 6 O ~ 6 O
o ] ....
~ t "-..
6 0 6 5 7 0 7 5 6O 8 5 9 0
A g e ( y e e ~ )
. . S t a t e I - H e n l ~ , y
- - - Sta~e 2 - C , n = ~ ~ A D
- - - S t a t e 3 - Inz~'a f r o m ~ D
- - S Z a ~ 4 - Oe.ed
FIGURE 10: Occupancy probabilities for females, healthy at age 60, high relative risks (m = I).
o
o
" , .
o ......,
. . . . .
o
d
6O 6 5 7 0 7 5 8O 8 5 9O
A g e I w ~ u ~ )
G e n o t y p e e 2 / e 2 & e 2 / e 3
o ,
. - . . . . . .
" , .
d "'..
g
o
o
6 0 6O ~1 7 5 6O 6O 6O
" % . .
6O 6.5 7O 7 5 8O 8 5 9O
A g e l y g m )
A g g r e g a t e d G e n o t y p e s
-=' I ' . . . . .
~1 "..
"2
6 0 6 5 7 0 7 5 8O 8 5 9 0
Genotype e3/e3 Genotype e2/e4
........,........
d "..
d "..
6O 6 5 7 0 7 5 8O 8 5 9 0
A g e ( ~ , z r = )
A g g r e g a t e Mo d e l
o
_ . . . .
o " " " '
, ~ " '
I i L
~ 0 6 5 7 0 7 5 8 0 8 5 9 0
A g e ( ~ m )
o
o =
==
o o
" . . . .
6O 0 5 7 0 7 5 8 0 8 5 DO
A g e (ye~r~)
. S t a t e 1 - H e a t h y
- - - SI.=W 2 - O n z e t ( ~ A D
- - - S t a t e 3 - Ir~.t'a f ~ m A D
- - S t a l e 4 - D e a a
FIGURE I1: Occupancy probabi l i t i es for females, healthy at age 60, low relative risks (m = 0.25).
A M A T H E M A T I C A L M O D E L O F A L Z H E I M E R ' S D I S E A S E A N D T H E A P O E G E N E ] 0 ]
G e n 0 t y p e e4/e4
, . . . . , . . " . . . " ,
6O 6 5 7O 7 5 I ~ 8 5 ~
G e n 0 t y p e e?./e2 & e2/ e3
o
. : . . . . . . . . . . . . . . . , , , . . . .
6

o
o
o
6O e5 70 75 80 85 go
rage ( W=' S)
o . . . . . . . . . . . . . .
~ ,...
o
~ "',...,
o
" . . . .
=
o ",,
o /
o
8 0 ~ ; ~ 7 5 ~ 8 5 ~
Gen0type e3/e3
o. . . . . . . . . . . . . . . . . . " " . .
o
'. . . .
5
I
60 6S 70 75 80 85 SO
Age (y,~,'=)
A g g r e g a t e M o d e l
O [ . . . . . . . . . . . . . " . .
o~l
6 0 6 5 7 0 7 5 8 0 8 5 9 0
Ag e (y,Mr=)
o . . . . " . . .
o .,
o "...
~ E~5 ~ 7 5 0 0 g 5 ~
S=a~o I - H~ee/ ~
- - - S~a~ 2 . C, nset o~ AD
- - - S l a ~ 3 - I n~l f n~n AO
- - St at e 2 & St ef = 3
FIGURE 12: Pr eval ence r at e o f Al z he i me r ' s di sease f or f emal es h e a h h y at age 60, hi gh r el at i ve r i sks ( m = I).
Genotype e4/e4
i l . . . . . . . . . . . . . . . . . . . . . .
G e n 0 t y p e e 2 / e 2 & e 2 . / e 3
, , . , .
o " . ,
=
o
o
o
o
i '
/
o
I . . . . . . . . . . . . . . . . . . . . . . " ' . .
Genotype e3/ e3
.=-I ................
~ { ...........
~1 ""
Aggregat e Model
~1 ......
~1 ................... .....
~,1 ".
'
o
o
o
d
o
o
Gen0type e2/e4
"....
60 (L5 70 75 B0 &5 g0
Ago (yoa,'=)
...... Scare I - Hoel e~
- - - St at e 2 . Co, si n ~ AD
- - - St a ~ 3 I n ' ~d I r cr n A.D
- - Sta~a 2 & Sta~o 3
FIGURE 13: Pr eval ence r a t e o f Al z he i me r ' s di sease f or f emal es he a l t hy at age 60, l ow r el at i ve r i sks (m = 0. 25) .
TABLE 9
F R E QUE NC I E S OF A P o E GENOTYP ES AMONG LIVES FREE OF AL Z HE I ME R ' S DI SEASE AT AGES 65, 70 AND 75,
ESTI MATED BY SOLVI NG T HE K O L MO G O R O V EQUATI ONS F ORWARD FROM AGE 60
P r o p o r t i o n GeneFrequenc~sinAD-free population
Gender ofrelative Age e2/e2 &
e4/e4 e3/e4 e3/e3 e2/E4
r~k, m e2/e3
M & F 1.00 65 0.0168 0.2103 0.6114 0.0258 0.1357
70 0.0151 0.2055 0.6168 0.0251 0.1374
75 0.0133 0.1978 0.6246 0.0241 0.1403
0.50 65 0.0174 0.2115 0.6100 0.0259 0.1353
70 0.0164 0.2090 0.6128 0.0255 0.1362
75 0.0154 0.2050 0.6170 0.0250 0.1377
0.25 65 0.0177 0.2121 0.6092 0.0259 0. 1351
70 0.0172 0.2109 0.6106 0.0258 0.1355
75 0.0166 0.2088 0.6128 0.0255 0.1363
1.00 65 0. 0171 0.2097 0.6116 0.0257 0.1358
70 0.0159 0. 2041 0.6176 0.0247 0.1377
75 0.0145 0. 1951 0.6263 0.0232 0.1409
0.50 65 0.0175 0.2112 0. 6101 0.0258 0.1354
70 0.0168 0. 2081 0.6133 0.0253 0.1364
75 0.0160 0.2033 0. 6181 0.0245 0. 1381
0.25 65 0.0177 0.2119 0.6093 0.0259 0.1351
70 0.0174 0.2104 0.6110 0.0256 0.1357
75 0.0169 0.2078 0.6135 0.0252 0.1365
1.00 65 0.0169 0.2120 0.6094 0.0260 0.1357
70 0.0153 0.2110 0.6105 0.0258 0.1373
75 0.0138 0.2097 0.6105 0.0257 0.1403
0.50 65 0.0175 0.2125 0.6088 0.0260 0.1352
70 0.0168 0.2120 0.6094 0.0259 0.1359
75 0.0160 0.2115 0.6094 0.0258 0.1372
0.25 65 0.0177 0.2126 0.6086 0.0260 0.1350
70 0.0174 0.2124 0.6089 0.0259 0.1354
75 0.0170 0.2122 0.6089 0.0259 0.1360
M
7. CONCLUSIONS
7. 1. The Model
We have specified and cal i brat ed a simple cont i nuous- t i me Mar kov model of
AD allowing for the variability of the ApoE gene, suitable for use in
insurance and ot her appl i cat i ons, which will be the subject of furt her studies
(for example, Macdonal d & Pri t chard (1999)). Much uncert ai nt y remains:
(a) No single st udy yet exists t hat woul d allow all the intensities in the model
to be est i mat ed si mul t aneousl y. The est i mat i on is based on a number of
different studies, some quite small, of different popul at i ons, with
different research pr ot ocol s and met hods of analysis, and very likely
different definitions of ' onset of AD' and ' i nst i t i t ut i onal i sat i on' .
(b) The relative risks of the ApoE genot ypes are based on case-based
studies, not prospect i ve popul at i on studies, and the risks associ at ed with
the e4 allele are al most certain to be lower than t hose est i mat ed to date.
We have been unabl e to do mor e t han to show what effect this
might have.
Nevertheless, certain features of our model ought to be robust . What ever
reduct i on in relative risks we have used, we have adj ust ed the genot ype-
specific incidence rates of AD so t hat the aggregat ed (popul at i on) incidence
rates are close to t hose actually observed. The latter is one of the few
reasonabl y reliable benchmar ks available. Furt her, our model pr oduces
prevalence rates of AD t hat fall within the range of t hose observed in many
studies.
As well as the intensities, we have est i mat ed the ApoE gene frequencies at
ages up to 75, in respect of lives unaffected by AD at these ages. These are
needed in (for example) insurance appl i cat i ons.
7. 2. Di scussi on
(a) The model specification is di ct at ed entirely by the event s st udi ed in the
medical and epidemiological literature, and not by the events t hat might
be of interest in any part i cul ar application. If it is the case t hat act uari al
model s might, in future, need to i ncor por at e more medical detail, it
woul d be very useful to try to col l abor at e with medical and ot her
researchers.
(b) The publ i shed concl usi ons of medical papers are usually in the form of
summar y statistics (means, medi ans, odds rat i os and confidence
intervals) and if age-related out comes are shown they are usually in
the form of graphs. These are not ideal for actuarial use. AD is a maj or
condi t i on, much studied, but we have had to make some crude
assumpt i ons in or der to cal i brat e the model from publ i shed dat a only.
There must be many medical dat a sets that coul d furnish age-related
104 ANGUS MAC DONAL D AND DELME P R I T C HAR D
est i mat es of incidence rates, if onl y they coul d be re-analysed. Again,
closer col l abor at i on bet ween act uari es and ot her researchers woul d be
valuable.
(c) Anot her common t ype of medical statistic is prevalence rates. The
difference bet ween prevalence rates and incidence rates is exactly the
difference bet ween the Manchest er Uni t y appr oach to modelling
Per manent Heal t h Insurance, and the multiple-state model approach.
Prevalence rates are oft en easier to estimate, as they can be based on
census-t ype surveys, but it woul d be helpful if the great er versatility of
incidence rates (transition intensities) was mor e widely appreci at ed.
(d) As a consequence of fitting the intensities using publ i shed summar y
statistics, it is impossible to est i mat e even crude confidence intervals for
them. In any appl i cat i on, therefore, sensitivity analysis is especially
i mport ant .
(e) Several epi demi ol ogi cal aut hor s have suggested the use of individual
pat i ent dat a rat her than summar y or publ i shed dat a, part l y to avoi d
publ i cat i on bias in met a-anal yses. Useful references are Pi ant adosi
(1997), Green, Benedetti & Cr owl ey (1997) and Fri edman, Fur ber g &
DeMet s (1998).
(f) It is now about six years since the r61e of the ApoE gene in AD was
confirmed. Since then, the gene has been intensively studied, to the poi nt
t hat met a-anal yses including t housands of lives have been published.
Even so, little is known about popul at i on risk, and dat a are very scarce
in places, so that:
(1) we have had to reduce relative risks rat her arbi t rari l y to allow for the
selectiveness of case-based studies; and
(2) the relative risks for males are suspect, with the e2 allele conferri ng
such st rong prot ect i on t hat carriers of the e4 allele are not necessarily
at higher risk overall.
If this is typical, it seems likely t hat the time bet ween the identification of
a gene di sorder, and the assessment of its i mpact on insurance, will be of
the or der of ten years.
(g) Despi t e the fact that AD is a much-st udi ed condi t i on, many studies
reach conflicting concl usi ons. When setting up an actuarial model, it is
necessary to consi der the body of medical research in its entirety (hence
the large number of references). Inevitably, some studies must be chosen
as the basis of the model, but to confine one' s at t ent i on onl y to these
risks over l ooki ng i mpor t ant features or sources of variation, and coul d
i mpai r the credibility of the results in the eyes of medical experts.
(h) ApoE is a relatively simple gene to consider, since it has onl y three
relevant alleles, hence six genot ypes. Ot her genes are more complex; for
example, the BRCAI gene (predi sposi ng to breast and ovari an cancer)
has several hundred known mut at i ons, some of which have onl y been
observed in a single family.
(i) We cannot stress t oo st rongl y the speed at which human genetics is
developing. This wor k started in late 1997, and since then the vol ume of
A MATHEMATICAL MODEL OF ALZHEIMER'S DISEASE AND THE APOE GENE 1 05
papers on AD and the Apo E gene has increased greatly, as the references
show. It is very likely that assessments o f the i mpact o f specific genet i c
tests on i nsurance will have to be revisited qui te frequentl y, i f they are to
remain credible.
Poi nt s (a) to (c) above suggest ways in whi ch medi cal data mi ght be made
more useful f or actuarial model s, but there is no a pr i or i reason why medi cal
studies shoul d be pl anned wi th that in mind. However, actuarial model s
derived from insurance practice, capabl e o f deal i ng wi th fairly general
payment s whi l e in di fferent states or on transi ti on bet ween di fferent states,
coul d make a useful cont ri but i on to health economi cs , and it woul d be
hel pful to pursue col l aborat i ons from that poi nt o f view.
ACKNOWLEDGEMENTS
Thi s work was supported by a research grant from the Facul t y and
Insti tute o f Actuari es, and one aut hor ( DJP) is in receipt o f a
Ph. D. studentshi p sponsored by Standard Li fe Assurance Company. We
woul d like to thank Ross Ai nsl i e, Dr Carol Brayne, Dr Peter Brett,
Robert Pl umb, Prof essor Andrew Read, Dr Virginia Warren, Prof essor
Howar d Waters and Dr Robert Wat s on for hel pful di scussi ons.
REFERENCES
BARCLAY, L.L., ZEMCOV, A., BLASS, J.P. and McDOWELL F.H. (1985a) Fa c t o r s a s s o c i a t e d wi t h
dur a t i o n o f survival in Al z he i me r ' s di s e as e . Biological Psychiatry 20, 86-93.
BARCLAY, L. L. , ZEMCOV, A., BLASS, J.P. and SANSONE, J. (1985b) Survival in AIzheimer's
di s e a s e and vas c ul ar de me nt i as . Neurology 35, 834-840.
BASUN, H., GRUT, M., WINBLAD, B. and LANNFELDT, L. (1995) Apolipoprotein e4 allele a nd
di s e a s e pr o g r e s s i o n in pat i e nt s wi t h l at e o ns e t Al z he i me r ' s di s e as e . Neuroscience Letters 183,
32-34.
BEARD, C.M., KOKMEN, E., O'BR[EN, P.C. and KORLAND, L.T. (1994) Ar e pat i e nt s wi t h
Al z he i me r ' s di s e a s e s ur vi vi ng l onge r in recent years? Neurology 44, 1869-1871.
BERG, L., MILLER, J.P., STORANOT, M., DUCHEK, J., MORRIS, J.C., RUBIN, E.H., BURKE, W.J.
a nd COBEN, L.A. (1988) Mild senile de me nt i a o f t he Al z h e i me r t ype: 2. Lo ng i t udi na l
as s e s s me nt . Annals of Neurology 23, 477-484.
BICKEBOLI,ER, H., CAMPION, D., BRICE, A., AMOUYEL, P., HANNEQUIN, D., DIDIERJEAN, O.,
PENEr, C., MARTIN, C., PEREZ-TUR, J., MICHON, A., DUBOIS, B., LrzDOZE, F., THOMAS-
ANTERION, C., PASQUIER, F. , PUEL, M. , DEMONET, J-F., MOREAUD, O., BABRON, M-C.,
MEULIEN, D., GuEz, D., CHARTIER-HARLIN, M-C., FREBOURG, T., AGIO, Y., MARTINEZ,
M. and CLERGET-DARPOUX~ F. (1997) Apolipoprotein E and AIzheimer's di s eas e: g e n o t y p e -
s peci f i c ri sks by age a nd sex. American Journal of Human Genetics 60, 439-446.
BONMUTO, S., MELE, M., GALLUZO, L. and GIANNANDREA, E. 0995) Survival and de me nt i a:
A 7-year f o l l o w up o f an I t al i an po pul a t i o n. Archives of Gerontology and Geriatrics 20,
105-113.
BRACCO, L., GALLATO, R., GRIGOLE'rTO, F., LlPPI, A., LEPORE, V., B[NO, G., LAZZARO, M.P.,
CARELLA, F., PICCOLO, T., POZZlLU, C., GIOMETTO, B. and AMADUCCI, L. (1994) Fa c t o r s
affecting c o ur s e and s ur vi val in Alzheimer's disease. Archives of Neurology 5 I, 1213-1219.
BRETEI, ER, M.B.B., CLAUS, J.J., VAN DUUN, C.M., LAUNER, L.J. and HOVMAN, A. (1992)
Epidemiology o f Al z he i me r ' s di s e as e . Epidemiologic Reviews 14, 59-82.
BRINDLE, N., SONG, Y., ROGAEVA, E., PREMKUMAR, S., LEVESQUE, G., YU, G., IKEDA, M.,
NISHIMURA, M., PATERSON, A., SORBI, S., DUARA, R., FARRER, L. and ST GEORGE-
HYSLOP, P. (1998) Analysis of the butyrycholinesterase gene and nearby chromosome 3
markers in Alzheimer's disease. Human Molecular Genetics 7, 933-935.
BURNS, A., LEWIS, G., JACOBY, R. and LEVY, R. (1991) Factors affecting survival in Alzheimer's
disease. Psychological Medicine 21,363-370.
CLAYTON, D. and H ILLS, M. (1993) Statistical methods in epidemiology. Oxford University Press.
CONTE, S.D. and DE BOOR, C. (1972) Elementary numerical analysis. McGraw-Hill,
New York.
COPELAND, J.R.M., DAVIDSON, I.A., DEWEY, M.E., GILMORE, C., LARKIN, B.A.,
McWILLIAM, C., SAUNDERS, P.A., SCOTT, A., SHARMA, V. and SULLIVAN, C. (1992)
Alzheimer's disease, other dementias, depression and pseudo-dementia: prevalence,
incidence and three-year outcome in Liverpool. British Journal of Ps3,chiatry 161,230-239.
CORDER, E.H., SAUNDERS, A.M., STRITTMATTER, W.J., SCHMECHEL, D.E., GASKELL, P.C.,
RIMMLER, J.B., LOCKE, P.A., CONNEALLY, P.M., SCHMADER, K.E.. SMALL, G.W.,
ROSES, A.D., HAINES, J.L. and PERICAK-VANCE, M.A. (1994) Protective effect of
apolipoprotein E type 2 allele for late onset Alzheimer's disease. Nature Genetics 7, 180-184.
CORDER, E.H., SAUNDERS, A.M., STRITTMATI'ER, W.J., SCHMECHEL, D.E., GASKELL, P.C.,
RIMMLER, J.B., LOCKE, P.A., CONNEALLY, P.M., SCHMADER, K.E., TANZI, R.E.,
GUSELLA, J.F., SMALL, G.W., ROSES, A.D., PERICAK-VANCE, M.A. and HAINES, J.L.
(1995) Apolipoprotein E, survival in Alzheimer's disease and the competing risks of death
and Alzheimer's disease. Neurology 45, 1323-1328.
DIESFELDT, H.F.A, VAN HOUTE, L.R. and MOERKENS, R.M. (1986) Duration of survival in
senile dementia. Acta Psychiatr Scandinavica 73, 366-371.
VAN DIJK, P.T.M., DIPPEL, D.W.J. and HABBEMA, J. DIK F. (1991) Survival of patients with
dementia. Journal of the American Geriatric Society 39, 603-610.
VAN DUIJN, C.M., DE KNIFF, P., WEHNERT, A., DE VOECI.IT, J., BRONZOVA, J.B.,
HAVEKES, L.M., HOFMAN, A. and VAN BROECKHOVEN, C. (1995) The apolipoprotein E
e4 allele is associated with an increased risk of early-onset Alzheimer's disease and a
reduced survival. Annab of Neurology 37, 605-610.
EVANS, D.A., FUNKENSTEIN, H., ALBERT, M.S., SCHERR, P.A., COOK, N.R., CHOWN, M.J.,
HEBERT, L.E., HENNEKENS, C.H. and TAYLOR, J.O. (1989) Prevalence of Alzheimer's
disease in a community population of older persons: Higher than previously reported.
Journal of the American Medical Association 262, 2551-2556.
EVANS, D.A., SMITH, L.A., SCHERR, P.A., ALBERT, M.S., FUNKENSTEIN, H.H. and
HEBERT, L.E. (1991) Risk of death from Alzheimer's disease in a community population
of older persons. American Journal of Epidemiology 134, 403-412.
EVANS, D.A., BECKE'rT, L.A., FIELD. T.S., FENG, L., ALBERT, M.S., BENNE'I'r, D.A.. TYCKO, B.
and MAYEUX R. (1997) Apolipoprotein E e4 and incidence of Alzheimer's disease in a
community population of elder persons. Journal of the American Medical Association 277,
822-824.
FARRER, L.A., CUPPLES, L.A., HAINES, J.L., HYMAN, B., KUKULL, W.A., MAYEUX, R.,
MYERS, R.H., PERICAK-VANCE, M.A., RISCH, N., VAN DUIJN, C.M. and APOE and
ALZHEIMER DISEASE META ANAI,YSIS CONSORTIUM (1997) Effects of age, gender and
ethnicity on the association between apolipoprotein E genotype and Alzheimer's disease.
Journal of the American Medical Association 278, 1349-1356.
FORFAR, D.O., McCUTCHEON, J.J. and WILKIE, A.D. (1988) On graduation by mathematical
formula. J.I.A. 115, 1-149 and (with discussion) T.F.A. 41, 97-269.
FRIEDMAN, L.M., FURBERG, C.D. and DEMETS, D.L. (1998) Fundamentals of clhlical trials.
Springer-Verlag, New York.
FRISONI, G.B., GOVONI, S., GEROLDI, C., BIANCHE'I'I-I, A., CALABRES, L., FRANCESCHINI, G.
and TRABUCCHI, M. (1995) Gene dose of the E4 allele of apolipoprotein E and disease
progression in sporodic late onset Alzheimer's disease. Annals of Neuralogy 37, 596-604.
GAO, S., HENDRIE, H.C., HALL, K.S. and HUl, S. (1998) The relationship between age, sex
and the incidence of dementia and Alzheimer's disease. Archives of General Psychiatry 55,
809-815.
GOMEZ-ISLA, T., WEST, H.L., REBECK, G.W., HARR, S.D., GOWDON, J.H., LOCASCIO, J.J.,
PERLS, T.T., LEPSITZ, L.A. and HYMAN, B.T. (1996) Clinical and pathological correlates of
apolipoprotein E E4 in AIzheimer's disease. Annals of Neurology 39, 62-?0.
GREEN, S., BENEDETTI, J. and CROWLEY, J. (1997) Clinical trials in oncology. Chapman and
Hall, London.
HAGNELL, O., OJESJO, L. and RORSMAN, B. (1992) Incidence of dernentia ill the Lundby study.
Neuroepidemiology I I, 61-66.
HEBERT, L.E., SCHERR, P.A., BECKE'gr, L.A., ALBERT, M.S., PILGmM, D.M., CHOWN, M.J.,
FUNKENSTEIN, H.H. and EVANS, D.A. (1995) Age specific incidence of Alzheimer's
disease in a community population. Journal of the American Medical Association 273,
1354-1359.
HEYMAN, A., PETERSON, B., FILLENBAUM, G. and PIEPER, C. (1996) The consortium to
establish a registry for Alzheimer's disease (CERAD) Part XIV: Demographic and clinical
predictors of survival in patients with Alzhcimer's disease. Neurology 46, 656-660.
HEYMAN, A., PETERSON, B., FiLLENRAUM, G. and P]EPER, C. (1997) Predictors of time to
institutionalization of patients with Alzheimer's discase: the CERAD experience part XVII.
Neurology 48, 1304-1309.
JARVtK, G.P., LARSON, E.B., GODDARD, K., KUKUt.L, W.A., SCHELLENBERG, G.D. and
WUSMAN, E.M. (1996) Influence of apolipoprotein E genotype on the transmission of
Alzheimer's disease in a community-based sample. American Journal of Human Genetics 58,
191-200.
JORM, A.F. and JOLLEY, D. (1998) The incidence of dementia - a meta analysis. Neurology 51,
728-733.
JOST, B.C. and GROSSBERG, G.T. (1995) The natural history of Alzheimer's disease: a brain
bank study. Journal of the American Geriatrics Society 43, 1248-1255.
KAHN, H. and SEMPOS, C.T. (1989) Statistical methods hi epidemiology. Oxford University
Press.
KOKMEN, E., CHANDRA, V. and SCHOENBERG, B. (1988) Trends in incidence of dementing
illness in Rochester, Minnesota, in three quinquennial periods, 1960-1974. Neurology 38,
975-980.
KOKMEN, E., BEARD, R.N., O'BRmN, P.C., OFFORD, M.S. and KURLAND, L.T. (1993) Is the
incidence of dementing illness changing'? Neurology 43, 1887-1892.
KUUSJSTO, J., KOIVISTO, K., MYKK,~NEN, L., HELKALA, E-L., VANHANEN, M., H,~NNINEN, T.,
PYOR,~LA, K., KESA, NIEMI, Y.A., RIEKKINON, P. and LAAKSO, K. (1994) Association of
apolipoprotein E phenotypes with late onset Alzheimer's disease: population based study.
British Medical Journal 309, 636-638.
LAMBERT, J-C., PASQUIER, F., COTI'EL, D., FRIGARD, B., AMOUYEL, P. and CHARTIER-
HARUN, M-C. (1998) A new polymorphism in the APOE promoter associated with risk
of developing Alzheimer's disease, Human Molecular Genetics 7, 533-540.
LEHMANN, D.J., JOHNSTON, C. and SMrrH, A.D. (1997) Synergy between the genes for
butyrylcholinesterase K variam and apolipoprotein E4 in late-onset confirmed Alzheimcr's
disease. Human Molecular Genetics 6, 1933-1936.
LEHTOVIRTA, M., HELISALMI, S., MANNERMAA, A., SOININEN, H., KOIVISTO, K., RYYN,~NEN,
M. and RIEKKiNEN SR. P. (1995) Apolipoprotein E polymorphism and Alzheimer's disease
in Eastern Finland. Neuroseience Letters 185, 13-15.
LEMAIRE, J., SUBRAMANIAN, K., ARMSTRONG, K. and ASCH, D.A. (2000) Pricing term
insurance in the presence of a family history of breast or ovarian cancer. To appear in North
American Actuarial Journal.
LETENNEUR, L., COMMENGES, D., DARTIGUES, J.F. and BARBERGERGATEAU, P. (1994)
Incidence of dementia and Alzheimer's disease in elderly community residents of south-
western France. International Journal of Epidemiology 23, 1256-1261.
LEVY-LAHAD, E. and BIRD, T.D. (1996) Genetic factors in Alzheimer's disease: a review of
recent advances. Annals of Neurology 40, 829-840.
LIDDELL, M., WILLIAMS, J., BAYER, A., KAISER, F. and OWEN, M. (1994) Confirmation of
association between the E4 allele of apolipoprotein E and Alzheimer's disease. Journal of
Medical Genetics 31, 197-200.
LILIENFELD, D.E. and STOLLEY, P.D. (1994) Foundations ofepidemiology. Oxford University
Press.
LOPEZ, O.L., LOPEZ.POUSA, S., KAMBOH, M.I', ADROER, R., OLIVA, R., LOZANO-GALLEGO, M.,
BECKER, T.B. and DEKOSKY, S.T. (1998) Apolipoprotein E polymorphism in Alzheimer's
disease: a comparative study of two research populations from Spain and the United States.
European Neurology 39, 229-233.
LUCOTTE, G., DAVID, F., LEORAND, C., TURPIN, J-C. and CLAVEL, C. (1997) Predictive value of
apolipoprotein E e4 allele genotyping in Alzheimer's disease. AIzheimer's Research 3, 7-9.
MACDONALD, A.S. (1996a) An actuarial survey of statistical models for decrement and
transition data i: Multiple state, Poisson and binomial models. British Actuarial Journal 2,
129-155.
MACDONALD, A.S. (1996b) An actuarial survey of statistical models for decrement and
transition data Ill: Counting process models. British Actuarial Journal 2, 703-726.
MACDONALD, A.S. (1997) How will improved forecasts of individual lifetimes affect
underwriting?. Philosophical Transactions of the Royal Society B, 352, 1067-1075, and
(with discussion) British Actuarial Journal 3, 1009-1025 and 1044-1058.
MACDONALD, A.S. (1999) Modeling the impact of genetics on insurance. North American
Actuarial Journal 3:1, 83-101.
MACDONALD, A.S. and PRtTCHARD, D.J. (1999) Genetics, Alzheimer's disease and long-term
care insurance. Genetics and Insurance Research Report No. 99/2, Department of Actuarial
Mathematics attd Statistics, Heriot- Watt University, Ediaburgh.
MASSULLO, C., DANIELE. A., SERIPA, D., FILIPPINI, V., GRAVINA, C., CARBONE, G.,
GAINOTTI, G. and FAZ[O, V.M. (1998) Apolipoprotein E genotype in sporadic early- and
late-onset Alzheimer's disease. Dementia and Geriatric Cognitive Disorders 9, 121-125.
MAYEUX, R., STERN, Y., OTTMAN, R., TATEMICHI, T.K., TANG, M-X., MAESTRE, G., NGAI, C.,
TYCKO, B, and GtNSBERG, H. (1993) The apolipoprotein E4 allele in patients with
Alzheimer's disease. Annals of Neurology 34, 752-754.
MOLSfi, P.K., MAR'r'rtLA, R.J. and RINNES, U.K. (1982) Epidemiology of dementia in a Finnish
population. Acta Neurologica Scandinavica 65, 541-552.
MOLS~,, P.K., MAR3"rILA, R.J. and RtNNES, U.K. (1986) Survival and cause of death
in Alzheimer's disease and multi-infarct dementia. Acta Neurologica Scandinavica 74,
103-107.
MYERS, R.H., SCHAEFFER, E.J., WILSON, P.W.F., D'AGOSTINO, R., ORDOVAS, J.M., ESPINO, A.,
WHITE, R.F., KNOEFEL, J.E., COBB, J.L., MCNULTY, K.A., BEISER, A. and WOLF, P.A.
(1996) Apolipoprotein E e4 association with dementia in a population based study: the
Framingham study. Neurology 46, 673-677.
NALBANTOGLU, J., GILFIX, B.M., BERTRAND, P., ROBITAILLE, Y., GAUTHIER, S.,
ROSENBLA'I-[', D.S. and POIRIER, J. (1994) Predictive value of apolipoprotein E genotyping
in Alzheimer's disease: results of an autopsy series and an analysis of several combined
studies. Annals of Neurology 36, 889-895.
NILSSON, L.V. (1984) Incidence of severe dementia in an urban sample followed from 70 to
79 years of age. Acta Psvchiatrica Scandinavica 70, 478-486.
NORBERG, R. (1995) Differential equations for moments of present values in life insurance.
Insurance: Mathematics and Economics 17, 171 - 180.
NORRMAN, J., ]~ROOKES, A.J., YATES, C. and CLAIR, D.S. (1995) Apolipoprotein E genotype
and its effect on duration and severity of early and late onset Alzheimer's disease. British
Journal of Psychiatry 167, 533-536.
O'CONNOR, O.W., POLLITT, P.A., HYDE, J.B. et al. (1989) The prevalence of dementia as
measured by the Cambridge Mental Disorders of the Elderly Examination. Acta
Psychiatrica Scandinavica 79, 190-198.
OTT, A., BRETELER, M.M.B., VAN HARSKAMP, F., STIJNEN T. and HOFMAN, A. (1998)
Incidence and risk of dementia. American Journal of Epidemiology 147, 574-580.
PAYAM[, H., SCHELLENBERG, G.D., ZAREPARSt, S., KAVE, J., SEXTON, G.J., HEAD, M.A.,
MATSUYAMA, S.S., JARVIK, L.F., MtLLER, B., MCMANUS, D.Q., BIRD, T.D., KATZMAN, R.,
HESTON, L., NORMAN, D. and SMALL, G.W. (1997) Evidence for association of
H LA-A2 allele with onset age of Alzheimer's disease. Neurology 48, 512-518.
PERICAK-VANCE, M.A., BEBOUT, J.L., GASKELL, P.C., YAMAOKA, L.H., HUNG, W.Y.,
ALBERTS, M.J., WALKER, A.P., BARTLEI"r, R.J., HAYNES, C.A., WELSH, K.A.,
EARL, N.L., HEYMAN, A., CLARK, C.M. and ROSES, A.D. (1991) Linkage studies in
familial Alzheimer's disease: evidence for chromosome 19 linkage. American Journal of
Human Genetics 48, 1034-1050.
PFEFFER, R.[., AFIFI, A.A. and CHANCE J.M. (1987) Prevalence of Alzheimer's disease in a
retirement community. American Journal of Epidemiology 125, 420-436.
PIANTADOSI, S. (1997) Clinical trials." A methodologieperspective. John Wiley and Sons, New York.
PODUSLO, S.E., NEAL, M., HERRING, K. and SIIELLY, J. (1998) The apolipoprotein CI A allele
as a risk factor for Alzheimer's disease. Neurochemical Research 23. 361-367.
PO1RIER, J., DAVIGNON, J., BOUTHILLIER. D., KOGAN, S., BERTRAND, P. and GAUTHIER, S.
(1993) Apolipoprotein E polylnorphism and Alzheimer's disease. Lancet 342, 697-699.
PRITCHARD, D.J. (I 997) Life assurance." financial implications of a change in hssuring behaviour
resulthag f rom #Mividuals' increased knowledge of their genetic predispositions. M.Sc.
dissertation, Heriot-Watt University, Edinburgh.
ROCEA, W.A., CHA, R.H., WARING, S.C. and KOKMEN, E. (1998) Incidence of dementia and
Alzheimer's disease: a reanalysis of data from Rochester, Minnesota, 1975-1984. American
Journal of Epidemiology 148, 51-62.
RORSMAN, B., HAGNELL, O. and LANKE, J. (1986) Prevalence and incidence of senile and multi-
infarct dementia in the Lundby study: a comparison between the time periods 1947-1957
and 1957-1972. Neuropsychobiology 15, 122-129.
ROSES, A.D. (1995) Apolipoprotein E genotyping in the differential diagnosis, not prediction, of
Alzheimer's disease. Annals of Neurology 38, 6-14.
SAYE'rTA, R.B. (1986) Rates of senile dementia- Alzheimer's type in the Baltimore longitudinal
study. Journal of Chronic Disabilities 39, 271-286.
SELVIN, S. (1996) Statistical analysis of epidemiologic data. Oxford University Press.
SEVERSON, M.A., SMrrH, G.E., TANGALOS, E.G., PETERSON, R.C., KOKMEN, E., IVNtK, R.J.,
ATKtNSON, E.J. and KURLAND, L.T. (1994) Patterns and predictors of institutionalization in
community-based demcntia patients. Journal o.[the American Geriatrics Society 42, 181-185.
SINGLETON. m., SMITH, G.. GIBSON, A.M., WOODWARD, R., PERRY, R.H., INCE, P.G.,
EDWARDSON, J.A. and MORRIS, C.M. (1998) No association between the K variant of the
butyrylcholinesterase gene and pathologically confirmed Alzheilner's disease. Haman
Molecular Genetics 7, 937-939.
SLOOTER, A.J.C. and VAN DUIJN, C.M. (1997) Genetic epidemiology of Alzheimer disease.
Epidemiologic Re views 19, 107- I 19.
SLOOTER, A.J.C., CRUTS, M,, KALMIJN, S., HOFMAN, A., BRETELER, M.M.B.,
VAN BROECKHOVEN, C. and VAN DUIJN, C.M. (1998) Risk estimates of dementia by
apolipoprotein E genotype from a population-based incidence study: the Rotterdam study.
Archives of Neurolog.v 55, 964-968.
STERN, G., TANG, M.X., DENARO, J. and MAYEUX, R. (1995) Increased risk of mortality in
Alzheimer's disease patients with more advanced educational and occupational :~ttainment.
Am~als of Neurology 37, 590-595.
STERN, Y., BRANDT, J,, ALBERT, M., JACOBS, D.M., Ltu, X., BELL, K., MARDER, K., SANO, M.,
ALBERT, S . , CASTENADA, C.D-C., BYLSMA, F., TYCKO, B. and MAYEUX, R. (1997) The
absence of an apolipoprotein c4 allele is associated with a more aggressive form of
Alzheimer's disease. Annals of Neurology 41,615-620.
STRACHAN, T. and READ, A.P. (1999) Human molecalar genetics (second edition). BIOS
Scientific Publishers. Oxford.
STRITTMA'r'rER, W.J., SAUNDERS, A.M., SCHMECHEL, D., PERICAK-VANCE, M., ENGHILD, J.,
SALVESEN, G.S. and ROSES, A.D. (1993) Apolipoprotein E: High-avidity binding to
f3-amyloid and increased frequency of type-4 allele in late-onset Alzheimer's
disease. Proceedings of the National Academy of Sciences of the United States of America
9 0 , 1977-1981.
SUBRAMANIAN, K. and LEMAiRE, J. (2000) Estimating adverse selection costs in a market with
genetic testing for breast and ovarian cancer: the case of life insurance. To appear in Journal
of Risk and Insurance.
1 l 0 ANGUS MACDONALD AND DELME PRITCHARD
TAN, K.W. (1997) The financial bnpact of genetic testing on annuities. M.Sc. dissertation,
Heriot-Watt University, Edinburgh.
TIERNEY, M.C., FISHER, R.H., LEWIS, A.J., ZorzrrTo, M.L., SNOW, W.G., REID, D.W. and
NleUWSTRATEN, P. (1988) The NINCDS-ADRDA work group criteria for the clinical
diagnosis of probable Alzheimer's disease: a clinicopathologic study of 57 cases. Neurology
38, 359-364.
TREVES, T., KORCZYN, A.D., ZILBER, N., KAHANA, E., LEIBOWlTZ. Y., ALTER, M. and
SCHOENBErG, B.S. (1986) Presenile dementia in Israel. Archives of Neurology 43, 26-29.
TSAI, M-S., TANGALOS, E.G., PETERSEN, R.C., SMITH, G.E., SCHMD, D.J., KOKMEN, E.,
IVNIK, R.J, and THmODEAU, S.N. (1994) Apolipoprotein E: risk factor for Alzheimer's
disease. American Journal of Human Genetics 54, 643-649.
WaLSH, J.S., WELCH, H.G. and LARSON, E.B. (1990) Survival of outpatients with Alzheimer-
type dementia. Annals of hzternal Medic#le 113, 429-434.
WARREN, V., BRETT, P., MACDONALD, A.S., PLUMB, R.H. and READ, A.P. (1999) Genetic tests
and future need for long-term care in the UK: report of a Work Group of the Conthming Care
Conference Genetic Tests and Long-term Care Study Group. Continuing Care Conference,
London.
WATERS, H.R. (1984) An approach to the study of multiple state models. Journal of the hTstitute
of Actuaries I I 1, 363-374.
WATERS, H.R. and WtLKIE, A.D. (1987) A short note on the construction of life tables and
multiple decrement tables. Journal of the hlstitute of Actuaries 114, 569-580.
WATERS, H.R, (1991) Computational procedures for the model. Continuous Mortality
Investigation Bureau Report 12, 79-96.
ANGUS MACDONALD AND DELME PRITCHARD
Department of Actuarial Mathematics and Statistics
Heriot- Watt University
Edinburgh EHI4 4AS
United Kingdom

Heriot-Watt University, Edinburgh: Angus Macdonald and Delme Pritchard

Uploaded by

Copyright:

Available Formats

You might also like

Heriot-Watt University, Edinburgh: Angus Macdonald and Delme Pritchard

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Heriot-Watt University, Edinburgh: Angus Macdonald and Delme Pritchard

Uploaded by

Copyright:

Available Formats

A MATHEMATICAL MODEL OF ALZHEIMER' S DISEASE AND

THE APOE GENE

You might also like