Microbial forecasting: How do we predict the impact of climate change on infectious diesases?

October 07, 2020

Information

Dr. Virginia Pitzer, Associate Professor of Epidemiology (Microbial Diseases)

EPH 570 Seminar in Climate Change and Health

ID5714

To CiteDCA Citation Guide

00:01- Hi everyone, welcome to our second seminars
00:06in this fall.
00:07This belongs to our Climate Change and Health Center
00:13four seminar series on climate change and health
00:17and today we're very fortunate to have Dr Virginia Pitzer.
00:23She's an Associate Professor of Epidemiology
00:25from the Microbial Disease Department.
00:29So she's also one of our pilot project awards winners.
00:35So her research mainly focused
00:37on the new mathematical modeling
00:40of the transformation dynamics of infectious diseases.
00:45So without further ado,
00:48the floor is Gina Pitzer.
00:52- Well thanks very much for the introduction
00:54and hopefully we can switch to my screen.
00:59I think you need to enable share screen,
01:02screen sharing for me.
01:13- Just a general reminder to everyone
01:16that you should mute yourself
01:18if you are not asking questions.
01:20So we will be greatly appreciated.
01:28- Gina, I think you're good to go.
01:30- Got it.
01:42Okay, can everyone see the presentation now?
01:47Okay, well thanks very much for the introduction
01:50and for the opportunity to speak with you all today
01:55and so as Kai said,
01:56I'm gonna be talking about some of the ways
01:59in which we use different models
02:02as well as our mechanistic understanding of relationships
02:06in order to predict the impacts of climate change,
02:09specifically for infectious diseases.
02:14And so when it comes to predicting the impacts
02:16of climate change on infectious diseases,
02:19often the way that this is done
02:21and let me see if I can get this working,
02:29is to take a model of climate projections over time
02:34and for example this is projections around variations
02:40in temperature from the current day up through 2100
02:49and to combine this with a model
02:51for the incidence of disease given climate
02:55and use this to get projections of the relative risk
02:59of different diseases over time and for example this is data
03:06on model projections of the relative risk of diarrhea
03:10over time given the model forecast
03:13for increases in temperature over time,
03:16suggesting that by 2010 to 2039,
03:21you should see a median increase
03:23or relative risk of around 1.1 and by 2040 to 2069,
03:29a relative increase, median increase
03:32of around a relative risk of 1.2 and upwards of 1.3
03:37by the 2070 to 2099 time frame
03:45and voila, I mean that's basically you know,
03:48one of the ways that people have gone about doing this.
03:52So you know, thank you.
03:53I'll take any questions now.
03:56But of course if it were as simple as that,
03:58this would be a very short talk
04:01and I'd argue that really the most difficult part
04:03of understanding the climate of this is
04:08really understanding the true climate disease relationship
04:12and in particular, the causal effects of climate change
04:16on infectious diseases which is really far
04:20from straight forward and typically cannot be determined
04:23by the simple regression type analyses
04:26that were used in that previous study
04:29and are often used in many types of analyses
04:32of relationships between climate
04:34and chronic diseases for example.
04:39And so I'd argue that in order
04:41to really have a true causal model for linking climate
04:51to infectious diseases, there are really three main criteria
04:55that are needed to be met and these include
04:59that the change in infectious disease incidents
05:01really must occur at the right time, in the right place
05:06and in the right direction in order to be causally linked
05:11to a change in climate
05:15and the last of these criteria
05:21really requires that you have a hypothesis
05:25about the mechanism through which climate impacts
05:29on infectious diseases.
05:31well the first tWo really involve the careful analysis
05:35of spatiotemporal data
05:38and so in this talk I'm really going to begin
05:42by talking about the mechanisms
05:45through which climate can have important impacts
05:48on infectious diseases,
05:50including both the direct effects of climate
05:53on infectious diseases as well as some of the indirect ways
05:56in which climate can impact on infectious diseases,
06:00and then I will talk about how we go about identifying
06:06and quantifying these associations
06:08between climate and infectious diseases,
06:11including the types of data that are often used
06:13to draw these associations
06:16and the various quantitative approaches
06:18that apply specifically to infectious diseases
06:23when trying to measure these associations
06:27between climate and disease transmission.
06:31And I'll largely be drawing on examples from my own work,
06:35particularly when talking
06:36about these quantitative approaches.
06:38And then finally I'm gonna end with just some challenges
06:41and some opportunities to really get further
06:45when it comes to making these predictions
06:49around climate change on infectious diseases.
06:53And so one of the main ways
06:56in which climate can have an impact
06:59on infectious diseases is through the effects of climate
07:04on pathogen survival and, or replication
07:07within the environment.
07:09And so one example here is work that's been done
07:14by researchers to understand the effects of temperature
07:21and humidity on the transmission of influenza
07:25where researchers use guinea pigs
07:27which are a great kind of model system
07:29for measuring influenza transmission,
07:32to examine how the level of transmission happened
07:37from an infected guinea pig to a susceptible
07:41and exposed guinea pig that was housed in a separate cage,
07:45but downwind of this infected guinea pig
07:49when they modulated the temperature
07:54and humidity of the cages
07:58in which these guinea pigs were housed.
08:01And generally what they found was that both survival
08:06and transmission of influenza virus was really enhanced
08:09at low temperatures and low relative humidities,
08:14and when colleagues went about and re-analyzed
08:17some of this data, what they were able to show was
08:21that it was really absolute humidity or vapor pressure
08:24which was even better at explaining
08:26some of these associations
08:28and in particular the combined effect of temperature
08:32and relative humidity and then based on this relationship,
08:39we were able to use some of this data
08:42and combine it with mathematical models of flu transmission
08:47to show that by incorporating the relationship
08:50between absolute humidity and flu transmission
08:54into these models, we could better forecast the timing
08:58of seasonal flu epidemics happening across the US each year
09:04where often the epidemic tended to be preceded
09:08by a dip in absolute humidity
09:10before each flu epidemic happening in each year.
09:17So another way
09:18in which climate might affect infectious disease incidents
09:23is through its impact on host defenses and host behavior
09:28and so you know, we've all been told to bundle up
09:31in the winter so that we don't catch a cold
09:34and of course we know that
09:36while we don't actually catch colds by being cold,
09:42there is actually some potential truth to this mechanism
09:48and so when it comes to colds
09:51which are caused by rhinoviruses as one example,
09:55it has been shown by work from Ellen Foxman
09:59and Akiko Iwasaka here at the med school
10:02that when the nasal cavities are exposed
10:09to warmer temperatures, they tend to exhibit higher levels
10:13of expression of Interferon gamma
10:16which is an important first line in defense
10:20against viruses such as rhinoviruses and other cold viruses.
10:25But these levels of Interferon gamma tend
10:29to be quite a bit lower when temperatures are lower,
10:35leading to potentially slightly impaired
10:37kind of first-line immune responses in the nasal cavities
10:41at colder temperatures
10:43which may be part of the reason why we do indeed tend
10:46to get colds more often during the winter season.
10:54And so another way in which climate can impact
10:59on infectious diseases is through its impacts
11:04on the risk of human exposure.
11:07And so for example it's known
11:09that flooding can increase the risk of exposure
11:13to various waterborne pathogens
11:16including Leptospirosis which is a febrile illness
11:20that's transmitted primarily through the urine of rats
11:24and so when you see these heavy rainfall events,
11:26often the rat urine can get washed into the street
11:33and into sewers where people are walking through
11:36and the bacteria can then enter into humans
11:40through small cuts on the feet
11:42when people are walking around in these floodwaters
11:45and mud that has been contaminated by this rat pee
11:49and this is something that has been studied
11:52by Albert Coe who's the department chair in EMD here
11:57at the School of Public Health
12:00and then finally for vector-borne diseases,
12:04it's very clear
12:05that climate can have really important impacts
12:08on the risk of disease often through its impacts
12:12on factors such as vector survival,
12:14vector fertility and development
12:18as well as the biting behavior of various vectors.
12:23And so this is why diseases like Dengue fever,
12:27Chikungunya and most recently Zika virus
12:30really tended to be confined primarily to the tropics
12:35since the adult Aedes aegypti mosquito
12:39as well as the Aedes albopictus mosquitoes,
12:42the survival of these mosquitoes is
12:44really temperature dependent
12:46and so at these warmer temperatures,
12:48they tend to live longer,
12:50allowing for the key time for these viruses
12:57to be acquired by the mosquitoes,
13:00develop within the mosquito gut and then be transmitted
13:03to a susceptible individual.
13:07Although these factors such as temperature
13:10really aren't the only factor that needs
13:12to be taken into account
13:14when predicting the risk of disease,
13:16since factors such as human behavior
13:20and the amount of time spent outdoors,
13:22housing development and whether or not there are screens
13:25on the windows and other actions that contribute
13:28to the prevention of mosquito breeding sites for example,
13:32can all play a really big role
13:34in the risk of vector-borne diseases
13:36across different climatic conditions
13:39and kind of across time.
13:43And another important factor is that
13:46while climate can affect the development rate of parasites
13:53and viruses within mosquito vectors
13:59and so for example the extrinsic incubation period
14:03of malaria often tends to be shorter at higher temperatures,
14:09another important factor which is
14:11often not necessarily taken into account is that variations
14:15around mean temperatures can also play
14:18a very important role.
14:21So it was found using experimental system
14:24that diurnal temperature variations
14:27or the variation in temperature between night and day
14:31really can play an important role
14:33in modulating both the development time of mosquitoes,
14:40the EIP as well as the survival rate of mosquitoes
14:45at different temperatures where at lower temperatures,
14:50larger diurnal temperature variations tended
14:53to increase the survival of mosquitoes
14:56and decrease the development time
14:59whereas at higher temperatures
15:01that tend to be you know,
15:02typically more conducive to survival of mosquitoes,
15:06when you take into account
15:07the diurnal temperature variations,
15:09it can actually lead to lower survival
15:12than might be predicted in a higher developmental time.
15:16And so you need to not only take into account
15:18just mean temperatures,
15:19but also often these variations in temperature
15:23around the mean.
15:27And then finally there are other both direct
15:30as well as indirect impacts of climate
15:32on infectious diseases and these include impacts of climate
15:37on the geographic range, population dynamics
15:40and behavior of zoonotic reservoir species
15:43as well as effects on human behavior
15:46such as seasonal migration that may be linked
15:49to agriculture and pastoralism that lead to kind of movement
15:54and aggregation of individuals
15:56in different areas at different times of the year.
15:59Finally, climatic events can cause displacement
16:02and aggregation particularly of climate refugees
16:06in different areas which can make them
16:07particularly vulnerable to various infectious diseases
16:11and then finally, climate can have important impacts
16:15on host susceptibility as we talked about earlier,
16:18but there are both climate related
16:20as well as unrelated causes of seasonal variation
16:24in host susceptibility
16:26for example linked to the length of day
16:31and how exposure to solar radiation can impact
16:35on vitamin D metabolism and such which plays an important,
16:38can be an important co factor in the immune system
16:43and so one of the ways in which we can identify
16:48and quantify the mechanistic impact of climate
16:52on infectious diseases is through experimentation
16:56and so for example, this is what was done
16:58with the guinea pig experiment that I talked about earlier
17:01where they looked at the effects of temperature
17:03and relative humidity on flu transmission.
17:07I also am showing here results of another experiment
17:11in which they looked at the effect of temperature
17:14on snail mortality which is an important host
17:18of Schistosomiasis and showed that the mortality rate
17:23of snails tended to be lowest
17:25when mean water temperatures were around 20 degrees Celsius
17:30in this experimental system,
17:32suggesting kind of the ideal climatic conditions
17:36for kind of greater survival of these snails
17:40which play an important role
17:41in the transmission cycle for Schistosomiasis.
17:49And another way to really identify
17:51and to quantify some of these mechanistic links
17:55between climate and infectious diseases is
17:59to use model-based approaches,
18:02but in this way, in this sort of fashion,
18:05it's really important to make sure
18:08that you're following supposed links
18:11within the causal pathway.
18:14And so for example, we have been working with researchers
18:17in Nepal based on some of the pilot funding
18:21that we received
18:21from the Climate Change and Health Initiative
18:24to try to quantify the impacts of rainfall
18:27on typhoid trans, typhoid fever transmission
18:30within the setting and to estimate the incidence
18:34of typhoid fever that might be attributable
18:36to rainfall in this setting,
18:40and you can see on the plot on the bottom left here
18:44that typhoid fever incidence tends to peak kind of
18:48during the rainy season within this particular setting,
18:50but there are also some of these important variations
18:53in seasonal incidents that are hard to explain
18:56just based on rainfall patterns alone.
19:00However, studies from our collaborators have shown
19:03that levels of bacterial DNA present in water sources
19:09in this region such as these wells that are often used
19:13by individuals to obtain water tend to,
19:18the levels of the bacterial DNA tend to be slightly higher
19:22following increases in rainfall
19:25or these big rainfall events.
19:28However when it comes to trying
19:32to quantify these associations
19:34between infectious disease incident and climate,
19:43there's a variety of different data types
19:45that are often used in order to do this
19:48and one of the most common types of data
19:51that is typically used to look at relationships
19:54between climate variables and infectious disease variables
19:58is data on seasonality,
19:59since often infectious diseases do exhibit
20:02these seasonal variations in incidents,
20:06however there often are a lot of things that vary seasonally
20:09and in these types of analysis,
20:11you need to be really careful to avoid confounding
20:15and just because things are correlated with each other,
20:18it doesn't necessarily mean
20:19that one thing is a cause of another.
20:22So for example, this is data on murders by steam,
20:28hot vapors and other hot objects in the US plotted
20:31in black here and the average age
20:33of the Miss America winner plotted in red
20:36which oddly enough are very highly correlated
20:39with one another, with a correlation coefficient of 87%,
20:43but I have a very hard time seeing how these,
20:47one thing could possibly be causally linked to another
20:51and so just because there are correlations present,
20:54doesn't necessarily mean that
20:55any of these correlations are necessarily causal.
21:01And so it's best if you can also link
21:04when looking at these seasonal relationships,
21:07link the between year variations in incidents
21:12and deviations from normal climatic conditions
21:16to anomalies in the infectious disease incidents.
21:20So for example, one of the things that we found
21:23in modeling the relationship between absolute humidity
21:28and influenza, seasonal influenza in the United States was
21:31that there tended to be these dips
21:34in the absolute humidity relative
21:38to kind of normal absolute humidity
21:40expected for that time of year
21:42and these dips often preceded the onset
21:45of the seasonal influenza epidemic in different US states
21:50by around seven to 14 days,
21:55and this sort of provides good evidence
21:56that there's actually sort of this relationship
21:59where in addition to the experimental evidence,
22:03that absolute humidity is really kind of pre
22:06or precipitating the influence epidemic each year.
22:12And another type of data that's often used
22:16and can potentially be a very strong way
22:19to link infectious disease incidents
22:22to climatic variables is to take advantage
22:25of multi-annual variations in both climate
22:29as well as infectious disease incidents,
22:32and one of the best known multi-annual cycles
22:35when it comes to climate is the El Nino phenomenon
22:39or the El Nino-Southern Oscillation
22:41which has been linked to variation
22:44in cholera cases happening in Bangladesh since the 1990s
22:52through late, or sorry 1980s through late 1990s
22:56where you typically tended to see higher peaks
23:00of cholera epidemics coinciding with years
23:04in which there were
23:05greater sea surface temperature anomalies happening
23:09and these are happening with a frequency
23:12of around five to six years.
23:17However, while these generally provide stronger evidence
23:23in favor of a climate disease link,
23:26since fewer things will vary
23:28at these kind of multi-annual frequencies,
23:31you still need to be careful
23:33when it comes to drawing these causal links
23:36between variations in climate and these variations
23:40in infectious disease incidents,
23:43since infectious diseases can
23:44often exhibit multi-annual cycles that are driven instead
23:49by the internal dynamics of immunity
23:52and susceptibility which I'm gonna touch on
23:54in a couple slides.
23:59And then finally spatial data can often be useful as well.
24:04In particular, the geographic range limits
24:06of a particular pathogen may help to tell you something
24:10about how climate affects its transmission.
24:14For example, this is a distribution map
24:17for the Ixodes scapularis tick
24:20which is the main vector of Lyme disease
24:23within the United States, showing that the
24:27sort of suitable ranges in which we would expect
24:30to see the tick species overlap
24:35with the observed distribution of the tick.
24:39Sorry, I'm accidentally going forward too quickly.
24:46And the one caveat with doing this though is
24:50that you need to be careful not to over interpret
24:53some of the data, since there may also be
24:55other factors involved including behavioral factors
24:59or it just may be possible
25:00that the pathogen hasn't been introduced yet,
25:03for example, to a region where you would predict the climate
25:09to be suitable, but you don't see presence
25:13of the particular pathogen there yet.
25:17And so when it comes to methods for drawing these links
25:24between climate and infectious diseases,
25:29one of the ways that this has traditionally been done
25:32for other diseases not necessarily infectious diseases is
25:37through the use of time series models.
25:41So for example, generalized linear models
25:44such as this Poisson type regression model
25:46which models the log of the number of cases at time t
25:50as a function of the baseline incidence
25:52as well as a variety of different predictors,
25:55some of which may be climatic variables,
25:59but the main limitations of this approach is
26:04that it really assumes that you have independent outcomes
26:09or in other words that the number of cases
26:12of the observed disease at time t is independent
26:17of the number of observed cases of disease
26:20at time t minus one and we know for infectious diseases
26:24that that's just not true
26:26because of the transmission process
26:28and because often the cases at time t minus one
26:32are actually causing the cases happening at time t.
26:39And so models that do not account
26:42for these underlying variations in susceptibility
26:45of the population may fail to identify
26:48some important climate disease relationships.
26:51And this is just an example plotted here
26:55in which we model the potential relationship
26:58between a climactic variable,
27:01in this case we're gonna say precipitation
27:04and we're gonna say that there's this link
27:06between precipitation and climate
27:08which we're modeling
27:10or rather the transmission rate, beta t,
27:13which we're modeling up on the top here
27:16where there is this biannual pattern of precipitation
27:22with two lengths a year causing these sort of two peaks
27:27in the transmission rate happening
27:29at different times of the year.
27:31So this large peak and then this minor peak
27:34in the transmission rate happening each year.
27:38And if you model the incidence of a disease
27:41in which you have a low r-0
27:44or a lower transmission rate within the population,
27:48you see kind of a similar predicted pattern
27:52of cases happening through time
27:56where you see a peak in cases happening
27:58following the peak in precipitation
28:01or the peak in the transmission rates,
28:04followed by a decrease
28:05and then a smaller peak happening coincident
28:08with the smaller peak in the transmission rate
28:10and this pattern kind of repeating over time
28:12where you just see this sort of lag between,
28:15for example your climatic variable
28:17which is shaping transmission here
28:19and your peak in incidence.
28:22But if you take the same model and simulate it
28:25with a higher r-0 or a higher baseline transmission rate,
28:30you can get into these patterns
28:32in which you see a very large epidemic happening,
28:35kind of the first time climate is
28:39sort of favorable to transmission,
28:42but then you've kind of overshot the susceptible population
28:45such that you don't have enough susceptible people around
28:47to cause an epidemic the next time climate is favorable
28:52to transmission happening
28:54and so there's no epidemic happening
28:55even though conditions are favorable this year
28:59and that you have to wait another year
29:02until climate conditions are both favorable
29:04as well as there's enough susceptible individuals around
29:07to have another epidemic occurring.
29:09And you can see that in this instance it would be
29:11very much more difficult to link your climate driver
29:14on top here to the observed incidence of cases happening
29:17in the population as modeled here.
29:23And so as a result, there's a variety of different methods
29:29that can be used and are often used
29:32when specifically looking
29:33at the climate disease relationship for infectious diseases.
29:37and these vary from the traditional statistical methods
29:40that I mentioned earlier
29:41including your generalized linear models
29:44through to models that do account for autocorrelation
29:48within data such as ARIMA models
29:50and time-varying coefficient models
29:53to methods such as time series decomposition and wavelets,
29:58semi-mechanistic models known as TSIR-type models,
30:02down through the fully transdynamic models
30:06such as transmission dynamic models
30:08or individual based models.
30:10And similarly, there are spatial methods
30:12that can be applied as well varying from static risk maps
30:15through to dynamic risk maps
30:18and I'm just gonna touch on
30:19a few of these different examples,
30:20kind of using some of our own work to illustrate it.
30:24So for example, one of the things
30:26that we're working on currently is to try
30:28and understand links between climate and diarrhea incidents
30:32across different districts within Ghana
30:35as modeled or as shown in this map here on the right
30:43where we have the observed incidents per 10,000 individuals
30:47on the left and the model predicted incidents on right
30:51where we're using
30:52a simple time series Poisson regression model
30:56where the log number of cases at time t is a function
30:58of the baseline incidence plus a function
31:02of the mean temperature in the given district at time t,
31:07the diurnal temperature variation,
31:09a model for wetness prevalence
31:13or the presence of wetness
31:15which incorporates precipitation data
31:17as well as often using harmonic terms
31:21for annual and possibly biannual variations in incidents
31:25where you can see the model provides a reasonably good fit
31:28to diarrhea incidents in Navrongo which is a city,
31:36a small city in the northern part of Ghana
31:39as well as Accra which is the main capital
31:42in the southern part of Ghana,
31:44but one of the interesting things when you look
31:46at actual correlations and the coefficients
31:49within these models is that you see opposite relationships
31:54between your climatic variables
31:57including the mean temperature in this panel,
32:02the second panel as well as the wetness prevalence
32:05or a measure of precipitation in the fourth panel here
32:09in the northern part of the country
32:11versus the southern part of the country,
32:13where here we're plotting
32:15the Pearson correlation coefficient
32:17across these different areas and showing
32:19that you see negative associations between temperature
32:23and diarrhea incidence in the north
32:25and positive associations in the south
32:28whereas you see the opposite pattern
32:29when it comes to wetness presence
32:32where it tends to be positive associations in the north
32:35and more negative associations found in the south.
32:39And so it's I think gonna be difficult
32:41to really kind of tease apart what are the main drivers
32:44of these differences and what are the other factors
32:46that are involved that really explain
32:49sort of the differences in climate,
32:51in the role that climatic factors play in diarrhea
32:56in this setting, and one of the ways that we can do this
32:59and that we're planning to do this
33:02is using spatiotemporal models and this is a previous study
33:08in which we use spatiotemporal hierarchical Bayesian models
33:12to look at diarrhea and the associations between climate
33:15and diarrhea incidents in Afghanistan
33:18and using these methods, we're really kind of able
33:21to show that higher diarrhea incidents
33:27which tended to be concentrated
33:29around the population centers in the northeast
33:33as well as in some of the other
33:36kind of northern outlying regions is really associated
33:40with both positively with aridity and fluctuations
33:46in mean daily temperature as well as negatively
33:49with changes in average annual temperature
33:53where colder parts of the country tended
33:58to have a higher incidence than might be expected
34:02kind of otherwise.
34:08And another way in which we can use different
34:12or another approach rather to using models
34:16to tease apart these climate disease relationships is
34:19to use what's called a TSIR type model
34:23which is a semi-mechanistic model
34:27which estimates the susceptible population
34:30through time at each time point
34:33as well as the affected population at each time
34:36and incorporates it into a regression type of framework
34:41such as this where we can kind of model out
34:45the transmission rate through time
34:47and make it a function of different climatic variables
34:51and this is an approach that we used along
34:54with colleagues from Princeton to examine the relationships
34:56between humidity, rainfall and cases
35:00of Respiratory syncytial virus or RSV
35:04across different parts of the US and Mexico
35:07under both current and future climates.
35:11And using this approach, we were able to show
35:14that the transmission rates of RSV
35:20which is indicated by the various colors here,
35:23tended to depend both on the level of humidity
35:28within the population where it tended
35:30to be higher transmission happening
35:33at lower specific humidity as well as
35:36on the level of precipitation within the population
35:39where particularly at kind of middle
35:41to higher specific humidity,
35:43precipitation played a larger role
35:45in modulating transmission of RSV
35:50and this really helped to explain
35:52some of the very different patterns that we see
35:54in sort of the seasonality of RSV
35:56across different parts of the US
35:58where we see this sort of biennial every other year pattern
36:01of large followed by small epidemics often happening
36:05in Upper Midwestern states such as Minnesota,
36:08kind of regular annual seasonal outbreaks happening
36:11in the winter in states
36:12such as New York and Connecticut, earlier epidemics
36:16with kind of more year-round transmissioning happening
36:18in Florida and these sort of biannual
36:20two peaks a year happening in parts of Mexico.
36:24And by linking this kind of specific relationship
36:27between the transmission rates to these climatic factors,
36:31we're able to make projections
36:33about the impacts of climate on future disease incidents
36:37which is shown on the plots over here
36:41on the map on the right in which we predict
36:45that overall transmission rates of RSV will be lower
36:52in the future in the Upper Midwest
36:54and Northeastern United States,
36:56but potentially higher seasonal differences
36:58in transmission in the west as well as the south,
37:02although there's a lot of uncertainty
37:03in some of these model predictions,
37:05partly related to uncertainty
37:08in rainfall predictions going forward.
37:11And we've also used this approach to look
37:13at the relationship between rainfall
37:18and typhoid fever using historical data from the US
37:23where we had data from 19 cities
37:25across different parts of the US
37:27and found this really kind of interesting differences
37:30in seasonal patterns between cities where
37:32for example in New York,
37:34we saw very strongly seasonal epidemics peaking
37:37of typhoid fever before,
37:39this is data from like the late 1880s, early 1900s
37:44where you saw these typhoid fever epidemics peaking
37:47every summer, early fall
37:50whereas in a city like Philadelphia
37:52which was right next door,
37:54there's very little kind of seasonal variation in climate
37:57and one of the things that we were able
37:59to identify oh sorry, seasonal variation
38:04in the typhoid transmission rate,
38:05and by teasing apart these variations
38:07in the transmission rate,
38:08one of the things that we identified was that the amount
38:13of seasonal variation in the transmission rate
38:15really tended to vary depending on the primary water source
38:19for the city where cities that relied on reservoirs,
38:25often reservoirs that were outside of the city
38:28such as the New York reservoir which is located
38:31in upstate, upstate New York
38:34as well as outside of Boston and in Baltimore,
38:38tended to exhibit these kind of stronger
38:41overall seasonal variations summarized in the plot
38:44on the bottom right here,
38:46compared to cities that relied on data
38:49or a water from nearby rivers or rivers that ran
38:54through the city such as in Philadelphia
38:57or cities that drove their water from the great lakes
39:00which actually had the lowest seasonal variation
39:02in transmission rates and so taking into part,
39:05into account kind of other factors
39:07such as water sources is really important
39:09in understanding some of these relationships.
39:12And overall, this relationship between
39:14kind of temperature and you know,
39:17why transmission rates tend to peak
39:19in the summer months was consistent with results
39:23of a systematic review that we conducted
39:25looking at associations between climate
39:27and typhoid fever incidents that generally showed
39:30that temperature on the bottom here was a stronger correlate
39:35of typhoid fever incidence at lags of zero
39:38to two months across different latitudes
39:41and studies conducted across different latitudes
39:44compared to rainfall which is really
39:46kind of only associated often in studies conducted
39:49in the monsoon belts where,
39:52and you often also saw potentially negative associations
39:55between rainfall and typhoid fever incidents
39:58in places such as the Middle East.
40:02And then finally, one of the last ways
40:05in which we can estimate the climate disease relationship is
40:09to incorporate climate models
40:12into fully mechanistic dynamic models
40:17which explicitly account for the susceptible, infected
40:21and recovered populations through time.
40:25And so for example the way this works is
40:26to assume in your population
40:29that all individuals are born susceptible
40:33to a particular disease and that they become infected
40:37at a rate which we're gonna call lambda
40:40and remain infectious for a period of time
40:46after which they recover and have some immunity
40:49to future infections of the disease
40:52and the important part about these models is
40:54that this lambda parameter or the rate
40:57from going to susceptible to infected depends
41:00on the current prevalence of infectious individuals
41:05within your population through time,
41:08such that the lambda at time t is gonna be a function
41:11of our transmission rate at time t,
41:13times the number of currently susceptible individuals
41:16and times the number of currently infectious individuals
41:19within our population.
41:21And so our incidence of new cases is dependent
41:24not just on the transmission rate or climatic variables
41:29which may affect the transmission rate,
41:31but also on the current prevalence
41:32of the infection within the population.
41:35And then within these models,
41:37we can decompose this transmission rate
41:39or this beta parameter at time t
41:42to be a function of various other factors
41:46and often the way we model it is as a function
41:49of sort of a baseline transmission rate
41:51plus some seasonal variation
41:54which we may not understand kind of all the factors leading
41:57into the seasonal variation, but using a harmonic term
42:00and then can incorporate our various climatic predictors
42:03as coefficients in this equation for our beta t parameter.
42:09And this is something that we've done to look at the impacts
42:13of climate on rotavirus diarrhea in particular in Bangladesh
42:19where we're using this slightly more complicated model
42:21specific to our understanding of immunity
42:25and natural history of rotavirus infections
42:28which is depicted on the left here
42:30and modeling our incidence rate at time t
42:33as a function, not only of sort of the baseline incidence
42:36and these harmonic terms accounting for
42:39kind of annual and bi-annual potential differences
42:43or changes in transmission rate,
42:44but also climatic terms
42:46including the diurnal temperature variation
42:49which is plotted in the middle here
42:53showing kind of a larger diurnal temperature variation
42:57happening in the kind of winter months
43:01or early parts of the year
43:02and less diurnal temperature variation
43:05in the middle of the year,
43:08as well as the wetness presence
43:11which tends to be more higher
43:15in the middle parts of the year
43:17coinciding with the monsoon season
43:19and you can see the rotavirus incidence
43:22in this particular setting, if you kind of average it
43:25over time shows these sort of bi-annual peaks
43:27where you have a larger peak happening
43:29kind of in the cooler, dry season
43:31and then a smaller secondary peak happening
43:34in the wet season over time
43:36and we can use kind of this relationship
43:38to try and tease apart some of those relationships
43:41and how they're associated with climate factors
43:45as well as other factors within the population
43:48and you can see that we have plotted here
43:52the incidence of rotavirus diarrhea from the 1990s
43:55to early 2000s as well as incidents
43:58over kind of a later time period from 2003 to 2013
44:03where if we fit models to the early time period
44:06and use it to predict the later time period
44:09which is shown in blue here,
44:12we can generally kind of capture some of these patterns
44:15in which we observe a stronger kind of bi-annual pattern
44:19or two peaks a year happening across the 1990s
44:22and early 2000s, but a much more kind of annual pattern
44:26that emerges in the later 2000s,
44:31in 2000 through 2013 where you're starting to see
44:35much greater predominance of this annual
44:39kind of winter dry season peak and it doesn't seem
44:42that this is necessarily related to variation
44:44in climate over time, but really is more driven
44:48by actually a decline in the birth rate
44:50within Bangladesh and particularly within Dhaka
44:54which is sort of interacting
44:57with the different climatic factors
45:00to change the sort of the predominant way
45:05in which the climate influences transmission of rotavirus
45:08within the setting, where kind of even modeling
45:11kind of same relationships between climate
45:14and rotavirus transmission over time,
45:17we can capture this shift from kind of more biannual peaks
45:22to greater predominance of annual peaks over time
45:31and so finally, I just wanna talk a little bit
45:33about how we can kind of pull everything together
45:37and how we can make these predictions
45:40around the impacts of climate change on infectious diseases.
45:44And again, the way that we do this is
45:46to really combine our climate model projections
45:49with a good understanding
45:51of the incidence of disease given climate,
45:56but then there's still a number of big challenges
45:59in doing this.
46:00Often the climate models have poor resolution
46:03and wide uncertainty which needs to be propagated
46:06throughout the relationships of predictions going forward
46:13and infectious diseases may often vary
46:15based not only on mean climate,
46:18but can also show important variability
46:20on shorter spatial and temporal scales
46:24such that things like diurnal temperature variation
46:26or changes in climate kind of through the day
46:29can have important impacts often on climate
46:32or on infectious disease transmission
46:35and then finally, the associations with climate are often,
46:39I would say non-linear, when it comes to infectious diseases
46:44and climate models are often bad at predicting
46:47some of the extremes and extremes in temperature
46:50and rainfall for example which have been shown
46:54to be kind of important drivers
46:57of a transmission of infectious diseases
47:01and then finally, another important caveat is
47:03that the impacts of climate change may be quite small
47:06relative to the impacts of human interventions
47:09and demographic changes as we saw
47:11with the rotavirus example,
47:13but as we also see with models of projections
47:17around malaria risk over time
47:20where if you were to just simply look back
47:22at the malaria risk map from the 1900s
47:26and compare it to the malaria risk map from 2007,
47:30you'll see that the overall regions
47:34in which you see malaria has contracted significantly
47:38where we no longer see malaria happening
47:40in parts of the US for example,
47:42but really being more confined to parts of Africa
47:46and Asia and other places.
47:49But at the same time, the climate has actually been warming
47:55and this would suggest that you would see
47:58more favorable conditions for malaria climate
48:00if you'd only take into account climate over time
48:05and so while what's actually been observed is
48:07mostly been driven by human interventions
48:10and changes in development and exposure to mosquitoes.
48:13If you don't take into account those changes,
48:15you would totally misunderstand
48:19or misrepresent the climate associations
48:20that we know are there.
48:23And so in terms of the way forward,
48:25what really needs to be done is to take on
48:29some of these climate disease relationships
48:31in which we have experimental systems set up
48:35and we have a better understanding of the relationship
48:38between climate and how it impacts on infectious diseases.
48:42Furthermore, some of the climate change productions are
48:44gonna be more reliable and so those infectious diseases
48:47that rely on or have been shown to vary
48:50based on climate variables which are more predictable,
48:53I think, are gonna be kind of more amenable
48:55to making predictions around the impacts of climate change.
49:01But overall, there's really I think an important need
49:04for interdisciplinary work between climate scientists,
49:07lab scientists and microbiologists who can help
49:12to test some of these mechanisms
49:14and infectious disease modelers
49:16who can quantify the relationships
49:18between climate and infectious diseases to really,
49:23I think, move forward some of the field
49:27when it comes to trying to make impacts
49:30or predictions around impacts of climate change
49:33on infectious diseases,
49:35and in doing so we really need to take into account factors
49:39such as human adaptation and the impacts
49:41that climate may have on human behavior
49:44and population distribution
49:46since these are often kind of greater drivers
49:50of infectious disease incidents
49:52than factors affecting pathogen survival for example.
49:57And so really there's this need to move beyond
49:59just the simple climate disease correlations
50:03that other I think previous attempts
50:06to predict the impacts of climate change have relied upon.
50:11And so finally I just want to quickly acknowledge
50:15some of the people who I've worked with
50:16on these various projects including collaborators
50:20and lab members here at Yale
50:23as well as collaborators elsewhere
50:24and funding from the
50:27Yale Climate Change and Health Initiative
50:28as well as NIH, the Gates Foundation
50:30and collaborators funding from Welcome Trust
50:33and James McDonald.
50:34So I'd be happy to take any questions.
50:41- Thank you.
50:42Very wonderful presentation.
50:45I think it covers all the aspects
50:47when we talk about conscientious infectious disease
50:50from modeling the climate disease relationship
50:53to how to better project the future impacts.
50:58So we do have a lot of questions from the students.
51:02So because we only have very limited time,
51:05so I will summarize two questions from the students
51:10and if we have more time,
51:12then maybe our audience can speak for their questions.
51:16- Great. - The first question is
51:18kind of follow up your later part talking
51:21about the inferences of the non-climatic drivers.
51:26You're showing that actually human intervention
51:30can have much larger impacts.
51:33So students are wondering how do you consider this
51:39in projecting the future climate change impacts?
51:43- Yeah I mean I think that that's often the difficulty
51:45when it comes to making predictions
51:48about the future is understanding how you know,
51:52you can make projections kind of assuming
51:55all other things remain the same
51:57and climate's the only thing that's changing,
51:59but the reality is that climate's never gonna be
52:00the only thing that changes over time
52:03and so you have to have either some other model
52:06for how human behavior may change over time
52:13or human development or things like that may change
52:16over time and that may just be sort of trying
52:18to make simple extrapolations or maybe
52:22based on sort of more sophisticated
52:26kind of sociological or sociopolitical models
52:29of say, development or things that other factors
52:32that may affect like interact the risk of disease.
52:38So for example when it comes to malaria,
52:41obviously sort of some of the developmental factors
52:44and industrialization that happened
52:49that led to kind of clearing of mosquito breeding sites
52:54and things like that played a huge role
52:56in kind of why we no longer see malaria
52:58in parts of the world where it was previously,
53:03but yeah, I mean I think you need a separate model
53:07to really account for
53:09how we see those things changing over time
53:12separate from or potentially related to climate.
53:18- Wonderful, so while Gina is answering questions,
53:24from the audience, if you do have any questions,
53:26you can type your questions in the chat box
53:29or if you are willing to speak,
53:32you can raise your hand
53:36and then we can ask you the questions.
53:38So I have actually a couple questions from the students
53:43given we are under you know the COVID 19 pandemic.
53:47We are very interested in like
53:49what's your answer to the climate inference
53:52on the transmission of COVID 19
53:55and what are the potential challenges
53:57in using the approaches that you are talking about today
54:03to study the relationship between COVID 19
54:06and all the climate drivers.
54:09- Yeah, I mean I think that that's definitely something
54:10that some people have tried
54:11to kind of tease apart using data
54:14I think mostly from kind of different locations
54:16and trying to understand kind of how
54:20perhaps how quickly the epidemic has taken off
54:24in different locations could potentially be explained
54:27by some of the differences in climate possibly
54:32and so I mean I think that that is potentially one approach
54:35to take, but really doesn't factor in
54:39all of the other things that may potentially affect
54:48whether it's climate that's driving
54:50how quickly the epidemic takes off across different places
54:53or whether it's other factors,
54:55for example just kind of the extent of social distancing,
55:00the extent of other interventions
55:01and how all of those things have played a role
55:04or just chance in terms of when the virus was introduced
55:07in different places in determining
55:08kind of how quickly the epidemic has occurred
55:10in different locations.
55:14Another approach that has been taken is
55:15to look at our understanding of other Coronaviruses
55:20within the human population
55:23and there are kind of
55:27at least two other human Coronavirus species
55:31that cause cold like illness every year
55:35that circulate within the US
55:37and we know that those other Coronaviruses
55:41tend to peak in the fall, early winter time period
55:46and likely the reasons behind why they peak in the fall
55:51and winter time period is really related
55:53to climate conditions favoring transmission,
55:57be it from what I talked about earlier
55:59in terms of host defenses
56:01and host defenses being slightly weakened at that time
56:03or potentially direct relationships
56:06with virus survival or potentially you know,
56:08seasonal differences in behavior such as aggregation of kids
56:11in schools in the fall period,
56:16but I think trying to bring that to bear
56:20and directly predicting the incidence
56:24of the SARS-CoV-2 virus
56:25at this time is gonna be very difficult
56:28because I think the biggest factor
56:30kind of underlying transmission right now is differences,
56:35I mean we have a virus in which everybody is susceptible
56:38and so it's gonna be able to spread efficiently
56:40kind of regardless of climate conditions
56:43across different settings and so I think
56:47that climate is gonna play kind of less of a role now
56:50in terms of determining when these seasonal peaks happen
56:54compared to just all the other factors
56:58in terms of social distancing and other interventions
57:01and when some of these things are relaxed,
57:05when people become complacent
57:07and stop you know taking all the precautions
57:09that they've been taking
57:10during the summer months for example,
57:13more so than the role that climate is gonna play right now.
57:17So I think it's really a situation we're gonna have
57:20to wait and see kind of what are the major climate drivers
57:23of the SARS-CoV-2 virus
57:25and how much is it really modulating transmission.
57:29- Thanks, that's very insightful.
57:32So I think we have some time from the audience
57:37to ask a questions.
57:38So there's one question was wondering,
57:43if the terms adjusted for the annual
57:45and the bi-annual pattern only account for seasonality,
57:50how about the long-term change?
57:54And a further question is
57:54if one disease shows a bioannual pattern,
57:57why still use the annual term in the model?
58:03- Okay, so that's a, I think very--
58:05- Very technical.
58:06- Yeah, technical question,
58:07it's a you know, difficult question to answer
58:11kind of directly related to,
58:13I'm not sure kind of which disease this is in relation to.
58:16If it's specifically around kind of rotavirus in Bangladesh,
58:24certainly I think that there is potentially
58:28also long-term trends happening
58:30in transmission rates over time
58:31and often we do need to account for these potential
58:35like linear or long-term trends happening
58:38in baseline incidents over time
58:40which may be important as well
58:41and it's often something that we do
58:43kind of explore incorporating into the models
58:47and is potentially able to explain
58:50some of these you know, unusual shifts
58:52that we might see for example from the biannual
58:55to more annual epidemics happening in Bangladesh
58:58in conjunction with the sort of the decrease in birth rate,
59:01we're probably also seeing a decrease
59:03in potentially transmission rates over time in that setting.
59:08But in the question around
59:10kind of why do you incorporate both biannual
59:13and annual terms in a model.
59:16The reason for that is often
59:18because if you're only incorporating a
59:21sort of biannual harmonic term
59:22that assumes inherently
59:24that the size of the two peaks is the same
59:29whereas when you add an annual harmonic,
59:31it allows for two peaks of varying size happening
59:36throughout the year and so it can sort of basically lead
59:40to a larger peak and a smaller peak throughout the year,
59:42whereas if you only have a biannual term,
59:44you can only have those two peaks
59:46by definition have to be the same size.
59:49- Great, thank you.
59:50So I think we have reached the end of this seminar
59:56and thank you Gina for this wonderful presentation
60:00and thank you all for coming.
01:00:02Our recording will be available later
01:00:06and we will have our next seminar in November.
01:00:11So looking forward to see you soon, bye.
01:00:15- Great, thank you.