Association of ambient temperature with mortality in community-based patients with severe mental disorders: a time-stratified case-crossover study in western China

Abstract

Background

Previous studies have indicated that high temperatures are associated with excess mortality among individuals with mental disorders, but comprehensive evaluations of the association between high and low temperatures, extreme temperature events and mortality in patients with severe mental disorders are limited.

Methods

A time-stratified case-crossover study was conducted using mortality data from 22,342 deaths among community-based patients with severe mental disorders in western China between 2006 and 2018 (11,235 during hot seasons and 11,107 during cold seasons). Individual-level exposure to high temperatures, heat waves, low temperatures, and cold spells was assessed, and the associations between these ambient temperatures and mortality were estimated using conditional logistic regression models.

Results

High temperatures during the hot seasons were associated with a 23.66% increased risk of all-cause mortality (95% CI, 15.17%–31.66%), with effects diminishing as the lag period increased. Low temperatures during the cold seasons showed a significant association with mortality at lag day 4, peaking at 15.25% (95% CI, 3.92%–25.72%) at lag day 6. Heat waves were associated with increased mortality risk, particularly with prolonged exposure and higher temperature thresholds. Cold spells did not show a similar pattern.

Conclusions

Both heat and cold related exposures are associated with higher mortality risk in community-based patients with severe mental disorders, but their temporal patterns differ—heat has an immediate effect, whereas cold acts with a delay. Our findings suggest that these patients should be prioritized in weather-related health policies, including heat–health and cold–weather warning systems, proactive follow-up by community mental health services, and tailored protection measures during forecasted extreme temperature events.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07672-9.

Background

Premature death among patients with mental disorders is an important public health issue of global concern. As of 2019, the number of people with mental disorders globally was approximately 970 million [1], with about 17% of cases occurring in China [2]. It is estimated that the number of people with severe mental disorders (SMDs) in China has exceeded 6.4 million [3], mainly including schizophrenia, schizoaffective disorder, paranoid psychosis, bipolar affective disorder, epilepsy-induced mental disorders, and mental retardation with mental disorders [4]. Due to severe complications and adverse behavioral habits [5, 6], the mortality rate among these patients is more than twice that of the general population [7, 8]. Furthermore, for individuals with SMDs, medical and social care expenses exceed those of non-SMDs patients by more than 2.5 times, while mental health service costs are as high as nine times greater [9]. Therefore, reducing the disease burden associated with the premature death of these SMDs patients is a critical area that deserves in-depth study.

It is well known that the factors influencing the mortality of mental disorders are numerous and complex, with environmental factors, particularly temperature, receiving increasing attention in recent years. As a key environmental exposure, ambient temperature has been shown in numerous studies to be associated with various adverse outcomes in mental disorders. Patients with SMDs may be particularly susceptible to temperature-related effects. Psychotropic medications can impair thermoregulation, and functional or cognitive limitations may slow their behavioral responses to heat or cold. In addition, common physical comorbidities in this population can be acutely worsened by temperature extremes, which can increase the risk of mortality. However, most previous studies have used the outpatient visit [10, 11] or hospitalization [12–14] of patients with mental disorders as the outcome of interest, and few studies have examined the impact on death.

Existing studies on the association between ambient temperature and mortality in patients with mental disorders have the following limitations. First, regarding research objectives, most studies have primarily focused on the effects of high temperatures on mortality in psychiatric populations [15–19]. Based on our search, studies that comprehensively assessed high temperatures, low temperatures, and extreme temperature events (ETEs) in relation to mortality in this population have not yet been reported. Second, concerning the study population, previous research have predominantly focused on patients with dementia [15, 18, 20, 21] or other general psychiatric disorders [16, 22]. To our knowledge, no study has examined the association between ambient temperature and mortality specifically in community-managed patients with SMDs. Finally, in terms of study outcomes, most existing studies have concentrated on the effects of temperatures on all-cause mortality in patients with mental disorders, while few have systematically evaluated the impact on both all-cause and cause-specific mortality [22].

To fill these research gaps, we conducted a time-stratified case-crossover study based on a large-scale prospective cohort survey in western China. To our knowledge, this is the first study to comprehensively assess high temperatures, low temperatures, and extreme temperature events in relation to mortality among community-based patients with severe mental disorders using individual-level temperature assignment. We analyzed the associations and lag effects of high and low temperatures with all-cause and cause-specific mortality among community-based patients with SMDs from 2006 to 2018, and further explored the effects of heat waves and cold spells on patient deaths.

In addition, considering that ambient temperature may have a more severe impact on special populations, we also stratified the analysis by demographic characteristics such as age, gender, and economic status. This study aims to provide scientific evidence and empirical support for public health authorities in managing and intervening with community-based patients with SMDs in the context of global climate change.

Methods

Study population

The dataset was from the Integrated Management Information Platform for Severe Mental Disorders (hereinafter referred to as the Platform) in western China, which was established in 2006. It utilizes the 10th revision of the International Classification of Diseases (ICD-10) to identify patients with SMDs. According to the work specification for the management and treatment of severe mental disorders [4], SMDs included in community-based comprehensive management encompass epilepsy-induced mental disorders, schizophrenia, paranoid psychosis, schizoaffective disorder, bipolar affective disorder, and mental retardation with mental disorders. The corresponding ICD-10 codes for each disease are provided in Table S1. Further details about the Platform, including its study design and subject selection, can be found in Li Y, et al. [23].

We used the de-identified data for the period from May 1, 2006 to December 31, 2018, with 451,915 community-based SMDs patients registered throughout the follow-up period, and a total of 29,816 deaths of patients were recorded after excluding cases with logical errors. Individuals with baseline address errors, missing addresses, and individuals who moved out of the area were further excluded. In addition, 7,443 deaths that occurred in other months (March, April, and October) were excluded because the main analyses focused on the hot season (May–September) and the cold season (November–February). The final analytic sample therefore included 22,342 deaths, of which 11,235 occurred in the hot season and 11,107 occurred in the cold season, and the screening flowchart is shown in Figure S1. This study was approved by the Research Ethics Committees of Sichuan University and the Sichuan Mental Health Center. At enrollment in the community-based management and treatment program for severe mental disorders, mental health staff explained the service content and the rights and obligations of patients and families; written informed consent was obtained from the patient or a legal guardian before registration, in accordance with the Declaration of Helsinki. For the present analysis, all potentially identifiable information was removed and a de-identified dataset was used.

Outcomes

The primary outcome was death between 1 May 2006 and 31 December 2018. Deaths were ascertained through routine platform follow-up: family members provided the official death certificate issued by a medical institution or public security authority, and follow-up personnel at the mental health centers recorded the date and cause of death. Cause of death was determined from the death certificate and then coded into six mutually exclusive categories according to prespecified rules: physical disease, complications related to mental illness, suicide, accident, homicide and other.

Exposure

We obtained daily mean temperatures at 2 m from the ERA5-Land [24] dataset in European Centre for Medium-Range Weather Forecasts (ECMWF) at a 9 × 9-km resolution (0.1° × 0.1°). The ERA5-Land dataset is a reanalysis dataset providing a consistent view of the evolution of land variables over several decades with enhanced spatial and temporal resolution. We linked temperatures to the 0.1° grid cell corresponding to each patient’s registered residential address to construct individual daily temperature time series. High temperature was defined as above the 99th percentile of hot-season temperatures, and low temperature as below the 1st percentile of cold-season temperatures. ETEs are generally defined as daily mean temperatures exceeding or falling below specific temperature thresholds for several consecutive days [25–29]. We defined ETEs by combining intensity (92.5th, 95th, and 97.5th percentile of the mean temperatures in hot seasons for heat waves and 10th, 5th, and 2.5th percentile of the mean temperatures in cold seasons for cold spells) and duration (lasting 2, 3, 4 days), 9 metrics were defined for heat waves and cold spells, respectively. The corresponding indicator variables of ETEs were set to 1 if a patient had experienced extreme weather and 0 if he had not.

Statistical analysis

We estimated the associations of high and low temperatures in the week before death and of ETEs with mortality using a time-stratified case-crossover design [30]. For each death day, we selected 3–4 referent days defined as the same weekday within the same calendar month, using a bidirectional time-stratified scheme that yields unbiased conditional-logistic estimates and controls for seasonality and long-term trends in environmental applications [30]. A minimum 7-day separation between death and referent days was imposed to reduce serial autocorrelation in exposures and outcomes [31]. For each death and referent day, we prespecified a 0–6-day lag window to capture immediate and short-term delayed effects commonly examined in temperature–mortality studies. We extracted the daily mean temperature for the index day (lag 0) and for the preceding 1–6 days (lags 1–6). Considering that the effects of temperatures on patient deaths may be different in the hot and cold seasons, our study respectively explored 11,235 deaths during the hot seasons (May to September) and 11,107 deaths during the cold seasons (November to February) in the main analysis according to the actual climatic conditions of the research area, as detailed in Figure S1.

In this study, the mean temperatures from the day of death to n days prior to death (n: 0–6 days) were used as the lag temperature exposures. A conditional logistic regression model was employed to estimate the association between high or low temperatures and mortality of SMDs patients. As previous studies have found a U- or J- shaped nonlinear relationship between temperatures and mortality [12, 32–35], we modeled the exposure-response curve using a restricted cubic spline with three knots at the 10th, 50th, and 90th percentiles. All models were also adjusted for whether the date of death fell on a national holiday, included as a dichotomous variable. The odds ratios (ORs) for death associated with high and low temperatures was reported, calculated as the odds at the 99th percentile of temperature during the hot seasons and the 1st percentile of temperature during the cold seasons, compared to the respective 50th percentile of temperature. To quantify the impact of heat waves and cold spells in addition to temperatures, the models adjusted the average temperatures during ETEs to specifically capture the impact of ETEs on the risk of death in patients with SMDs [36]. Further, considering the interpretability of the findings from a public health significance, results were expressed in terms of both the attributable risk (AR) and the absolute risk difference (ARD) [37–39] of death associated with high temperatures, low temperatures, and ETEs among patients with SMDs. AR was defined as (OR − 1)/ OR × 100%, and ARD was defined as α×(OR-1)/OR, where α denotes the baseline daily mortality rate, and OR is the estimated OR values. α was calculated as the total number of deaths during the hot and cold seasons among all SMDs patients, divided by the total person-days experienced by these patients during the respective seasons, and multiplied by 100,000 [38]. Monte Carlo simulations were used to obtain 95% empirical confidence intervals (eCIs).

For the sensitivity analyses, we implemented the following strategies: (1) redefined high temperatures in the hot seasons as those above the 95th and 90th percentiles and low temperatures in the cold seasons as those below the 10th and 5th percentiles. (2) changed the reference temperatures for the hot seasons to the 75th percentile and for the cold seasons to the 25th percentile. (3) adjusted the knot positions of the restricted cubic spline and adjusted the number of knots to four. (4) used alternative thresholds to define heat waves (above the 90th percentile) and cold spells (below the 7.5th percentile). (5) modified the definitions for the hot and cold seasons, defining the hot seasons as June to September and the cold seasons as November to March, or conducted analyses without distinguishing different seasons, including all deaths of SMDs patients throughout the year. (6) additionally adjusted for daily mean relative humidity and PM2.5 as covariates in the conditional logistic models.

The outcome for the main analysis above was all-cause mortality of SMDs patients. In addition, we repeated the analyses based on different causes of death among SMDs patients. We also conducted the stratified analyses by disease type, sex, age, marital status, residential area, education level, economic status, and disease duration. Furthermore, 2-sample z-tests were applied to test the differences between the two subgroups [37, 40]. All P values were 2-tailed and less than 0.05 was considered statistically significant. Subgroup analyses were exploratory; therefore, no formal adjustment for multiple comparisons was performed. All statistical analyses were done with R (version 4.4.0).

Results

The main analysis of this study included 22,342 SMDs patients who died between 2006 and 2018, of whom 11,235 died during the hot seasons and 11,107 during the cold seasons. This study population had a mean (SD) age of 60.89 (16.13) years and a mean (SD) illness duration of 18.79 (14.75) years, and 54.8% were male. The leading causes of death were physical diseases (12,582 cases, 56.3%), complications related to mental disorders (1,980 cases, 8.9%), and accidents (2,047 cases, 9.2%), as shown in Table 1. The mean temperature (SD) on the day of death was 24.17 °C (3.66 °C) in the hot seasons and 8.76 °C (3.80 °C) in the cold seasons. The distribution of temperatures during the lag period is detailed in Table S2. In this study, 12.74% and 25.30% of the total study population experienced heat waves and cold spells during the hot and cold seasons, respectively, as shown in Table S3.

Table 1.

Characteristics for the death cohort of SMDs patients

	Total a	Hot seasons b	Cold seasons b
Case	22,342	11,235	11,107
Control	75,855	38,663	37,192
Age, mean (SD), years	60.89(16.13)	60.31(16.14)	61.47(16.10)
Sex
Male	12,243(54.8%)	6153(54.8%)	6090(54.8%)
Female	10,099(45.2%)	5082(45.2%)	5017(45.2%)
Duration of illness, mean (SD), years	18.79(14.75)	18.82(14.67)	18.77(14.83)
Marriage
Married	12,471(55.8%)	6350(56.5%)	6121(55.2%)
Unmarried c	8941(40.0%)	4428(39.4%)	4513(40.6%)
Not available d	930(4.2%)	457(4.1%)	473(4.3%)
Education level
Educated	11,417(51.1%)	5865(52.2%)	5552(50.0%)
Uneducated e	9668(43.3%)	4735(42.1%)	4933(44.4%)
Not available	1257(5.6%)	635(5.7%)	622(5.6%)
Economics
Poverty	14,012(62.7%)	7018(62.5%)	6994(63.0%)
Non-poverty	4321(19.3%)	2213(19.7%)	2108(19.0%)
Not available	4009(17.9%)	2004(17.8%)	2005((18.0%)
Rural-Urban
Rural	17,899(80.1%)	8994(80.1%)	8905(80.2%)
Urban	3142(14.1%)	1590(14.2%)	1552(14.0%)
Not available	1301(5.8%)	651(5.7%)	650(5.8%)
Disease type
Epilepsy-induced mental disorders	1261(5.6%)	661(5.9%)	600(5.4%)
Schizophrenia	18,032(80.7%)	9045(80.5%)	8987(80.9%)
Paranoid psychosis	73(0.3%)	43(0.4%)	30(0.3%)
Schizoaffective disorder	310(1.4%)	161(1.4%)	149(1.3%)
Bipolar affective disorder	701(3.1%)	351(3.1%)	350(3.2%)
Mental retardation with mental disorders	1965(8.8%)	974(8.7%)	991(8.9%)
Cause of death
Physical disease	12,582(56.3%)	6200(55.2%)	6382(57.5%)
Complications related to mental illness	1980(8.9%)	934(8.3%)	1046(9.4%)
Suicide	800(3.6%)	435(3.9%)	365(3.3%)
Accident	2047(9.2%)	1179(10.5%)	868(7.8%)
Homicide	14(0.1%)	8(0.1%)	6(0.1%)
Other	4919(22.0%)	2479(22.1%)	2440(22.0%)

^a Patients with six different disease types died in both the hot and cold seasons

^b The hot seasons are defined as the period from May to September and the cold seasons as the period from November to February each year

^c Unmarried on behalf of unmarried, widowed and divorced

^d Demographic characteristics of this group were not collected and therefore were excluded from the subgroup analyses

^e Uneducated stands for illiterate and semi-literate

Figure 1 shows the exposure–response curves for different lag days in the hot and cold seasons, and confirms a nonlinear relationship between ambient temperature and all-cause mortality in this population. In the hot season, the curve is relatively flat at lower temperatures and then begins to rise more sharply once the daily mean temperature exceeds about 24.2 °C, indicating a high-temperature threshold. In the cold season, the mortality risk remains nearly stable at milder cold-season temperatures but begins to increase when temperatures fall below about 8.5 °C, suggesting a low-temperature threshold. Figure 2A shows that in the hot seasons, the association between high temperatures and all-cause mortality is strongest on lag day 0 (OR = 1.31, 95% CI, 1.18–1.46), and the effect diminishes with increasing lag days. In contrast, during the cold seasons, the effects of low temperatures on mortality first appears at lag day 4 (OR = 1.13, 95% CI, 1.00–1.28), and increases as the number of lag days growing, reaching the maximum at lag day 6 (OR = 1.18 (95% CI, 1.04–1.34). For cause-specific analyses, Table S4 presents a similar pattern of effects to the main analysis for deaths due to physical diseases, and no significant effects for other causes.

Fig. 1.

Exposure-response curves for temperatures with varying lag days in hot and cold seasons associated with mortality in SMDs patients. HT means high temperature. LT means low temperature. lag0-n indicates that the mean values of temperature from the day of death to n days before the onset of death are used as the exposure. All high temperatures are analyzed in the hot seasons (from May to September) and low temperatures are analyzed in the cold seasons (from November to February)

Fig. 2.

Association between temperatures and extreme temperature events and the risk of all-cause mortality in SMDs patients. We show the lag effects of high and low temperatures on all-cause mortality in SMD patients in panel A and the impact of heat waves and cold spells in panel B. HT means high temperature, LT means low temperature. lag0–n indicates that the mean values of temperature from the day of death to n days before the onset of death are used as the exposure. Heat waves and cold spells are defined using the combination of intensity and duration; for example, HW P92.5 2d denotes heat waves with daily mean temperature ≥ 92.5th percentile of the temperature distribution and with ≥ 2 consecutive days. CS P10 2d denotes cold spells with daily mean temperature ≤ 10th percentile of temperature distribution and with ≥ 2 consecutive days. All high temperatures and heat waves are analyzed in the hot seasons (from May to September), low temperatures and cold spells are analyzed in the cold seasons (from November to February)

Figure 2B shows that heat waves and cold spells are linked to higher all-cause mortality in SMDs patients. The risk of death in SMDs patients is generally higher when exposed to heat waves with higher temperature thresholds. Furthermore, when heat waves are defined using the same thresholds, the effects on mortality increase with longer exposure, particularly at more stringent thresholds (P95, P97.5). In contrast, this pattern is not observed for cold spells. For cause-specific analyses (Table S5), this study finds that heat waves significantly increase the risk of deaths from physical diseases and accidental deaths, whereas cold spells are only associated with deaths from physical diseases.

For high and low temperatures, high temperatures at lag 0 days and low temperatures at lag 6 days were associated with the largest effects. At lag 0 days, high temperature had an AR of 23.66% (95% eCI, 15.17%–31.66%); at lag 6 days, low temperature had an AR of 15.25% (95% eCI, 3.92%–25.72%). The corresponding ARDs were 1.30 (95% eCI, 0.83–1.74) and 1.03 (95% eCI, 0.26–1.74) per 100,000 person-days, respectively. The associations remained significant when we examined heat waves and cold spells. The largest effects were observed for heat waves defined as temperatures above the 97.5th percentile of daily mean temperatures for four consecutive days, and cold spells defined as temperatures below the 10th percentile of daily mean temperatures for three consecutive days, with ARs of 22.12% (95% eCI, 8.34%–34.36%) and 11.19% (95% eCI, 5.06%–17.20%), and ARDs of 1.21 (95% eCI, 0.46–1.89) and 0.76 (95% eCI, 0.34–1.16) per 100,000 person-days, respectively (Table 2). The results of the remaining scenarios are detailed in Table S6.

Table 2.

Odds ratio and attributable risk at maximum mortality risk ^a

	OR b (95%CI)	AR (95%eCI)	ARD (95%eCI)
High temperatures	1.31 (1.18–1.46)	23.66% (15.17%-31.66%)	1.30 (0.83–1.74)
Low temperatures	1.18 (1.04–1.34)	15.25% (3.92%-25.72%)	1.03 (0.26–1.74)
Heat waves	1.28 (1.09–1.51)	22.12% (8.34%-34.36%)	1.21 (0.46–1.89)
Cold spells	1.13 (1.05–1.21)	11.19% (5.06%-17.20%)	0.76 (0.34–1.16)

OR, odds ratio. CI, confidence interval. AR, attributable risk. ARD, absolute risk difference. eCI, empirical confidence interval

^a The table presents results for high temperatures at lag 0 days (HT lag 0–0), low temperatures at lag 6 days (LT lag 0–6), heat waves defined as continuous 4-day periods with temperatures above the 97.5th percentile (HW P97.5 4d), and cold spells defined as continuous 3-day periods with temperatures below the 10th percentile (CS P10 3d). High temperatures and heatwaves are analyzed in the hot seasons (from May to September), while low temperatures and cold spells are analyzed in the cold seasons (from November to February)

^b The ORs of high and low temperatures are computed by comparing the 99th and 1st percentile temperatures with the respective 50th percentile temperatures

Subgroup analyses are summarized in Fig. 3 and Tables S7–S10. Although point estimates varied across disease types and demographic subgroups, formal between-group comparisons using 2-sample z-tests did not detect statistically significant heterogeneity (Table S11). Sensitivity analyses yielded similar inferences across alternative specifications (Tables S12–S16; Figures S2–S3).

Fig. 3.

Stratified analysis of temperature and extreme temperature events on death in SMD patients. For high and low temperatures, we present (HT lag 0–0), (LT lag 0–6); and for ETEs, we show (HW P97.5 4d), (CS P10 3d). Unmarried* on behalf of unmarried, widowed and divorced. The effects of paranoid psychosis are not estimated because the number of deaths is too small.

Discussion

As far as we know, this is the first comprehensive study to assess the association of different types of ambient temperature with all-cause and cause-specific mortality in SMDs patients. Our findings suggest that both high and low temperatures, as well as ETEs including heat waves and cold spells, are linked to an increased risk of mortality in SMDs patients. Specifically, the impact of high temperatures is immediate, whereas low temperatures show delayed effects. Heat waves, characterized by higher thresholds and longer durations, are generally associated with higher mortality risks for SMDs patients.

First, regarding the impact of high temperatures and heat waves, our results are consistent with previous studies reporting a positive temperature–mortality relationship in people with mental disorders. A 2018 study in the UK [18] reported that for every 1 °C rise above the 93th percentile temperature, the mortality rate of mental disorder patients increased by 4.9% (95%CI, 2.0%-7.8%). Similarly, a 2020 Italian study [22] found a 5.5% (95%CI, 2.4%-8.6%) rise in mortality risk for mental disorder patients per 1 °C increase when temperature exceeded 24 °C. A 2021 systematic review [17] also indicated that for each 1 °C rise in temperature, mental health-related mortality increased (RR = 1.022, 95%CI, 1.015–1.029). In our study, compared to the 50th percentile temperature (24.28 °C), the AR of death at the 99th percentile (31.75 °C) was 23.66% (95% eCI, 15.17%-31.66%) and the ARD was 1.30 (95% eCI, 0.83–1.74). This means that, compared to the reference, 23.66% of the deaths can be attributed to high temperatures, and the number of deaths exceeds the reference by 1.30 deaths per 100,000 person-days in SMDs patients. Although the effect metrics are not identical to those of the above studies, the direction is consistent. Regarding heat waves, our study finds that the mortality risk for SMDs patients increased, with the maximum AR being 22.12% (95% CI, 8.34%-34.36%). Similar findings have been reported in Australia [16, 41], Sweden [42], and China [15]. Several possible mechanisms have been reported in previous studies and may provide biological plausibility for our findings. High temperatures may disturb the balance of neurotransmitters such as serotonin and dopamine, which are important for mood regulation, cognition and complex tasks, and this could make patients with severe mental disorders more unstable [17, 43, 44]. During heat waves, physiological and behavioral adaptation may also be impaired, which could contribute to higher risks of irritability, psychological distress, aggression, violence, and suicide [45–47]. The medications that patients with mental disorders use (e.g., risperidone, olanzapine) may alter their thermoregulation capacity through pharmacologic effects on parasympathetic pathway. This may reduce their sensitivity to environmental heat signals, delays or diminishes the response to heat stress, which could in turn increase their risk of adverse outcomes [16, 48, 49]. In addition, human experimental studies and animal models have reported adverse effects on cognitive function, emotional state, and psychological performance in humans after experiencing heat exposure [50]. For example, significant neuroinflammatory responses and neuronal cell death occurred in the hippocampus of the brain in mice after heat exposure [51] and in the hypothalamic region of the rat [52]. These findings indirectly indicate that heat-related neural changes may lower patients’ awareness or capacity to cope with extreme temperatures, making them more susceptible to heat [16, 53].

Second, the results of this study suggest that low temperatures and cold spells also are associated with a higher risk of death in patients with SMDs. Most existing research on the effects of cold on mental disorders have focused on hospital admissions due to the exacerbation of symptoms [13, 54, 55]. However, the available evidence generally supports the idea that cold exposure may lead to hospital admissions as patients’ conditions worsen, which in turn increases the risk of death. This finding is consistent with our results. Existing studies on cold exposure and mortality in patients with mental disorders are only mentioned in some evaluations of the burden of death due to low temperatures or cold spells. A 2013 Swedish study [42] reported that the risk of death for patients with a history of hospitalization for mental disorders, who were exposed to low temperatures during the cold seasons, was 1.065 (95% CI, 1.019–1.113). Similarly, a 2021 study from Guangzhou, China [56] found that the risk of death from cold spells in individuals with mental disorders was 2.02 (95% CI, 0.78–5.22). However, the underlying mechanisms of how cold exposure affects patients with mental disorders remain unclear. Some studies suggest that low temperatures may trigger inflammatory responses and oxidative stress, which could contribute to the development of mental disorders [57, 58]. Additionally, patients with mental disorders often have comorbid physical diseases [59] and may be more sensitive to cold exposure. Combined with common self-management deficits and lower socioeconomic status, these factors can hinder timely care and may further increase the risk of death [60].

The third concern of this study is the lag effects of ambient temperature on mortality. Similar to the findings of previous studies [61–63], our results reveal that in the hot seasons, the effects of high temperatures appeared and maximized on the day of death and tended to decrease with increasing lag days. In contrast, the effects of low temperatures in the cold seasons occurred with a lag, and the effects tended to strengthen as the number of lag days increased [28, 64–66]. One potential explanation for the difference in patient response to the cold and heat effects could be attributed to the underlying causes of death, with high temperatures likely to be primarily associated with more acute causes of death, such as cardiovascular disease [67], whereas the effects of low temperatures on health risks may be more associated with subacute causes of death, such as respiratory disease [68]. Due to the lack of disease-specific information on the causes of death from physical diseases in this study, further exploration was not possible. Notably, the mechanism of delayed cold effects might be that brief exposure to low temperatures can induce peripheral vasoconstriction and sympathetic activation, increasing blood pressure, hemoconcentration, and platelet activity, thereby heightening susceptibility to thrombotic events over subsequent days [28, 65, 69]. Concurrently, more time spent indoors, greater viral stability at low temperatures, and impaired mucociliary clearance favor respiratory outcomes that manifest after a lag [70].

Additionally, previous research has shown a positive correlation between the duration and intensity of ETEs and their negative impact on patients [20, 26]. In our study, this association is observed in heat waves but not in cold spells. This discrepancy may be explained by several factors. First, winters in our study region are relatively mild, so percentile-based cold cut-offs correspond to modest absolute temperatures, making cold effects harder to detect. Second, routine cold-protective behaviors (e.g., wearing warmer clothing, using indoor heating, spending more time indoors) likely reduce both the intensity and variability of personal exposure, so ambient outdoor temperature may not reflect actual exposure, further attenuating observable effects. Third, cold-related impacts are often delayed and protracted; within our 0–6-day lag window, such associations may not be observed.

This study has three main strengths. First, it is, to our knowledge, the first study to systematically examine high temperatures, low temperatures, and extreme temperature events in relation to both all-cause and cause-specific mortality in patients with SMDs. Second, we used 12 years of mortality data from a large community-based SMD registry and linked it with individual-level temperature exposure, which improves exposure accuracy compared with studies using area-level monitoring data. Third, we applied a time-stratified case-crossover design and conducted multiple sensitivity analyses, which helped control individual-level confounding and confirmed the robustness of our findings.

However, there are several limitations in interpreting the results of this study. First, the mortality data were collected from only one region in western China, so caution should be exercised when extrapolating the findings to other regions with different climatic conditions. Second, when analyzing different causes of death and SMDs subtypes, the sample sizes of some categories were small, which may have resulted in insufficient statistical power. In the future, we plan to conduct a multicenter study to increase the sample size for each category and further explore these associations in greater depth. Third, ambient temperature was assigned according to patients’ registered residential addresses, and we assumed that patients stayed at or near their registered address during the relevant days. Therefore, individual mobility and temporary stays elsewhere could not be captured, which may have led to a certain degree of exposure misclassification. Fourth, we used moving average temperatures to assess lagged effects, which assumes a similar impact of temperature across days within the lag window. This simplification may smooth day-specific associations. Finally, although many confounders were controlled by the study design and we further adjusted for relative humidity and PM 2.5 in sensitivity analyses, residual time-varying confounding cannot be ruled out. Study-wide, spatiotemporally compatible data for other air pollutants were unavailable and thus could not be included. In addition, due to data availability, we could not compare temperature-related vulnerability between patients with SMDs and those with less severe mental disorders; future work incorporating severity indicators and broader patient populations is needed to address this question.

Conclusions

This case-crossover study found that high temperatures, low temperatures, and extreme temperature events were all associated with a higher risk of death among community-based SMDs patients. Because the data were derived from a single region in western China with its own climatic and service context, these findings are most directly applicable to similar settings. As global climate change becomes more frequent and intense, our findings may provide a scientific basis for government agencies, mental health care providers, and community managers to develop intervention policies for patients with severe mental disorders during adverse climatic conditions. These policies may help minimize the adverse effects of ambient temperature on these patients and reduce the mortality burden associated with temperature-related conditions.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

B Yi, C Zuo, Y Liu, X Zhao, and X Liu were involved in the study design. S Chen, X Yang, J Liu and R Fan participated in data collection. B Yi, C Zuo and Y Liu developed the analysis plan. B Yi, C Zuo, Y Liu, X Zhao, and X Liu had access to all the data and performed the data analyses. B Yi and C Zuo verified the underlying data. All authors write up of and reviewed the manuscript. B Yi, C Zuo, Y Liu, X Zhao, and X Liu and X Yang had final responsibility for the decision to submit for publication.

Funding

Science and Technology Program of Sichuan province [grant number 2025ZNSFSC0782]. National Natural Science Foundation of China [grant number 81903414].

Data availability

The data used in this study come from the community management platform for severe mental disorders and contain identifiable or potentially re-identifiable health information. According to the regulations of Sichuan University and the Sichuan Mental Health Center, these data cannot be made publicly available. De-identified data may be made available from the corresponding author on reasonable request and with permission from the data-holding institution.

Declarations

Ethics approval and consent to participate

The study protocol was reviewed and approved by the Research Ethics Committee of Sichuan University, West China School of Public Health/West China Fourth Hospital (Ethics approval No. HXGW-EC-Y2025046). All procedures contributing to this study complied with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration. Signed informed consent was provided by all participants. Strict criteria for the use of anonymized data were followed throughout the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Clinical trial number

Not applicable.