Many results of this study align with hypothesized effects and relationships based on biological plausibility, and evidence from similar research on respiratory illness [38, 51], asthma [37], enteritis [18], and direct heat-related illnesses [17, 18, 25, 28, 34, 51]. In the primary analysis, EHEs were found to increase the rate of hospital admissions due to general respiratory illnesses and asthma; general infectious and parasitic diseases, lower respiratory infections, and enteritis. Similarly, rates of ED visits due to asthma; heat-related illnesses, heatstroke, dehydration, and lower respiratory infections were positively associated with EHEs.
Interestingly, EHEs were found to reduce rates of hospital admissions and ED visits for general injuries and transportation related injuries, and ED visits due to falls. This novel finding contrasts with those of similar studies in settings including New York City [52, 53] and England [54] in which injuries amongst children were found to increase during EHEs. This may reflect the efficacy of heat warnings in mitigating effects of EHEs on child health through activity modifications like abstaining from sports and physical activities, for example. The same theory may help explain why sex stratification showed that during EHEs males reduced rates of injuries, falls, and transportation-related injuries as well as hospitalizations due to renal disease. This protective effect contrasts with results of a study in New York City which found the highest risk of unintentional injury among males ages 5 to 9 years old [52]. When stratified by age, the 0–4 and 13–18 year age groups exhibited the most positive associations. Previous studies have also found that, when compared to older age groups, children under 5 have higher rates of emergency department visits [53], asthma [37], respiratory illness [38, 51], and renal diseases [51]. This may be due to heightened caution for the 0–4 age group by caregivers and healthcare providers increasing the likelihood of an ED visit patient being admitted, and due to the 13–18 age group being more physically active and independent which can heighten susceptibility to EHEs.
Adjusting for relative humidity did not alter primary results. Associations, either positive or negative, were often strongest in the primary analysis and approached the null in sensitivity analyses with lower threshold temperatures and lag days from exposure. This monotonically decreasing trend effect demonstrates an exposure-response relationship between heat and rates of pediatric emergency healthcare. However, although RR of most outcomes were generally highest on same-day exposure to EHEs, hospital admissions and ED visits due to enteritis were highest the day after an EHE (lag 1), reflecting typical incubation periods of common foodborne and waterborne pathogens [55] including nontyphoidal Salmonella spp. (12–96 h) [56], norovirus (12–48 h) [57], and Campylobacter spp. (2–4 days) [58]. Similarly, RR of renal and infectious and parasitic disease hospital admissions were highest 2 days after an EHE (lag 2), possibly reflecting the cumulative effects of days of dehydration on renal and immune system function.
EHEs do not have a universal definition. Rather, they vary in threshold, temperature, and length making relative each of the terms “extreme”, “heat” and “event”. This consequently complicates comparisons. When aptly compared to the results of primary or sensitivity analyses herein, similar literature has made similar findings. Pediatric hospital admissions due to respiratory illnesses [25, 38], asthma [37], and heat [25] were positively associated with EHEs in Australia, the US, and China. It is also important to note that studies used varied age categories for children [15, 18], including several which used coarse age categories of 0–14 and 15–64 years [51], or 0–4 and 5–65 [17].
In two Ontario hospitals, Wilk et al. found pediatric infectious and respiratory diseases ED visits were positively associated with EHEs. This is the only prior study of pediatric ED visits in Canada, and it included only these two outcome measures. Respiratory illnesses [51], asthma [34], heat-related illnesses [26, 29, 44, 45], dehydration [17, 34, 51, 59], renal diseases [25], otitis [18], and bacterial enteritis [18, 34] have been positively associated with EHEs in studies outside of Canada. Of these outcomes, our study similarly found that EHEs increased rates of ED visits for asthma, heat-related illnesses, and dehydration in primary analysis, as well as with renal diseases and otitis in 1-day lag, and renal diseases under the 97.5th and 95th percentile definitions.
To date, this study was the largest and most comprehensive examination globally of EHEs and pediatric healthcare use, and the first study in Canada to assess associations of EHEs and pediatric hospital admissions. The data spanned the warm months of 10 years, including 284,939 hospital admissions, and 5,875,119 ED visits. The large scale of this analysis increases the power and representativeness of the study, especially for general, broadly defined causes of admissions and ED visits such as respiratory disease and injury. Given that Ontario provides universal free healthcare to residents and reports all admissions and ED visits to NACRS and DAD, the data used in this study is comprehensive and includes all pediatric cases. The space-time stratified case crossover design applied in this analysis controls for characteristics that do not vary over time or vary only slowly (e.g., age, race, sex) and for time-trend and seasonal patterns. Conditional quasi-Poisson regression, common in similar studies [15, 60,61,62,63,64,65,66], was applied instead of the traditionally used conditional logistic regression because it is simpler and faster to code, and allows for adjusting for overdispersion and auto-correlation [40]. Poisson distribution models rate counts given exposure, whereas conditional logistic regression models odds of a binary outcome given a one unit increase in exposure. To create a binary outcome to use conditional logistic regression, the data would need to be expanded such that each stratum contains only one case, or semi-expanded such that each stratum contains only one case day and was weighted by the number of cases that day. This study also benefits from the application of Daymet daily estimates of temperature aggregated to residential FSA, which are a more accurate measurement of exposure than the alternative of regional weather station data.
Despite the above strengths, since effects were measured over the whole study period conclusions cannot be drawn concerning possible changes in these effects over time. It is possible that the effects may differ year to year or between early and late season. Further, to maximize power this study focused on two-day EHEs and was not able to differentiate and assess impacts of rarer more extended EHEs. This limitation leaves a gap in research on the specific impacts of extended EHEs on child health. Similarly, extended hospitalizations may give rise to another limitation due to carryover bias. Specifically, hospitalizations that began during the first week of a month and lasted the entire month would have invalid control days (same day of the week in subsequent weeks) since the child would still be hospitalized. Although this impact cannot be ascertained without access to length of stay data, the rarity of this occurring likely minimizes the potential impact on the overall results. Another limitation is having only examined same day (0-day lag) and 1- and 2-day lagged associations of exposure to 2 consecutive extreme heat days. Though equally short lag periods are common in studies of pediatric morbidity and extreme heat [17, 19, 21,22,23, 25], some studies have found associations at up to 7-day lag periods [18, 21, 67]. Given that the largest effects were observed at lag 0, declines in effects were observed at lags 1 and 2, and our evaluation of multiple outcomes and exposure measures, we focused on a limited lag period. Further, the time stratified case crossover design of this study uses 3–4 controls days on the same day of week in the same month for each case day. Thus, a 7-day or longer lag period would cause the case and a control day to overlap, potentially biasing results towards the null. Additionally, the ecological design of this study limits generalizability. Individual level factors, like SES and comorbidities, may modify the effect of EHEs on pediatric emergency healthcare use but were not included in this study. Built environments, housing and neighborhood characteristics may affect susceptibility to EHEs within FSAs [68]. Moreover, children who live in remote areas, areas where ED wait times are long, or who have low SES may not be well represented in this study. However, this is unlikely to have systematically biased results since the distribution of urban and rural FSAs of study participants was comparable to the distribution of urban and rural FSAs of children in Ontario, according to the 2011 census [69]. In fact, children with rural FSAs made up a slightly higher proportion (7%) of the distribution within this study than was observed in the 2011 census.
ECCC’s heat alert protocols in Canada identify conditions in which increased all-cause mortality of the general population are likely to occur [2]. In Ontario, regional alert criteria are 2 or more consecutive days with daytime maximum temperatures above 29 °C in Northern Ontario, or 31 °C in South and Southwestern Ontario [4]. These temperatures are most similar to those of the 95th percentile of temperature (see Additional File 1 Table A3). The results of this study may suggest that ECCC’s heat warnings are effective in preventing emergency healthcare use for pediatric injuries as evidenced by the consistently protective effect of EHEs on injuries observed in all analyses, and a contrasting harmful association in adults who are largely unable to change their activities (ex. outdoor labour, commuting to work) [32, 33]. Despite these regional alerts, in analyses using the 95th percentile, EHEs increased rates of pediatric ED visits for heat-related illnesses, heatstroke, dehydration, renal diseases, and otitis; as well as hospital admissions for renal diseases, lower respiratory infections, and bacterial enteritis. It is also important to consider that with increasing intensity and frequency of EHEs due to climate change, today’s 99th percentile temperatures may be the 95th percentile in the near future [2]. Given the suggested efficacy of heat warnings in preventing pediatric injuries, these outcomes may be preventable with more stringent child-specific heat warnings at lower temperatures along with messaging specifying the increased risk of these health impacts.