Dietz - SMMR - 1993

Statistical Methods in Medical Research
http://smm.sagepub.com/
The estimation of the basic reproduction number for infectious diseases

K. Dietz
Stat Methods Med Res 1993 2: 23
DOI: 10.1177/096228029300200103
The online version of this article can be found at:

http://smm.sagepub.com/content/2/1/23
Published by:
http://www.sagepublications.com
Additional services and information for Statistical Methods in Medical Research can be found at:
Email Alerts: http://smm.sagepub.com/cgi/alerts
Subscriptions: http://smm.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://smm.sagepub.com/content/2/1/23.refs.html
>> Version of Record - Mar 1, 1993
What is This?
Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

23
The estimation of the basic reproduction number

for infectious diseases
K Dietz Department of Medical Biometry, University of Tubingen, Germany
The basic reproduction number R o is the number of secondary cases which one case would produce in a
completely susceptible population. It depends on the duration of the infectious period, the probability of
infecting a susceptible individual during one contact, and the number of new susceptible individuals
o may vary considerably for different infectious diseases but also for
contacted per unit of time. Therefore R
the same disease in different populations. The key threshold result of epidemic theory associates the
outbreaks of epidemics and the persistence of endemic levels with basic reproduction numbers greater than
one. Because the magnitude of 0 allows one to determine the amount of effort which is necessary either to
R
o for a given
prevent an epidemic or to eliminate an infection from a population, it is crucial to estimate R
disease in a particular population. The present paper gives a survey about the various estimation methods
available.
1 What is Ro ?
The concept goes back to demography where it is usually called the ’net reproduction
rate’. The head of the Statistical Office of Berlin, Richard B6ckh’ calculated in 1886
what he called the ’total propagation of the population’ (’die totale Fortpflanzung der
Bevolkerung’). Using a life-table for females of the year 1879 he summed the products of
the survival probabilities for all reproductive years between 14 and 53 and the rates of
giving birth to a girl. He concluded that 2.172 female babies would be born to a
representative women who throughout her life would be subject to current age-specific
mortality and fertility rates. The formula for this quantity was given by Sharpe and
Lotka2 in 1911, but without symbol and name. At the time of writing this survey it is not
clear to the author where exactly Lotka introduced the notation Ro and the name ’net
reproduction rate’ for the first time. In 1939 he published a summary of his contribu-
tions to demography in French. In a comparison of the advantages and disadvantages of
various indices for natural increase he writes: ’La reproductivit6 nette, Ro, introduite
par Boeckh, a plus de m6rite, car elle donne une mesure essentiellement ind6pendante
de la r6partition de la population par ages.’
Let p (a) denote the probability of a woman surviving age a and let r3(a) be the rate of
giving birth to a girl for an individual of age a then
(The gross reproduction rate is given by
Address for correspondence: Professor K Dietz, Department of Medical Biometry, University of Tübingen,
Westbahnhofstr. 55, D-7400 Tübingen, Germany.

24
i.e. it is ’the number of female children that a female just born may expect to bear during
her reproductive life, ignoring the possibility of mortality’.4)
The index zero of Ro is explained by the notation’ .
i.e. it is the moment of order zero of the net maternity function <f>(a) p(a) r3(a). =
The ’natural rate of increase’ of the population is usually denoted by r. It satisfies the
transcendental equation
The sign of r equals the sign of R° -1...

The demographic term ’basic reproduction rate’ was introduced to epidemiology in
1952 by Macdonald6 in the context of malaria: ’The number of infections distributed in a
community as the direct result of the presence in it of a single primary non-immune
case.’ Macdonald’s basic reproduction number was applied by Smith’ to the transmis-
sion of arboviruses. He uses it to calculate the minimum proportion to be vaccinated and
proposes to estimate it by the inverse of the proportion of susceptibles after an epidemic.
The concept of the basic reproduction number assumes that one infective case is
introduced into a large susceptible population, such that the initial spread of the
epidemic can be approximated by a branching process where one neglects the decline of
the number of susceptibles in subsequent generations of the epidemic. The first model
using a branching process approximation to epidemics is due to Bharucha-Reid. This
was followed up by Neyman and Scott’ and Bartoszyfiski. ~° Beckerll applied branching
processes in order to evaluate vaccination strategies for the control of epidemics. In this
framework it is clear that the key parameter is the number of secondary cases generated
by the initial case. If this number is =~1, one will get only minor outbreaks with a
probability 1 of extinction. If this number is > 1, then there is a positive chance of a large
outbreak affecting nearly the total population.
The concept of the basic reproduction rate for directly transmitted infections has
independently been introduced by Dietz 12 and Hethcote.l3 (Hethcote used the phrase
’infectious contact number’.) Diekmann et al. 14 provide a mathematically rigorous
framework for the definition of Ro in the context of infectious disease models and
propose to call this quantity either ’basic reproduction number’ or ’basic reproduction
ratio’. The term ’rate’ should be avoided since Ro is dimensionless and in epidemiology
this expression should refer to quantities with dimension ’per unit of time’, like ’hazard
rate’ or ’incidence rate’ .1s
For the evaluation of control programs it is not sufficient to find out that Ro is greater
than one. It is important to know whether Ro is much or little above the threshold value
of 1, because this will determine the necessary amount of control needed to reduce the
corresponding parameters sufficiently to achieve a specified goal. Therefore Nåsell16
calls R° , or rather Ro , the ’eradication effort’.
This brings up the question how Ro should be defined in the case of vector-borne
infections like malaria or yellow fever. Traditionally Ro is defined as the number of

25
human cases which one case could generate during the infectious period via the vector
population. The paper of Diekmann et al.,14 however, defines Ro as the number of
secondary cases per generation, i.e. for host-vector diseases one would take the square
root of the classical basic reproduction number. In terms of the threshold condition
Ro = 1 this makes no difference. If one defines Ro like Macdonald,6 the minimum
proportion to be vaccinated in the human population if one had a vaccine for the
prevention of a vector-borne disease, would be given by the expression’
If however one defines Ro like Diekmann et a1.14 the corresponding minimum

proportion to be vaccinated would be
The same equations which describe the spread of malaria between humans and
mosquitoes can also be used to describe the spread of a sexually transmitted disease
(STD) in a heterosexual population. This analogy has already been noticed by ROSSI7:
’The venereal diseases may be looked upon as metaxenous diseases in which the two
sexes take the part of the two hosts.’ .
Hethcote and Y orke18 use the malaria equations of Ross for the description of the
heterosexual transmission of gonorrhea. They speak of the ’second generation contact
number’ which ’is the average number of women (second generation) adequately
contacted by men (first generation) who were adequately contacted by an average
infectious woman during her infectious period.’ If one retains the notion of Ro as the
average number of secondary cases per generation, then one has to determine the
geometric mean of the number of secondary cases if the opposite sex generated by one
newly infected case. Similarly one would also define the basic reproduction number in
the case of vector-borne diseases as the geometric mean of the reproduction numbers per
generation.
The definition of the basic reproduction number does not depend on the assumptions
about the immunity after the infectious period. This means that Ro is the same for the
so-called SIS and SIR models, i.e. for models where an individual is either susceptible
again after the infection or completely immune (’resistant’).
Usually the transmission dynamics are described by a nonlinear system of differential
equations which is then linearized. Let K denote the number of persons contacted per
unit of time by one infectious individual. The proportion of contacted persons who are
infected given a contact is denoted by h. The rate of tranfer to a non-infectious state
(either susceptible or immune) is denoted by y. Let the initial size of the population be
denoted by No and let the initial number of infectious individuals be equal to one. Then
the following differential equation describes the initial spread of the epidemic, where it
is assumed that No is large such that it is justified to apply differential equations. Let Y
denote the number of infectives and X the number of susceptibles then
Since X equals No - 1 initially, we see immediately that Y will increase if and only if

26
Ro Kh/y is greater than one. The duration of the infectious period with mean D equals
=
1/y and the total number of persons contacted during this period is EcD. The fraction h of
those individuals contacted will be infected. This deterministic model implictly assumes
an exponentially distributed infectious period, and the rate of generating secondary cases
r3 Kh defines a Poisson process as long as the individual is infectious. This means that
=
the number of secondary cases per initial case has a geometric distribution:
with mean value Ro and variance R° ( 1 + R° ).

’
2 Why would one need to know Ro?

.
2.1 Elimination or eradication .. ~ - .
In the epidemiological literature the terms elimination and eradication are often used
synonymously. Both refer to a state of zero prevalence of the infection. But a prevalence
of zero may refer to a stable or an unstable equilibrium. Here stability refers to the result
of an introduction of an infectious case into the population.
If the zero equilibrium is stable then the introduction of one case will at most lead to a
small number of generations of secondary cases, and therefore to a return to the original
prevalence of zero. This situation can either occur naturally, without special interven-
tions because of low local contact rates and infection probabilities per contact, or because
of permanent interventions like vaccination programmes which reduce the proportion
of X/N susceptibles sufficiently such that Ro XIN is less than one. Such a stable
prevalence of zero is called elimination.
If however the zero equilibrium is unstable the introduction of an infectious case will
lead to a major epidemic and potentially to a subsequent endemic state, depending on
the size of the population and the rate of introduction of new susceptibles. A prevalence
of zero would be the result of a time-limited intervention after which Ro is allowed to go
back to its original level. Such a programme is called eradication. A unique successful
example is the global smallpox eradication programme. Presently it is planned to
eradicate polio globally by the year 2000. Measles are targeted for elimination by the year
2000 in the USA and in Europe. Because of the ongoing risk of introduction of cases one
would have to continue vaccination for the maintenance of a stable prevalence of zero
until a global measles eradication programme was successful - which may mean forever.
2.2 Vaccination programmes

If one wishes to use Ro in order to determine the minimum coverage required for
elimination or eradication assuming a non-selective programme, one has to take into
account the effect of a vaccine on an individual. In general the susceptibility of an
individual is reduced by a certain factor. In addition a vaccine could also reduce the
infectious period and the infectivity. In the following we concentrate however on the
reduction in susceptibility. The paper by Svensson19 assumes a general distribution
function for the susceptibility among the vaccinated. Special cases have been discussed
in the literature. For example Greenland and Frerichs2° assume that a fraction r is

27
completely immunized, i.e. the susceptibility is reduced to zero. For the remaining
fraction 1- r, the susceptibility is reduced to the value 1- s. If c denotes the coverage of
the vaccination programme, i.e. the proportion of the population vaccinated, then the
vaccination programme reduces the reproduction number to the value
From this it follows that the minimum coverage for elimination or eradication is given by
For many vaccines it may be appropriate to set s 0, i.e. the vaccine has an all or
=
nothing effect: either the vaccine completely protects the individual (which happens
with probability r) or the individual is completely susceptible (which happens with
probability 1- r). In the terminology of Smith et al.21 this corresponds to a ’Model 2’
vaccine. The other extreme assumes r 0 such that all vaccinated individuals have the
=
same reduced susceptibility 1- s. Such a model may be appropriate for inactivated polio
vaccines. The corresponding assumption is referred to as ‘Model 1’. If r and s are smaller
than 1 then the lower bound for the coverage may be greater than 1, which would mean
that even for 100% coverage eradication or elimination is not possible. From the lower
bound given above this conclusion depends very much on the value of the basic
reproduction number.
2.3 Chemotherapy or reduction of the contact rate
If we consider a control programme which reduces either the duration of the
infectious period by chemotherapy or the contact rate by changes in behaviour then the
reproduction number R * is given by the following formula ,1, .
where r and rD denote the reductions for the corresponding parameters indicated in the
index. For elimination or eradication R* has to be less than one. Therefore the rK rD has
to be greater than Ro . This inequality clearly shows that Ro can be interpreted as the
minimum absolute elimination or eradication effort, if we are dealing with a
homogeneous population, and a control method which effects everybody in a non-
selective way. This means that the minimum proportional reduction of the susceptible
fraction is given by
which is highly non-linear function of Ro . For Ro > 10, control programmes have to be
a
nearly perfect requiring a reduction of the parameters by more than 90%; whereas for
values of Ro between 1 and 2 a reduction of the parameters by less than 50% will already
be successful. This non-linear relationship between the minimum proportional reduc-
tion in the transmission factors K and D, and the basic reproduction number Ro, is the
key to understanding the puzzle that apparently the same control programme may be
successful in one situation and not at all in other situations.

28
For the evaluation of control strategies it is not only important to estimate by how
much Ro exceeds 1 but also to determine how sensitive Ro is with respect to changes of
individual parameters entering into a formula of Ro . In the simple formula given above
this is straightforward. But if individual parameters enter in a nonlinear way, then
important conclusions may be drawn about the relative effect of different control
strategies. We shall illustrate this point in the next section in comparing larvicides versus
adulticides for malaria control, and use of condoms versus reduction of the number of
partners in the prevention of STDs.
3 Estimation of Ro using data about individual parameters .
As we have seen above, Ro depends on three parameters: ’
1 The contact rate

2 The duration of the infectious period and
3 The probability that a contact between an infective and a susceptible individual leads
to an infection.
The last parameter is sometimes broken down into two factors describing infectivity and
a factor describing the susceptibility.22 Obviously if all individuals are alike then one
cannot identify the individual components of this product. If however there is
heterogeneity with respect to infectivity and susceptibility, then the ratios between the
values in the corresponding subgroups may be estimated.
The estimation of Ro on the basis of the individual parameters will be particularly
relevant in populations where the infection is absent and there is a problem of assessing
the risk of an epidemic. It is obvious that an attempt to estimate Ro is only meaningful
for diseases, where contacts are clearly defined such that they can be counted. This
excludes whole classes of infectious diseases like measles and cholera, where the spread
is either airborne or due to contamination of food and/or water. Because of this
limitation, attemps to determine Ro directly from individual parameters have been
restricted: on the one hand to vector-borne infections based on estimates of the number
of human bloodmeals which one vector takes per unit of time and the number of vectors
contacting one human host per unit of time; and on the other side for STDs based on
estimates of the number of new partners per person per unit of time and the number of
contacts per partner. .
3.1 Estimation of Ro for malaria

The formula of Ro for malaria involves the following parameters: let m be the number
of female mosquitoes per human host and K be the number of human bloodmeals which
one female mosquito takes per day. This involves the rate of taking a bloodmeal and the
probability of choosing man as a host. In the mosquito there is a latency period I2, the
so-called extrinsic cycle, which is necessary for the development of sporozoites after the
mosquito was infected by gametocytes. Because of the high death rate ~2 ? only a small
fraction exp ( - ~,212 ) survives this latency period. Let D1 be the duration of the
infectious period in the human host and D2 be the duration of the infectious period in the
vector, and let h, and h2 be the probabilities per bloodmeal of infecting a susceptible

29
human host by an infectious mosquito and vice versa, respectively. Putting all these
parameters together one arrives at the following formula
Macdonald23 provides an example based partly on entomological observations and partly

on assumed values: a mosquito density in relation to man m 10, a daily feeding =
frequency on man K 0.5, daily

=
a survival probability of the mosquito of 0.9, which
corresponds to a life expectancy D2 of 9.5 days, and a survival probability of 0.39 for a 9
day long extrinsic cycle. For the duration D1 of the infectious period in man an interval
of 80 days is assumed. Putting all these parameters together and assuming h, h2 = 1 =
Macdonald comes value for what we would now call Ro of roughly

to a
735 ===10-(0.5)~-0.39-80-9.5 which gives a value of Ro = 27.1. Since it is assumed that
the infection probability per contact in each direction equals one, this means that the
estimate represents an upper bound.
With the exception of the parameter Di all the other parameters are determined by
entomological measurements. Garrett-Jones24 introduced the expression ’vectorial
capacity’ for the product
..:< .
which he interpreted as ’the average number of inoculations with a specified parasite

originating from one case of malaria in unit time that the population would distribute to
men if all the vector females biting the case became infected’. Garrett-Jones clearly
stressed that the vectorial capacity ’is a characteristic of a mosquito population related to
the ecological conditions and may be assessed irrespective of whether any cases of
malaria are present at the given place and time’. A comprehensive survey of methods to
estimate the individual components of the vectorial capacity is given by Molineaux et
al. 21 (See Dye 26,* who raises the question of whether one must measure all the
components.)
Najera 27 compared the methods of estimating Ro on the basis of individual parameters
with other methods based on equilibrium assumptions, and found great discrepancies
which we shall explain later.
There have been some attempts to assess the effects of modifying certain assumptions
on which the formula for Ro given above is based. The present model assumes that there
is an age-independent mortality in the mosquito. Clements and Paterson 2’ however
estimate the effect on the basic reproduction number if one takes into account that the
vector’s death-rate may increase with the vector’s age. They assume an exponential
increase according to a Gompertz function. It is shown that the neglect of increasing
mortality may overestimate Ro considerably.
It is very difficult to collect reliable data for the estimation of the individual
entomological parameters. The product mK for instance, the so-called ’man-biting-rate’
is estimated by exposing human baits during whole nights and catching all mosquitoes
landing on their legs and trying to collect a bloodmeal. The daily survival probability is
estimated from the interval between two ovipositions and the proportion of mosquitoes
having laid eggs at least once, i.e. the proportion of parous mosquitoes. Particularly
difficult is the estimation of the probability of choosing man as a host. One collects
*Dye C. The analysis of parasite transmission by bloodsucking insects. Annu Rev Entomol 1992; 37: 1-19.

30
mosquitoes in their resting-places and determines the proportion of human bloodmeals

which they are digesting. Since one is usually collecting mosquitoes near human
dwellings the mosquito population sampled may give a highly biased estimate for the
total mosquito population which is in contact with man. The greatest uncertainty is
associated with the infection probabilities h, and h2. In an endemic situation one can
compare the so-called ’entomological inoculation rate’ with the ’epidemiological inocula-
tion rate’. The former would be the man-biting-rate by mosquitoes with sporozoites in
their salivary glands. The latter is determined by fitting catalytic models to age-specific
prevalance curves of infants in order to estimate the daily force of infection, i.e. the rate
at which infections are acquired per unit of time (see Section 6.2 below). The ratio of the
epidemiological and the entomological inoculation rate is an estimate of h 1. In certain
places this may be as low as 0.01. But such an estimate may not be relevant for
calculating Ro because it may be affected by immunity in the human population.
In summary it is fair to say that it is impossible to give absolute estimates of Ro based
on individual parameters for malaria. However attempts to estimate the vectorial
capacity, i.e. the contact rates between human hosts via the vector, are still valuable in
order to provide relative estimates which explain differences of incidence and prevalence
in different populations. Expressing Ro in terms of the individual entomological
parameters which may be affected by alternative control strategies helps to assess the
different effect of larvicides and adulticides, i.e. the reduction of the mosquito density
without effect on mosquito survival and the reudction of mosquito survival.
Let V2 denote the number of newly emerging female mosquitoes per human host per
unit of time. Then the mosquito density m can be written as
Larvicides would reduce the emergence rate V2 and alduticides would reduce the life
expectancy D2 of the vector. The corresponding reductions are denoted by r v and rD2.
We can express R* on the one hand as a function of rV2 and alternatively as a function of
rD2 . Because the two expressions have to be equal, one can derive the following formula
which expresses r,, in terms of rDz :
All parameters which are not affected by the two control methods can be cancelled. It is
clear that the situation without control (r&dquo;2 rn = 1) must be a solution of this equation.
=
But we also see that the reduction in mosquito emergence through larvicides r, which
would bring about the same value of R* as a given reduction in mosquito survival
described by rD is much greater. For rD 2 we obtain r, --:= 10 and for rD
=
4 we have =
r, 274. This means that a reduction of the mosquito’s life expectancy

-==
50% or by by
7~% brings about the same value of R* as a reduction of mosquito density by 90% or by
99.6%, respectively. ’The single most important insight into public health from
modelling has probably been in the consequences of different insecticidal methods for
transmission of mosquito-borne disease agents with an extrinsic developmental cycle in
the insect’ (Bradley).29
Molineaux3° points out however two caveats with respect to this approach:
1 ’The identification of the weakest point on which to concentrate control efforts does
notfollow automatically from the sensitivity analysis; it obiously depends also on the

31
cost and effectiveness of the intervention methods by which we can affect the various
factors.’
2 ’Models commonly assume more uniformity than there is in reality, and in terms of
control this usually leads to exaggerated optimism.’
3.2 Estimation of Ro for AIDS

The AIDS epidemic has tremendously stimulated the development of mathematical
models of infectious diseases. Among the numerous problems addressed we restrict
ourselves to the estimation of Ro for the heterosexual population. In Europe the
proportion of AIDS cases which are attributed to heterosexual transmission outside the
group of IV drug users is (still) very low. Some authors3~-33 have raised the question
whether Ro for the heterosexual population is greater or smaller than 1. They use data on
sexual behaviour for the heterosexual population to estimate a mean and variance of the
number of new partners per year. They also use estimates for the average duration of the
infectious period and for the infection probability per partner. On the assumption that
individuals with different rates of acquiring new partners choose their partners
randomly (proportional mixing), Ro is proportional to the ratio of the second moment of
the partner-acquisition rate and the mean.34 Bonneaux and Howeling3 ’ and Blower et
al. 32 obtain a value of Ro which may be slightly below or slightly above the critical value
of one. Stigum et al., 33 using a much more complicated model, come to the conclusion
that ’with present sexual behaviour the indigenous spread of HIV is not likely to sustain
an epidemic in the Norwegian heterosexual population.’ It must however be stressed
that these estimates depend crucially not only on the underlying parameter values,
which are sometimes based on rather small samples, but also on the model assumptions
which determine the formula for Ro which is used. One aspect for example which is
neglected in the present models is the age preference in sexual choice. Knolle 31 shows in
a simplified model how this factor may greatly influence the estimate of Ro .
The model of May and Anderson34 assumes that the infection probability per partner
does not depend on the number of contacts per partner. Later they justify this
assumption by reference to data by Peterman et a1.36 where no relationship could be
found between the risk of HIV transmission and the number of contacts at risk.
Kaplan 31 presents a possible explanation by postulating that individuals with a high rate
of partner change, and therefore a low number of contacts per partner, would tend to
practise riskier sex acts. The model with a constant infection probability per contact is
not consistent with the data of Peterman et al. One has to assume either variability
between the individual couples or within a couple during the course of the infection.
Diekmann et al.38 provide a framework for calculating Ro for AIDS taking into
account simultaneously variable infectivity and pair formation. The number of contacts
per partner is modelled in a dynamic way. In models with pairing, i.e. multiple contacts
with one partner, one has to take into account the time spent by the index case with the
infecting partner. From the model equations it is clear that the relevant parameters
which enter the formulae have to be estimated from data describing the behaviour of
HIV infected individuals. What matters is the number of new partners which a newly
infected individual acquires during the total infectious period, and the total number of
contacts and how they are distributed among the various partners. The model takes into
account that a partnership may have a higher risk of separation due to the disease-
induced death of a partner. This raises the question to what extent data collected in the
general population among susceptibles is representative for HIV infected individuals.

32
Also the number of contacts during the infectious period may undergo a disease-induced
reduction. Therefore the present estimates for Ro in the papers quoted above have to be
considered very tentatively. New models have to be developed which would allow one to
derive estimates of Ro if one had the appropriate data. Also, further data are needed to
improve estimates of the infection probability per contact and the duration of the
infectious period. The models for Ro also need extension in terms of behavioural
variables such that alternative interventions can be compared with respect to their
impact in reducing Ro .
Models which take partnership duration into account produce highly nonlinear
formulae with respect to the parameters describing HIV transmission. For constant
infection probability per contact h one gets the following formula:
where
and
Here N is the number of new partners during the infectious period and hp is the infection
probability per partner. The individual parameters have the following definitions (in the
order of appearance in the formula for N):
p the rate of acquiring a new partner for a single individual; .,
)JLi the death-rate of an infected individual;

cr the separation rate of a partnership;
Ro the death-rate of a susceptible individual;
.
r3 the rate of sexual contact during a partnership; ..:. .~ .

h the infection probability per contact.
The formula for the infection probability per partner can be considerably simplified if
one introduces the probability generating function f (x ) for the number of contacts per
partner:
.
The summation starts with k 1 since by definition of a partner at least

=
one contact
takes place. If the number of contacts during a partnership equals k, then
and in general one gets39

33
In the simple case of partnerships with an exponentially distributed duration and a

Poisson process of contacts with rate r3, one obtains a zero-truncated geometric
distribution:
where c is the average number of contacts per partner. The expected number of contacts
of an infected partner with a susceptible partner is given by
because the average duration of such a partnership is ([Lo + ~,1 + a) - 1. Let C denote the
total number of expected contacts with susceptible partners during the infectious period:
Then the formula for Ro may be simplified as follows:
This simple formula clearly shows how Ro depends jointly on the number of partners
and the number of contacts during the infectious period. If C N, i.e. if only one =
contact takes place per partner, then
For a given number of total contacts C, this formula provides a lower bound for the
number of partners for Ro > 1:
This inequality is only meaningful for
This shows that for a per-contact-infection probability h of 0.2% and C =

1000(100
contacts per year for 10 years) only 2 partner in 10 years are required to bring Ro above
1.
Let rh and rN denote the reductions for the per-contact-infection probability h and the
number of partners N, respectively. It is assumed that the total number of contacts C is
kept fixed. In order to compare the relative efficiency of reducing h and N we can first
express the reduced R* as a function of rn and rN, respectively, and subsequently express
rN as a function of rh :
where

34
Obviously wehave rN 1 for rh

=
1. Whether a given proportional reduction of N is
=
more effective in reducing R* than the same reduction of h depends on q. For q > 1; i.e.
for N> hC, the best advice would be to reduce h, i.e. to use condoms. For q< 1; i.e. for
N < hC, a given proportional reduction of the number of partners N would reduce Ro
more than a similar reduction in h. In applying this conclusion the two caveats of
Molineaux3° given in Section 3.1 should be kept in mind.
4 Estimating Ro from the initial rate of growth of the epidemic

When the AIDS epidemic started to attract the attention of the press, the increase of the
cumulative number of cases seemed to be exponential with doubling times in the order
of less than one year. This caused some authors to derive relationships between the
initial growth and R° . For a very simple model, Jacquez and Simon4° use the following
formula of Anderson et a1.41 which yields a value of Ro of 19.7 for the San Francisco gay
community:
Here D is the duration of the infectious period and td is the initial doubling time.
Cairns42 applies a more detailed model to establish a relationship between the initial rate
of growth and Ro which is now no longer explicitly determined because it has to be
calculated as the eigenvalue of a certain matrix, taking into account the heterogeneity in
the contact rates and variable infectivity.
5 Estimates of Ro based on the final size of an epidemic

Smith’ proposed to estimate Ro by the inverse of the proportion of susceptibles after the
termination of an epidemic. As an example he refers to urban yellow fever epidemics
which stopped when 48-65% of the population had become immune. Accordingly the
estimate of Ro would be between about 2 and 3. According to the theory of the
deterministic general epidemic one gets the following simple formulal2 between the
proportion ux of susceptibles after the termination of an epidemic and Ro : .
This means for the values reported by Smith the estimates of Ro would be between 1.4
and 1.6, with the consequence that immunizing about 38% of the population would have
been sufficient to prevent the epidemic. (If one takes the inverse of the proportion of
susceptibles one would require immunizing about 3 of the population.) It should be
stressed however that the formula given above assumes that the total population was
susceptible at the beginning of the epidemic. In general one has the following formula43:
where uo denotes the initial proportion of susceptibles. If for example due to previous

35
epidemics already 25% of the population would be immune, i.e. uo 0.75, then Ro =
would be 1.9. The differences of the various estimates do not appear to be large, but one
has to keep in mind that small differences of Ro between 1 and 2 imply important
differences of the resulting prevalence.
In 1967 an outbreak of smallpox occurred in a socially isolated community in
Abakaliki in south-eastern Nigeria. It was investigated by Thompson and Foege.44 It
would be interesting to find out more details about this epidemic because starting with
Bailey and Thomas45 many statistical analyses of this data set have been published (see
Becker46 for references). Bailey and Thomas 41 list the number of days between 30 cases.
Part of the population consisted of 120 members of a religious group who in principle
refused to be vaccinated. If one assumes that initially all individuals except the
introductory case were susceptible then the formula given above yields an estimate of
1.15 for Ro. Bailey and ThomaS41 mention however that ’about a quarter of them
appeared from scars in fact to have been vaccinated at some time or other’. This suggests
that the initial proportion of suceptibles was 75%, which would bring the final
proportion of susceptibles to 50% which corresponds to Ro of 1.62. This religious group
lived together with other individuals who did not belong to this group and who were
vaccinated. A re-analysis of this epidemic, based on the original report, is in preparation.
The survey paper by Becker 47 in the present issue contains a description of his
estimation approach for Ro using martingale theory.
6 Estimating Ro using data from equilibrium situations

6.1 Estimation of Ro based on age-independent prevalence data
As indicated above the definition of Ro does not depend on the assumptions about
immunity subsequent to the infectious period. The concept is based on a hypothetical
situation where one infective is introduced into a large susceptible population. In
practice such situations hardly ever occur, such that one has to base estimates of Ro on
data available in endemic situations where the infection is present for a long time.
However, this requires detailed knowledge about the density-dependent regulation of
the transmission process which determines the equilibrium endemic level. It also
assumes knowledge about the heterogeneities in transmission patterns, as it is obvious
that estimates of Ro based on equilibrium data are highly dependent on the availability of
this required information or on the degree to which the assumed mechanisms reflect
reality. Macdonald 41 was the first to derive an estimate of Ro, or rather Rõ as discussed
above, from data on equilibrium prevalence. In Macdonald’s model the only regulating
factor was the prevention of superinfections in the mosquito, i.e. only susceptible
mosquitoes could be infected. He allowed superinfections in man, and the infection risk
of the mosquito was not proportional to the prevalence of human infections but to the
mean number of infections per human host. (It has been pointed out earlier by Fine49
that Macdonald’s mathematical equations did not correspond to his biological assump-
tions.) In the notation of the present paper, Macdonald’s formula for RÕ reads as follows:
where R2 and 12 have the same meaning as above and Y2 is the equilibrium prevalence of
mosquitoes with sporozoites in their salivary glands, i.e. the proportion of infectious

36
mosquitoes. Baseddata from Tanganyika he found a value of R 2 0 of 1.15 which is

on
based probability of surviving the extrinsic cycle of 0.39 and a proportion of
on a
sporozoite positive mosquitoes of 5%. Macdonald 48 attributed the discrepancy of the
present estimate and the previous estimate to the factors hl and h2, i.e. the infection
probabilities. In a further paper Macdonald et al. 5° derived the following formula for Ro
which is based on the force of infection À-1 (see next section) and on the prevalenceyl of
infectious individuals in the human host:
DietzSl,52 provides a system of formulae which relate the equilibrium variables of the
prevalence of infection in the human and the vector host and the force of infection with
the square of the basic reproduction number. The five models differ with respect to their
assumptions about density dependent regulation:
Model I: No superinfections in man and mosquito.
’
Model II: Superinfections in man but infectivity to vector proportional to prevalence

of infection; no superinfection in the vector.
Model III: Superinfection in man; infectivity to vector proportional to number of
infections; no superinfection in vector.
Model IV: No superinfection in man; superinfections in the vector and infectivity to
man proportional to number of infections.
Model V: Superinfections in man; infectivity to vector proportional to prevalence of
.
infection; superinfections in vector and infectivity proportional to number

of infections.
From the formula for Ro based on the force on infection one can see that it corresponds
to Model II in the list given above. Najera27 used these formulae with data collected in a
field project in Nigeria, which was carried out to test whether malaria could be
eradicated in Africa with a combination of spraying DDT and mass drug administration.
For example, in the treated zone for the period mid-July to mid-September 1967 he gets
the following three estimates for Ro :
0 &dquo;
,.....
1.02
(based on the proportion of infectious mosquitoes);
..
(based on the force of infection and the proportion infectious in the human
2.88
population);
135.77 (based on the vectorial capacity). ..
It is
no surprise that the first two estimates differ because they are based on different
assumptions about the density-dependent regulation. The first one is based on Model III
and the second one on Model II.
The observation that estimates of Ro based on prevalences in equilibrium situations in
the human and the vector host tend to be smaller than expected from direct observations
of the transmission intensity has also been discussed by Barbour. 53 He considers a model
for schistosomiasis which mathematically speaking is equivalent to Model III given
above. According to this model the basic reproduction number is only determined by the
proportion of susceptible snails. Since these proportions are usually in the order of 1 to
10% the corresponding values for Ro are only slightly above 1. Because this is considered

37
to unrealistic, Barbour takes into account immunity in the human host and derives
be
equations which would allow one to estimate RÕ by the inverse of the product of the
proportion of susceptible human hosts and susceptible vectors, suitably modified to
adjust for temporary immunity. By using data from the field he obtains an estimate of
R 2= 6.5. Similar attempts to estimate R 2 for other vector transmitted diseases have
been carried out by Roger S51 for the African trypanosomiases and by Hasibeder et al.55
for canine leishmaniasis. The particular emphasis of the second paper is the develop-
ment of methods for assessing the influence of heterogeneous biting rates of sandfiies on
dogs. Formulae are derived which allows one to estimate Ro if the proportion of
susceptibles in the various subgroups and the ratios of the contact rates between the
subgroups are known. Effects of heterogeneity on the relationship between the
equilibrium prevalence and the heterogeneity in contact rates has also been investigated
by Dietz.51 Assuming that there is a gamma distribution of contact rates in the human
population and that vectors choose hosts randomly proportional to their contact rates
(which is called ’proportional mixing’ in the epidemiological literature) then the
prevalence in equilibrium always decreases as the variance increases. If Q2 denotes the
variance of the contact rate it is shown that the slope of the variance with respect to the
square of the basic reproduction number equals ( 1 + 2Q2 )-1, i.e. it decreases for
increasing variance. This means that values for Ro which are based on the assumption of
homogeneous mixing, and which use the equilibrium proportion of susceptibles,
invariably underestimate Ro .
6.2 Estimation of Ro based on age-specific prevalence data
In the epidemiology of infectious diseases the data available are often cross-sectional
surveys of populations with the aim of determining the age specific prevalence of
antibodies with respect to a particular infection indicating past infection. This means
that one has only censored data in the statistical sense, i.e. one only knows that for a
certain age of an individual the age of infection is less than the current age if antibodies
are present, and that the age of infection is greater than the current age if antibodies are
absent. Such age specific prevalence data show a typical increase which can be described
by so-called catalytic models according to Muench. S6 Muench introduced the term ’force
of infection’ for the hazard function of a susceptible to be infected in an endemic
situation. Dietz12 showed that this hazard rate can be related to the basic reproduction
number in the following way:
where Vt denotes the death rate of the population and x is the force of infection. In this
model it is assumed that there is a constant death rate and an age independent force of
infection. If one calculates the average age A at infection one gets the following
expression: . _
-
If one denotes the life expectancy of a human host by eo I-L -1 then the formula for the
=
force of infection yields the following simple expression for Ro :1

38
This means that the basic reproduction number Ro can simply be estimated by the ratio
of the life expectancy and the average age at infection. In the original paper Dietz 12
denoted the average age at infection by X-’ which yields the formula Ro 1 + eo/A. =
Usually jjL is negligible compared to X such that the difference can be neglected.
Anderson and May 57 provide numerous estimates of Ro based on this formula for
common childhood diseases in industrialized and developing countries.
In general, for age-dependent mortality and age-specific force of infection, one gets
the following formula for the average age A at infection:
where
and
If the force of infection is in an equilibrium situation then this reflects

age-dependent
heterogeneous contact rates between different age groups. In principle one cannot
measure the age-specific contact rates directly in order to calculate R° . One would need
to know the elements of the contact matrix if one would discretize the population into a
finite number of age strata. Anderson and May 58 and Schenzle59 study various
assumptions about the structure of the contact matrix. Dietz and Schenzle6° consider the
case of proportional mixing. Under the assumption that the length of the infectious
period can be neglected compared to the life expectancy of the individual they derive the
following formula for Ro based on the age-dependent hazard rate x(a)
This shows that Ro can no longer be simply estimated by the inverse of the proportion of
susceptibles in the population. This formula leads to the statistical problem of estimating
the force of infection on the basis of a data set where all observations are censored.
Grenfell and Anderson 61 use the maximum-likelihood method for some polynomial
function describing the age-specific force of infection.
Recently, Keiding62 presented a nonparametric method to estimate the age-specific

39
prevalence and the corresponding smoothed force of infection. He applies this method to
a data set for hepatitis A in Bulgaria and obtains Ro 3.8. =
7 Concluding remarks
The present review tries to give a summary of attempts to define Ro in a meaningful way,
to derive estimates of Ro based on epidemic and endemic situations and to interpret Ro
for the evaluation of control strategies. During the last twenty years Ro has emerged as a
basic concept in infectious disease epidemiology, but it has also become apparent how
difficult it is to apply in actual field situations. It is hoped that this survey will stimulate
further research in this direction.
Acknowledgements
This work has been supported in part by SIMS (Societal Institute of the Mathematical
Sciences) under support from the US National Institute on Drug Abuse (NIDA grant
DA 04722).
I thank Christopher Dye, Chin Long Chiang, Paul Fine, Helmut Knolle, Bob May
and particularly Herb Hethcote for useful comments on the first draft of this paper, and
Michael Haber for his help in locating the report on the smallpox epidemic in Abakaliki.
The article is based on a paper presented at the ASA meeting in Atlanta, 1991.
References Basis of Medicine. Ann. Rev. 1964; 125-50.

8 Bharucha-Reid AT. On the stochastic theory of
1 Böckh R. Statistisches Jahrbuch der Stadt Berlin, epidemics. In: Neyman J ed. Proceedings of the
Zwölfter Jahrgang. Statistik des Jahres 1884. Third Berkeley Symposium on Mathematical
Berlin: P Stankiewicz, 1886: 30-31. Statistics and Probability. Volume IV: Biology
2 Sharpe FR, Lotka AJ. A problem in age and Problems of Health. Berkeley: University of
distribution. Philosophical Magazine 1911; California Press, 1956: 111-19.
Series 6, 21: 435-8. (Reprinted in: Smith D, 9 Neyman J, Scott EL. A stochastic model of
Keyfitz N. Mathematical demography, selected epidemics. In: Gurland J ed. Stochastic models
papers, Berlin: Springer, 1977: 97-100). in medicine and biology. Madison: The
3 Lotka AJ. Théorie analytique des associations University of Wisconsin Press, 1964: 45-83.
biologiques. Deuxième partie. Analyse 10 Bartoszy&nacute;ski R. Branching processes and models
démographique avec application particuliére à of epidemics. (Dissertationes Mathematicae).
l’espèce humaine. Paris: Hermann, 1939: 102. Warszawa: Pa&nacute;stwowe Wydawnictwo
4 Impagliazzo J. Deterministic aspects of Naukowe, 1969.
mathematical demography. An investigation of the 11 Becker N. The use of mathematical models in
stable theory of population including an analysis of determining vaccination policies. Bulletin of the
the population statistics of Denmark. Berlin: International Statistical Institute 1975; 46: 478-
Springer, 1985: 139-40. 90.
5 Dublin LI, Lotka AJ. On the true rate of 12 Dietz K. Transmission and control of
natural increase as exemplified by the arboviruses. In: Ludwig D, Cooke KL eds.
population of the United States, 1920. Journal Proceedings of the SIMS conference on
of the American Statistical Association 1925; 20: epidemiology. Philadelphia: Society for
305-39. Industrial and Applied Mathematics, 1975:
6 Macdonald G. The analysis of equilibrium in 104-21.
malaria. Tropical Diseases Bulletin 1952; 49: 13 Hethcote HW. Mathematical models for the
813-29. spread of infectious diseases. In: Ludwig D,
7 Smith CEG. Factors in the transmission of Cooke KL eds. Proceedings of the SIMS
virus infections from animals to man. Scientific conference on epidemiology. Philadelphia: Society

40
for Industrial and Applied Mathematics, 1975: 29 Bradley DJ. Epidemiological models - theory
122-31. and reality. In: Anderson RM ed. The
14 Diekmann O, Heesterbeek JAP, Metz JAJ. On population dynamics of infectious diseases: theory
the definition and the computation of the basic and applications. London: Chapman and Hall,
reproduction ratio R o in models for infectious 1982: 324.
diseases in heterogeneous populations. Journal 30 Molineaux L. The pros and cons of modelling
of Mathematical Biology 1990; 28: 365-82. malaria transmission. Transactions of the Royal
15 Elandt-Johnson RC. Definition of rates: some Society of Tropical Medicine and Hygiene 1985;
remarks on their use and misuse. American : 743-47.
79
Journal of Epidemiology 1975; : 267-71.
102 31 Bonneux L, Howeling H. Is een epidemie von
16 Nåsell I. Hybrid models of tropical infections. heteroseksueel overgedragen HIV-infectie
Berlin: Springer, 1985. mogelijk in Europa? Nederlandse Organisatie
17 Ross, R. The prevention of malaria. Second van Tijdschrift-uit Geneeskd 1989; 133: 1922-26.
edition with an addendum on the theory of 32 Blower SM, Anderson RM, Wallace P.
happening. London: John Murray, 1911: 685. Loglinear models, sexual behavior and HIV:
18 Hethcote HW, Yorke JA. Gonorrhea: epidemiological implications of heterosexual
transmission dynamics and control. Berlin: transmission. Journal of AIDS 1990; 3: 763-72.
Springer, 1984: 53. 33 Stigum H, Grønnesby JK, Magnus P, Sundet
19 Svensson Å. Analyzing effects of vaccines. JM, Bakketeig LS. The potential of spread of
Mathematical Biosciences 1991; : 407-12.
107 HIV in the heterosexual population in Norway:
20 Greenland S, Frerichs RR. On measures and a model study. Statistics in Medicine 1991; 10:
models for the effectiveness of vaccines and 1003-23.
vaccination programmes. International Journal 34 Knolle H. Age preference in sexual choice and
of Epidemiology 1988; 17: 456-63. the basic reproduction number of HIV/AIDS.
21 Smith PG, Rodrigues LC, Fine PEM. The 1990;
Biometrical Journal 32: 243-56.
Assessment of the protective efficacy of 35 May RM, Anderson RM. Transmission
vaccines against common diseases using case- dynamics of HIV infection. Nature 1987; 326:
control and cohort studies. International Journal 137-42.
of Epidemiology 1984; : 87-93.
13 36 Peterman TA, Stoneburner RL, Allen JR, Jaffe
22 Scalia-Tomba G. Asymptotic final size HW,Curran JW. Risk of HIV transmission
distribution of the multitype Reed-Frost from heterosexual adults with transfusion-
process. Journal of Applied Probability 1986; associated infections. Journal of American
: 563-84.
23 Medical Association 1988; 259: 55-63.
23 Macdonald G. Theory of the eradication of 37 Kaplan EH. Modelling HIV infectivity: Must
malaria. Bulletin of the World Health sex acts be counted. Journal of AIDS 1990; 3:
Organisation 1956; 15: 369-87. 55-61.

24 Garrett-Jones C. Prognosis for interruption of 38 Diekmann O, Dietz K, Heesterbeek JAP. The
malaria-transmission through assessment of the basic reproduction ratio for sexually
mosquito’s vectorial capacity. Nature 1964; transmitted diseases: I. Theoretical
:1173-75.
204 considerations. Mathematical Biosciences 1991;
25 Molineaux L, Muir DA, Spencer HC, 107: 325-39.
Wernsdorfer WH. The epidemiology of malaria 39 Dietz K, Schenzle D. Mathematical models for
and its measurement. In: Wernsdorfer WH, infectious disease statistics. In: Atkinson AC,
McGregor I eds. Malaria, principles and practice Fienberg SE eds. A celebration of statistics: the
of malariology. Edinburgh: Churchill ISI centenary volume. New York: Springer,
Livingstone, 1985: 999-1089. 1985:167-204.
26 Dye C. Vectorial capacity: must we measure all 40 Jacquez JA, Simon CP. The reproduction
its components. Parasitology Today 1986; 2: number in deterministic models of contagious
203-209. diseases. Comments on Theoretical Biology 1991;
27 Najera JA. A critical review of the field 2: 159-209.
application of a mathematical model of malaria 41 Anderson RM, Medley GF, May RM, Johnson
eradication. Bulletin of the World Health AM. A preliminary study of the transmission of
Organisation 1974; 50: 449-57. the human immunodeficiency virus (HIV), the
28 Clements AN, Paterson GD. The analysis of causative agent of AIDS. IMA Journal of
mortality and survival rates in wild populations Mathematics Applied to Medicine and Biology
of mosquitoes. Journal of Applied Ecology 1981; : 229-63.
3
1986;
: 373-99.
18 42 Cairns AJG. Model fitting and projection of the

41
AIDS epidemic. Mathematical Biosciences 1991; diseases:

theory and applications. London:
107: 451-89. Chapman and Hall, 1982: 180-208.
43 Hethcote HW, Van Ark JW. Epidemiological 54 Rogers DJ. The dynamics of vector-
models for heterogeneous populations: transmitted diseases in human communities.
proportionate mixing, parameter estimation, Philosophical Transactions, Royal Society of
and immunization programs. Mathematical London 1988; B321
: 513-39.
Biosciences 1987: 84: 85-118. 55 Hasibeder G, Dye C, Carpenter J.
44 Thomson D, Foege W. Faith tabernacle Mathematical modelling and theory for
smallpox epidemic, Abakaliki, Virginia. WHO/ estimating the basic reproduction number of
SE/68.3, 1968: 8. canine leishmaniasis. Parasitology 1992; 105:
45 Bailey NTJ, Thomas AS. The estimation of 43-53.
parameters from population data on the general 56 Muench H. Catalytic models in epidemiology.
stochastic epidemic. Theoretical Population Cambridge, MA: Harvard University Press,
Biology 1971; 2: 253-70. 1959.
46 Becker NG. Analysis of infectious disease data. 57 Anderson RM, May RM. Infectious diseases of
London: Chapman and Hall, 1989. humans: dynamics and control. Oxford: Oxford
47 Becker NG. Martingale methods. Statistical University Press, 1991.
Methods in Medical Research 1993; 2: 93-112. 58Anderson RM, May RM. Age-related changes
48 Macdonald G. The measurement of malaria in the rate of disease transmission: implications
transmission. Proceedings of the Royal Society of for the design of vaccination programmes.
Medicine 1955; 48: 295-301. : 365-436.
Journal of Hygiene 1985;
94
49 Fine PEM. Superinfection - a problem in 59 Schenzle D. Control of virus transmission in
formulating a problem (an historical critique of age-structured populations. In: Capasso V,
Macdonald’s theory). Tropical Diseases Bulletin Grosso E, Paveri-Fontana SL eds. Mathematics
:475-88.
72
1975; in biology and medicine, proceedings Bari 1983.
50 Macdonald G, Cuellar CB, Foll CV. The Berlin: Springer, 1985: 171-78.
dynamics of malaria. Bulletin of the World 60 Dietz K, Schenzle D. Proportionate mixing
Health Organisation 1968; 38: 743-55. models for age-dependent infection
51 Dietz K. Mathematical models for transmission transmission. Journal of Mathematical Biology
and control of malaria. In: Wernsdorfer WH, : 117-20.
22
1985;
McGregor I eds. Malaria, principles and practice 61 Grenfell BT, Anderson RM. The estimation of
of malariology. Edinburgh: Churchill age-related rates of infection from case
Livingstone, 1985: 1091-133. notification and serological data. Journal of
52 Dietz K. Density dependence in parasite Hygiene 1985; 95: 419-36.
transmission dynamics. Parasitology Today 62 Keiding N. Age-specific incidence and
: 91-97.
4
1988; prevalence: a statistical perspective. Journal of
53 Barbour AD. Schistosomiasis. In: Andersom the Royal Statistical Society A 1991; 154: 371-
RM ed. The population dynamics of infectious 412.

Dietz - SMMR - 1993

Uploaded by

Copyright:

Available Formats

Dietz - SMMR - 1993

Uploaded by

Document Information

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

Dietz - SMMR - 1993

Uploaded by

Copyright:

Available Formats

Statistical Methods in Medical Research

The estimation of the basic reproduction number for infectious diseases

The online version of this article can be found at:

Email Alerts: http://smm.sagepub.com/cgi/alerts

>> Version of Record - Mar 1, 1993

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

The estimation of the basic reproduction number

(The gross reproduction rate is given by

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

The sign of r equals the sign of R° -1...

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

If however one defines Ro like Diekmann et a1.14 the corresponding minimum

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

with mean value Ro and variance R° ( 1 + R° ).

2 Why would one need to know Ro?

2.1 Elimination or eradication .. ~ - .

2.2 Vaccination programmes

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

3 Estimation of Ro using data about individual parameters .

As we have seen above, Ro depends on three parameters: ’

1 The contact rate

3.1 Estimation of Ro for malaria

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

Macdonald23 provides an example based partly on entomological observations and partly

frequency on man K 0.5, daily

Macdonald comes value for what we would now call Ro of roughly

which he interpreted as ’the average number of inoculations with a specified parasite

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

mosquitoes in their resting-places and determines the proportion of human bloodmeals

r, 274. This means that a reduction of the mosquito’s life expectancy

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

3.2 Estimation of Ro for AIDS

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

p the rate of acquiring a new partner for a single individual; .,

)JLi the death-rate of an infected individual;

r3 the rate of sexual contact during a partnership; ..:. .~ .

The summation starts with k 1 since by definition of a partner at least

and in general one gets39

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

In the simple case of partnerships with an exponentially distributed duration and a

Then the formula for Ro may be simplified as follows:

contact takes place per partner, then

This inequality is only meaningful for

This shows that for a per-contact-infection probability h of 0.2% and C =

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

Obviously wehave rN 1 for rh

4 Estimating Ro from the initial rate of growth of the epidemic

5 Estimates of Ro based on the final size of an epidemic

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

6 Estimating Ro using data from equilibrium situations

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

mosquitoes. Baseddata from Tanganyika he found a value of R 2 0 of 1.15 which is

Model II: Superinfections in man but infectivity to vector proportional to prevalence

infection; superinfections in vector and infectivity proportional to number

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

force of infection yields the following simple expression for Ro :1

Downloaded from smm.sagepub.com at UNIV TORONTO on November 5, 2012

If the force of infection is in an equilibrium situation then this reflects