Daily temperature variability and mental health-related hospital visits in New York State

Abstract

Background:

Despite plausible behavioral and physiological pathways, limited evidence exists on how daily temperature variability is associated with acute mental health-related episodes.

Objectives:

We aimed to explore associations between daily temperature range (DTR) and mental health-related hospital visits using data of all hospital records in New York State during 1995–2014. We further examined factors that may modify these associations, including age, sex, hospital visit type and season.

Methods:

Using a case-crossover design with distributed lag non-linear DTR terms (0–6 days), we estimated associations between ZIP Code-level DTR and hospital visits for mood (4.6 million hospital visits), anxiety (2.4 million), adjustment (~368,000), and schizophrenia disorders (~211,000), controlling for daily mean temperature, via conditional logistic regression models. We assessed potential heterogeneity by age, sex, hospital visit type (in-patient vs. out-patient), and season (summer, winter, and transition seasons).

Results:

For all included outcomes, we observed positive associations from period minimum DTR (0.1 °C) until 25th percentile (5.2 °C) and between mean DTR (7.7 °C) and 90th percentile (12.2 °C), beyond which we observed negative associations. For mood disorders, an increase in DTR from 0.1 °C to 12.2 °C was associated with a cumulative 16.0% increase (95% confidence interval [CI]: 12.8, 19.2%) in hospital visit rates. This increase was highest during transition seasons (32.5%; 95%CI: 26.4, 39.0%) compared with summer (10.7%; 95%CI: 4.8, 16.8%) and winter (−1.6%; 95%CI: −7.6, 4.7%). For adjustment and schizophrenia disorders, the strongest associations were seen among the youngest group (0–24 years) with almost no association in the oldest group (65+ years). We observed no evidence for modification by sex and hospital visit type.

Discussion:

Daily temperature variability was positively associated with mental health-related hospital visits within specific DTR ranges in New York State, after controlling for daily mean temperature. Given uncertainty on how climate change modifies temperature variability, additional research is crucial to comprehend the implications of these findings, particularly under different scenarios of future temperature variability.

1. Introduction

Mental health conditions, a major contributor to disability world-wide, account for nearly 15% of global years lived with disability, as reported by the Global Burden of Disease study (Global, 2019). In the United States, 52.9 million adults were estimated to live with a mental health condition in 2020; among them, 14.2 million live with a severe mental condition, representing 5.6% of all United States adults (National Institute of Mental Health, 2021). In New York State (NYS), the fourth largest state by population in the country,(US Census Bureau) each year one of five individuals experiences symptoms related to mental health conditions. (New York State Department of Health) In New York City, where almost half of the state’s population lives, there are 676 psychiatric hospital visits per 100,000 residents every year. (New York City Mayor)

Among the potential modifiable risk factors that may worsen or ameliorate mental health are ambient environmental exposures (Kioumourtzoglou, 2019). Given the rapid global climate change that our world is facing there is a particular interest in the effect of weather-related exposures on mental health (Organization, 2022; Clayton, 2021). Several meta-analysis have shown a positive association between ambient temperature and poor mental health outcomes, including suicide, (Gao et al., 2019; Frangione et al., 2022; Thompson et al., 2023) mental health-related mortality (Liu et al., 2021) and hospital visits, (Thompson et al., 2023; Liu et al., 2021) and poor community mental health and wellbeing (Thompson et al., 2023). In the United States, higher temperatures were shown to be associated with higher rates of suicide (Burke et al., 2018) and emergency department (ED) visits for mental health-related conditions (Nori-et al., 2022). Analyses from NYS have also shown positive associations between heat exposure and ED visits for total and specific mental health-related conditions,(Yoo et al., 2021) as well as hospital visits for alcohol- and substance-related disorders, (Parks et al., 2023) and mental health-related ED visits among young individuals in New York City (Niu et al., 2023). However, despite this extensive evidence, limited evidence exists on how daily temperature variability is associated with acute mental health outcomes.

Daily temperature variability, commonly measured by the difference between the daily maximum and minimum, and referred to as diurnal or daily temperature range (DTR), is considered an important meteorological indicator, reflecting short-term weather stability (Guo et al., 2021; Kotz et al., 2021; Lindvall and Svensson, 2015). While the rising temperature of the Earth’s surface is a fundamental effect of climate change,(IPCC et al., 2022) the long-term progression of DTR is not completely clear (Guo et al., 2021). Conflicting evidence exists regarding the direction of the asymmetric warming trend of daily maximum and minimum over the last decades, which underlies DTR alterations: Some studies indicate a smaller warming rate of maximum than minimum temperature, resulting in a decrease in global DTR,(Sun et al., 2019) while others indicate an opposite trend leading to DTR expansion (Zhong et al., 2023). Additionally, DTR greatly varies across different geographical regions, (Guo et al., 2021) and projected to increase in some places of the world (e.g., South America, southern Africa, Central US) while decrease in others (e.g., Northern latitudes and the Sahara) (Guo et al., 2021; Lindvall and Svensson, 2015).

Daily temperature variability has been associated with various health outcomes including all-cause (Guo et al., 2016; Ma et al., 2019; Yu et al., 2022) and cause-specific mortality, (Ma et al., 2019; Yu et al., 2022; Kan et al., 2007) respiratory disease, (Zhao et al., 2018; Rahman et al., 2022; Xu et al., 2020) cardiovascular disease, (Rahman et al., 2022; Rowland et al., 2021) stroke (Lei et al., 2020) and incidence of infectious diseases, including infectious diseases hospitalizations, (Xu et al., 2020) gastrointestinal infections (Ghazani et al., 2018; Wang et al., 2023) and childhood hand, foot, and mouth disease (Xu et al., 2020; Ghazani et al., 2018; Yin et al., 2017). However, in the context of mental health evidence is lacking. Plausible pathways imply that individuals with preexisting mental health conditions may have difficulties to adjust themselves behaviorally and physiologically to rapid temperature change, which may lead to exacerbation of their chronic condition (Martin-et al., 2007; Lim et al., 2012). A few analyses examined the association between daily temperature variability and mental health outcomes, including suicide mortality, (Holopainen et al., 2014) schizophrenia admissions (Zhao et al., 2016) and self-assessed mental health score (Xue et al., 2019). These analyses were limited in terms of outcomes examined and their geographical representation (Finland (Holopainen et al., 2014) and China (Zhao et al., 2016; Xue et al., 2019)).

In the current investigation we aimed to evaluate (a) the association between DTR, an indicator of daily temperature variability, and mental health-related hospital visits for four major causes—mood, anxiety, adjustment, and schizophrenia and other psychotic disorders—leveraging data of all hospital visits in NYS during 1995–2014, and (b) how these associations varied by age group, sex, hospital visit type (in-patient vs. out-patient), and season (summer, winter, or transition seasons).

2. Methods

2.1. Study population

We obtained hospital records across NYS from 1995 to 2014 from the New York Department of Health Statewide Planning and Research Cooperative System (SPARCS) (https://www.health.ny.gov/statistics/sparcs/). SPARCS includes administrative data collected from all non-military acute care facilities in NYS, covering ~98% of all NYS hospital visits; as of 2015, SPARCS included 222 acute care facilities (Rowland et al., 2021). We included both in- and out-patient visits in analyses. For each hospital visit record, we obtained International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes, along with patient residential ZIP Code, date of hospital visit, age, sex, and visit type (in-patient vs. out-patient).

This study was approved by the Institutional Review Board at the Columbia Mailman School of Public Health and was classified as exempt from needing to obtain Informed Consent (Protocol IRB-AAAR0877).

2.2. Outcomes

We identified the four mental health outcomes most commonly occurring in NYS—anxiety, mood, adjustment, and schizophrenia and other psychotic disorders (referred to as schizophrenia disorders hereafter)—from the first four ICD-9-CM diagnostic position codes in each hospital visit record. Classifications were based on the Clinical Classifications Software (CCS) algorithm,(Elixhauser et al., 2014) commonly used in epidemiologic studies to group ICD codes into clinically meaningful categories (Web Table 1). (Rowland et al., 2021) For each cause, we counted a hospital visit as a case if it included at least one matching code in the first four ICD-9-CM codes categories, i.e., a single hospital visit could be attributed to several causes.

2.3. Exposure

We obtained daily average temperature information from the North American Land Data Assimilation System, NLDAS-2 Forcing,(Cosgrove et al., 2003) with a complete space and time coverage over the study period. NLDAS-2 estimates hourly mean weather values within 0.125° grids (~11 km × 14 km in NYS). Similar to previous work,(Rowland et al., 2021) we intersected gridded temperature daily averages with census tract-level population from 2010 US Census data. We then computed population-weighted averages at the ZIP Code Tabulation Area (ZCTA) level, a consistent geographic representation of ZIP Codes (https://www.census.gov/programs-surveys/geography/guidance/geo-areas/zctas.html), referred to as ZIP Code hereafter. Based on temperature data, we calculated DTR as the difference between daily maxima and minima, representing daily temperature variability. We also calculated standard deviation (SD) of hourly measures as an additional exposure metric. Due to the very high correlation between the two measures (Pearson’s r = 0.98) and the more intuitive interpretation of DTR, we chose the latter as the main exposure variable in the current analysis.

2.4. Statistical analysis

Web Fig. 1 schematically illustrates conceptual pathways for the health effects of temperature variability among individuals with mental health conditions. We employed a time-stratified case-crossover design (Maclure, 1991). In this design, we compared DTR of the day of hospital visit and relevant preceding days (case period) with the DTR of sets of days when the hospital visit did not occur (control periods). Control periods were selected based on matching year, month, and day of the week to eliminate potential confounding by long-term, seasonal, and day-of-week trends. For example, if a case occurred on Monday, January 11th, 2010, the matched control days would include all other Mondays in January 2010 (i.e., January 4th, January 18th, and January 25th, 2010). By comparing hospitalized individuals to themselves during other periods when they were not hospitalized, this design removes confounding due to factors that vary across individuals. The choice of case-crossover design over time series was primarily to avoid excess of zero counts in the time series, since mental health-related hospital visits are less common compared to other causes of hospital visits (Salah et al., 2021). Case-crossover design, focusing on within-individual comparisons, facilitates the handling of rare events, while still enabling the assessment of acute exposure-outcome relationships (Maclure, 1991). We employed outcome-specific weighted conditional logistic regressions (Rothman et al., 2008) to quantify the association between temperature variability and hospital visit rates, coupled with distributed lag non-linear model (dlnm) terms to estimate cumulative associations following the exposure, while controlling for the serial correlation across examined lags (Gasparrini et al., 2010). We chose to examine the cumulative association over seven days following the exposure (lags 0–6) to represent short-term associations. The rationale behind this choice was as follows: Given the uncertainty regarding the specific time period affected by the exposure, we selected a period longer than a single day (lag >0), while limiting this period to 7 days (up to lag 6), to prevent case and control periods to overlap. Overlapping periods might reduce exposure contrast, potentially complicating the differentiation between cases and controls and thus reducing statistical power. The study’s period minimum DTR served as the reference point. We assigned weights based on the number of hospital visits in each ZIP Code for a specific date, and adjusted the models for daily mean temperature, incorporating distributed lag terms, following a structure similar to that of the DTR terms. We assessed non-linear relationships using natural spline terms. Specifically, via a logit function, we modelled the log-odds of hospital visit according to the following formula:

$$\mathit{logit}\left[\mathit{Pr}\left({\mathit{Y}}_{\mathit{ci}}=1\right)\right]={\mathit{\alpha }}_{\mathit{c}}+{\sum }_{\mathit{l}=0}^6\mathit{s}(\mathit{DTR},\mathit{df}{)}_{\mathit{lci}}+{\sum }_{\mathit{l}=0}^6\mathit{s}(\mathit{T},\mathit{df}{)}_{\mathit{lci}}$$

where ${\mathrm{Y}}_{\mathit{ci}}$ denotes whether subject $i$ in matched stratum c was hospitalized, i.e., $c$ represents a group of a case and its matched controls; ${\mathit{\alpha }}_{\mathit{c}}$ the matched stratum-specific intercepts (not estimated in conditional logistic models); $\mathit{s}(\mathit{DTR},\mathit{df}{)}_{\mathit{lci}}$ the lag-specific natural spline terms as part of the dlnm terms for DTR; and $\mathit{s}(\mathit{T},\mathit{df}{)}_{\mathit{lci}}$ the lag-specific natural spline terms as part of the dlnm terms for daily mean temperature.

To select the optimal fit for the non-linear dlnm terms, we fitted models for the four outcomes separately using a variety of plausible degrees of freedom (dfs) to model the lag-specific exposure–response function (df_var), as well as the function of the association over the examined lags (df_lag). We considered a range of 1–4 for df_var, along with a range of 3–6 for df_lag, and selected the df_lag–df_var combination with the lowest Akaike Information Criteria (AIC) values (Snipes and Taylor, 2014). For each cause, the df_var = 4 and df_lag = 5 combination yielded the lowest AIC values and was therefore chosen as the combination of dfs for the main models.

In addition to the main analysis, which examined all hospital visits together for each outcome, we further conducted a stratified analysis to assess whether estimated effects varied by age group (0–24, 25–44, 45–64, or 65+ years), sex (female or male), visit type (in-patient or out-patient), or season (summer: June–September; winter: December–March; transition seasons: April, May, October, November). We used models with the same parameterization described above for the stratified analysis. Previous related work (Parks et al., 2023; Rowland et al., 2021) demonstrated that adjusting for relative humidity in the model did not impact the estimated exposure-response curves while considerably increasing model processing times, so we did not add those terms in this analysis.

Our method for selecting controls was based on incidence density sampling, i.e., controls are sampled from the same underlying population and time periods as the cases for a case-crossover. Therefore, the estimated rate ratio, which is referred to as incidence density ratio, provides a measure of the relative risk associated with the exposure during the case period compared to control periods (Knol et al., 2008). We present results as cumulative percentage change (that is, in the subsequent seven days following the exposure, lag day 0 to lag day 6) in hospital visit rates at the quoted DTR relative to the minimum DTR (0.1°C) in NYS throughout the study period. We performed all statistical analyses in R software, version 4.2.0.

3. Results

3.1. Hospital visits

Mood disorders were the leading cause of mental health-related hospital visits in NYS during the study period (1995–2014) with 4,680,294 total hospital visits recorded, followed by anxiety (2,445,837), adjustment (368,206), and schizophrenia disorders (211,973) (Table 1). The 45–64 years age group had the largest percentage for mood (37%) and schizophrenia (38%) disorders, the 25–44 years group had the largest for anxiety disorders (36%), and the 0–24 years group had the largest for adjustment disorders (37%). Female patients had the largest proportion of hospital visits for mood (64%), anxiety (65%), and adjustment disorders (60%), whereas for schizophrenia disorders, the number of hospital visits were equally distributed between sexes. Most hospital visits were out-patient for mood (64%), anxiety (75%), and adjustment disorders (66%), but not for schizophrenia disorders (39%). For all causes, the percentage of hospital visits recorded in summer, winter, and transition seasons was similar (32%–34%). Over time, the number of hospital visits for all outcomes increased across all categories (Web Figs. 2–4), except for adjustment disorders, for which we observed a decrease during 1995–2003, followed by an increase over the remaining follow-up period (2004–2014).

Table 1.

Demographic characteristics for mental health-related hospital visits in New York State for 1995–2014.

Characteristic	Mood disorders, n = 4,680,294a	Anxiety disorders, n = 2,445,837a	Adjustment disorders, n = 368,206a	Schizophrenia and other psychotic disorders, N = 211,973a
Age group (years)
0–24	811,439 (17%)	447,286 (18%)	134,839 (37%)	27,888 (13%)
25–44	1,490,554 (32%)	876,791 (36%)	122,725 (33%)	76,684 (36%)
45–64	1,735,812 (37%)	792,254 (32%)	84,484 (23%)	79,501 (38%)
65+	642,489 (14%)	329,506 (13%)	26,158 (7%)	27,900 (13%)
Sex
Female	2,982,684 (64%)	1,588,542 (65%)	219,472 (60%)	106,731 (50%)
Male	1,697,610 (36%)	857,295 (35%)	148,734 (40%)	105,242 (50%)
Hospital visit type
In-patient	1,667,994 (36%)	599,914 (25%)	126,481 (34%)	129,117 (61%)
Out-patient	3,012,300 (64%)	1,845,923 (75%)	241,725 (66%)	82,856 (39%)
Seasonb
Summer	1,558,524 (33%)	830,029 (34%)	119,904 (33%)	72,140 (34%)
Winter	1,527,121 (33%)	784,655 (32%)	121,633 (33%)	67,886 (32%)
Transition seasons	1,594,649 (34%)	831,153 (34%)	126,669 (34%)	71,947 (34%)

^an (%).

^bSummer: June-September; Winter: December-March; Transition seasons: April, May, October, November.

3.2. Temperature variability

Across case and control periods, the mean DTR was 7.7 °C (median 7.5 °C), ranging from 0.1 °C to 23.4 °C (10th– 90th percentiles: 3.5°C–12.2 °C). On average, DTR was lower at winter (mean 6.2 °C, 10th–90th percentiles: 2.8°C–10.3 °C) compared to summer (8.8 °C, 5.0°C–12.6 °C) and transition seasons (8.2 °C, 3.8°C–13.0 °C) (Web Fig. 5). DTR varied across the state with lower values observed in the north and middle regions, and higher values in the south (except for Suffolk County, the eastern half of Long Island, that had low DTR values during the entire year) (Figure 1). The maximal period mean DTR (10.7 °C) was observed in Upper West Side Uptown Manhattan while the lowest (2.2 °C) was seen in Remsenburg, Long Island.

Fig. 1.

Map of average daily temperature range by ZIP Code in New York State for 1995–2014, overall and by season.

3.3. Association between temperature variability and mental health-related hospital visits

For all four causes, the temperature-adjusted associations between DTR and hospital visit rates were consistently characterized by an M-shaped exposure-response curve, centered around the mean exposure value (7.7 °C) (Fig. 2 and Web Fig. 6). Starting from the minimum period DTR (0.1 °C), we observed a cumulative increase in hospital visit rates until the 25th percentile (5.2 °C). Subsequently, there was a decline until the mean DTR value (7.7 °C), followed by another increase, with the peak occurring at the 90th percentile (12.2 °C), and then a decrease for DTR levels exceeding this threshold.

Fig. 2.

Exposure-response curves of cumulative percentage change in hospital visit rates relative to minimum DTR (0.1 °C) for cause-specific mental health-related hospital visits in the lag days (0–6 days after exposure). Black lines show the point estimates and purple ribbons represent 95% confidence intervals. The histogram at the bottom shows DTR distribution across case and control days during the entire study period (1995–2014). In all panels, lowest and highest exposure percentiles (1st and 99th) were trimmed for visualization purposes. DTR; daily temperature range.

For example, for mood disorders, an increase in lag 0–6 DTR from the minimum to the 25th percentile (0.1 °C–5.2 °C) was associated with a cumulative 12.9% increase (95% confidence interval [CI]: 9.7, 16.3%) in hospital visit rates. A larger increase in DTR to the 90th percentile (0.1 °C–12.2 °C) was associated with a 16.0% increase (95% CI: 12.8, 19.2%), while an increase to the 99th percentile (15.9 °C) resulted in a smaller magnitude increase of 9.1% (95% CI: 6.0, 12.3%). Similarly, for adjustment disorders, an increase in DTR from the minimum to the 25th percentile was associated with a 26.7% increase in hospital visit rates (95% CI: 14.7, 40.0%), while increases to the 90th and 99th percentiles were associated with increases of 31.6% (95% CI: 19.6, 44.7%) and 20.0% (95% CI: 8.7, 32.5%), respectively. Percentile-specific estimates for all outcomes are shown in Web Table 2. Lag-specific associations are presented in Web Fig. 8.

3.4. Association between temperature variability and hospital visits by individual characteristics

For adjustment and schizophrenia disorders, associations varied across age groups, with a marked distinction between the youngest (0–24 years) and oldest (65+ years) age groups (Fig. 3). Generally, for both causes, across almost entire DTR distribution, the cumulative increase in hospital visit rates was more pronounced as age decreased, yet with overlapping CIs for most of the age groups, except those of the lower and upper age ranges. The most apparent differences were seen for schizophrenia disorders: an increase in DTR from the minimum to 25th percentile (0.1 °C–5.2 °C) was associated with a cumulative increase of 49.1% (95% CI: 26.8, 75.3%) in hospital visit rates among the 0–24 years group, 20.2% (95% CI: 2.4, 41.2%) among the 25–44 years group, 5.6% (95% CI: −10.3, 24.2%) among the 45–64 years group and −6.5% (95% CI: −20.5, 10.0%) among the 65+ years group. For anxiety disorders there were higher increases among the 0–24 years group, but with no clear-cut differences from the other age groups, while for mood disorders, we did not observe any differences across groups. The stratified analyses by sex and hospital visit type have not provided conclusive evidence of an effect modification (Web Figs. 8 and 9).

Fig. 3.

Cumulative percentage change in hospital visit rates by selected DTR percentiles relative to minimum DTR (0.1 °C) for cause-specific mental health-related hospital visits in the lag days (0–6 days after exposure) by age group. Points show the point estimates and whiskers represent 95% confidence intervals. DTR; daily temperature range.

3.5. Association between temperature variability and hospital visits by season

For all causes, the association between DTR and hospital visit rates was stronger in transition seasons compared with summer and winter (Fig. 4). However, these differences were particularly apparent for mood disorders, indicating distinct patterns across seasons. For example, an increase in lag 0–6 DTR from the minimum to the 90th percentile (0.1 °C–12.2 °C) was associated with a cumulative increase of 32.5% (95%CI: 26.4, 39.0%) in transition seasons compared with 10.7% (95% CI: 4.8, 16.8%) in winter and −1.6% (95%CI: −7.6, 4.7%) in summer for mood disorders; 12.2% (95%CI: 5.4, 19.5%) in transition seasons compared with 6.8% (95%CI: −0.5, 14.7%) in winter and −0.2% (95% CI: −8.0, 8.2%) in summer for anxiety disorders; 52.2% (95%CI: 29.2, 79.1%) in transition seasons, compared with 13.7% (95%CI: −5.5, 36.8%) in winter and 44.7% (95%CI: 15.8, 80.9%) in summer for adjustment disorders; and 39.5% (95%CI: 13.1, 72.0%) in transition seasons, compared with 4.0% (95%CI: −18.2, 32.2%) in winter and 26.2% (95%CI: −4.2, 66.3%) in summer for schizophrenia disorders.

Fig. 4.

Cumulative percentage change in hospital visit rates by selected DTR percentiles relative to minimum DTR (0.1 °C) for cause-specific mental health-related hospital visits in the lag days (0–6 days after exposure) by season. Points show the point estimates and whiskers represent 95% confidence intervals. DTR; daily temperature range.

4. Discussion

In this investigation exploring the relationship between DTR and hospital visit rates for mood, anxiety, adjustment, and schizophrenia disorders in NYS from 1995 to 2014, we observed non-linear associations, characterized by distinct patterns across different DTR levels. We found specific ranges of positive associations, beyond which direction of the association reversed. Strongest associations for most outcomes were seen among young individuals (0–24 years) and during transition seasons.

4.1. Plausibility of results

It is plausible that non-optimum daily temperature changes, as indicated by DTR, may exacerbate existing mental health conditions. In general, DTR-related health effects may result from a failure of circadian body temperature rhythm to synchronize with the external thermal environment (Weinert, 2010). Sharp changes in daily temperature within a few hours may disrupt the balance of the of autonomic nervous system,(Cheng et al., 2014) due to insufficient adaptation of the automatic thermoregulation system to the sudden temperature change (Zanobetti et al., 2012). An efficient temperature regulation system and appropriate behavioral measures are supposed to enable healthy individuals to cope with stress from daily temperature variation (Cheng et al., 2014). However, individuals with preexisting mental health conditions might have lower adaptation capacities,(Lim et al., 2012) and therefore may be more vulnerable to stress due to temperature change (Shiloh et al., 2009). Many medications used in psychiatry have pharmacologic effects on the parasympathetic pathway, which may impair the body’s ability to control its temperature (Martin-et al., 2007). Antidepressants, for example, such as selective serotonin-reuptake inhibitors which are used to reduce depression, anxiety and insomnia –common conditions in anxiety, mood, adjustment and schizophrenia disorders– (National Institute of Mental Health, 2023) may interfere with hypothalamic thermoregulation (Sorensen and Hess, 2022). Antipsychotics drugs, which are widely used in the treatment of schizophrenia (National Institute of Mental Health, 2023) and several mood disorders,(Rybakowski, 2023) may also interfere with hypothalamic thermoregulation,⁵⁴while anticholinergics drugs that are used to treat side effect of the antipsychotics drugs, (Ogino et al., 2014) decrease sweat production (Sorensen and Hess, 2022). Impairment of cognitive awareness is another side effect of several psychiatric medications, including sedatives (e.g., benzodiazepines), used to treat anxiety and insomnia, and mood stabilizers (e.g., lithium) used to treat several mood disorders (such as bipolar disorder) (National Institute of Mental Health, 2023). These medications may impair cognitive awareness, by reducing alertness and affecting judgment and perception of temperature,(National Institute of Mental Health, 2023) which are fundamental abilities for adaptation behaviors, such as increased fluid intake or wearing appropriate clothing (National Institute of Mental Health, 2023; Hansen et al., 2008). Given that comorbidity in psychiatric disorders is common, i.e., many symptoms overlap with those of differential diagnoses, (McElroy et al., 2001) it is plausible that DTR will have a similar effect on different conditions, through a common pathway. For example, one of the pathophysiological mechanisms underlying acute psychosis –a condition related to several mental health conditions, such as schizophrenia, bipolar disorder, and severe depression (National Institute of Mental Health, 2023)– is altered immune system function (Korhonen et al., 2023). Given that acute exposure to DTR was shown to be associated with changes in immune system function,(Lin et al., 2022) it may serve as a possible pathway through which DTR triggers acute psychosis, thereby exacerbating preexisting mental health conditions.

4.2. Previous studies

There is limited evidence of the relationship between mental health conditions and temperature variability to support our findings. A study in China found that increase in temperature variability (measured as SD of daily temperature) was associated with declined self-rated mental health, specifically with feeling nervous, upset, hopeless, and meaningless (Xue et al., 2019). A time-series analysis of DTR and suicide records in the Helsinki, Finland, reported an association between springtime DTR and suicide rates among men (Holopainen et al., 2014). In Hefei, China, where DTR distribution was slightly higher than current analysis (median DTR 8.3 °C), a positive association between DTR and schizophrenia admissions was observed from the minimum DTR until approximately 7.0 °C and then from the 75th percentile (11.0 °C) up to the end of the distribution (Zhao et al., 2016). The main difference from our exposure-response curve is in extreme DTR values, where the positive association has reversed. This contrasting finding may be explained by the different model structure which yielded less flexible exposure-response curves. The higher flexibility of the models in the current analysis enabled us to detect nuances of the curve, i.e., more specific ranges of positive associations.

The association of temperature variability and mental health conditions might occur in two ways: elevated DTR may have a direct effect on exacerbation of the mental health condition and an indirect effect through triggering a new or an existing physical condition, which eventually may lead to exacerbation of the mental health-related condition (M et al., 2011). In both scenarios, other outcomes than mental health hospital visits (e.g., non-mental hospital visits or death) may occur. These outcomes should be considered while interpreting our results, particularly in relation to the steep decline in association observed at extremely high DTR levels. This pattern of a negative association at extreme weather, which has been also previously reported for the association between temperature and injuries-related hospital admissions, (Rizmie et al., 2022) may reflect a competing risks situation: a decrease in hospital visits that not necessarily reflect a protective effect, but outcomes that “replace” the outcome of interest. Previous analyses have shown that the impact of extreme weather on mortality is not paralleled by similar magnitude increases in hospital admissions,(Dang et al., 2019; Kovats et al., 2004) which supports the hypothesis that many weather-related deaths occur rapidly or among isolated people before they seek medical attention. The negative association we observed at high DTR may be explained by other outcomes that prevent individuals with mental health conditions to arrive to the hospital and receive psychiatric treatment.

Results of our subgroup analysis provide additional insights. We observed the strongest association in transition seasons compared to summer and winter. A similar pattern was observed in South Korea for DTR-mortality association, with a stronger association in the fall, (Lim et al., 2012) and in Finland for DTR-suicide association detected only in spring (Holopainen et al., 2014). It is plausible that marked variability in daily temperature during transition seasons, when day-to-day variability is already high, makes the physiological and behavioral adjustment even more difficult, particularly for high-risk populations. We also observed differences across age groups, with strongest associations seen among the youngest group (0–24 years) and almost no association seen in the oldest group (65+ years), particularly for adjustment and schizophrenia disorders. The mechanism behind this finding is not clear. However, as major differences exist in clinical characteristics of adolescents vs. older adults hospitalized with these conditions (such as pre-existing comorbidities and medical treatment), (Greenberg et al., 1995; Folsom et al., 2006) it is possible that one or more of these factors are involved with a physiological or a behavioral pathway through which DTR may trigger these conditions. Future analysis may benefit from finer scale age classification, conducted with shrinkage and statistical power considerations, to detect the most vulnerable age groups in the population. We haven’t observed a differential association between in-patient and out-patient visits. As severity of the mental health condition is a major factor predicting referral to in-patient rather than out-patient treatment,(Way et al., 1992; Al et al., 2012) thus expected to increase one’s vulnerability, this result is not completely clear. However, it is possible that similar to absolute temperature that was shown to induce both out-patient (Zhang et al., 2020) and in-patient (Sung et al., 2013) visits, also DTR serves as a triggering factor for exacerbation of mental health condition on different scales of severity.

5. Strengths and limitations

Leveraging complete hospital records data from NYS for 20 years along with a comprehensive record of ZIP Code-level daily temperatures, the current study explored the association between daily temperature variability and mental health outcomes. Several limitations of the current investigation should be acknowledged. First, although the NLDAS-2 model well-captures temporal contrasts in exposure, which are crucial for conducting within-individual comparisons in a case-crossover analysis, (Cheung et al., 2019) its spatial resolution (~11 km × 14 km in NYS) may overlook important fine-scale spatial variations of temperature in urban areas, particularly those due to the urban heat island effect (Bernhard et al., 2015). Relying on coarse spatial resolution of temperature predictions might introduce exposure measurement error, (Spangler et al., 2019) which may be primarily attributed to Berkson error. Berkson error may arise due to employing a single value of temperature to represent the exposure of numerous individuals residing across municipalities in which there is spatial variability in daily temperatures, and is expected to lead to decreased precision of effect estimates (Weinberger et al., 2019). Several techniques have been developed to overcome this challenge, including satellite-based land surface temperature prediction, machine learning models, as well as ensemble models that combine these approaches (Jin et al., 2022). Future analysis may benefit from using these approaches to improve accuracy of temperature prediction and thus provide a more comprehensive understanding of heat dynamics in epidemiological settings.

Ascertainment bias is also of concern. Considering that only cases that ended in hospital visit are addressed, more minor cases that do not reach out for treatment (e.g., less severe symptom exacerbation), cases treated in primary care, or most severe cases that have resulted in deaths without arriving to medical attention, are not included in the current analysis. As a result, the outcome may not fully capture the spectrum of outcomes related to mental health-related conditions. In addition, since information on medical history of study’s participants is not available, uncertainty exists about whether study population may be considered a vulnerable population of individuals with preexisting mental health conditions. However, given that most of mental health conditions are managed in primary care, cases included in current analysis are likely to be severe cases that could not be managed outside the hospital,(Hamer et al., 2008) and therefore assumed to be of a group of chronic sufferers. Another limitation arises since the exposure metric we used ignores the temporal sequence of the temperature values during a 24-h period, considered an important component of temperature variability (i.e., the pattern of the change of hourly temperature within one day) (Rowland et al., 2021). However, as high correlation was previously reported between exposure estimates derived from different temperature variability metrics in NYS (accounting and not accounting for temperature temporal sequence), with all methods pointing to associations in the same directions, (Rowland et al., 2021) we also expect that our results would be similar had we used other metrics. Also, residual confounding should be taken into consideration. Although the inherent nature of the design of current study, which uses a time-stratified case-crossover structure, controls for factors that vary across individuals, as well as day of the week and month (which covary with both hospital visit rates and ZIP-Code level DTR), residual confounding by unknown or unmeasured factors cannot be ruled out. For example, solar parameters such as daylength and sunshine duration which are associated with both DTR (van den et al., 2015) and mental health conditions,(Raza et al., 2024; Ji et al., 2023) were not explicitly examined as an effect modifier of the exposure-response relationship, though implicitly controlled for by the case-crossover design. Future studies may benefit from considering these variables as potential effect modifiers in the DTR-mental health association. Finally, this study was focused on NYS. As DTR greatly varies across different geographical regions,(Guo et al., 2021) the association observed here may not be generalizable to other communities with a different DTR distribution as well as with different distribution of potential modifiers; future investigations should explore DTR-mental health relationship in other climates and communities.

Our work highlights how mental health-related hospital visits may be impacted by changes in daily temperature variability, highlighting the increased vulnerability of individuals with preexisting mental health conditions to the adverse effects of weather. Given uncertainty on how climate change modifies temperature variability, additional research is crucial to comprehend the implications of these findings, particularly under different scenarios of future temperature variability. It is a public health priority to enhance population resilience for changing climate particularly among vulnerable populations. Daily temperature variability, alongside other weather-related exposures, is likely a potential risk factor for worsening preexisting mental health conditions.

Supplementary Material

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envres.2024.119238.

Glossary

CCS

Clinical Classifications Software

DTR

Daily temperature range

DLNM

Distributed lag non-linear model

ED

Emergency department

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

NYS

New York State

NLDAS

North American Land Data Assimilation System

SD

Standard deviation

SPARCS

Statewide Planning and Research Cooperative System

ZCTA

ZIP Code Tabulation Area

ZIP

Zone Improvement Plan

Data availability

Temperature data are downloadable from NASA website. Health data can be requested through submission of a proposal to the NYS Department of Health. Code will be publicly available via GitHub.