Warmer Weather and the Risk for Urinary Tract Infections in Women

Abstract

Background:

The incidence of UTIs is seasonal, peaking in summer months. One possible mechanism for the observed seasonality of UTIs is warmer weather.

Materials and Methods:

We identified all UTI cases located in approximately 400 metropolitan statistical areas (MSAs) in the contiguous United States between 2001 and 2015 using the Truven Health MarketScan databases. A total of 167,078,882 person-years were included in this dataset, and a total of 15,876,030 UTI events were identified by ICD-9 code 599·0. Weather data for each MSA and date were obtained from the National Centers for Environmental Information. We computed the mean temperature during the period zero to seven days prior to the UTI diagnosis. We used a quasi-Poisson generalized linear model. The primary outcome was the number of UTIs each day in an MSA in each age group. Covariates considered included group, day-of-week, year, and the temperature during the previous seven days.

Results:

Warmer weather increases the risk for UTIs among women treated in outpatient settings in a dose-response fashion. On days when the prior week’s average temperature was between 25 and 30°C, the incidence of UTIs was increased by 20–30% relative to when the prior week’s temperature was 5 to 7·5°C.

Conclusions:

The incidence of UTIs increases with the prior week’s temperature. Our results indicate that warmer weather is a risk factor for UTIs. Furthermore, as temperatures rise, the morbidity attributable to UTIs may increase.

Introduction

Urinary tract infections (UTI) are common bacterial infections treated in outpatient settings.¹ UTIs are especially common among women² as are UTI recurrences.³ Almost half of all women have a UTI by middle age.⁴ Worldwide, UTIs affect an estimated 150 million people each year.⁵ In the United States, over 10 million office visits and 2 million emergency- department visits are attributable to UTIs annually⁶ as well as an estimated 100,000 hospitalizations.¹ In addition, UTI-related hospitalizations have been increasing, perhaps due to increasing rates of antimicrobial resistance against oral antimicrobial agents.⁷ Thus, understanding the risk factors associated with UTIs is of growing clinical and public-health importance.

The known individual-level risk factors for UTIs include a previous history of UTIs,³ a new sexual partner,^{3, 8} recent sexual intercourse,⁸ family history of UTIs,⁹ urologic abnormalities,¹⁰ diabetes,¹¹ and limited fluid intake.¹² Environmental-level risk factors for UTIs have also been described. For example, several studies describe how both inpatient and outpatient UTIs occur with greater incidence in summer and early fall relative to other times of the year ^{7, 13, 14} and the warmer temperatures associated with these months are commonly postulated as the cause of much of the seasonal variation.^{13, 15–17} Increased exposure to warmer temperatures increases fluid loss from perspiration, thus decreasing extracellular fluid, resulting in dehydration, and decreased urine output, which decreases the mechanical clearance of bacteria from the urinary tract, ^{18, 19} and concentrates bacteria in urine.^18–20 However, the seasonality in the incidence of UTIs could be due to other behaviors associated with warmer weather.¹³ Indeed, while the seasonal UTI pattern has been described across a spectrum of ages¹⁴, the association is strongest in young adults, for whom behaviors may change the most in the warmer months, and is weakest in adults over 70 in whom UTIs are the most common.¹⁴

Our purpose is to investigate the effect of warmer weather on the risk for UTIs. We used data on a daily time scale across diverse geographic regions to capture a range of varying heat exposures.

Material and Methods

Data

We used healthcare utilization data from the Truven Health Analytics Marketscan Commercial Claims and Encounters and Medicare Supplemental databases from 2001 to September 30, 2015.

We computed daily counts of patients in the Truven/Marketscan databases in each of approximately 400 metropolitan statistical areas (MSAs) based on their enrollment dates. Claims for UTIs in an outpatient setting were identified by matching ICD-9 code 599·0 (“urinary tract infection, site not specified”). We included all claims originating from outpatient visits and excluded hospital-acquired UTIs or those developing in other inpatient settings (e.g., assisted living, long-term-care, hospice, inpatient psychiatric facilities, intermediate care/mental disability care facilities, residential substance abuse or psychiatric treatment centers). We also excluded claims that originated from independent laboratories. Next, we aggregated claims to the patient-day level. Because patients could have multiple office visits for the same UTI event, we required at least 14 UTI-claim-free days between any two subsequent UTI diagnosis claims in order to avoid double counting of a single UTI event. The number of people with a diagnosis of UTI by day for each of the Truven MSAs was counted after stratification by sex (male/female) and age (five-year bins). We restricted our analysis to UTIs in females over the age of 18 because the clinical presentation of and risk factors for UTIs in children and in males differ from adult females.

Using weather data for each MSA from the National Centers for Environmental Information’s Integrated Surface Database, we collected hourly temperature observations at thousands of weather stations across the United States. We geocoded the names of the MSAs in Truven using the Google Maps Geocoding Application Programming Interface and used weather stations within 100 km of the MSA center to compute the weather for each MSA. We computed the average daily mean temperature for each day at each weather station and then averaged across all weather stations within 100 km to construct an estimate of weather for that MSA on that day. We weighted the averages by the number of measurements recorded at that weather station on that day to reduce the impact of potential missed measurements on daily temperature summaries.

Model

We used a quasi-Poisson generalized linear model with a log link function. The primary outcome was the number of UTIs diagnosed on a day in an MSA in each age group. Covariates are the age group, day-of-week, year, and the temperature during the prior zero to seven days. We included the day-of-week indicators as our data records clinic visits with a diagnosis of UTI and not the onset of UTI symptoms. While UTI symptom onset should occur uniformly each day of the week, clinic visits do not. Many clinics are closed or have restricted hours on the weekend relative to the weekday, and as a result, incidence, as measured by clinic visits, decreases on the weekend. We divided the temperature into separate bins based on whether the mean temperature was below 0°C, or over 37·5°C, and we divided temperatures between 0°C and 37·5°C into 2·5-degree bins. The reference level was temperatures >5°C and <= 7·5°C. Because the number of women at risk varied among the different age groups, MSAs, and over time, we included the log of the daily number of women at risk as an offset in the model. Because temperature and season are collinear, we did not include an adjustment for seasonality or month in our model.

Sensitivity Analyses

Grouping by Seasonal Fluctuation

We constructed an ordinary-least-squares model for each MSA using all weather data for that MSA (average of 73·3 years, range of 12·0 to 85·0 years) to predict the average temperature for each date in that MSA. We modeled average daily temperature as the combination of a linear time trend, sine and cosine seasonal patterns, and a series of indicators for day-of-week. The average model had an R² of 0·79 (inner-quartile range 0·76 to 0·82). The lowest R² values generally occurred in places with the least seasonal patterns in weather (e.g., Miami, FL)

We then computed the average seasonal fluctuation in temperature as the difference between the average coldest day’s average temperature and the average warmest day’s average temperature. The MSAs were then divided into deciles according to this difference.

To determine if the observed effect of temperature was caused by omitted seasonal variables, we re-estimated the primary model using only data from the cities in each seasonal-fluctuation decile. If the observed temperature effect in our main model were driven by omitted seasonal factors, the estimation of the model in locations with limited seasonal fluctuation should show reduced-to-no response to temperature while locations with greater seasonal patterns in temperature should show greater response to temperature.

We performed all analyses using R version 3 (The R Project for Statistical Computing.) Studies of this type are deemed non-human subjects research by the University of Iowa Institutional Review Board.

Results

There are a total of 15,876,030 UTI events over 167,078,882 person-years at risk in the cohort after applying exclusion criteria (Table 1). The mean age of the enrollees with UTIs was 48·6 (SD = 18·9) years.

Table 1:

Sample Size by Restriction

Restriction	Number of Cases of UTI	Percent of Total Claims
Patient Days with claims for UTIs in outpatient, independent living, non-lab settings	30,534,050	100·0%
Remove claims less than 14 days since last UTI claim	25,418,807	83·5%
Remove claims that do not occur in an identified MSA	20,670,319	67·7%
Remove claims that occurred in men	17,707,273	60·0%
Remove claims in patients 17 or younger	15,880,058	52·0%
Has Complete Weather History for 7 Days	15,876,030	52·0%

Unadjusted incidence of UTIs exhibited a dose-response pattern to low, mean and high temperature (Figure 1). As temperatures rise above 10°C, there appears to be a rapid non-linear increase in incidence. This relationship is still present when the unadjusted incidence rate is stratified by month-of-year (Figure 2).

Figure 1: Incidence of UTI Office Visits by Temperature Zero to Seven Days Prior.

Only data points where at least 1,000,000 person-days were at risk were included to remove poorly populated and uncommon temperature levels.

Figure 2: UTI Incidence by Temperature Zero to Seven Days Prior Stratified by Month of Year.

Only data points with at least 1,000,000 person-days at risk and with temperatures in the range 0 to 35° C are shown. The dose-response increase in UTI incidence with temperature is present in all months and has a largely similar slope.

There were 26,088,703 MSA-age-stratum-days with at least one woman at risk and complete temperature data for the zero to seven days prior. Results for the primary model are summarized in Table 2. Incidence (relative risk) is significantly greater at warmer temperatures than at mild or cool temperatures after adjustment for day-of-week, age and year effects.

Table 2:

Primary Model Parameter Estimates. Temperature, after adjusting for age, year and day of week is associated with an increase in UTI incidence. Increasing temperature from 5.0 – 7.5° C to 25.0 – 30° C during the previous zero to seven days is associated with a 20–30% increase in UTI incidence.

Variable	Incidence Rate Ratio	95% Confidence Interval
Age
18–24	1·00	Reference
25–29	1·00	0·99, 1·00
30–34	0·93	0·93, 0·93
35–39	0·88	0·88, 0·89
40–44	0·84	0·84, 0·84
45–49	0·82	0·82, 0·83
50–54	0·86	0·86, 0·86
55–59	0·93	0·93, 0·93
60–64	1·05	1·05, 1·05
65–69	1·07	1·07, 1·08
70–74	1·30	1·30, 1·31
75–79	1·58	1·58, 1·58
80+	2·10	2·09, 2·10
Day of Week
Sunday	1·00	Reference
Monday	5·01	4·99, 5·02
Tuesday	4·95	4·94, 4·97
Wednesday	4·54	4·53, 4·55
Thursday	4·48	4·47, 4·49
Friday	4·15	4·13, 4·16
Saturday	1·43	1·43, 1·44
Year
2001	1·00	Reference
2002	1·01	1·00, 1·02
2003	1·01	1·01, 1·02
2004	1·04	1·03, 1·04
2005	1·10	1·09, 1·10
2006	1·19	1·18, 1·19
2007	1·20	1·19, 1·20
2008	1·14	1·13, 1·14
2009	1·37	1·36, 1·38
2010	1·36	1·35, 1·36
2011	1·38	1·37, 1·39
2012	1·41	1·41, 1·42
2013	1·36	1·35, 1·37
2014	1·38	1·37, 1·39
2015	1·39	1·38, 1·40
Average Temperature 0–7 Days Prior (°C)
<= 0	0·95	0·95, 0·96
0–2·5	0·97	0·96, 0·97
2·5–5	0·98	0·98, 0·99
5–7·5	1·00	Reference
7·5–10	1·01	1·01, 1·01
10–12·5	1·01	1·01, 1·01
12·5–15	1·02	1·02, 1·03
15–17·5	1·05	1·04, 1·05
17·5–20	1·08	1·08, 1·09
20–22·5	1·10	1·10, 1·11
22·5–25	1·12	1·11, 1·12
25–27·5	1·17	1·17, 1·17
27·5–30	1·23	1·23, 1·23
30–32·5	1·29	1·28, 1·30
32·5–35	1·23	1·22, 1·24
35–37·5	1·18	1·16, 1·21
> 37·5	0·98	0·87, 1·10

The climate range for each of the Truven MSAs is depicted in Figure 3. There is substantial variability in seasonal temperature changes across the United States (US): coastal regions have reduced variability while inland MSAs have very large temperature variability. The data were split by decile with the first decile having the lowest seasonal temperature range and decile 10 having the greatest. Responses varied from 1·05 in the second-highest decile to 1·24 in the lowest decile, Table 3. Overall, the cities where temperatures change greatly over the year have similar increases in UTI incidence in response to increased temperature relative to cities where temperatures do not change much.

Figure 3: Difference Between the Temperature on the Hottest and Coldest Days, on Average, by Location.

The temperature range is the difference between the average temperature on the hottest and coldest days during the year. Regional differences exist across the US with cities on the West Coast, Gulf Coast and Florida having limited temperature ranges while cities in the Midwest have very large temperature swings. Not shown in this map are MSAs in Hawaii and Puerto Rico (very limited, under 5° C temperature range) and Alaska (relatively large, over 35° C temperature range).

Table 3:

Linear Incidence Rate Ratio by City Climate Decile. Each model was fit using only data from the cities in that particular climate decile group and adjusts for age group, day-of-week, and year. Lower decile numbers are associated with smaller seasonal temperature ranges while larger decile numbers are associated with larger seasonal temperature ranges.

Seasonal Temperature Range Decile	Incidence Rate Ratio per 10° C	95% CI
1	1·24	1·24, 1·25
2	1·14	1·14, 1·14
3	1·09	1·09, 1·10
4	1·06	1·06, 1·06
5	1·06	1·06, 1·07
6	1·05	1·05, 1·05
7	1·06	1·06, 1·07
8	1·05	1·05, 1·05
9	1·04	1·04, 1·04
10	1·05	1·05, 1·05

Discussion

Our results demonstrate that warmer weather increases the risk for UTIs among women treated in outpatient settings. For example, we found on days when the prior week’s average temperature was very warm (i.e., similar to summer temperatures in Oklahoma City or St. Louis with average temperatures around 30°C), the incidence of UTIs was increased by 20–30% relative to when the prior week’s temperature was much cooler (e.g., around 5°C). The plateau observed for temperatures above 30°C or 35°C may result from behavioral changes, such as remaining in air conditioned buildings. Indeed, these periods would typically be accompanied by heat advisories or excessive heat warnings by the National Weather Service directing people to remain in air-conditioned locations, drink extra fluids, reschedule outdoor activities, and remain in the shade to avoid heat-related illnesses. While this interpretation seems plausible, we are unable to assess such potential behavior changes in our data. Our results were consistent across different sensitivity analyses designed to clarify the relationship between warmer weather and UTI risk. Additionally, the temperature effect on UTI risk is similar in cities with limited temperature ranges and in cities with very large temperature ranges. Thus, our results help confirm that warmer weather is a risk factor. Because UTIs are so common, even the relatively low level of absolute risk that we attribute to warmer weather has a substantial effect on overall morbidity. Furthermore, as temperature patterns change, the morbidity attributable to UTIs may increase. Interestingly, the day-of-week is a significant predictor of UTI incidence with the lowest volume on Sunday, followed by Saturday, and the highest on Monday. The weekend decrease is likely the result of decreased access to clinics on the weekend with the people who would present on Saturday or Sunday instead arriving in the clinic on Monday or Tuesday.

Several different infectious diseases are seasonal.^21–23 The seasonality of many infections has been known for centuries,²³ but the seasonality in the incidence of UTIs was described relatively recently.¹³ Subsequently, a number of investigators reported the seasonality of UTIs in various geographic regions.⁷ A variety of different factors have been proposed to describe reasons for this seasonality, including behavioral factors, and seasonal antibiotic prescribing patterns.¹³ For example, recent sexual activity is a risk factor for UTIs in women,^{3, 8} and seasonal patterns for a range of sexually transmitted infections have been reported.²⁴ Warmer weather leading to dehydration and corresponding lower urine output has been suggested as a possible reason for the seasonality of UTIs by a number of different investigators.^{18, 19, 25}

Several factors suggest that dehydration and lower urine output are associated with risk for UTI. Workers in clean rooms, employed in electronics manufacturing, have high rates of UTIs and interventions designed to increase fluid intake have reduced UTI rates.²⁵ Also, in a randomized controlled trial, women with recurrent UTI and low fluid intake were assigned to either an increase in one to five liters of fluid per day or no change in fluid intake. In the following year there were twice as many cases of UTI in the control group as in the added fluid group.¹² These results collectively highlight the importance of increasing water consumption, especially during warmer weather. However, warmer weather may also have other effects associated with increasing the risk for UTIs. For example, the bacterial burden near the opening to the female urethrae increases during warmer weather,¹⁷ as does bacterial skin colonization.²⁶ Although almost always caused by different organisms than UTIs, warmer weather is also associated with higher rates of skin and soft tissue infections including surgical-site infections.²⁷

Our results are subject to several limitations. First, for the index date of the UTI, we use the date associated with the diagnosis of the UTI: we do not know the specific date when patients developed UTI symptoms. However, by using a zero-to-seven-day window prior to the index date, we should capture the date symptoms appeared, along with the weather exposure prior to symptom development. Second, because we are using administrative data, we do not have access to microbiology results. Thus, we cannot validate the diagnosis against culture results. However, cultures are not routinely used to diagnose uncomplicated outpatient UTIs. Additionally, detection methods are designed to specifically target E. coli, the cause of a little more than half of UTIs,²⁸ at the expense of other potential pathogens²⁹ potentially reducing the value of urine microbiology reports. Other potentially relevant clinical measures about the hypothesized mechanism, such as urine concentration or hydration level, are not completely understood. We also do not have data on race or ethnicity. Third, we do not have a way of establishing exposures to outdoor temperatures: we do not know to what extent people are outside or have access to air conditioning. Fourth, temperature and humidity are strongly correlated and humidity may contribute to bacterial colonization on the skin. Our analysis is unable to address this complex relationship. Finally, we restricted our analysis to only adult women, and thus we cannot generalize our findings to other populations.

Future work is needed to address the limitations associated with our work. For example, a better understanding of when symptoms first occurred will precisely identify environmental conditions that increase the risk for UTIs. In addition, a more precise notion of exposure (e.g., time spent outdoors, or exposure to heat, indoors and outdoors), along with some notion of fluid intake would help improve our understanding of how the environment affects UTI risk. Future work should also incorporate microbiological results, including the newly-realized urobiome,³⁰ in order to understand the role of the environment on different pathogens that can cause UTIs. Finally, hospital admissions for UTIs are seasonal,¹⁵ but it is not clear that warmer weather plays a role in severity of UTIs, nor is it clear to what extent the environment is associated with recurrent UTIs.

Conclusion

We demonstrated that warmer weather is strongly associated with an increased risk of UTIs. Thus, as temperatures rise, the morbidity associated with UTIs may increase. Indeed, future investigations of risk factors associated with UTIs should consider the role of the environment for this common infection.

Abbreviations

MSA

metropolitan-statistical area

US

United States

UTI

urinary-tract infection