ncidence functions and population thresholds Jamie Lloyd-mith Center for nfectious Disease Dynamics Pennsylvania tate University Outline ncidence functions Density dependence Population thresholds - nvasion - Persistence - Thresholds and host extinction The incidence rate Density-dependent transmission ncidence rate = f (,) ncidence rate = f (,) ncidence term in models describes the rate that new infections arise. f(,) = Force of infection Force of infection, λ = c(n) p /N c (N) = contact rate (possibly density-dependent) o p = probability of transmission given contact /N = prob. that randomly-chosen partner is infectious f (, ) = c( N ) p N f (, ) = c( N ) p N f contact rate is linearly density-dependent: c(n) = kn Then f(,) = kn p /N = β MA where β MA = kp Mass action transmission. Also known as density-dependent or, confusingly, pseudo-mass action (see McCallum et al, 2) Frequency-dependent transmission ncidence rate = f (,) f (, ) = c( N ) p N f contact rate is constant with respect to density: c(n) = c Then f(,) = c p /N = β FD /N where β FD = c p Frequency-dependent transmission. Also known as the standard incidence or, confusingly, true mass action (see McCallum et al, 2) McCallum et al (2) Trends Ecol Evol 6: 295-3.
aturating transmission Many choices what to do? Classically it was assumed that transmission rate increases with population size, because contacts increase with crowding. mass action (β) was dominant transmission term Hethcote and others argued that rates of sexual contact are determined more by behaviour and social norms than by density, and favoured frequency-dependent transmission for TDs. ince the 99s, this has been a topic of active research using experimental epidemics, field systems, and epidemiological data. Deredec et al (23) Ann Zool Fenn 4: 5-3. Detecting density dependence Evidence for FD vs MA transmission How can we test for density dependence in transmission? Fit models with different transmission functions to epidemic time series. Look at indicators for transmission N in epidemiological data: With increased transmission rate, we expect: estimates of R exponential growth rate of epidemic, r proportion susceptible following epidemic, or at steady state mean age of infection in endemic setting Fitting models to data from cowpox in bank voles and wood mice FD model is better fit than MA (though neither is perfect) Begon et al (999) Proc Roy oc B 266: 939-945. Evidence for FD vs MA transmission Measles in England and Wales R is ~ constant vs population size Evidence for FD vs MA transmission 9 8 Model results: density-dependent transmission 9 8 Lepto data roughly FD transmission (recall that MA predicts that R N) Mean age of infection (y) 7 6 5 4 3 2 p<. Mean age of infection (y) 7 6 5 4 3 2 p=.54 p=.35 p<..6.8.2.4.6.6.8.2.4.6 Population size, N x 5 Population size, N x 5 Leptospirosis in California sea lions Mean age of infection does not decrease with N Bjornstad et al (22) Ecol Monog. 72: 69-84 transmission not density-dependent. 2
Evidence for FD vs MA transmission Model results: density-dependent transmission Lepto data Evidence for FD vs MA transmission neither?.35.3 p=5.9e-6, slope C = (4.7e-7, 8.6e-7).7.6 p=.9, slope C = (-4.4e-6, e-6) Exponential growth rate.25.2.5. Exponential growth rate.5.4.3.2.5..5.5 2 2.5 3 ize of mixing population x 5.4.6.8.2.4.6 ize of mixing population x 5 Leptospirosis in California sea lions Epidemic growth rate does not increase with N transmission not density-dependent. PiGV in Plodia (ndian meal moth) Transmission rate is not FD or MA need complex functional forms. nterpret in terms of host heterogeneity and effects of density on behaviour. o what should we do? Despite its fundamental importance, the issue of how to formulate the transmission term in simple models is unresolved. ome pointers: FD transmission is generally thought to be more appropriate than MA in large well-mixed populations. n quite small populations, transmission is generally thought to exhibit some density dependence and MA is acceptable. Think about population structure and mechanisms of mixing at the scales of space and time you re thinking about. s a very simple model appropriate? (more on this in the next lecture) Population thresholds in epidemic dynamics R has been the central concept in epidemic dynamics since ~98, thanks largely to the work of Anderson & May. (see the history of R by Heesterbeek 22, Acta Biotheoretica) Long before this, people studying epidemic dynamics have focused on population thresholds. Population threshold for invasion (Kermack & McKendrick 927): host population size below which parasite cannot invade. Population threshold for persistence, or the critical community size (Bartlett 957, Black 966): host population size below which parasite cannot persist long-term. Population threshold for disease invasion Under density-dependent transmission, R = βnd or in fact R any increasing function of N. R > corresponds to a population threshold N>N T. Population threshold for disease invasion Under frequency-dependent transmission, R = βd. No threshold N for R >. Probability of extinction.8.6.4.2 R < Disease dies out R > Disease can invade/persist 2 3 4 5 Basic reproductive N T number, R R Host population size, N R > Disease can invade/persist Host population size, N 3
usceptibility threshold for disease invasion Recall: under any form of transmission, R effective = R /N. For R effective >, must have /N > /R. R eff < Disease dies out R eff > Disease can invade/persist Population thresholds for invasion: evidence Despite its conceptual simplicity, real-world evidence for invasion thresholds is hard to find, for several reasons failed invasions are difficult to observe demographic stochasticity leads to variation in outbreak sizes when R <, limited chains of transmission can still occur when R >, epidemic can still die out by chance. N= R =.9 R =.5 /R Proportion susceptible, /N This phenomenon is the basis for herd immunity. Frequency Outbreak size Outbreak size tochastic variation in outbreak size tochastic variation in outbreak size Measles outbreaks in vaccinated populations, UK Lloyd-mith et al (25) Trends Ecol Evol 2: 5-59. Outbreak size Posterior distribution on R eff under two models Branching process models allow analysis of outbreak size to make inference about the effective reproductive number. Farrington et al (23) Biostatistics 4: 279-295. Population thresholds for persistence Even if parasite is able to invade (R >), this does not guarantee its persistence in the long term. There are two broad mechanisms whereby a disease can fail to persist, or fade out: Endemic fadeout: random fluctuations around the endemic equilibrium can cause extinction of the parasite. Epidemic fadeout: following a major epidemic, the susceptible pool is depleted and the parasite runs out of individuals to infect. Critical Community ize is population size above which a disease can persist long-term (yes, this definition is vague). Persistence thresholds another view Broken chains of transmission can arise in two ways:. usceptible 2. Birth Lost immunity nfected. usceptible bottleneck 2. Transmission bottleneck Recovered R usceptible Courtesy of Ottar Bjornstad Transmission Epidemic fadeout: Parasite extinction occurring because susceptible numbers are so low immediately following an epidemic that small stochastic fluctuations can remove all parasites. (usceptible bottleneck) Endemic fadeout: Parasite extinction occurring because endemic numbers of infected individuals are so low that small stochastic fluctuations can remove all parasites. (Transmission bottleneck) 4
Endemic fadeout tochastic R model with FD transmission and R =4. simulations are shown. + signs show times when disease fades out. Population size 4 usceptible,.8 3.6 2.4.2 nfectious, 2 4 6 Time Fraction fadeout Endemic fadeout Mean time to endemic fadeout for stochastic R model with R =4 and different rates of demographic turnover.. No sharp threshold in N; there s a gradual trend of longer persistence. 2. Demographic rates are as important as N, if not more so. Mean time to disease fadeout 2 5 (i) Fast turnover 5 low turnover (iii) 5 5 2 Population size (N ) Lloyd-mith et al (25) Trends Ecol Evol 2: 5-59. (ii) Endemic fadeout Quasi-stationary distribution: distirbution of conditioned on non-extinction. Fast births and deaths Epidemic fadeout tochastic R model with FD transmission and R =4 (exact same model as for endemic fadeout, but now started from = instead of =*.) simulations are shown. + signs show times when disease fades out. 6 25 low births and deaths Quasi-stationary distribution of usceptible hosts ( ) 5 4 3 2 2 5 5 nfected hosts () ee work by. Nasell for mathematical development. Quasi-stationary distribution of 2 4 6 8 Time Epidemic fadeout Probability that disease persists through the first post-epidemic trough. No sharp threshold in N; there s a gradual trend of longer persistence. 2. Demographic rates are as important as N, if not more so. Probability of persistence.8.6.4.2 (i) Fast turnover 5 5 2 25 (ii) Population size (N ) (iii) low turnover The classic example of epidemic fadeout: measles Note how measles is not endemic in celand, but instead has periodic outbreaks dependent on re-introduction of the virus. UK ~ 6M people Denmark ~ 5M people celand ~.3M people 5
The classic example of epidemic fadeout: measles Extinction risk: can a disease drive its host extinct?.7 Density-dependent transmission Frequency-dependent transmission Proportion of time extinct.6.5.4.3.2. Recurrent Epidemics Endemic Critical Community ize ~3-5k Population threshold N T protects host from disease-induced extinction. R < R > No threshold diseaseinduced extinction is possible (Getz & Pickering 984). R > N T population size Measles in England and Wales Direct extinction due to disease in single host is unlikely, but diseases can cause bottlenecks such that genetic diversity and Allee effects become important. Extinction risk: can a disease drive its host extinct? Extinction risk: multiple host species and spillover pillover from reservoir can threaten endangered populations What if the host population itself has a threshold density below which it cannot persist? Then the outcome depends on the relative values of the threshold population size for disease extinction vs host extinction. Deredec & Courchamp (23) Ann. Zool. Fenn. 4:5-3. 997 Mediterranean monk seal die-off in Mauritania. > monk seals died (~/3 of global population), probably due to dolphin morbillivirus (a relative of measles) that spilled over from another species. Woodroffe, 999 6