Abstract
Background Environmentally-mediated protozoan diseases like cryptosporidiosis and giardiasis are likely to be highly impacted by extreme weather, as climate-related conditions like temperature and precipitation have been linked to their survival, distribution, and overall transmission success.
Objectives Our aim was to investigate the relationship between extreme temperature and precipitation and cryptosporidiosis and giardiasis infection using monthly weather data and case reports from Colorado counties over a twenty-one year period.
Methods Data on reportable diseases and weather among Colorado counties were collected using the Colorado Electronic Disease Reporting System (CEDRS) and the Daily Surface Weather and Climatological Summaries (Daymet) Version 3 dataset, respectively. We used a conditional Poisson distributed-lag nonlinear modeling approach to estimate the lagged association (between 0 and 12-months) between relative temperature and precipitation extremes and the risk of cryptosporidiosis and giardiasis infection in Colorado counties between 1997 – 2017, relative to the risk found at average values of temperature and precipitation for a given county and month.
Results We found a consistent, significant increase in the relative risk of cryptosporidiosis and giardiasis infection when the maximum or minimum monthly temperature in the 4-6 months prior were low (1st percentile) for a given county and calendar month. High precipitation (90th percentile for a given county and calendar month) 2-12 months prior was associated with a significant decrease in the relative risk of cryptosporidiosis, while high precipitation (90th percentile) 12-months prior was associated with a significant increase in giardiasis infection risk.
Discussion Our study presents novel insights on the influence that extreme low temperatures can have on parasitic disease transmission in real-world settings. Additionally, we present preliminary evidence that the standard lag periods that are typically used in epidemiological studies to assess the impacts of extreme weather on cryptosporidiosis and giardiasis may not be capturing the entire relevant period.
Introduction
In recent years, a diverse array of extreme weather events across the globe have been inextricably linked to human-induced climate change – including heatwaves, heavy precipitation, flooding and droughts – all of which are likely to impact the transmission of environmentally-mediated illnesses such as those transmitted by parasitic protozoa. 1–4 Protozoan diarrheal diseases such as cryptosporidiosis and giardiasis are a leading cause of the estimated 1.7 billion cases of diarrheal disease that occur every year, 8–12 with an estimated 33,900 deaths and 2.94 million disability-adjusted life-years (DALYs) lost to illness caused by enteric protozoa each year. 13 Cryptosporidium and Giardia are both zoonotic protozoan parasites that are transmitted by animals and humans via the fecal-oral route, and are most frequently acquired by consuming contaminated water or food. 14–17 As such, regions of the world that have limited access to clean water, are prone to water scarcity, or lack infrastructure to support widespread access to improved sanitation tend to be the most at risk of protozoan disease transmission. 18 Nevertheless, climate change is also expected to increase waterborne enteric disease risk in high income settings as a result of projected increases in the frequency and severity of extreme weather events. 1,4–7
Across the globe, both periods of drought and heavy rainfall can cause inefficiencies and reduced effectiveness of water treatment systems. 2,19–23 For example, studies conducted in Australia and the Netherlands have found that periods of drought were associated with decreased water quality, 25,26 and an increased risk of cryptosporidiosis. 24 Meanwhile, periods of heavy rainfall can also cause sewage system overflow, contamination of irrigation systems, wells and private water supplies, and associated food and waterborne illnesses. 2,27 Protozoa are particularly susceptible to mobilization and transmission following heavy rainfall events due to their ability to persist in soil and water for months, their capacity to infect individuals at very small doses, and their ability to be transmitted by humans, domestic animals and livestock. 12,28–30 For instance, researchers in Vancouver, Canada found that water turbidity and the risk of runoff-related cryptosporidiosis and giardiasis incidence was highest when a period of excessive dryness was followed by heavy rainfall events. 23 Exposure via runoff and waste water system failures may be particularly important for parasitic protozoan diseases, which are more resilient than most bacteria and viruses in the face of conventional water treatment methods, 32 and are less commonly the target of operational surveillance systems. 4
When it comes to the impacts of temperature extremes on diarrheal diseases, protozoan infections have been less studied than bacterial or viral infections, 33,34 though associations between maximum and minimum temperature and cryptosporidiosis and giardiasis have been suggested in a few instances. A review by Ikiroma & Pollock (2021) on the impacts of weather on cryptosporidiosis identified studies highlighting a positive association between maximum temperature and cryptosporidiosis, 35–39 and to a lesser extent, giardiasis. 38 While further mechanistic studies are needed to determine why rising temperatures are associated with an increase in protozoan disease risk in some contexts, possible explanations include increased potential for transmission via alternative vectors at higher temperatures, or the lengthening of the transmission season when temperatures are unseasonably warm during the offseason. 40–43 By contrast, very few studies have investigated the impacts of minimum temperature, and the results have been inconsistent. 39 For example, one study from Victoria, Australia found that an increase in monthly minimum temperature in the current month (no lag) was associated with an increase in cryptosporidiosis incidence in metropolitan areas, whereas the 3-month lagged monthly minimum temperature in rural areas was associated with a decrease in cryptosporidiosis incidence. 44 Further complicating our understanding of the influence of temperature on enteric protozoan disease transmission is the fact that these relationships are pathogen-specific and highly dependent on the local climate system and season. 27,38,45 For instance, a meta-analysis conducted by Jagai and colleagues (2009) demonstrated that while temperature was more strongly associated with cryptosporidiosis in temperate climates, precipitation was a stronger predictor in tropical climates, though notably, neither temperature nor precipitation was significantly associated with cryptosporidiosis in arid or semi-arid climates. 46
Although the majority of studies assessing the effects of weather and climate on cryptosporidiosis and giardiasis have assessed lag periods of between one and three months, 35–38,44 pathogen survival in the ambient environment is likely longer under optimal conditions. Survival time of Cryptosporidium oocysts and Giardia cysts will vary substantially with local environmental conditions such as ultraviolet light (UV), temperature, humidity, precipitation, soil composition, season, daylight hours, excrement moisture, water velocity, water turbidity, etc. 12,28,29,40,41,47–51 As such, determining the maximum survival time in the ambient environment can be challenging, though controlled studies have demonstrated that Giardia cysts submerged in river and lake water at winter temperatures maintained viability at 12-weeks post-baseline, 47 while Cryptosporidium oocyst infectivity can remain after being stored in water for over five months at 5-15°C, 51 and up to a year when stored in low turbidity water. 52 Given remaining uncertainty regarding the maximum survival of Cryptosporidium and Giardia and their demonstrated capacity for prolonged survival under optimal conditions, epidemiological studies that assess the potential for longer lagged effects are needed.
In light of the many influences that weather and environment have on the transmission of protozoan pathogens, as well as the remaining uncertainty surrounding the time- and location-specific conditions that may amplify or reduce risk, localized investigations of weather extremes and their lagged effects on transmission are needed. Supplemental Table S1 provides examples from the literature and a summary of some of the hypothesized mechanisms by which extreme precipitation and temperature can influence cryptosporidiosis and giardiasis transmission. Overall, despite mixed evidence on the direction of association between precipitation extremes and protozoan disease transmission, and the limited body of epidemiological evidence on the relationship between temperature extremes and cryptosporidiosis and giardiasis risk, there remain several plausible mechanisms by which weather extremes could affect the viability, distribution and overall transmission potential of Cryptosporidium and Giardia.
In this study, we investigated the lagged effects (between 0 and 12 months) that low and high temperature and precipitation values – defined in this study as the 1st, 5th, 10th, 90th, 95th and 99th percentile values of our weather variables – each have on the risk of cryptosporidiosis and giardiasis infection in Colorado counties between 1997 – 2017, relative to the risk found at average values of temperature and precipitation. In so doing, we provide valuable insight on the Colorado-specific lagged effects that precipitation and temperature extremes have on two important, but often overlooked sources of waterborne illness. As extreme weather events are likely to increase in frequency and strength in the decades to come, our findings contribute to the larger body of science on climate and health, helping to characterize the diverse and complex effects that climate change can have on human health.
Methods
Study design
In this study, we assessed the impacts of monthly precipitation and temperatures on reported cases of cryptosporidiosis and giardiasis in Colorado counties between 1997 and 2017. Specifically, our primary objective was examining the effects of relative temperature and precipitation extremes – mean maximum monthly temperature (MAXT), mean minimum monthly temperature (MINT), and total monthly precipitation (PREC), each mean-centered by county and calendar month – on the monthly case counts of human cryptosporidiosis and giardiasis. As secondary goals, we also identified relevant lag periods over which extreme weather may impact cryptosporidiosis and giardiasis in Colorado, and evaluated whether the effects of precipitation and temperature on monthly case counts varied by season. We used a distributed lag nonlinear modeling (DLNM) approach, conditioned on total case counts within each of the twelve calendar-months for a given county. 53,54 The Colorado Multiple Institution Review Board (COMIRB) reviewed and approved of this study.
Cryptosporidiosis and giardiasis incidence
Data on reportable diseases in Colorado were collected using the Colorado Electronic Disease Reporting System (CEDRS), which is made available upon request from the Colorado Department of Public Health and Environment (CDPHE). Our study spanned the 21-year period between January 1997 through December 2017, using monthly case reports of cryptosporidiosis and giardiasis by county as our two outcomes of interest. As both cryptosporidiosis and giardiasis are mandatory reportable diseases and CDPHE conducted lab audits to ensure proper monthly reporting, for the purposes of this study we assumed that cases of sufficient severity to warrant medical attention in Colorado counties for each month between 1997 and 2017 were captured in the CEDRS dataset.
In order to estimate incidence, we used annual county population estimates from the Colorado Department of Local Affairs. 55 Although Broomfield County was approved by voters as Colorado’s 64th county in 1998, it was not included as its own county in this analysis until 2000, as prior to this point, the Colorado Department of Local Affairs census data recorded Broomfield’s population as a part of Boulder County’s population. The natural log of each county’s annual population estimate was included as an offset term in the statistical models.
Climatological variables
To obtain monthly estimates of temperature and precipitation for each Colorado county, we used the Daily Surface Weather and Climatological Summaries (Daymet) Version 3 dataset derived by the National Center for Atmospheric Research (NCAR), which provides daily estimates of weather parameters across a 1-kilometer grid spacing for North America. 56 For each month between January 1996 and December 2017, MAXT (°C) and the MINT (°C) were each averaged across all 1-km grid points within a given county, such that each county-month had a single estimated average value. PREC for each county-month was estimated using total daily precipitation (mm) measured at each 1-km grid point within a county’s borders, summed across all days in each month and averaged across all grid points in the county. All forms of precipitation were included (e.g., snow, sleet, rain) and converted to a water-equivalent depth in millimeters. Notably, while our outcome data begins in 1997, the data for the explanatory variables begins in January 1996 to allow for up to 12-month lagged effects for each of our weather parameters.
To make the definition of low and high most relevant to the local weather conditions found in Colorado counties, we mean-centered our weather variables by county and calendar-month, such that values could be interpreted as hot/cold or wet/dry for the specific location and time of year. All three weather variables were subsequently mean-centered by county and calendar-month, allowing MAXT, MINT and PREC to be defined relative to their location and time. For example, the average MAXT was 6.3°C for Adams County in December 2017, while the mean-centered (MC)-MAXT was +2.6°C (as compared to other Decembers between 1996 – 2017 in Adams County), indicating a warmer than average December in Adams County in 2017.
In this analysis, we were ultimately interested in the association between higher and lower than average temperature and precipitation values on cryptosporidiosis and giardiasis incidence. Thus, to define the set of relevant predictor values to be evaluated as the numerator in our relative risk calculations, we used the 1st, 5th, 10th, 90th, 95th and 99th percentile values of our mean-centered weather variables, referred to respectively as extreme low, very low, low, high, very high and extreme high values throughout this analysis, as detailed in Supplemental Table S2.
Analysis
To assess the lagged effects of MC-MAXT, MC-MINT and MC-PREC on human cryptosporidiosis and giardiasis cases in Colorado counties between 1997-2017, we used a conditional Poisson DLNM approach. 53,54 DLNMs were used because they provide a flexible platform for exploring nonlinear outcome-predictor relationships and they permit lagged effects to be spread over time, two characteristics that are desirable when exploring the impacts of highly variable environmental conditions on health outcomes. As such, we investigated a wide range of potential lags between 0- and 12-months by varying the number of internal knots (0-6) for the spline functions used to define the lag space and comparing model performance across these formulations of the lag space. Whereas 2-month and 4-month lags were selected because they aligned with the typical lag range indicated in the literature, 35–39,44 6-month and 12-month lags was also examined in detail in order to determine whether extreme weather could have more prolonged impacts of on giardiasis or cryptosporidiosis case counts in Colorado counties between 1997 – 2017. Ultimately, this design allowed us to identify potential longer-term trends that might have otherwise been missed if only short-term lags were assessed.The impacts of MC-MAXT, MC-MINT and MC-PREC on monthly cryptosporidiosis and giardiasis county case counts were each assessed independently, making a total of six outcome-predictor combinations. The steps that were taken to conduct this analysis for each of the six outcome-predictor pairs of interest to this study are outlined in Figure 1.
The 8-Step analytical process that was used for each outcome and predictor of interest in this study is outlined above. The two outcomes in this study were case counts of cryptosporidiosis, and case counts of giardiasis. The three predictors were the average maximum monthly temperature, the average minimum monthly temperature, and the total monthly precipitation. We therefore repeated Steps 1-8 a total of six times for this study, once for each outcome-predictor pair.
In Step 1, initial exploratory analyses were conducted to summarize between-season differences in our outcome-predictor pairs across our study period. A three-month definition of season was also used in our exploratory assessment to investigate the distribution of the data across the calendar year. Winter was defined as December – February; spring as March – May; summer as June – August; and fall as September – November.
Next, we developed several different cross-basis structures to investigate potential non-linearities in the lagged predictor-outcome relationship over time and space, using the “dlnm” package in R. 58,59 A cross-basis is a bi-dimensional space of functions that encapsulates both the shape of the relationship between the outcome and predictors (termed the “predictor space”), as well as the distributed lagged effects (i.e. the “lag space”). 53 To develop a cross-basis with suitable functions defining both the predictor space and the lag space, we investigated the fit of different third-degree spline functions (natural spine, B-spline and penalized spline) with incrementally increasing degrees of freedom (maximum of seven) and a maximum lag of 12-months for the lag space, while the predictor space was held constant using a simple linear basis function. The default settings for knot placements were used for each spline function such that internal knots were placed at equally spaced points along the lag-space while the boundary knots were set to 0- and 12-months for the natural and B-spline functions. A conditional Poisson generalized non-linear modeling (GNM) approach was used to compare model fit across the different lag space functions used in the cross-basis. All models also included an intercept, a continuous year term, an offset for county population and county-month as the stratum (i.e., conditional) variable. Each model’s Bayesian information criterion (BIC) was used to identify the three best lag space definitions for the cross-basis.
Once suitable functions were identified for the lag space, Step 3 was to develop the predictor space using a similar process for the lag space. That is, the fit of different third-degree spline functions with incrementally increasing degrees of freedom (up to a maximum of five) were compared for the predictor space, while we defined the lag space using the top three previously identified lag space formulations. In addition to assessing different variations of third-degree spline functions, threshold functions were also assessed for the predictor space. The BIC value was used again to select the three best fit models across all the different cross-basis formulations.
We looked for evidence of potential effect modification by season in Step 4. Four separate cross-basis terms (one for each three-month season) were generated, using the top performing predictor space and lag space definitions previously identified through Steps 2 and 3. In total, the procedure outlined in steps 2-4 yielded a set of 50 models for each outcome-predictor pair, which was then compared a final time using both the Akaike information criterion (AIC) and BIC values to select a final subset of six models that were reserved for further exploration. To demonstrate the analytical process that was used to build suitable cross-basis structures in Steps 2-4, Supplemental Table S3 provides an example from one outcome-predictor pair, detailing all model variants that were developed for MC-MINT and cryptosporidiosis.
The predictive capacity of each of the six AIC/BIC-selected models for a given outcome-predictor was then assessed using cross-validation in Step 5. For each of the six models, repeated training and testing was conducted using a 21-fold cross-validation. 60 Each fold contained a year of observations, such that, within each fold, 20 years of data were used to train the model while the remaining year of county-month observations was used for validation. This training-validation process was repeated a total of twenty-one times, once for each year between 1997 and 2017. The mean RMSE value summarizing the performance across all tuning and validation iterations was then used as an indicator of the overall predictive skill of the model. The model with the lowest mean RMSE was selected as the final model (Supplemental Table S4).
The final model was re-run using a conditional quasi-Poisson GNM variant to obtain adjusted confidence intervals to account for overdispersion (Step 6). As was done with for the conditional Poisson models, the quasi-conditional Poisson models included an intercept outside of the cross-basis, a continuous year term to account for long-term trends, county-month as the stratum variable, and an offset for county population. In Step 7, we used these models to generate estimates of the relative risk (RR) with 95% confidence intervals (95%CI) of cryptosporidiosis or giardiasis across a matrix of relevant lag and predictor values, relative to the risk of disease found at zero (i.e., the mean for a given county and calendar-month). The lag periods that were used in these relative risk calculations were 2-months, 4-months, 6-months, and 12-months.
Finally, to visualize our relative risk estimates and the entire predicted outcome surface across all predictor and lag values, we created a set of plots depicting the predictor-lag-outcome surface, using a reference value of zero (i.e., the mean) in all cases. We used a p-value of <0.05 to indicate statistical significance in this study. Stata 15 (Stata Statistical Software: Release 15 (2017). StataCorp LP, College Station, TX) and R Studio 4.0 (RStudio Team (2020). RStudio: Integrated Development Environment for R. RStudio, PBC, Boston, MA URL http://www.rstudio.com/) were used for all analyses. 59,61
Results
Between 1997 and 2017, the total population in Colorado nearly doubled, increasing from approximately 2.9 million to 5.6 million (Table 1). For all counties, PREC was highest in the spring and summer, and lowest in the fall and winter. Cryptosporidiosis and giardiasis cases tended to be highest in the summer (38.2% of all cryptosporidiosis cases; 30.6% of all giardiasis cases) and the fall (32.1% of all cryptosporidiosis cases; 30.3% of all giardiasis cases) across all Colorado counties (Table 1). Reported cryptosporidiosis cases increased over the study period while giardiasis cases decreased (Figure 2).
Depiction of th annual count of cryptosporidiosis and giardiasis cases across all Colorado counties between 1997 – 2017. Whereas there was a modest decrease in the total number of cases of giardiasis, there was an increase in the total number of cryptosporidiosis cases in Colorado cases between 1997 and 2017.
Each of the final models selected via cross-validation for modeling the impacts of weather extremes on cryptosporidiosis cases used natural or B-splines in both the lag and predictor spaces, while the models for giardiasis cases ultimately employed a natural or B-spline in the lag space and a linear function to model the predictor space. Although some of the top six AIC/BIC-selected models included interaction terms for season, none of these models were selected as the final model via cross-validation for any outcome-predictor pair. Supplemental Table S4 details the cross-basis structure and the cross-validation results for the top six models of each outcome-predictor pair.
Cryptosporidiosis
Temperature
Our final models indicated that there was an increase in cryptosporidiosis case counts when relative MC-MAXT or MC-MINT in the 4-6 months prior was at an extreme low (Figure 3, Table 2). When MC-MAXT was 4.5°C lower (1st percentile) than the county and calendar-month mean 4-months prior, there was a 5.6% increase in the risk of cryptosporidiosis (RR 1.056; 95% CI 1.007 – 1.107), while at 6-months prior, there was an 8.4% increase in the risk of cryptosporidiosis relative to the mean (RR 1.084; 95% CI 1.030 – 1.141). Similarly, when MC-MINT was 3.8°C below average 6-months prior (1st percentile value), there was a 45.3% increase in reported cryptosporidiosis relative to the risk at the county and calendar-month mean (RR=1.453; 95% CI 1.347 – 1.568). At even lower temperatures (< 1st percentile), the increase in the risk of cryptosporidiosis was even more pronounced, as demonstrated in Figure 3.
3D illustrations of the best performing model for the lagged relationship between cryptosporidiosis cases and MC-MAXT, MC-MINT and MC-PREC. For each predictor-outcome pair, the 3D graphic depicts the change in the relative risk of cryptosporidiosis across the range of observed temperature and precipitation values for lags of 0 – 12 months, relative to the risk found at the county and calendar-month averages for temperature and precipitation.
By contrast, when MC-MAXT was very low or low (5th and 10th percentiles, respectively), or when MC-MINT was low (10th percentile) 2-6 months prior, there was a decrease in cryptosporidiosis risk relative to the risk at the mean (Table 2). However, when a 12-month lag was used, the relationship between cryptosporidiosis and low temperatures (10th percentile) mimicked that of extreme low temperatures, wherein the risk of cryptosporidiosis increased relative to the risk at the mean 12-months prior (MC-MAXT 10th percentile: RR 1.132; 95% CI 1.085 – 1.132; MC-MINT 10th percentile: RR 1.054; 95% CI 1.010 – 1.100).
Although the risk of cryptosporidiosis was not significantly different (p<0.05) when comparing extreme high MC-MAXT or MC-MINT (99th percentile) to the mean, when either MC-MAXT or MC-MINT was high or very high (90th and 95th percentiles) 2-6 months prior, there was a significant increase in cryptosporidiosis risk relative to the risk at the mean (Table 2).
Precipitation
Very high and extreme high (95th and 99th percentile values) MC-PREC in the 2-6 months prior were associated with a significant decrease in the relative risk of cryptosporidiosis (Figure 3, Table 2). For example, when MC-PREC 4-months prior was at the 99th percentile value (84 mm higher than the county and calendar-month average), the relative risk of cryptosporidiosis was 19.2% lower than the risk at the average (RR 0.808; 95% CI 0.749 – 0.872).
When MC-PREC was lower than average during the 2-6 months prior, cryptosporidiosis risk was not significantly different than the risk at the mean (Table 2). However, when a 12-month lag was used, lower than average MC-PREC (1st – 10th percentiles) was associated with a decrease in cryptosporidiosis risk as compared to the risk at the mean (e.g., MC-PREC 1st percentile: RR 0.727; 95% CI 0.659 – 0.802).
Giardiasis
Temperature
Our models highlighted a negative linear relationship between giardiasis and temperature, wherein the risk of giardiasis was highest when 2-6 month lagged temperatures were lower than average, and lowest when 2-6 month lagged temperatures were higher than average (Figure 4). When MC-MAXT was low, very low or extremely low relative to the average, there was a 1-2% increase in the 2-6 month lagged risk of giardiasis (Table 3). For MC-MINT, the change in risk was more pronounced: there was a 2-5% increase in the risk of giardiasis at low to extreme low values of MC-MINT, relative to the risk at the county and calendar-month mean MC-MINT (Table 3). Similarly, when MC-MAXT and MC-MINT were high to extremely high (90th – 99th percentiles), there was a 1-2% increase in the 2-6 month lagged risk of giardiasis relative to mean MC-MAXT, and a 2-4% increase in the 2-6 month lagged risk of giardiasis relative to mean MC-MINT. The risk of giardiasis at high and low temperature extremes was not statistically significant (p < 0.05) when a 12-month lag was used.
3D illustrations of the best performing model for the lagged relationship between giardiasis cases and MC-MAXT, MC-MINT and MC-PREC. For each predictor-outcome pair, the 3D graphic depicts the change in the relative risk of giardiasis across the range of observed temperature and precipitation values for lags of 0 – 12 months, relative to the risk found at the county and calendar-month averages for temperature and precipitation.
Precipitation
Neither the extreme high nor extreme low values of 2-6 month lagged MC-PREC was significantly associated with a change in giardiasis risk, relative to the county and calendar-month mean values (Table 3). However, there was consistent evidence of 12-month lagged effects for giardiasis. We found that higher than average MC-PREC 12-months prior (at the 90th, 95th and 99th percentile values) was associated with a decrease in giardiasis risk, relative to the risk at the mean (90th percentile: RR= 1.020, 95% CI 1.007 – 1.034; 95th percentile: RR= 1.030, 95% CI 1.010 – 1.051; 99th percentile: RR= 1.056, 95% CI 1.019 – 1.094). Similarly, lower than average MC-PREC 12-months prior (at the 1st, 5th and 10th percentile values) was associated with a decrease in giardiasis risk, relative to the risk at the mean (1st percentile: RR= 0.968, 95% CI 0.947 – 0.989; 5th percentile: RR= 0.977, 95% CI 0.962 – 0.992; 10th percentile: RR= 0.982, 95% CI 0.970 – 0.994).
Discussion
In this analysis, we found that cryptosporidiosis and giardiasis responded similarly to unseasonably low temperatures in Colorado counties between 1997 and 2017. There was a significant increase in the relative risk of both cryptosporidiosis and giardiasis infection when temperatures 4-6 months prior were extremely low for a given county and calendar-month (i.e. at the 1st percentile value, corresponding with -4.5°C below average for MC-MAXT, and -3.8°C below average for MC-MINT). As far as we are aware, this is a novel finding that has not been replicated in other epidemiological studies, perhaps because both giardiasis and cryptosporidiosis incidence tend to peak in summer or late summer in the United States, 45 leading most investigations to focus primarily on warmer temperatures and transmission potential. Nevertheless, experimental studies have highlighted that both Cryptosporidium and Giardia can withstand cool temperatures, with the viability of (oo)cysts improving when stored at 4°C as compared to room temperature. 62 One study identified developmental stage parasites in mice exposed to Cryptosporidium oocysts and stored at -20°C, -15°C, -10°C and 5°C for between 1-168 hours before thawing, 50 highlighting oocyst persistence even when temperatures are lower that average.
By contrast, when temperatures were moderately high (90th – 95th percentiles) 2-6 months prior, we found a statistically significant decrease in giardiasis cases in Colorado counties, while there was a significant increase in cryptosporidiosis cases. For giardiasis that relationship held at the 99th percentile values of MC-MAXT and MC-MINT, while there was no evidence of an association between extreme high temperatures (99th percentile) and cryptosporidiosis cases. The fact that higher than average temperatures was associated with a decrease in giardiasis cases at all levels (90, 95th and 99th percentiles) but an increase in cryptosporidiosis at moderate levels (90th and 95th percentiles), may be related to the two organisms having different thresholds for tolerating above average temperatures. Both in vitro and in vivo experiments have suggested that Giardia cysts demonstrate greater temperature-dependence and higher die-off rates than Cryptosporidium oocysts at both high and low temperatures. 29,41 For example, one recent study focused on the potential for foodborne transmission of giardiasis and cryptosporidiosis found that Cryptosporidium oocysts were stable at both room temperature and refrigeration conditions (4°C), whereas Giardia cysts saw a 50% die-off rate at room temperature over a period of 24-hours. 62 Thus, the fact that both moderate and extreme increases in temperature relative to the county and calendar-month mean were associated with a decrease in giardiasis risk may potentially reflect the greater temperature sensitivity of Giardia. Meanwhile, the increase in cryptosporidiosis cases with moderately high (90th – 95th percentiles), but not extremely high temperatures (99th percentiles) could be indicative of a threshold effect, wherein moderately high temperatures lengthen the transmission season and/or allow for transmission via alternative vectors, while extremely high temperatures lead to exponential oocysts die-off and reduced transmission. 40
Another possible explanation for why Cryptosporidium and Giardia respond differently to higher-than-average temperatures is that the primary hosts associated with each pathogen are different. For example, a study conducted in Georgia, USA found that synanthropic flies living alongside wildlife and livestock could serve as mechanical vectors of Cryptosporidium and Giardia, with viable Cryptosporidium being isolated from 56% of sampled flies, while Giardia was isolated from just 8% of those sampled. 42 As warm-weather, ectothermic species, the development time of the studied fly species tends to decrease as temperatures increase, up until a threshold temperature when development will cease – for example, above 34°C for similar species of blowflies found in Asia. 63 Thus, if cattle and synanthropic flies were in fact a major part of the transmission pathway for cryptosporidiosis, but not giardiasis (as has been found in other contexts 42,64), then the association between cryptosporidiosis and moderate, but not extreme increases in maximum and minimum temperature is consistent with what would be expected. However, further study is still needed to determine whether flies are in fact an important vector in either the cryptosporidiosis or giardiasis transmission cycles in Colorado.
In our study, there was a significant decrease in the relative risk of cryptosporidiosis when total monthly precipitation in the 2-6 months prior was relatively high. However, neither relatively high nor low monthly precipitation were found to significantly impact giardiasis using a 2-, 4- or 6-month lag. One possible explanation for this difference could be that the primary underlying exposure pathways of Cryptosporidium and Giardia transmission differ between the two pathogens. For example, if surface water contamination via zoonotic livestock or wildlife species were the primary source of cryptosporidiosis transmission, but not giardiasis, then the overall concentration of cryptosporidium in surface water sources might be diluted following higher than average precipitation. As an example, this was believed to be the case in one study conducted in Canada which observed lower odds of human cryptosporidiosis when water levels were higher than average 19-20 days prior. 64 While the lag periods assessed in our study were of a considerably longer duration than those studied by Brankston and colleagues, we did find a similarly negative association between cryptosporidiosis and the 90th, 95th and 99th percentiles of total monthly precipitation at both 2-month and 4-month lags, a finding that held for the 95th and 99th percentile at the 6-month lag. Meanwhile only the 99th percentile value of precipitation was still negatively associated with cryptosporidiosis when lagged 12-months. Notably, a recent review that assessed the body of evidence in support of the concentration-dilution hypothesis – that is, that conflicting findings on the impacts of rainfall on diarrhea result from underlying differences in background rain levels – found four studies that identified dilution (specifically, rainfall following wet periods) as a potential mechanism explaining an inverse association between rainfall and diarrhea. 66 While an investigation of the impacts of rainfall following wetter than average weather was outside of the scope of this study, future investigations of the impacts of this phenomenon on giardiasis and cryptosporidiosis, as well as the underlying sources of contamination for each pathogen within Colorado counties is warranted.
Whereas other studies assessing the effects of weather and climate on cryptosporidiosis and giardiasis have most frequently assessed lag periods of between one and three months, 35–38,44 in our study, we allowed for a lag of up to 12-months. This decision was based on the body of evidence from a small set of experimental studies which have suggested that under the right conditions, Cryptosporidium oocysts and Giardia cysts may be able to survive and maintain viability for much longer than the period typically assessed in the epidemiological literature – potentially as long as six months to a year. 51,52 Whereas the relative risk of maximum monthly temperature and minimum monthly temperature extremes relative to the averages tended to only be statistically significant for giardiasis and (to a slightly lesser extent) cryptosporidiosis in the 2-6 month lag range, the 12-month lagged estimates of the relative risk of precipitation extremes (compared to the mean) were statistically significant in all but one instance (See Tables 2 & 3). Moreover, lower than average precipitation relative to the mean was only significantly associated with giardiasis or cryptosporidiosis when lagged 12-months, suggesting that the transmission of these pathogens in Colorado counties may be sensitive to longer-term periods of excess dryness or drought, as has been found with diarrheal pathogens in other regions of the world. 23,24,67 While our study was not designed to assess the impacts of rainfall after prolonged periods of drought on transmission, future studies aimed at investigating this potential in Colorado and the American West at large are greatly needed.
Although giardiasis and cryptosporidiosis cases reported in Colorado counties between 1997 – 2017 tended to follow the standard seasonal pattern that have been previously identified for temperate climates, wherein case counts tend to peak in the summer or early fall 45 (See Table 1), the final models selected via cross-validation for each of our outcome-predictor pairs did not ultimately include an interaction for season. This is notable, as season-specific effects have been demonstrated in other contexts, 31,68 which are believed to arise as a result of season-specific exposure opportunities such as local agricultural practices or recreational water uses. 68,69 While it is possible that the relationship between precipitation and temperature and cryptosporidiosis and giardiasis truly does not vary substantially by season in Colorado counties, it is also possible that our models were underpowered for detecting seasonal effects.
Thus, a limitation of this study was the potential for being underpowered when it came to the seasonal effect modification assessment, due to the panel structure of the data, our conditioning on county-month, the complex lag-predictor cross-basis structures already included in each model, and the fact that we had a relatively small number of observations for each county (252 month-years for 63 counties, and 216 for Broomfield County). Another limitation to this study was that the smallest time-unit possible given the data available was month, making it impossible for us to determine the effects of short-term weather events. Additionally, because county was the smallest geographic-unit available in this study, we were unable to investigate location-specific nuances such as variations in elevation, access to healthcare and local population factors such as popular recreational activities. While including a grouping term in our cross-basis and conditioning on county-calendar-month helps to account for unmeasured intra-county variation, we remain unable to assess inner-county differences. Another limitation to this study is that the CEDRS database that was used for obtaining monthly case counts of cryptosporidiosis and giardiasis by county does not necessarily record the location where the pathogen was acquired, only the county of residence for an infected individual.
This study relied on clinically reported cryptosporiosis and giardiasis, and, due to the variable severity of disease, many infections with these protozoa go unreported. Meanwhile, among cases of sufficient severity to warrant medical attention, changes in the use, accuracy and types of diagnostic tests that were available for the clinical detection of cryptosporidiosis and giardiasis across the study period also have the potential to bias our results. While cryptosporidiosis and giardiasis diagnostic tests remained relatively unchanged in the 90s and early 2000s, 70 the more widespread introduction of multiplex tests means that healthcare providers could more easily test for an array of diseases in patients exhibiting gastrointestinal symptoms. Being able to easily test for multiple diseases could increase detection of suspected disease, as well as increase incidental findings of cryptosporidiosis and giardiasis. In Figure 2, a plot of the total case counts by year across all Colorado counties shows that giardiasis cases appear to have a slight downward trend over time, whereas cryptosporidiosis cases have shown a steady increase across the study period. Thus, in the case of cryptosporidiosis, it is possible that increasing use of multiplex testing is influencing the upward trend in case reports. A linear term for year was included in each of our models to help account for these linear time trends.
Our analysis has shown that the risk of cryptosporidiosis and giardiasis in Colorado counties is likely to change as weather patterns in Colorado shift in the coming years. Warmer temperatures and an increase in weather extremes (i.e., blizzards, droughts, etc.) can be expected. In the last 30 years, Colorado’s average temperature has increased by 2°F (3.6 °C), with current estimates projecting that the average temperature for the state could increase another 5°F (9°C) before 2050. 72 As such, it is possible that the coming years will see a continued reduction in the transmission of giardiasis (Figure 2), as our study has highlighted that extreme high (i.e., 99th percentile) values of maximum and minimum temperatures tend to be associated with reduced giardiasis infection risk (relative to the risk at average values) in Colorado counties. However, it is possible that rising temperatures will increase cryptosporidium risk, as we found temperature extremes at the 90th and 95th percentile values were associated with a statistically significant increase in cryptosporidiosis infection risk. These discordant findings indicate that the impact of climate change may not be the same for all protozoan pathogens and highlight the need for further study of the locations-specific environmental conditions that could be simultaneously promoting the transmission potential of one protozoan pathogen while suppressing another. Further investigation into pathogen-specific environmental contamination sources in Colorado counties, as well as the potential impacts of extreme weather following prolonged periods of wet weather or prolonged periods of drought are also needed. Colorado-specific climate projections are also essential pre-requisites for quantifying the associated increased risk of infectious disease transmission over the coming decades. Ultimately, combining the findings of future climate projections and environmental risk assessments with our findings on the Colorado-specific impacts of extreme temperatures and precipitation on two important, but often overlooked parasitic diseases will help to inform the development of education and prevention messaging for high-risk populations, as well as aid in the development of tailored climate change adaptation and mitigation protocols.
Data Sharing
Data on reportable diseases in Colorado were collected using the Colorado Electronic Disease Reporting System, which is made available upon request from the Colorado Department of Public Health and Environment. The Daily Surface Weather and Climatological Summaries (Daymet) Version 3 dataset that was used in this analysis is available upon request from the National Center for Atmospheric Research.
Declaration of competing financial interests
The authors declare that they have no actual or potential competing financial interests. There is no grant support to disclose.
Acknowledgements
This work was made possible by the Colorado Department of Public Health and Environment and the National Center for Atmospheric Research, who collected and shared the data used in this analysis.