SESSION 3 MEASURES OF POPULATION IMPACT AND INFECTIOUSNESS

Objectives By the end of this session, students should be able to: define, calculate and understand the applications of measures of population impact: the population attributable risk (PAR) and the population attributable risk fraction (PAF) remember forever that measures of effect compare exposed and unexposed; and measures of impact compare population and unexposed decide which questions are answered by using measures of effect or measures of population impact and understand the assumptions behind the use of these measures understand the concepts of transmissibility of infectious diseases; define and calculate the net reproduction number and the basic reproduction number define and calculate the herd immunity threshold

See Hennekens & Buring Chapter 4.

Measures of population impact estimate the expected impact on a population of removing or changing the distribution of risk factors in that population. They take into account the strength of the association (estimated by a measure of effect, like the rate ratio) and the distribution of the risk factor in the population and assume that we have established that the association between disease and risk factor is causal. If assumptions are true, these estimates measure the population impact, i.e. estimate how much of the disease in the population is caused by the suspected risk factor. (But only if assumptions are true). Measures of population impact are specific to the population studied, and can only be generalised to populations with exactly the same distribution of risk factors. Risk factors that are strongly associations but which are rare, like being exposed to an X ray in pregnancy and leukaemia in childhood, may have a large measure of effect but small measure of impact. There are two main measures of population impact: population attributable risk (PAR) and population attributable risk fraction (PAF). Population attributable risk (PAR): is the absolute difference between the risk (or rate) in the whole population and the risk (or rate) in the unexposed group. It is a particular type of risk (or rate) difference that contrasts the risk (or rate) in the unexposed group (r0) with the risk in the total study population (r) (exposed and unexposed combined). (From Session 2 we learned that the risk (or rate) difference contrasts the risk (or rate) in the unexposed (r0) with that in the exposed group (r1)).

L3.1

As the risk (or rate) in a population depends on how common risk factors are in that population, PAR varies according to how common exposure to the risk factor is in the population. It is also called the population risk difference. Population attributable risk (PAR) is calculated as: PAR = r - r0, or alternatively, if we know the risks among the exposed (r1) and unexposed (r0), and the prevalence of exposure in the population (p): PAR = p (r1-r0) Population attributable risk fraction (PAF): the proportion of all cases in the whole study population (exposed and unexposed) that may be attributed to the exposure, assuming a causal association. It is also called the aetiologic fraction, the percentage population attributable risk and the attributable fraction. The population attributable risk fraction (PAF) is estimated by dividing the population attributable risk by the risk in the total population (r). i) PAF = PAR/r Substituting PAR = r - r0 in this equation, you get ii) PAF = (r - r0) / r When only the risk ratio (RR) and the prevalence of exposure in the population (which can be estimated by the exposure prevalence among controls in a case-control study) (p) are known, PAF can also be calculated by: iii) PAF = p (RR-1) / [{p (RR-1)}+1] If you know the prevalence of exposure among cases (p1) there is a very useful formula for PAF which can be used with risk or rate ratios that have been adjusted for confounding: iv) PAF = [p1 (RR-1)] / RR You can learn more about this in Term 3, in the Advanced Statistical Methods in Epidemiology study unit. It is possible to calculate PAR and PAF based on rates rather than risks. As usual, when diseases are rare, these measures are numerically very similar. The most important use of measures of population impact is for the planning of public health interventions, to predict the impact of removing, or changing, the distribution of the risk factor on the frequency of disease in the population. This can also be interpreted as the fraction of all cases in the population that is caused by the risk factor; and also as the probability that any case in the population was caused by the risk factor (and has been used as such in court cases).

L3.2

Causality and other assumptions; and more than one risk factor There are caveats to these interpretations. The main one is that the calculation of PAF assumes that all of the association between the risk factor and the disease is causal; i.e. that there was complete control of confounding. Other assumptions are also necessary: that both risk factor and frequency of disease were measured accurately, that removal of the risk factor actually removes the risk (i.e. that those previously exposed and now unexposed have the same risk as those never exposed), that the risk factor is actually removable, etc. Things get more complicated when we think about causes that are necessary but not sufficient, effect modification, etc. This is well covered in Term 3). The first three PAF formulae (above) can only be used when the estimated risk ratio is from a single variable (unadjusted) analysis. When there are many risk factors under study, and risk ratios are derived from multivariable analysis, PAFs can still be estimated using different formulae (see Bruzzi et al; 1985 and Coughlin et al, 1991) including the fourth formula above, but the interpretation is more problematic. Examples of measures of effect and measures of impact The table below shows the mortality in the first year of life of babies born to married parents by occupational social class of father in a country in 1999, excluding congenital malformations and pre-term births. Table 1 Social class of father r0 r1 r Non-manual Manual Total Births (number) 640396 944297 1584693 Births (proportion) 40.5% 59.5% 100% Rate per 1,000 births 2.75 4.08 3.55

Deaths 1762 3857 5619

If we define the outcome of interest as mortality in the first year of life and the potential risk factor as having a father in a manual occupation: Using the same nomenclature, i.e. r1 = rate in exposed; r0 = rate in unexposed; r = rate in the population; and p = prevalence of exposure in the population, the measures of effect and population impact of fathers social class on mortality in the first year of life can be calculated. These are presented in Tables 2 and 3 respectively. Table 2 Summary of measures of effect Rate ratio Rate difference Rate difference percent RR RD RD% r1/r0 r1-r0 (r1-r0)/r1 4.08/2.75 4.08-2.75 (4.08-2.75)/4.08 1.48 1.33 per 1,000 births 33%

L3.3

The interpretation of the measures in Table 2 is that having a father working in a manual occupation compared to a non-manual occupation (or some factor associated with this) increased the risk of death in the first year of life by roughly one and a half times and caused death in the first year of life of 1.3 of every 1,000 babies born to fathers in manual occupations. Furthermore, the increased risk for babies born to fathers in manual occupations was responsible for about a third of all deaths in the first year of life of babies of fathers in manual occupations. Table 3 Measures of impact
Population attributable risk PAR PAR Population attributable risk fraction PAF r-r0 p (r1-r0) PAR/r (r r0)/r p(RR-1) / [p(RR-1)+1] [p1(RR-1)] / RR 3.55-2.75 0.596 (1.33) 0.80/3.55 0.80 per 1,000 births 0.80 per 1,000 births 22.5%

PAF PAF PAF

(3.55-2.75)/3.55 0.596 (1.48-1) / [0.596(1.48-1)+1] [(3857/5619)(1.48-1)] / 1.48

22.5% 22.2% 22.3%

The interpretation of these measures is that, assuming causality, having a father working in a manual occupation (or the living conditions that are associated with it) is the cause of the death in the first year of life of 0.8 of every 1,000 babies in the country, and for just over a fifth of all deaths in the first year of life.

How infectious? Measures of infectiousness or transmissibility Infectious diseases are characterised by the fact that they are caused by infectious organisms that can be transmitted and lead to new cases. In the study of infectious diseases, and particularly in diseases where there is person-to-person transmission, it is possible to estimate a measure of transmissibility of an infection in a specified population. Infectiousness depends on factors unique to each organism, such the length of time an infected individual remains infections and the probability of transmission given contact between an infectious and an uninfected individual, as well as characteristics of the environment, such as the type and number of contacts in the population. So measles will be more infectious than mumps in any society, and both measles and mumps will be more infectious in crowded urban environments than in isolated rural societies. Some of the measures of infectiousness are: the attack rate and the case reproduction numbers. Attack rates The attack rate and the secondary attack rate are used only in specific situations: attack rate is used in the context of outbreaks or epidemics of infectious diseases and secondary attack rate is used in the study of the spread of infectious diseases in small communities like households and schools. Attack rate is measured over a short period of interest the duration of the outbreak or epidemic.

L3.4

Attack rate (AR) is the number of new cases (d) occurring in the duration of the outbreak among the population at risk in the start of the outbreak (N). The secondary attack rate (2nd AR) is the proportion of those exposed to the primary case that develop disease as a result of the exposure. The secondary attack rate is a measure of incidence, of new cases over those at risk. Case reproduction numbers The case reproduction numbers are measures of new cases to new cases in the chain of transmission ( and ignores denominators). There are two case reproduction numbers: the basic case reproduction number and the net case reproduction number. Case reproduction numbers estimate the average number of secondary cases originated by each case. The basic case reproduction number estimates the maximum potential for transmission (i.e. the average number of secondary cases originated by each case when a disease is introduced in a population where all are susceptible, and so all contacts can result in a new infection), while the net reproduction number estimates the actual, effective transmission, given the fact that a proportion of the population is immune or already infected. Basic case reproduction number (R0): Average number of secondary cases per case (number of successful transmissions per case) in a totally susceptible population. The figure illustrates a basic reproduction number of 2. Notice that all potential contacts lead to a new case as the population is totally susceptible. The basic case reproduction number can be measured directly, by counting the average number of secondary cases after an infection is introduced in a totally susceptible population. Examples of this situation include, for example, an infection re-introduced in an island where it has been absent for generations, or the spread of a newly-emerging infection, e.g. SARS. When a disease has been present in a population for some time, a fraction of the population becomes immune or is already infected. In this situation, not all potential contacts lead to transmission, as some contacts will take place with individuals who are immune (or already infected). Thus, the average number of secondary cases per case decreases - this is the net case reproduction number, R.

L3.5

Net case reproduction number (R): average number of new cases per case (number of successful transmissions per case) in a population. At the School, we use the notation R for the net case reproduction number, but others have used Rn or Rt. In the figure, subjects inside a box are immune and do not become cases. The net case reproduction number (R) illustrated is 1. Half the subjects are immune, so each case results in successful transmission to (on average) one other individual. By contrast, the basic case reproduction number (R0) is 2. The net case reproduction number is the basic case reproduction number multiplied by the fraction of the host population that is susceptible (x): R = R0*x or R0 = R/x This requires some assumptions about the population. The most important is of homogenous mixing, i.e. that infectious cases are as likely to mix with immune persons as with susceptibles. When the population and the disease are stable - and thus the number of cases in the population is constant - each case leads to the occurrence of a single new case. In this situation the net case reproduction number, R, is 1, and transmission is stable. For example, for a stable disease to which 90% of the population is immune and 10% susceptible: R = 1 (because disease is stable) x = 0.1 (10% susceptible as 90% of the population is immune) R = R0*x 1 = R0*0.1 R0 = 1/0.1 R0 = 10 When there is a good serological correlate of immunity, the proportion of the population that is immune can be estimated directly through serological surveys. For diseases that result in life-long immunity, the proportion susceptible in the population, x, (in the absence of vaccination and in a stable population) is a function of the average life expectancy and the average age of infection: If: L = average life expectancy, and a = average age of infection in unvaccinated population, then x, the proportion susceptible in the population is x = a/L As R0 = R/x, when the infection is stable (R=1) then R0 can also be expressed as: R0 = L/a

L3.6

For example, imagine a stable population where life expectancy is 80 years, and the average age for acquiring a particular disease is 3 years. The incidence and average age of the disease has not changed for some time. There is no vaccine, and the disease causes life-long immunity. L=80 a=3 x=3/80 or 0.036 (3.6% or 0.036 of this population is susceptible) R=1 (disease stable) R0= 1/0.036 or R0=80/3 =26.6 (in a totally susceptible population each case will give origin to 26.6 cases).
Average age of disease 3 Average age of death 80

susceptible susceptible

immune

Herd immunity is the proportion of the population immune to an infection; for a given infection there is a threshold of herd immunity above which there are not enough susceptibles to maintain the infection in a population; when this threshold is reached (for example, by vaccination) each case leads to a single new case (R=1) and transmission is constant; if this threshold is surpassed, transmission decreases (on average, each case leads to less than one new case and R<1), leading eventually to elimination (but note that the above assumptions, e.g. homogeneous mixing of cases and contacts, apply). Herd immunity threshold (HIT) is thus the percentage of the population that need to be immune for a disease to become stable ( for R =1): for every case to give origin to only one new case. If the proportion of the population that is immune is higher than this threshold, R<1, and the disease dies out. This is used in the planning stages of programmes to control or eradicate infectious diseases. When each case gives rise to, on average, one new case, R=1, so HIT can be derived from the basic case reproduction number as follows: HIT = (R0-1)/R0 or 1 - (1/R0). For example, if in a totally susceptible population one case of measles leads to 15 new cases, for the disease to become stable 14 out of 15 potential contacts should be immune. If in a totally susceptible population one case of rubella leads to 5 new cases, 4 out of 5 contacts should be immune. Let's look at the measles example: R0 = 15 HIT = (15-1)/15

L3.7

HIT = 14/15 = 0.93 or 93% Thus, vaccination coverage (and the vaccine efficacy) should be high enough to render 14/15, or 93%, of the subjects immune, in order for the disease to become stable. For the disease to die out, the level should be even higher.

L3.8

References:

Vynycky E, White RG. And introduction to infectious disease modelling. Oxford University Press. 2010. Anderson R, May R. Infectious diseases in humans: dynamics and control. Oxford Science Publications, 1992. Chapters 1 to 5

Beaglehole et al. Basic epidemiology. WHO, Geneva, 1993. Chapter 2, pp 14-30 Kleinbaum DG et al. Epidemiologic research. New York: Van Nostrand Cia, 1982. Chapter 8 Last JM. A dictionary of epidemiology (4th Edition). Oxford: OUP, 2001 Orenstein WA, Bernier RH, Dondero TJ et al. Field evaluation of vaccine efficacy. Bull WHO 1985;65:1055-68 Rockhill B, Newman B, Weinberg C. Use and misuse of population attributable fractions. Am J Public Health 1998; 88; 15-19 More advanced reading: Bruzzi P, Green SB, Byar DP et al. Estimating the population attributable risk for multiple risk factors using case-control data. Am J Epidemiol 1985;122:904-914 Coughling SS, Nass CC, Pickle LW et al. Regression methods for estimating attributable risk in population based case-control studies: a comparison of additive and multiplicative models. Am J Epidemiol 1991;133:305-313. Rothman KJ, Greenland S Lash T. Modern Epidemiology. Lippincott Raven, Philadelphia, 2008.

L3.9