Through this study, we found and characterised a general trend for a lagged, non-linear association between Tapp and in-hospital CVD mortalities in Puducherry, India. We found that the MMT temperature over the study period for Puducherry was approximately 34oC, with respect to CVDs, corresponding to an optimal temperature range between 33oC and 35oC. The MMT occurred at around the 60th percentile of the Tapp.
The overall temperature-mortality association follows a U-shaped curve as described previously with both cold and hot Tapp contributing to increasing the risk of CVD related mortality attributable to non-optimal temperatures with respect to the MMT [32-34]. While the overall AF for temperature was 20.2%, majority of it was attributable to cold non-optimal temperature (11.2%) and only 9.1% could be attributed to hot non-optimal temperature. This seems to be in line with similar studies from around the world as seen in a study spanning 750 locations across 43 countries found that out of 9.4% of all-cause excess deaths attributable to non-optimal temperatures, 8.5% were cold related while 0.9% were heat related . Thus, it is important to consider cold exposure as an important contributor to mortality, even in inherently hot regions like Puducherry.
We found that cold exposure had a bi-level response with a sharp, immediate increase in mortality risk followed by a brief protective effect that then culminated in a second peak of increased risk in CVD mortality, most likely due to the long-term effects of cold. On the other hand, heat exposure showed a 4-to-5 day lag before contributing to the CVD response. This differs from other studies, which found an immediate effect due to heat and a more lagged cold response [22, 32, 36, 37]. The harvesting effect or mortality displacement, when the most vulnerable people are affected earlier than the healthier members of the population, thereby bringing mortality forward in time, could explain the immediate increase in cold related mortalities followed by the slightly reduced risk till about lag day 7 [38, 39]. Since the average Tapp in Puducherry is 34oC, the population is likely more adapted to temperatures above 30oC. Repeated exposures to temperatures above 30oC, considered as hot in many places around the world, could induce a form of thermal pre-conditioning. This sub-lethal, frequent heat exposure could help to build tolerance and confer protection against further lethal thermal stress brought on by extremely high temperatures . The thermal pre-conditioning effect has been found to set in within hours of exposure and can last up to 5 days, possibly explaining the 5-day lag seen for hot temperatures in Puducherry .
Additionally, there is a greater proportion of ‘cold’ days compared to ‘hot’ days or ‘extremely hot’ days. The Indian Meteorological Department as several definitions of heatwaves. For the coastal regions, a heatwave is declared when the maximum temperature rises above 37oC and is a departure of 4.5oC or more from the normal temperature . There are few days with temperatures this high in Puducherry. In addition, there are relatively few consecutive ‘extremely hot’ days, while there is often a cold spell lasting for several days, especially during the winter months, causing people to suffer more, especially if they are unaccustomed to it. Indoor heating systems are also uncommon in the southern part of India where temperatures rarely drop below 20°C. However, the IMD definition of a cold wave in coastal areas is when the minimum temperature is <15oC or a departure of 4-5oC from minimum temperature, meaning that there has been no cold wave recorded in Puducherry for several years [41, 42]. Thus, our results also highlight the importance of regionally defining cold and heat from a health perspective using the MMT. Puducherry is one of the most urbanized territories in India with 68.3% of the population considered as urban according to the 2011 Census . Typical urban characteristics that modify the temperature effect on health, such as tightly packed spaces and living quarters, population density, air pollution and green spaces, might contribute to the overall relatively high heat AF we found .
One of the advantages of using the case-crossover method was that we were able to identify the differences in the temperature-CVD mortality association between sexes and age simultaneously, which to our knowledge has not been identified in the Indian context before. Our results demonstrate that age and sex together act as effect modifiers. All males were more susceptible to cold compared to heat. Males aged above 60 years were more vulnerable to cold non-optimal temperatures than females in the same age bracket, who were more susceptible to hot non-optimal temperatures and seem to withstand cold better. Meanwhile, females below 60 years were affected by both hot and cold non-optimal temperatures. We postulated several possible explanations for this phenomenon. Overall, age is a common risk factor for CVD mortality with older people, especially women, being more susceptible [13, 44-47]. Most women over the age of 50 years have undergone menopausal transition, which has long been associated with decreased cardio-protection and an increase in the risk of developing CVDs and vulnerability to heat . Sex differences in thermoregulation could also be a factor for these findings. For example, the temperature threshold, above which sweating is induced, is higher in women than in men while their overall sweat output is lesser, resulting in reduced heat tolerance [49, 50]. On the other hand, men have found to have a greater decrease in core body temperatures when exposed to cold compared to women, leading to a higher cold intolerance or sensitivity [51, 52]. A study by Achebak et al. in Spain reported a similar relationship between older females and males being more susceptible to CVD mortality from heat and cold respectively . The context of Puducherry might also play a role. Many of the men could be engaged in manual outdoor labour including agriculture and construction, helping them build tolerance to higher temperature. Traditionally, females, especially older females, are more likely to spend a larger part of the day indoors where the urban island effect, inadequate air conditioning and physiological factors could make them more vulnerable to heat . There is need for further studies to be done on this topic in Puducherry to identify the extent to which age and sex together act as effect modifiers for the association between temperature and CVD mortality.
The findings from our cause-specific analysis compare to a recent study by Schulte et al. in Switzerland which found limited risks of temperature attributable risk of myocardial infarction (part of IHD in our study), with a protective heat effect . They also found the risk of mortality from strokes (CVA in our study) increases with heat. The findings are also similar to the Fu et al. study from India, which found a protective effect for IHD with heat and smaller cold-attributable risk in addition to a U-shaped curve for CVAs as seen in Figure 5 of our study .
Many studies look at the temperature-mortality association, but few look at CVDs in particular. The MMTs for all-cause mortality are derived as a function of disease-specific MMT . In fact, temperature-CVD mortality associations have been found to be U or J-shaped while various patterns including the inverse U or reverse-J shape have been associated with infectious diseases [32, 33, 55, 56].The association between temperature-CVDs also varies by region and latitude, with different regions within a country reporting different relationships [16, 32, 57, 58]. For example, a study from Tianjin found a 1oC increase in temperatures above the MMT of 25.1oC associated with a 2.8% increase in CVD deaths while in Brisbane, Australia, a 1oC rise above the MMT of 24oC lead to an increase in CVD mortality by 3.5% [36, 59]. While most studies have found an increase in CVD events due to heat exposure, a study done across China found that the bigger burden of CVD mortality can be attributed to cold temperatures . Therefore, a one-measure-fits-all approach cannot be used to describe the temperature-CVD mortality or all-cause mortality relationship .
As of 2016, 28.1% of all deaths in India were due to CVDs as compared to 15.2% in 1990 and this burden is projected to increase along with the level of epidemiological transition (ETL) . Puducherry falls in the higher-middle ETL bracket with 53.1% of deaths below 70 and 46.9% of total deaths above 70 years due to CVDs , making it a severe public health issue.
We found few studies that looked at the temperature-mortality association in India; however, none of them were from Puducherry. In addition, only one paper by Fu et al. considered CVDs separately but this focused on the Köppen-Geiger climate regions pan-India and therefore could not represent the microclimatic associations and capture the aforementioned regional trends . The association they found between temperature and all-cause mortality is consistent with the U-shaped association we found considering the same temperature range, with cold-related deaths having a higher AF than heat. Their model predicted a MMT of 30oC for the entire country with a mean temperature range of 0.4oC to ~40oC, which when compared with our model MMT of around 34oC and a smaller temperature range, highlights the microclimatic, socio-cultural and demographic effects on acclimatization and mortality .
While there are relatively fewer “heat wave” days in Puducherry, if the warming trend continues as projected, the temperatures for Puducherry could increase or lead to erratic extreme temperatures. It could lead to either a potential right-shift of the optimal temperatures, if this occurs gradually or a significant increase in the AF for CVD mortality due to hot temperatures if there are more erratic extreme days. For example, a study from Hyderabad, a city with higher mean temperatures than Puducherry, found an increase in all-cause mortality by 16% and 17% for maximum temperatures above 40oC and heat index > 54oC respectively . The pattern of anthropogenic climate change over India is a complex one. Mean temperatures in the South Asian region have been decreasing in the past decades and India has not seen an increase in the maximum temperature trends since the 1970s [35, 62]. From a health perspective, Tapp, which accounts for humidity, is better at measuring the health effects. The increase in humidity in India has led to Tapps increasing in India and thereby the severity and occurrence of heat has increased . In the future, pollution control measures and a slower pace of irrigation expansion will likely counter the present cooling effects being seen and as humidity is projected to increase, the net effect will be a gradual rise in hot temperatures, especially during heat waves . It is difficult to assess whether the rise in temperatures might be accompanied by a decrease in the AF for cold-related mortalities or whether only the severity and frequency of heat waves will increase. The absolute number of CVD mortalities attributable to non-optimal temperatures are likely to increase, however, since more people will be at risk or have CVDs in the future.
A recent multi-country, multi-community study found that most excess deaths occur in eastern/southern Asia, especially in coastal cities, highlighting the difficulties to protect, react and reduce adverse temperature effects in these regions, partially due to the large and dense population . As there are several large cities both within this region and along the extensive coastline of India, it is imperative that further research is done on how temperature affects the health of the local population. There is also a need to develop a tailored temperature-health impact management and adaptation plan to reduce the burden of CVD mortalities due to non-optimal temperatures that accounts for regional demographics. These preliminary estimates can be used as a basis to support further detailed research on this topic in Puducherry or elsewhere.
Our study has several strengths. First, it demonstrates how both relatively cold and hot temperatures affect CVD mortality in the tropical region of Puducherry. The high quality of patient level data allowed for examining the effects of age-and-sex grouped together, which has not been explored in the Indian context. It highlights the added vulnerability of older women to extreme heat. Second, the case-crossover approach adjusted for stable within subject and residual individual confounders, particularly from variables that may not have been recorded, by design and allowed us to preserve individual characteristics. We could thus conduct individual-level and inter-individual analysis through subgrouping. Third, ss this is the first study of its kind in this region, we were able to show how regional and demographic variations play an important role in determining the fraction of CVD mortalities attributable to non-optimal temperatures over a relatively long time period. Additionally, the small size of Puducherry coupled with a single multi-speciality state government hospital and robust health system means that we were able to capture the general trends from the main state government hospital which caters to majority of the population within Puducherry. Finally, we were able to demonstrate that cold temperatures have a larger AF to CVD mortalities compared to heat, consistent with other Indian studies as shown. Overall, our study is comparable to global studies from different climate zones and areas, implying a greater contribution of population, genetics and acclimatization to the temperature-CVD mortality relationship. The results from our sensitivity analyses using only 5 year data from the whole hospital, or changing the knot placement were all insensitive to the changes in the model, supporting the robustness of our findings about the association in Puducherry.
The study has several limitations that we offer for consideration. The small sample size that we managed to obtain greatly reduces the certainty with which we can draw inferences, particularly regarding subgroups. Since the data comes from a state-run government hospital, we cannot account for patients of a higher socio-economic stratum or those who might have chosen to seek treatment in a private hospital or travelled to neighbouring states. We also did not include air pollution in our study, although temperature has been shown to have a relationship that is independent from the effects of air pollution. This study assumes that the effects of temperature on CVD mortality are through an acute exposure (the effect on CVD is assumed to only happen over the 21 day-lag that was modelled) as opposed to the chronic nature of temperature exposure. Finally, as there is no way to separate the temperature effects on the CV system from the medical interventions or hospital conditions that might counter the actual effects and work to prolong life, used the outcome of ‘fatal admission’ as opposed to simply mortality, as is commonly used. This allowed us to assume that exposure lasted only until hospital admission, following which treatments and climate control in the hospital could be expected to modify the exposure-mortality association, especially since patients who were admitted for more than 48 hours spent an average of 7 days in hospital before dying. We have performed sensitivity analysis including only those who spent a maximum of 10 days in hospital (Supplement Figure 7). While we tried to assess the cause-specific risk of morality, there were some limitations, for example as most of the patients presented with multiple CVDs, the overall risk from individual CVDs cannot be confidently assessed and there is likely a mixed effect.