[Page under construction]

Taking the temperature of our cities: A story from São Paulo, Brazil

The work discussed below is based on our work published in the Journal of Urban Health. Refer to the paper for more information or contact the corresponding author.

Cite as: Roca-Barceló, A., Fecht, D., Pirani, M. et al. Trends in Temperature-associated Mortality in São Paulo (Brazil) between 2000 and 2018: an Example of Disparities in Adaptation to Cold and Heat. J Urban Health (2022). [https://doi.org/10.1007/s11524-022-00695-7]


Why do we care about temperature?

A bit of background

Humans are a homeothermic species. This means that for the correct functioning of our bod, we need our body temperature to be within a certain temperature range regardless of the air temperature. For humans, this temperature range is between 36⁰C and 38⁰C (98.6F). If this temperature is not met, the function of our bodies is compromised.

The question is, how do we do that? The answer is that evolution has provided us with thermoregulation. It is a physiologic mechanism which allows our body to internally modulate its temperature. It comprises several processes that are activated when our body temperature diverts too much from the optimal temperature of 37⁰C. Sweating, for example, is one such process. When our temperature increases, our body starts sweating, a form of evapotranspiration, which allows us to expel heat from the body. When our body temperature drops too much, on the other hand, we start shivering, which produces localized heat, slightly raising our body temperature. But as you can imagine, and have probably experienced several times, these processes are not enough when the deviations from the optimal temperature are too large. In those situations, external factors start playing a role. According to Hondulas and colleagues, these external or non-physiological elements can be classified into behavioural, technological or infrastructural factors, depending on whether the adaption occurs thanks to changes in the behaviour patterns, technological advances, or infrastructure, respectively. In most cases, it is a combination of all the factors.

What do we mean by adaptation?

The examples we discussed above, depict acute responses to air temperatures. But what about long term shifts in temperature? Say you grew up in the Bahamas and you are moving to Helsinki, Finland. Will your body react in the same way as that of someone who has lived in Helsinki all their live? Probably your gut feeling tells you that the answer is no. And you are right. We know from several studies that people develop mechanisms that are tuned to the usual temperatures they are exposed to. Which makes total sense, right? If I live in a cold place, my metabolism, behaviours, and the infrastructure I have access to, will be tuned to the needs of that specific place. I will not, for example, walk around in a swimsuit (behaviours), nor have the air conditioning on (infrastructure) and my body will for sure, not be sweating (physiological). But this is not only the case when we move country. The same happens if it is not us who moves but the environment around us that changes? This is exactly what climate change is. Temperature and weather patterns have changed and will continue to change due to climate change. Not only that but our societies are constantly changing and evolving; new technologies are developed, awareness about the risks and prevention options increases, access to infrastructure and facilities changes, and so on. As we have seen above, all this shapes our response to temperature. This is what we call adaptation.

Adaptation. The process of adjustment to actual or expected 
climate and its effects. In human systems, adaptation seeks
o moderate or avoid harm or exploit beneficial opportunities.
In some natural systems, human intervention may facilitate 
adjustment to expected climate and its effects. 

What happens when these mechanisms fail?

When exposed to temperatures outside of our range of comfort, our bodies are exposed to some stress which can lead to a wide range of health consequences if not well managed. The images below summarize some of the main health consequences of both extreme heat and cold and their main clinical manifestations. These effects are more acute and can happen earlier in people who have a lower capacity of adaptation. This can include people with a compromised thermoregulation mechanism such as the eldest, people with diabetes or babies; people with no access to the right infrastructure or technological adaptations such as people at risk of poverty, outdoor physical workers, etc. So, it is important to pay closer attention to such groups.

Figure. Schematic diagram of the main medical manisfestations of extreme heat and cold. Adapted from Centers for Disease Control and Prevention (CDC).

Why is this important?

If we want to empower the population of act in situations of hazardous temperatures and give stakeholders and politicians the tools to appropriately protect their populations, we need to know what temperatures dangerous and which population groups are at higher risk. As we have seen above, temperatures change and the underlying factors for adaptation also change (e.g., behaviours, infrastructure, technology, and physiology). Hence, we also need to study these temporal changes to have the full picture. This information is essential to feed in public health policies and interventions such as Early Warning Systems. Early Warning Systems are comprehensive systems that combine weather forecast information with epidemiological evidence to define when to act, who to direct the actions primarily and what actions to take. Basically, when temperatures reach values above said value, the alarm system is sounded, and it activates an action plan. The action plan is specific for each setting and includes things like recommendations or to open public indoor spaces for people to take refuge in the event of the extreme temperatures. Multiple thresholds of temperature may exist to capture different levels of risk, associated to action plans of increasing magnitude and urgency. As you can imagine, the way these interventions are defined and the temperature at which they are activated is specific to each setting and population. However, regardless of the exact intervention there are two pieces of information that are essential.

At what temperature do we activate these interventions?

Who are they addressed to primarily?

To answer these two questions, we need to model the link between temperature and mortality risk for the setting of interest. Not only that but we need to make sure that what the evidence we use represents the most the reality of our population. Epidemiological studies investigating how these association changes over time and by population group is vital. And this is exactly what we did.


Our study

The aim: What did we want to know?

We collected data on all deaths and the average daily temperature between 2000 and 2018 in the municipality of São Paulo and. We then calculated the risk of death linked to temperature for each year and measured whether there was any clear trend over time. We did this for the whole population and focusing on different vulnerable populations, namely by age, gender, and ethnicity.

The study area: São Paulo

Located in the south-east of Brazil, near the Atlantic coast (see the image below), São Paulo is one of the seven megacities in the world, with over 11 million inhabitants. Its population is expected to raise in the next years and so is temperature. Therefore, providing up to date information to inform public health intervention has the potential of saving a lot of lives.

Figure. Map of the area under study.

The data: temperature and mortality

For this study we collected information on temperature from a reference meteorological station located in the city centre (marked as a white dot in the image below). From that station we also collected information on relative humidity as it is a factor involved in how temperature impacts health and so, we wanted to make sure we included it in our model so that we could isolate the effects of temperature. We did the same with air pollution, which we obtained from the air pollution ground stations network of the city managed by CETESB. To evaluate the impacts on the risk of death, we needed information on all deaths occurred between 2000 and 2018. We were interested only in accidental deaths. For each day, we calculated the number of deaths occurred across different population groups. The groups we were interested in were females, males, aged between 65 and 79 years old, above 80 years old, people of colour, white population, and combination of those groups. These groups were selected because there is suggestive evidence that they are more vulnerable to the effects of temperature.

Measures of adaptation: comfort and extreme temperature

Visualizing the link between temperature and the risk of dying. In epidemiology, we can visualize the link between an environmental factor and a health outcome as a 2D linear plot. In the X axis, we have the environmental factor, in our case temperature. In the Y axis, we have the health outcome, in our case the risk of dying.

We can represent the association as a line where each point defines the risk of dying linked to a certain temperature. For the case of temperature and mortality, it is important to know that this line is not straight but instead follows a U or V-shape where the lowest point –i.e., where the risk of dying is the lowest—is referred to as the minimum mortality temperature, MMT. We use the MMT as our reference assigning it a value of 1 and give the risks relative to it. Therefore, a relative risk of 2 means that for that temperature, the risk of dying is double to the lowest risk of death for that population, i.e., the MMT.

In the figure below we see a representation of this. If you think about it, this makes sense. We have just described how our bodies are adapted to the usual weather of the area we live in. Hence, when the temperature moves around that range of comfort the risk of death is small or minimum. As temperatures move away from this comfort zone, the risk of dying goes up because our bodies and tools are not ready to cope. And that is exactly what happens. This translates into a U-shape or V-shape risk curve we saw just above, that is, it is non-linear.

In addition, a key thing to consider here is that temperature does not have an immediate impact on us but instead, we see that the effects can be delayed for several days for heat and up to two weeks for cold. This means that when we look at the mortality rate associated to the temperature registered at a given day, we need to consider several days after as well.

A statistical framework that enables us to model both delayed and non-linear effects is the distributed lag non-linear model (DLNM) framework developed by Gasparrini et al. This is what we used in our study.

Annual specific adaptation indicators. This framework not only allows us to model these non-linear and delayed associations, but also whether they are changing over time by adding an interaction term in the formula between temperature and time. In this way, we can model the risk associated to each year and explore if it has changed or not.

To monitor changes, we need to pick some indicators for adaptation. In this study, we have selected changes in the MMT and changes in the risk associated to a given extreme temperature. These indicators should capture changes in the comfort temperature and changes in the vulnerability to extreme temperatures. Let’s try to have a look at how through two simple scenarios. The figure below shows two time periods, an early one (yellow) and a later one (purple). On the left hand-side, we see that the risk (Y axis) associated to a given temperature (X axis) has decreased in the later period. This would suggest that this population has decreased the risk of death when exposed to extreme heat (right of MMT) and cold (left of MMT), indicating adaptation. On the right-hand side, we see that the MMT has shifted to the right, that is, it has increased, suggesting that this population has now higher comfort temperatures.

Figure. Diagram of the different adaptation indicators used in this study. For two imaginary time points (1, yellow and 2, purple), we represent two scenarios. On the left, the flattening in the cumulative relative risk (cRR) curve. On the right a shift of the minimum mortality temperature (MMT).

Our findings

Note that all visualizations are interactive. You can zoom in and out and select different variables to be plotted by clicking on them.

Describing the data

Meteorological data. Meteorological data. Below you can explore the temperature distribution for each year in the study period in an interactive way. Select the years you want to plot and compare their distribution. You will see that there the distributions seem to flatten in later years shifting from one unique peak to two. This suggests that there are more days with high and low temperatures in later years than early years. However, the changes are not substantial. We can observe the same if we plot it using a slightly different visualization (boxplot).

Figure. Histograms of annual temperature for São Paulo between 2000 and 2018. Interactive plot.

Figure. Boxplot of annual temperature for São Paulo between 2000 and 2018. The red line shows the average temperature for the whole study period. The points at the extreme are days with extremeley different temperatures or outliers. Interactive plot.

Below you can see the daily mean values for daily mean temperature, relative humidity, and air pollution (only large particles, PM10). The last two variables were included in the model to ensure that the relationships that we found were related to temperature only and not affected by the presence of other co-occurring factors.

Figure. Annual daily average of temperature, relative humidity and air pollution (PM10) in São Paulo between 2000 and 2018. Interactive plot.

Mortality data. The plot below shows daily death counts per population subgroup. Things to note are the clear seasonality and the long-term trends specially for some population groups. Both were included in the model to make sure they didn’t interfere our results.

Figure. Death counts per population group under stduy. Interactive plot.

Adaptation patterns

There were 3 main findings. We will explore them in detail below:

Finding #1: The temperature-mortality association changes over time confirming the presence of adaptation and/or maladaptation. The temperature-mortality relationship in São Paulo varied significantly over time, confirming the dynamic nature of this association. To see this, we will be using the plots we described above. In this plot, we show the risk of dying for the entire population by year (a different color for each year). One thing you will notice, is that the relationship between temperature and risk of dying follows indeed a U-shape, as we expected. You can also see that there are multiple dashed lines. These mark the lowest risk of death, and if we follow them to the X axis, they give us the temperature at which they occur, which is known as the minimum mortlaity temperature.

If we compare the shape of the different years, we can see that the risk varies, especially on the extremes. The tendency for all population is for the risk to extreme heat (right part of th plot). The risk for cold remains rather similar (left side of the plot). FInaly, the minimum mortality temperature (dashed vertical lines) increases (shifts to the right). Overall, these findings suggest that in general, if we don’t look at specific population groups, the population seems to be have got used to higher temperatures but responds poorly to extreme heat.

Figure. Risk of dying (y axis) linked to temperature (x axis) for all non-external causes of deaths and coloured by year. Interactive plot with select/deselect variables and zoom features enabled.

Finding #2: There are disparities in the presence and magnitude of adaptation by population groups and adaptation indicator

If we focus on the comfort temperature, we see that it has changed over time, and it has done so differently for each population group. In the figure below, we have plotted the annual MMT by population group. You can select and zoom in to appreciate the differences. As you can notice, for most of the groups, the MMT increased slightly over the study period, indicating that people is slowly getting more comfortable with temperatures a bit higher. However, the speed and magnitude of this change varied substantially across groups. Not only that, but we also see that for some groups, the trend was actually going down, indicating a poorer adaptation to higher temperatures. This is the case for males of colour and aged 65 to 79 years old, population aged 65-79 years old of colour and population aged more than 80 years old and white.

Figure. Minimum mortality temperature for (a–s) all non-external causes of deaths by age, sex, ethnic groups, and combination of those, between 2000 and 2018. Interactive plot*

If we now focus on the temperature extremes the differences are also present. To look at it we have plotted the curves of risk for each year by population group. we can see that the heat-related mortality risk (on the right side of the plot) went up for most groups except for males, meaning that they are better adapted to these heat extremes. The cold-related mortality risk (left of the plot) barely changed for most groups except females who seem to adapt better (lower risk in later times).

Figure. Risk of dying (y axis) linked to extreme heat temperatures (red) and cold (blue) for each year (x axis). Each plots shows a different population group. The shaded part shows our intervals of confidence. Interactive plot with select/deselect variables and zoom features enabled.*

Finding #3: Non-climatic factors have a relevant role in adaptation

The next question we may be interest in is to try to figure out whether the adapation and changes observed in findings 1 and 2 are linked to changes in the temperature the population is exposed to or other factors, such as behavioural, technological or infrastructure. To do so, we plotted the the minimum mortality temperature values for each year which capture the comfort temperature of the population against the annual mean temperature (see plot below). If the temperature that the population is exposed each year was affecting the comfort temperautre we would expect that as the annual mean temperature increases or decreases the minimum mortality temperature does the same. That is, we would expect for these two indicators to be coupled. However, when we look at the plot below, we see that this is not the case. This, togehter with the fact that the changes in risk are different by population group (previous findings) suggest that there are several factors, not necessarily climate-related, whcich are playing a role. Some of these factors could include the use of air conditioning, a change in lifestyle and quality of life, job market shift to larger tertiary sector…

Figure. Plot showing the link between the annual mean temperature (X axis) and the minimum mortality temperature (Y axis). The line shows the trend fitted to the points. Plots provided for each population group under study. Interactive plot with select/deselect variables and zoom features enabled.*


Cite our work

Roca-Barceló, A., Fecht, D., Pirani, M. et al. Trends in Temperature-associated Mortality in São Paulo (Brazil) between 2000 and 2018: an Example of Disparities in Adaptation to Cold and Heat. J Urban Health (2022). https://doi.org/10.1007/s11524-022-00695-7