Climatic and meteorological exposure and mental and behavioral health: A systematic review and meta-analysis

Abstract

As climate change exerts wide ranging health impacts, there is a surge of interest in the associations between climatic factors and mental and behavioral disorders (MBDs). Existing quantitative syntheses focus mainly on heat and high temperature exposure, neglecting the effects of other climatic factors and their synergies. The objective of this study is to conduct a systematic review and meta-analysis of the evidence of associations between climatic exposure and combined mental and behavioral health conditions and specific mental disorders (e.g., schizophrenia, dementia).

A systematic search was conducted April 11–16, 2022 using Web of Science, Medline, ProQuest, EMBASE, PsycINFO, CINAHL, and Environment Complete. Screening and eligibility screening followed inclusion criteria based on population, exposure, comparator, and outcome guidelines. Risk of bias assessment was performed, a narrative synthesis was first presented for all studies, and random-effect meta-analyses were performed when at least three studies were available for a specific exposure-outcome pair. Certainty of evidence was evaluated following the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool.

The search process yielded 7696 initial results, from which we identified 88 studies to include in the review set. Climatic factors reported included air temperature, solar radiation/sunshine, barometric pressure, precipitation, relative humidity, wind direction/speed, and thermal index. Outcomes including MBD incidences (e.g., schizophrenia, mood disorders, neurotic disorders), mental health-related mortality, and self-reported psychological states. Meta-analysis showed that heatwaves (pooled RR = 1.05, 95 % CI = 1.02–1.08) and extreme high temperatures (99th percentile: pooled RR = 1.18, 95 % CI = 1.08–1.29) were associated with higher risk of MBD. Cold extremes, however, were not associated with MBD risk. The findings further identified an association between increases in a thermal index (i.e., apparent temperature) and elevated risk of MBD (pooled RR = 1.06, 95 % CI = 1.03–1.12); specifically, a 99th percentile high temperature was associated with increased schizophrenia risk (pooled RR = 1.07, 95 % CI = 1.01–1.12).

Risk of bias assessment showed most studies to have low or moderately low risks, while a few studies were rated probably high in confounding, selection bias, outcome measurement, and reporting bias. GRADE evaluation revealed moderate certainty of evidence on thermal comfort index and MBD, but low certainty related to air temperature or sunshine duration. These findings call attention to the heterogeneity of exposure measures and the utility of thermal indices that consider the synergistic effects of meteorological factors. Methodological concerns such as the linearity assumption and cumulative effects are discussed.

GRAPHICAL ABSTRACT

1. Introduction

Climate extremes have adverse effects on health. With climate change, climate extremes are expected to increase in intensity, frequency, and health impact (Petkova et al., 2013; Guo et al., 2018). Such extreme events are not isolated incidences, but are reflective of larger-scale, often persistent changes in the thermodynamic environment across cities and countryside (Trenberth et al., 2015). In fact, climate change is already affecting every inhabited region worldwide, causing changes in ecosystem services (Shaw et al., 2011) and ambient factors to which humans are routinely exposed in everyday life, such as temperature, relative humidity, and terrestrial radiation. For example, observed global surface temperature increased by >1.2 °C from 1850 to 2020, with the steepest increase occurring after 1990 (Masson-Delmotte et al., 2021). At the microclimate scale, this increase in surface temperature is directly linked to rising mean radiant temperature and human body heat flux in the environment (Gál and Kántor, 2020). Strong evidence has associated climate extremes with cardiovascular disease, respiratory disease, infectious disease, and other physical diseases (Braga et al., 2002; Medina-Ramón et al., 2006; Wu et al., 2016; Obradovich et al., 2018a). Moreover, climate change may exacerbate health disparities by affecting minority groups and socially-disadvantaged groups disproportionately (Voelkel et al., 2018; Li et al., 2022a). Thus, a comprehensive understanding of the relationships between meteorological factors and health is a research priority to the urgent societal goals of climate change adaptation and disparity reduction.

In the recent decade, epidemiological studies have revealed critical links between meteorological conditions, especially temperature extremes, and mental illness (Hayes et al., 2018; Berry et al., 2010). Research has identified impacts of three types of climate-related events (i.e., acute, subacute, and long-lasting changes) on mental health (Palinkas and Wong, 2020). For example, extreme high and low temperatures might be factors that contribute to mental disorders (Obradovich et al., 2018b; Shiloh et al., 2005; Williams et al., 2012; Yoo et al., 2021; Zhang et al., 2020), and heat events that occur due to incremental temperature changes in summer could pose a salient risk to human mental conditions. To date, several systematic reviews have assessed the associations between extreme weather events and human mental health. Rataj et al. (Rataj et al., 2016) investigated the prevalence of mental disorders in developing countries (except African regions) during extreme weather events (i.e., storms and flooding) with 17 articles published by 2014. Rother et al. (Rother et al., 2021) filled the geographic gap of research to some extent by examining existing findings on the impact of flooding on child and adolescent mental health in Africa. Focusing on European countries, Cruz et al. (Cruz et al., 2020) and Weilnhammer et al. (Weilnhammer et al., 2021) examined the findings from 21 and 35 articles published up to 2019–2020, respectively, and synthesized the mental health impacts arising from extreme weather events involving heat/cold waves, droughts, wildfires, and floods. Furthermore, systematic reviews focusing on extreme heat have summarized the epidemiological evidence of the impact of high temperature and heatwaves on mental health-related morbidity and mortality (Thompson et al., 2018; Liu et al., 2021). Thompson and colleagues summarized the findings from 34 articles published through 2017 and presented strong links between high temperatures and suicide risk (Thompson et al., 2018). Notably, Liu et al. (Liu et al., 2021) conducted a systematic review of 53 articles published between 1990 and 2020 and a meta-analysis of 41 by pooling the effect sizes of high ambient temperatures and reported a 0.9 % increase in mental health-related morbidity for every 1 °C increase in temperature.

The recent surge of published research in climate determinants of mental health warrants a comprehensive quantitative synthesis. Specifically, several major research gaps remain to be addressed with a systematic review. Existing reviews often focus on hot weather or extreme heat conditions as the sole exposure of interest. Although their findings provide solid ground for a relationship between ambient temperatures and mental illness, we have a limited understanding of the influences of multiple, or the combined effects, of meteorological variables on human-environment heat flux and health outcomes. According to biometeorology, human heat and cold regulation depends on a balanced heat budget (Ji et al., 2022; Brown and Gillespie, 1986). Environmental factors that influence human heat/cold stress include: radiation, temperature, vapor pressure, and diffusion conductance to heat and vapor. As such, widely accepted thermophysiological comfort and heat/cold stress models consider four meteorological determinants — air temperature, humidity, wind speed, and solar radiation (Ji et al., 2022), along with individual characteristics such as BMI, physical activity, and clothing insulation (Zhao et al., 2021; Höppe, 1999a). Research has revealed that, controlling for air temperature, other climate factors such as humidity and radiation affects human thermal sensation and comfort. The relationships among temperature, humidity, and radiation are non-linear (Li et al., 2018) and their relative importance in their contribution to thermal sensation vary across seasons (e.g., air temperature contributes strongly in summer while radiation contributes strongly in winter) (Liu et al., 2016). As such, a comprehensive review that critically assesses the full range of meteorological factors associated with mental health is warranted.

It is also worth noting that previous empirical articles and reviews outlining the importance of climate and weather conditions often differentiate morbidity/hospital admissions and mortality, but consider mental disorder as one outcome category without differentiating the specific diagnoses/subdiagnoses (Williams et al., 2012; Charlson et al., 2021). Such an approach may not capture the complexities relating to the various causes and symptomology of each different types of MBD (e.g., schizophrenia, depression, and post-traumatic stress disorder). In a recent review and meta-analysis, Liu et al. presented different effect sizes between high temperatures and specific MBDs, suggesting variations in the presence, magnitude, and possibly even direction and mechanism of the relationships (Liu et al., 2021). In addition to temperature, some mental disorders (e.g., bipolar disorders, depression, and schizophrenia) may be strongly related to one or a combination of multiple climatic conditions but not others, such as sunlight intensity, humidity, and wind conditions (Gu et al., 2019; Kim et al., 2021; Lee et al., 2007; Molin et al., 1996). As such, estimated effect sizes may differ across different types of mental health conditions, requiring a more nuanced review that considers each exposure-outcome pair. In addition, as the contemporary definition of health emphasizes not only the absence of disease but also positive mood states and happiness (World Health Organization, 2002), evidence on mental health outcomes other than morbidity and mortality should be included.

Furthermore, climate change and climatic factors affect each region differently. It has been suggested that the relative risk of mental health-related mortality might vary in different climate zones, e.g., higher in tropical zones than in subtropical or continental zones (Liu et al., 2021). Given such potential geographical variability, it is critical to conduct a review that identifies the geographical and climate zones that yet remain underrepresented.

To our knowledge, no review has provided a quantitative synthesis regarding the relationships of the full set of climate and meteorological factors with mental health and MBD. Considering the growing body of evidence on this topic, the current study aims to depict a complete picture of these relationships, model pooled effect sizes, discuss research gaps and risk of bias, and propose future directions. The specific research questions examined include:

1. What are the geographical, temporal, and topical trends in the literature on climate factors and mental health?

2. What are the direction and magnitude of the relationship between each climate/meteorological factor and mental and behavioral disorders?

3. How strong and consistent are the relationships when each mental and behavioral disorder is assessed?

4. What are the knowledge gaps in the existing literature, and what are the methodological concerns relating to study bias?

2. Methods

2.1. PECO framework and literature search

The protocol was registered at the International Prospective Register of Systematic Reviews database (PROSPERO, registered ID: CRD42022321928). We developed the systematic review protocol following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (Moher et al., 2009), which set standards for a four-step process to conduct rigorous systematic reviews: identification, screening, eligibility, and inclusion.

The study search was conducted in April 11–16, 2022, in seven electronic databases: Web of Science, Medline, ProQuest, EMBASE, PsycINFO, CINAHL, and Environment Complete. These databases were selected so as to best cover the literature from a broad array of fields such as climate science and meteorology, environmental science, psychology and psychiatry, public health, and medicine. We did not restrict the search period, so the search identified articles from the inception of each database until the search date. We adopted the population, exposure, comparator, and outcome (PECO) framework to guide the search and eligibility criteria.

2.1.1. Population

We included studies reporting the general human population exposed to varying levels of climate/meteorological conditions. Animal studies were excluded. In order to identify a complete set of studies, we did not restrict the geographical areas or demographic and socioeconomic characteristics.

2.1.2. Exposure

Climatic and meteorological indicators that are relevant to human health and well-being were considered as exposure. To ensure comparability, studies that reported objectively measured climate/meteorology/weather conditions were included, while those only considering perceived ambient conditions were excluded. To ensure the review encompassed relevant exposure types, we used existing biometeorological models of human-environment energy exchange such as the PET model (Höppe, 1999b) and the COMFA outdoor thermal comfort model (Brown and Gillespie, 1986). They estimate human heat flux in an environment based on ambient factors such as air temperature, humidity, wind speed, and solar radiation. In addition, other climate conditions such as barometric pressure, precipitation (e.g., rain, snow), and fog were also considered. A detailed table listing the exposure domains is presented in Supplementary material S1.

2.1.3. Comparators

A comparable population exposed to different levels of climate conditions needs to be presented in order to estimate the risks of mental health disorders or the values of reported psychological conditions. As such, only studies involving a control population or reporting varying exposure of the same population were included.

2.1.4. Outcome

Mental health-related outcomes in this study include mortality, morbidity, and self-reported mental health and emotional states. Mortality and morbidity related terms were selected based on the International Classification of Diseases Tenth Revision (ICD-10) mental, behavioral, and neurodevelopmental disorders (F00–99) classification, including subcategory terms. These terms encompass the entire MBD domain, including dementia and organic disorders (F00–09), psychoactive substance use (F10–19), schizophrenia (F20–29), mood disorders (F30–39), neurotic disorders (F40–49), behavioral syndromes (F50–59), personality disorder (F60–69), intellectual disabilities (F70–79), developmental disorders (F80–89), childhood behavioral disorders (F90–98), and other mental disorders (F99). Equivalent definitions using ICD-9 or the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) were included and cross-walked to ICD-10 classes. In addition, we included general terms such as “mental health” and terms that describe subjective psychological state, such as “happiness.” and “sadness”. A detailed table listing the outcome domains is presented in Supplementary material S1.

The search syntax was developed based on the PECO framework. We used wildcards to account for varying forms of the keywords. In addition, forward and backward searches using eligible articles and previously conducted systematic reviews were also performed. Example search syntax used for the Web of Science is provided in Supplementary material S2.

2.2. Study selection

After importing the retrieved articles into Endnote 20 and removing duplicates, we performed the screening and eligibility steps by examining the titles, abstracts, and full texts for concurrence with the eligibility criteria defined based on the PECO framework. The selection decision for each record was made independently by two researchers, with any disagreement resolved through discussion and then by consulting a third researcher when necessary. Studies were included in the review if they met the following criteria.

1. Study reported original empirical research and was published in a peer-reviewed journal

2. Study was written in English

3. Study reported an observational study on a human population (see Section 2.1.1)

4. Study included objective climate measurements as the exposure (see Section 2.1.2)

5. Study included mental health or behavior as the outcome (see Section 2.1.3)

6. Studies involved a comparison population or varying exposure of the same population (see Section 2.1.4)

7. Study was quantitative and reported at least one effect estimate of a climate-MBD pair

Studies were excluded if they: 1) were not peer-reviewed journal articles or were non-empirical such as a review or commentary piece; 2) were not written in English; 3) reported an experimental or quasiexperimental study, 4) targeted non-human beings as research subjects (e.g., animal studies); 5) comprised a case study/case report of a single subject; 6) did not have a relevant exposure or outcome variable; 7) examined climate impacts under a specific program, training, or occupational environment (e.g., military training); 7) investigated artificially designed/controlled ambient conditions (e.g., hospitals or classrooms with different levels of lighting); 8) used the season or self-reports of perceived climate as exposure without objective measures of climate factors; 9) reported a non-time-series study with meteorological factors compared at a spatial scale that was too coarse (cross-continent or cross-country); 10) reported all-cause mortality or morbidity due to climate exposure but not mental health-related outcomes; 11) were qualitative or did not conduct any estimate of the exposure-outcome relationship; and 12) examined reproductive health or linked-life pairs (e.g., a mother-child dyad).

The search strategy and selection procedure guided by PRISMA (Moher et al., 2009) is presented in Fig. 1.

Fig. 1.

PRIMA flowchart of the study selection process.

2.3. Data collection and narrative synthesis

Information extraction was conducted by three researchers independently and then cross-checked. A descriptive information spreadsheet and a meta-analysis sheet were developed in Microsoft Excel to extract and tabulate information from the included studies. The following study characteristics were extracted in the descriptive information sheet: authors, citation details, country, population, Köppen Climate Zone of study area, study type, sample size, spatial resolution, temporal resolution, climate/meteorological measures, mental health measures, statistical analysis, effect size, lag period, and moderator/mediator. The climatic/meteorological measures category included check boxes for thermal comfort index, air temperature, wind direction/speed, solar radiation, relative humidity, barometric pressure, precipitation, among others. We also took notes of whether a working definition of a climate/weather disaster was used, detailed descriptions of the exposure measures, and data sources. The MBD measures included type of data (e.g., mortality/morbidity record, medication dispensation, self/caregiver report), detailed descriptions of outcome measures/scales, check boxes for each MBD type, time points of measurement, and data sources. The check boxes included categories corresponding to ICD-10 F01–F09, F10–F19, F20–F29, F30–F39, F40–F48, F50–F59, F60–F69, F70–F79, F80–F89, F90–F98, and F99 and equivalent ICD-9 and DSM categories.

We recoded and tabulated all data collected using Microsoft Excel Pivot Table and R packages. Descriptive plots were produced for data synthesis. A map showing geolocation and climate zone was produced in ArcGIS Pro. Studies were grouped based on the spatiotemporal resolutions, population characteristics, and exposure and outcomes measures and charts such as circular bar chart were produced using R packages.

2.4. Risk of bias assessment

As the included studies ranged in type from ecological time-series to individual-level cross-sectional, case-control, and cohort studies, we developed a risk of bias assessment (RoB) rubric by integrating relevant items from assessment tools commonly used for assessing studies on environmental exposure and health outcome at individual or aggregated population levels. These tools included the National Heart, Lung, and Blood Institutes Study Quality Assessment Tools (National Heart, Lung, and Blood Institute, 2014a; National Heart, Lung, and Blood Institute, 2014b), the JBI’s Critical Appraisal Tools (Moola et al., 2017), and the risk of a bias assessment instrument for systematic reviews informing World Health Organization global air quality guidelines (World Health Organization, 2020). Specifically, six domains with twelve items were considered: selection bias, exposure assessment, outcome measurement, confounding, missing data/attrition bias, and reporting bias. Each domain included one to three sub-items, which were rated on a four-point scale: low risk, probably low risk, probably high risk, and high risk. For example, for confounding, we assessed whether time-invariant factors (e.g., age, sex, socioeconomic conditions) and time-variant factors (e.g., air pollution, noise, seasonality) potentially related to the outcomes were considered, measured using valid approaches, and statistically adjusted for in the models. Studies that did not consider relevant control variables, did not adjust for these variables in the statistical models estimating the exposure-outcome relationships were rated as probably high/high in risk of bias for the respective rubric items under confounding. For selection bias, we considered whether study population was clearly defined and recruitment/inclusion criteria consistent and specified. Individual-based studies that used convenient samples or online panels, and ecological studies that used data from a single hospital or agency without describing the population served and coverage were rated probably high/high in outcome measure. For the outcome measurement, we assessed whether the health outcome assessors were blinded to the exposure status of participants and whether the measures were clearly defined, valid, reliable, and implemented consistently across study populations. Studies that used unvalidated scales of outcome variables were considered high in risk of bias. Studies that only reported significant results and omitted other findings from planned analysis were evaluated as probably high in reporting bias. The details of the assessment instrument are presented in Supplementary material S3. The graphics were generated using a web-based visualization tool named RoB2 (Sterne et al., 2019).

Each article was independently assessed by two researchers; conflicts were resolved by discussion and then consulting a third researcher. The item scores were then averaged and rounded to generate domain scores. We did not assign overall risk of bias scores due to the arbitrariness of assigning weights to different domains of risk. Instead, we present the byitem and -domain RoB scores, which retain transparency and better reflect the methodological strengths and weaknesses of each article. Accordingly, we also did not exclude any studies based on RoB results but instead discussed biases and provided recommendations for future studies.

2.5. Meta-analysis

Meta-analysis was performed to estimate the pooled effect sizes of the associations between climate exposures and mental health. For the meta-analysis data extraction sheet, we collected information on each exposure-outcome pair including exposure variable, outcome variable, statistical model, type of effect size statistic, effect size estimate, standard error, 95 % confidence interval lower bound and upper bound, and detailed steps of data processing. The natural logarithms of the relative risk (RR) estimates and the corresponding standard errors were calculated and used in the meta-analysis.

When multiple estimates on the same exposure-outcome pair were reported in one article, we extracted only one effect size following the approaches adopted in previous environmental epidemiology-related metaanalyses (Chen and Hoek, 2020; Khreis et al., 2017). Studies reporting odds ratio (OR) were converted into relative risk (RR) using Eq. 1 when data were available, and by assuming equality when risks in the control group were extremely low (Harrer et al., 2021). When an article reported two or more estimates for subgroups (e.g., by region or population group), we calculated the pooled effect size using a fixed-effects meta-analysis model. When multiple outcomes that belonged to the same ICD-10 subclass were tested separately, we used Eq. 2 to calculate the combined effect size (Harrer et al., 2021). When multiple estimates existed due to model specification, function, estimator, or lag duration, we followed the following decision rules in sequential order to select the model results: 1) select the authors’ favored model (main model or model following main conceptual framework/highlighted in the Abstract, Highlights, Results, or Main findings section of the Discussion), 2) select the fully adjusted model rather than the crude model, and finally 3) select the model with the smallest lag value. Pooled effects were not computed for self-reported health or psychological conditions because of the small number of studies and the high heterogeneity of psychometric scales used.

$$RR=\frac{OR}{\left(1-\frac{c}{n_{\mathit{\text{control}}}}\right)+\left(\frac{c}{n_{\mathit{\text{control}}}}\times OR\right)},$$ (1)

where c represents the events in the control group, ncontrol is the total number of participants in the control group, and OR is the estimated odds ratio, RR is the relative risk calculated.

$$\overline{Y}=\frac{1}{2}\left(Y_1+Y_2\right);V_{\overline{Y}}=\frac{1}{4}\left(V_{Y_1}+V_{Y_2}+2r\sqrt{V_{Y_1}}\sqrt{V_{Y_2}}\right),$$ (2)

where Y₁ and Y₂ are the two outcomes and V_Y1 and V_Y2 are the variance, Y is the merged effect size.

Heterogeneity of effect estimates across studies was assessed using τ², Cochran’s Q, and I² statistics. τ² is the variance of the distribution of the true effect size across studies and is easy to interpret based on the original measurement metric, but τ² and Cochran’s Q both heavily reflect statistical power while I² is not as sensitive to the number of studies; as such, presenting both Q and I² helps in depicting a complete picture of between-study heterogeneity. The rule of thumb employed for interpreting I² was as follows: up to 25 % as low heterogeneity, 50 % for moderate heterogeneity, and 75 % for substantial heterogeneity (Higgins and Thompson, 2002). Due to the small number of studies on each exposure-outcome pair and high overall heterogeneity, we did not remove outliers in the meta-analysis or perform sensitivity analyses but flagged on the forest plot those studies reporting either extremely small effects (when the upper bound of the 95 % CI is lower than the lower bound of the pooled effect) or extremely large effects (when the lower bound of the CI is higher than the upper bound of the pooled effect). Subgroup analysis based on geography or population was not possible due to small number of exposure-outcome pairs. The R package metafor (Viechtbauer, 2010) was used to perform the random-effects meta-analysis modeling, identify outliers, and produce forest plots.

2.6. Evaluation of certainty of evidence

We utilized the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework to assess the overall quality of evidence (Guyatt et al., 2008; Balshem et al., 2011). A baseline of moderate certainty was initially applied, and subsequently downgraded or upgraded based on each of the GRADE domains. Factors that decreased the certainty of evidence included D1) Risk of bias across studies, D2) Indirectness, D3) Inconsistency, D4) Imprecision, and D5) Publication bias. Factors that increased certainty included U1) Large magnitude of effects, U2) Consistent dose-response gradient, and U3) Confounding minimizes effect. The detailed criteria for downgrading and upgrading are as follows.

Downgrading. D1) The certainty of evidence was downgraded by one level if at least one study that had a non-negligible weight in the pooled effect size estimate showed at least three high or probably high bias ratings in the RoB evaluations. In our case, because the number of studies in each exposure-outcome pair was small, downgrading was performed when any study exhibited at least three high or probably high bias ratings. D2) The certainty of evidence was downgraded if studies did not adhere to the population, exposure, comparator, and outcome specified for the research question, for example not measuring outcomes directly but using surrogate outcome measures. Here, this applied when a study defined MBD cases based not on clinical diagnosis, but on outcomes from a preliminary screening tool or evaluations by researcher(s) whose credentials were not reported. D3) The certainty of evidence was downgraded when very large heterogeneity or variability in results was detected. Specifically, as observational studies are demonstrated to have moderate to large heterogeneity across geographies and populations (Chen and Hoek, 2020; Schwingshackl et al., 2021), we downgraded by one or two levels if the prediction interval was more than twice the confidence interval and studies presented the association as bidirectional (both positive and negative). D4) The certainty of evidence was downgraded if the number of participants and person/population-time were small (n < 500, determined based on previous studies discussing sample size in epidemiology (Vergouwe et al., 2005; Rigby and Vail, 1998)). D5) The certainty of evidence was downgraded if publication bias was identified based on visual assessment of funnel plots. However, funnel plots and Egger’s test had limited value due to the fact that the number of studies was smaller than the empirical threshold of 10.

Upgrading. U1) The certainty of evidence was upgraded by one level if the pooled effect size was large. Defining the threshold for “large” was challenging, especially given that RRs were mostly low compared to the threshold values used in other medical studies. U2) The certainty of evidence was upgraded by one level if studies presented a biological dose-response gradient (e.g., higher RR associated with higher exposure). U3). The certainty of evidence was upgraded when possible residual confounders would reduce the demonstrated effect.

3. Results

3.1. Literature search and selection results

The initial search yielded 7696 articles, to which 59 were added from forward/backward searches and other literature. After duplicate removal, abstract/title screening, and full-text review, the final review set comprised 88 articles published 1972–2022 (Table 1). A post-2016 surge in publications was evident. The vast majority of included studies (n = 84, 95.5 %) examined all ages or adults (e.g., 18+); only three (3.4 %) focused on older adults (e.g., 50+), and only one (1.1 %) on children.

Table 1.

Study characteristics (N = 88)

Citation	Population	Sample size	Study periods	Type of study	Spatial unit of analysis	Temporal unit of analysis	Microclimate exposure	Type of mental health outcome	Mental health outcome	Statistical analysis	Included in meta-analysis
Goldstein, 1972 (Goldstein, 1972)	College students: A cohort of students in an introductory psychology course at a community college in U.S.	22 persons	11 days	Cohort	City/Town level	Daily	Air temperature, Humidity, Barometric pressure	Self-reported affect/mental health	F30-F39	Correlation	No
Persinger, 1975 (Persinger, 1975)	College students: A cohort of university students from a class at Laurentian University, Canada	10 persons	90 days (9 January – 8 April 1974)	Cohort	City/Town level	Daily	Air temperature, Wind speed, Sunshine, Relative humidity, Barometric pressure,	Self-reported affect/mental health	F30-F39	Correlation, (Generalized) linear model	No
Howarth & Hoffman, 1984 (Howarth & Hoffman, 1984)	College students: A cohort of university students participating in the study for course requirement in Canada	24 persons	11 days	Cohort	City/Town level	Daily	Wind direction/speed, Sunshine, Humidity, Barometric pressure, Precipitation	Self-reported affect/mental health	F30-F39	Correlation, (Generalized) linear model	No
Mawson and Smith, 1981 (Mawson & Smith, 1981)	General population: Mental diseases-related hospital admissions within the area of the Greater London Council recorded by the Statistics and Research Department of the DHSS	2126 records	1975	Time series	City/Town level	Daily	Barometric pressure, Relative humidity	Medical records/Clinical diagnosis	F30-F39	Correlation	No
Barnston, 1988 (Barnston, 1988)	College students: Undergraduate psychology classes at the University of Illinois at Urbana, U.S.	62 persons	September 16-October 27, 1974	Cohort	City/Town level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity, Wind direction/speed	Self-reported affect/mental health	General psychological health	Correlation	No
Carney et al., 1988 (Carney et al., 1988)	General population: Mental disorders-related hospital admissions University Department of Psychiatry at the Regional Hospital, Galway.	104 records	1980–1984	Time series	Neighborhood/Hospital catchment level	Monthly	Air temperature, Solar radiation/Sunshine	Medical records/Clinical diagnosis	F30-F39	Correlation	No
Molin et al., 1996 (Molin et al., 1996)	A cohort of patients diagnosed with depression	126 persons	1991–1994	Cohort	City/Town level	Weekly	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation	Symptom worsening/New episode		(Generalized) linear model	No
Salib et al., 1999 (Salib & Sharp, 1999)	General population: admission to a large psychiatric hospital (Winwick Hospital) in North Cheshire	2070 records	1993	Time series	County level	Daily	Air temperature, Solar radiation/Sunshine, Relative humidity, Precipitation	Medical records/Clinical diagnosis	F00-F09	Correlation	No
Lee et al., 2002 (Lee et al., 2002)	General population: Admission to the psychiatric unit of the two hospitals affiliated with the Korea University Medical Center	152 persons	1996–1999	Cohort	Neighborhood/Hospital catchment level	Monthly	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity	Medical records/Clinical diagnosis	F30-F39	Correlation	No
Salib and Sharp, 2002 (Salib & Sharp, 2002)	General population: Admission to a large psychiatric hospital (Winwick Hospital) in North Cheshire	1084 records	1993	Time series	City/Town level	Daily	Air temperature, Solar radiation/Sunshine, Relative humidity, Precipitation	Medical records/Clinical diagnosis	F10–19, F20-F29, F30-F39, F40-F49	Correlation	No
Cornali et al., 2004 (Cornali et al., 2004)	Dementia patients: Patients diagnosed with dementia admitted to the Alzheimer Rehabilitation Unit, Richiedei Medical Center, Palazzolo s/O–Brescia, Italy.	25 persons	June14–21, 2002	Case-control	Neighborhood/Hospital catchment level	Daily	Air temperature	Symptom worsening/New episode & Drug dispensation	Dementia	(Generalized) linear model	No
Bulbena et al., 2005 (Bulbena et al., 2005)	General population: Psychiatric emergencies at del Mar Hospital, Barcelona	368 records	2002	Other ecological	City/Town level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity, Wind direction/speed	Medical records/Clinical diagnosis	F40-F49	(Generalized) linear model	No
Keller et al., 2005 (Keller et al., 2005)	General population: A cohort of population living in Ann Arbor, Michigan, U.S.	605persons	April 5-June 15, 2001, April 16-July 27, 2003, January-December 2002	Cross-sectional	City/Town level	Daily	Air temperature, Barometric pressure	Self-reported affect/mental health	F30-F39	(Generalized) linear model	No
Shiloh et al., 2005 (Shiloh et al., 2005)	General population aged 18+: Hospital admissions and psychiatric diagnoses recorded by the Israeli National Psychiatric Registry (INPR), Department of Information and Evaluation, Mental Health Services (Ministry of Health, Jerusalem, Israel)	33614 records	1981–1991	Time series	City/Town level	Daily	Thermal comfort index, Air temperature, Solar radiation/Sunshine, Barometric pressure, Relative humidity	Medical records/Clinical diagnosis	F20-F29	Correlation	No
Hartig et al., 2007 (Hartig et al., 2007)	General population: Dispensation of selective serotonin reuptake inhibitors (SSRIs) recorded by the Apoteket AB in Sweden	NR	1991–1998	Time series	National level	Monthly	Air temperature	Drug dispensation	Depression	Auto regressive integrated moving average	No
Lee et al., 2007 (Lee et al., 2007)	General population: Hospitalization record in the Taiwan National Health Insurance Research Database	15060 records	1999–2003	Time series	National level	Monthly	Air temperature, Solar radiation/Sunshine, Precipitation	Medical records/Clinical diagnosis	F30-F39, F99	Auto regressive integrated moving average	No
Christensen et al., 2008 (Christensen et al., 2008)	Bipolar patients: Bipolar patients admitted to the departments of psychiatry in three Copenhagen University Hospitals, aged 18 and 75 years old	56 persons	1990–1993	Cohort	Neighborhood/Hospital catchment level	Multi-month	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity, Wind direction/speed	Medical records/Clinical diagnosis	F30-F39	Correlation	No
Denissen et al., 2008 (Denissen et al., 2008)	General population: A cohort of population signing up to an online diary study in Germany	1233 persons	July 2005-February 2007	Cohort	National level	Daily	Air temperature, Wind speed, Sunlight, Barometric pressure, Precipitation	Self-reported affect/mental health	F30-F39	(Generalized) linear model	No
Hansen et al., 2008 (Hansen et al., 2008)	General population: Hospitalizations for mental and behavioral disorders or death records associated with mental and behavioral disorders in Adelaide, South Australia collected by the South Australian	171614 records	1993–2006	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	(Generalized) linear model	Yes
Bulbena et al., 2009 (Bulbena et al., 2009)	General population: Psychiatric emergencies at Hospital del Mar hospital and Institut Municipal Psiquiatria hospital, Barcelona	872 records	2003 summer days	Time series	Neighborhood/Hospital catchment level	Daily	Air temperature	Medical records/Clinical diagnosis	F10-F19, F20-F29, F30-F39, F40-F49, F60-F69	(Generalized) linear model	Yes
Huibers et al., 2010 (Huibers et al., 2010)	General population (aged 18–65): Reported presence of mental disorders in southern Netherlands	14478 persons	2005–2007	Time series	National level	Daily	Air temperature, Solar radiation/Sunshine, Precipitation	Medical records/Clinical diagnosis	F30-F39	(Generalized) linear model	Yes
Khalaj et al., 2010 (Khalaj et al., 2010)	General population: Hospital admissions in five regions (Sydney East; Sydney West; Gosford Yong; Newcastle and Illawarra) of New South Wales	1497655 records	spring and summer days (September-February) of 1998–2006	Time series	Regional level	Daily	Thermal comfort index, Air temperature	Medical records/Clinical diagnosis	F00-F99	(Generalized) linear model	Yes
Radua et al., 2010 (Radua et al., 2010)	General population: Admissions to the Acute Unit at the Department of Psychiatry in Bellvitge University Hospital, Barcelona, Spain	421 persons	1997–2004	Time series	City/Town level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F30-F39	Autoregressive Integrated Moving Average	No
Klimstra et al., 2011 (Klimstra et al., 2011)	Adolescents and their mothers: A cohort of adolescents and their mothers enrolled in an ongoing longitudinal project in The Netherlands, entitled Research on Adolescent Development and Relationships (RADAR).	823 persons	6 weeks	Cohort	National level	Daily	Air temperature, Sunlight, Precipitation	Self-reported affect/mental health	F30-F39	Correlation	No
Sung et al., 2011 (Sung et al., 2011)	General population: Medical records of psychiatric hospital admissions in Psychiatric Inpatient Medical Claim (PIMC) dataset of the National Health Insurance Research Database in Taiwan	41023 records	1996–2007	Time series	National level	Daily	Air temperature	Medical records/Clinical diagnosis	F20-F29	(Generalized) linear model	Yes
Yackerson et al., 2011 (Yackerson et al., 2011)	General population: Mental diseases-related emergency visits and hospitalization in the regional Mental Health Center (MHC) of Ben-Gurion University, Israel	4325 persons	2001–2003	Time series	Neighborhood/Hospital catchment level	Weekly	Air temperature, Wind speed/direction, Relative humidity	Medical records/Clinical diagnosis	F20–F29	Correlation	No
Gasparrini et al., 2012 (Gasparrini et al., 2012)	General population: Death records associated with heat recorded by the Office for National Statistics in England and Wales	92439 records	summers (June-September) of 1993–2006	Time series	Regional level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10-F19, F20-F29	(Generalized) linear model	Yes
Vida et al., 2012 (Vida et al., 2012)	General population: Emergency department visits for mental disorders in three geographic areas of Québec, aged 15 years and over	347552 records	1995–2007	Other ecological	Climate zone	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10-F19, F20-F29, F30-F39	(Generalized) linear model	Yes
Williams et al., 2012 (Williams et al., 2012)	General population: Mortality, hospital admissions, and emergency department visits collected by the South Australian Department of Health	NR	1993–2009	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	GEE	Yes
Alexander, 2013 (Alexander, 2013)	General population: Psychiatric disease-related calls to the public emergency service of the city of Buenos Aires, Argentina	80724 records	1999–2004	Time series	City/Town level	Monthly	Thermal comfort index, Air temperature, Precipitation, Relative humidity	Medical records/Clinical diagnosis	General mental health	Correlation	No
McWillimas et al., 2013 (McWilliams et al., 2013)	General population: Admissions to psychiatric hospitals in the Republic of Ireland	48347 records	1971–2002	Time series	Regional level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Wind direction/speed	Medical records/Clinical diagnosis	F20-F29	Autoregressive Integrated Moving Average; (Generalized) linear model	No
Sung et al., 2013 (Sung et al., 2013)	General population: Medical records of psychiatric hospital admissions in Psychiatric Inpatient Medical Claim (PIMC) dataset of the National Health Insurance Research Database in Taiwan	9071 records	1996–2007	Time series	National level	Daily	Air temperature, Precipitation	Medical records/Clinical diagnosis	F30-F39	(Generalized) linear model	Yes
Tsutsui, 2013 (Tsutsui, 2013)	College students: A cohort of university students recruited on campus grounds as well as through a website at Osaka University, Japan	75 persons	516 days (1 November 2006–31 March 2008)	Cohort	City/Town level	Daily	Air temperature, Wind speed, Sunshine, Humidity, Precipitation	Self-reported affect/mental health	General psychological health, F30-F39, F40-F49, F50-F59	Correlation, (Generalized) linear model	No
Vaneckova et al., 2013 (Vaneckova & Bambrick, 2013)	General population: Admissions to all private and public hospitals located in the Sydney	930322 records	1991–2009	Case-crossover	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F20-F29, F40-F49, F70-F79	(Generalized) linear model	Yes
Wilson et al., 2013 (Wilson et al., 2013)	General population: Medical records of hospital admissions or death records in the New South Wales Department of Health	NR	1997–2007 and 1997–2010	Case-crossover	City/Town level	Daily	Thermal comfort index, Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Henriquez-Sanchez et al., 2014 (Henriquez-Sanchez et al., 2014)	College Students: A cohort of Spanish university graduates participating in the SUN project from Seguimiento University of Navarra	13938 persons	1999–2009	Cohort	Regional level	Multi-year	Air temperature, Solar radiation/Sunshine, Precipitation	Medical records/Clinical diagnosis	F30-F39	Cox (Generalized) linear model	Yes
McWillimas et al., 2014 (McWilliams et al., 2014)	General population: Admissions to psychiatric hospitals in the Republic of Ireland	34465 records	1971–2002	Time series	Regional level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Wind direction/speed	Medical records/Clinical diagnosis	F30-F39	Autoregressive Integrated Moving Average; (Generalized) linear model	No
Obrien et al., 2014 (Obrien et al., 2014)	Population aged 15+: A cohort of population from the Household, Income and Labour Dynamics in Australia (HILDA) Survey	5012 persons	2007–2008	Cross-sectional (one wave of a longitudinal dataset)	National level	Monthly	Precipitation	Self-reported affect/mental health	General psychological health	(Generalized) linear model	No
Simons et al., 2014 (Simons et al., 2014)	General population: A cohort of population admitted to 3 medical centers (Radboud University Nijmegen Medical Centre in Nijmegen, University Medical Centre in Utrecht, and Jeroen Bosch Hospital, ‘s-Hertogenbosch) in The Netherlands	3198 persons	2008–2012	Cohort	Neighborhood/Hospital catchment level	Monthly	Solar radiation/Sunshine	Medical records/Clinical diagnosis	F00-F09	(Generalized) linear model	Yes
Wang et al., 2014 (Wang et al., 2014)	General population: Emergency room visits for mental illness in National Ambulatory Care Reporting System in Toronto that captured over 97% of the ER visits in the province of Ontario and had good reabstraction accuracy	271746 records	2002–2010	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Kim et al., 2015 (Kim et al., 2015)	General population: Death record provided by Statistics Korea	50055 records	1992–2009	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99, F10-F19, F20-F29	(Generalized) linear model	Yes
Beecher et al., 2016 (Beecher et al., 2016)	Mental health distress patients: A cohort of university students participating in mental health treatment at Brigham Young University, U.S.	16452 persons	2008–2014	Other	City/Town level	Daily	Sunshine	Self-reported affect/mental health	General mental health	(Generalized) linear model	No
Ding et al., 2016 (Ding et al., 2016)	Population aged 45+ A cohort of residents aged 45 and over of the southeast Australian state of New South Wales, Australia.	53144 persons	2006–2008	Cohort	State/Province level	Daily	Air temperature, Relative humidity	Self-reported affect/mental health	General psychological health	(Generalized) linear model	No
Noelke et al., 2016 (Noelke et al., 2016)	General population aged 18+: participants reporting presence of mental disorders in the Gallup G1K dataset	1854746 persons	2008–2013	Other	National level	Daily	Air temperature	Self-reported affect/mental health	General psychological health	(Generalized) linear model	No
O’Hare et al., 2016 (O’Hare et al., 2016)	Population aged 50+: A cohort of population from the first wave of The Irish Longitudinal Study on Ageing (TILDA)	8027 persons	late 2009 to early 2011	Cross-sectional (one wave of a longitudinal dataset)	National level	Monthly	Air temperature, Solar radiation/Sunshine, Precipitation	Medical records/Clinical diagnosis	F30-F39	(Generalized) linear model	No
Trang, Rocklöv, Giang, Kullgren, et al., 2016 (Trang, Rocklöv, Giang, Kullgren, et al., 2016)	General population: Mental disorders-related hospital admissions in a database from Hanoi Mental Hospital (one of mental hospitals in Hanoi City) in Northern Vietnam	21443 records	2008–2012	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F70-F79	(Generalized) linear model	Yes
Shiue, Perkins, & Bearman, 2016 (Shiue et al., 2016)	General population: Mental behavioral disorders-related hospital admissions in German hospitals	NR	2009–2011	Time series	Regional level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F01-F09, D10-F19, F20-F29, F30-F39, F40-F49, F50-F51, F98	Polynomial regression model	No
Trang, Rocklöv, Giang, & Nilsson, 2016 (Trang, Rocklöv, Giang, & Nilsson, 2016)	Population with medical records of mental disorders-related hospital admissions in a database from Hanoi Mental Hospital (one of the mental hospitals in Hanoi City) in Northern Vietnam	23525 records	2008–2012	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10–19, F20-F29, F30-F39, F40-F49, F50-F59, F70-F79, F99	(Generalized) linear model	Yes
Yi-Fan et al., 2016 (Yi-Fan et al., 2016)	General population: Participants of the International Social Survey Program (ISSP) reporting psychological conditions from 29 counties	5420 records	NR	Cross-sectional	National level	Daily	Air temperature, Relative humidity, Wind direction/speed	Self-reported affect/mental health	General psychological health	(Generalized) linear model	No
Bullock et al., 2017 (Bullock et al., 2017)	Bipolar patients: Patients diagnosed with bipolar disorder from a public health service in regional Victoria, Australia	11 persons	An average of 130 consecutive days (range: 14–231 days)	Case-crossover	Regional level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Relative humidity	Symptom worsening/New episode	Mood	(Generalized) linear model	No
Linares et al., 2017 (Linares et al., 2017)	General population: Mental disorders-related emergency admissions to municipal hospitals in Madrid	1175 records	2001–2009	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09	(Generalized) linear model	Yes
Peng et al., 2017 (Peng et al., 2017)	General population: Participated in Health Insurance System of Shanghai with medical records of hospital admissions for mental disorders in Shanghai, China	93971 records	2008–2015	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Sarran et al., 2017 (Sarran et al., 2017)	Patients with seasonal affective disorder A cohort of patients diagnosed with seasonal affective disorder at the University Center of Psychiatry at the University Medical Center Groningen, Netherlands	291 persons	2003–2009	Cohort	Neighborhood/Hospital catchment level	Weekly	Air temperature, Solar radiation/Sunshine, Barometric pressure, Relative humidity	Symptom worsening/New episode	Depression	(Generalized) linear model	No
Basu et al., 2018 (Basu et al., 2018)	General population: Medical records of mental disorders-related emergency room visits by California Office of Statewide Health Planning and Development	219942 records	2005–2013	Time series	State/Province level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F20-F29, F40-F49	(Generalized) linear model	Yes
Chan et al., 2018 (Chan et al., 2018)	General population: Mental disorders-related hospital admissions collected by the Hospital Authority of Hong Kong contained more than 99% of cases dues to mental disorders in Hong Kong	44600 records	2002–2011	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10-F19, F20-F29, F30-F39, F40-F49, F99	DLNM	Yes
M. Lee et al., 2018 (M. Lee et al., 2018)	General population: A corhort of population participating the Japanese Health Diary Study in Japan	4548 persons	October of 2013	Cohort	National level	Daily	Air temperature, Relative humidity	Medical records/Clinical diagnosis	F40-F49	(Generalized) linear model	No
S. Lee et al., 2018 (S. Lee et al., 2018)	General population: Emergency admissions collected by the Korean National Health Insurance Corporation containing medical information for almost 100% of the Korean population	166578 records	2003–2013	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F20-F29, F30-F39, F40-F49	DLNM	Yes
Obradovich et al., 2018 (Obradovich et al., 2018)	General population aged 18+: Participants completing the Behavioral Risk Factor Surveillance System (BRFSS) under the Centers for Disease Control and Prevention (CDC)	1961743 persons	2002–2012	Other	National level	Monthly	Air temperature, Precipitation	Self-reported affect/mental health	General mental health	(Generalized) linear model	No
Sherbakov et al., 2018 (Sherbakov et al., 2018)	General population: Hospitalizations in California in the Office of Statewide Health Planning and Development (OSHPD) Patient Discharge Data (PDD)	130065 records	May-October of 1999–2009	Time series	Climate zone	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Tapak et al., 2018 (Tapak et al., 2018)	General population: hospitalization in Farshchian hospital (the only psychiatric hospital across Hamadan Province) in western Iran	20406 persons	2005–2017	Time series	Neighborhood/Hospital catchment level	Daily	Solar radiation/Sunshine, Barometric pressure, Precipitation	Medical records/Clinical diagnosis	F20-F29, F30-F39	(Generalized) linear model	Yes
Wang et al., 2018 (Wang et al., 2018)	General population: Admissions to the Anhui Mental Health Center in China	17744 records	warm season (May-October) in 2005–2014	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F20-F29	DLNM	Yes
Xu et al., 2019 (Xu et al., 2019)	General population: Emergency department visits recorded by the Queensland Health	NR	2013–2015	Time series	Neighborhood/Hospital catchment level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Almendra et al., 2019 (Almendra et al., 2019)	General population: Mental disorders-related hospital admissions in the Diagnosis Related Groups general database provided by the Portuguese Health System Central Administration	30139 records	2008–2014	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10-F19, F20-F29, F30-F39, F40-F49, F50-F59, F60-F69	DLNM	Yes
Gu et al., 2019 (Gu et al., 2019)	General population: Mental disorders-related hospital admission at the biggest psychiatric hospital in Ningbo, China	10132 records	2012–2016	Time series	City/Town level	Daily	Solar radiation/Sunshine	Medical records/Clinical diagnosis	F20-F29	DLNM	Yes
Ho et al., 2019 (Ho & Wong, 2019)	General population: Death records in a mortality dataset providing information of all decedents in Hong Kong	133359 records	2007–2014	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	(Generalized) linear model	Yes
Liu et al., 2019 (Liu et al., 2019)	General population: Mental diseases-related hospital admissions at the Mental Health Center of Shandong province in China	19569 persons	14 days (14 June 14–17, June 28–30, July 4–7, and July 29–31) in 2010	Case-crossover	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	(Generalized) linear model	No
Min et al., 2019 (Min et al., 2019)	General population: Mental disorders-related hospital admissions in the hospital medical record systems of Yancheng city, China	8438 records	2014–2017	Time series	City/Town level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Mullins & White, 2019 (Mullins & White, 2019)	General population: Mental disorders-related emergency department visits and hospitalizations collected by the California’s Office of Statewide Health Planning and Development & participants reporting mental health by interviews from the Behavioral Risk Factor Surveillance System (BRFSS) under the Centers for Disease Control and Prevention (CDC), aged 18 years and older	5996037 records	2005–2016	Time series	County level	Daily	Air temperature	Medical records/Clinical diagnosis & Self-reported affect/mental health	F00-F99	(Generalized) linear model	No
Pan et al., 2019 (Pan et al., 2019)	General population: Mental diseases-related hospital admissions collected by the Anhui Mental Health Center in Hefei, China	30022 records	2005–2014	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F20-F29	DLNM	No
Xu et al., 2019 (Xu et al., 2018)	Children aged 6–11: A cohort of children aged 6 to 11 years old, participating in the Longitudinal Study of Australian Children	6875 records	2008–2014	Cohort	National level	Yearly	Air temperature, Precipitation	Self-reported affect/mental health	F00-F99	(Generalized) linear model	No
Yi et al., 2019 (Yi et al., 2019)	General population: Emergency admissions in the Anhui Mental Health Center in China	36607 records	2005–2014	Time series	City/Town level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F20-F29	DLNM	Yes
da Silva et al., 2020 (da Silva et al., 2020)	General population: Hospitalizations for mental and behavioral disorders in the public Single System of Health (SUS) in Curitiba, Brazil	5397 records	2010–2016	Time series	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Li et al., 2020 (Li et al., 2020)	General population aged 18+: Participants of the Behavioral Risk Factor Surveillance System (BRFSS) under the Centers for Disease Control and Prevention (CDC)	3060158 persons	1993–2010	Other ecological	County level	Daily	Air temperature	Self-reported affect/mental health	General mental health	(Generalized) linear model	No
Liu et al., 2020 (Liu et al., 2020)	General population: Death records collected by the Hong Kong Census and Statistics Department	19534 records	2006–2016	Time series	National level	Daily	Air temperature, Relative humidity	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Niu et al., 2020 (Niu et al., 2020)	General population: Mental disorders-related emergency admissions in 30 hospitals in Beijing recorded by Beijing Municipal Health Commission Information Center that covered all admissions	16606 records	2016–2018	Time series	City/Town level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F10-F19, F20-F29, F30-F39, F40-F49	DLNM	Yes
Oh et al., 2020 (Oh et al., 2020)	General population: Emergency department visits for panic attacks in the National Emergency Department Information System (NEDIS) in Seoul, South Korea	1926 persons	2008–2014	case-crossover	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F40-F49	(Generalized) linear model	Yes
Zhang et al., 2020 (Zhang et al., 2020)	General population: Admissions to the publicly funded and authoritative psychiatric specialist hospitals for mental disorders in three Chinese cities (Shenzhen, Zhaoqing, and Huizhou)	1133220 records	2013–2018	Case-crossover	City/Town level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F20-F29, F30-F39, F40-F49	DLNM	Yes
Bundo et al., 2021 (Bundo et al., 2021)	General population: Mental disorders-related hospital admissions in the University Psychiatric Hospital in Bern, Switzerland	88996 records	1973–2017	Case-crossover	Neighborhood/Hospital catchment level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09, F10-F19, F20-F29, F30-F39, F40-F49, F60-F69, F70-F79, F80-F89	Distributed lag linear model	Yes
Burdett, Davillas, & Etheridge, 2021 (Burdett et al., 2021)	General population: A cohort of population from the UK Household Longitudinal Study (UKHLS) on mental health	NR	April-July in 2020	Cohort	National level	Daily	Air temperature, Sunshine, Precipitation	Self-reported affect/mental health	General mental health	(Generalized) linear model	No
Eun-hye et al., 2021 (Eun-hye et al., 2021)	General population: Mental disorders-related emergency room visits in the Statewide Planning and Research Cooperative System operated by the New York State Department of Health.	92627 records	2009–2015	Time series	State/Province level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Jahan & Wraith, 2021 (Jahan & Wraith, 2021)	General population: Mental disorders-related hospital admissions in the Queensland Hospital Admitted Patient Data Collection that collects information from all public and private hospitals in Queensland, Australia	132088 records	1996–2015	Time series	Regional level	Daily	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity	Medical records/Clinical diagnosis	F20-F29	DLNM	No
Kim et al., 2021 (Kim et al., 2021)	General population: A cohort that underwent health examinations from the Korean National Health Insurance Service	25589 persons	2002–2013	Case-control	National level	Multi-day	Air temperature, Solar radiation/Sunshine, Barometric pressure, Precipitation, Relative humidity	Medical records/Clinical diagnosis	F30-F39	(Generalized) linear model	Yes
Middleton et al., 2021 (Middleton et al., 2021)	General population: Mental health-related clinic visits at five community clinics in Nunatsiavut	228104 records	2012–2018	Time series	Neighborhood/Hospital catchment level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	(Generalized) linear model	No
Son et al., 2021 (Son & Shin, 2021)	General population: Mental disorders from Health Insurance Review and Assessment Service-National Patient Sample provided by the National Health Insurance of Korea	531342 records	2016	Time series	National level	Daily	Solar radiation/Sunshine	Medical records/Clinical diagnosis	F30-F39	(Generalized) linear model	Yes
Tang et al., 2021 (Tang et al., 2021)	General population: Medical records of hospitalization in the largest psychiatric hospital (Anhui Mental Health Center) in Anhui Province, China	53288 records	2005–2019	Time series	Neighborhood/Hospital catchment level	Daily	Thermal comfort index	Medical records/Clinical diagnosis	F20-F29	(Generalized) linear model	No
Yoo et al., 2021 (Yoo et al., 2021)	General population: International Social Survey Progental disorders-related emergency room visits in the Statewide Planning and Research Cooperative System operated by New York State Department of Health	2893794 records	2009–2016	Time series	Regional level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F99	DLNM	Yes
Zapata et al., 2021 (Zapata, 2021)	General population: Participants of the employment survey of Ecuador conducted by the National Institute of Statistics and Census (INEC)	54541 persons	NR	Cross-sectional	National level	Monthly	Thermal comfort index, Air temperature, Precipitation, Relative humidity	Self-reported affect/mental health	General psychological health	(Generalized) linear model	No
Gong et al., 2022 (Gong et al., 2022)	General population: Mental disorders-related emergency room visits in the National Health Service (NHS) Digital in England	NR	1998–2009	Time series	Regional level	Daily	Air temperature	Medical records/Clinical diagnosis	F00-F09	Distributed lag linear model	Yes

Note. NR: not reported; record: the number of medical records (e.g., mental disorder-related admissions, hospitalization, or emergency visit); person: the number of individuals. GEE: generalized estimating equation; DLNM: distributed lag non-linear model.

3.2. Narrative synthesis of study characteristics

3.2.1. Geographical and climate zone distribution

The included studies represented 23 countries/regions, predominantly in Asia (n = 29, 33.0 %) and Europe (n = 27, 30.7 %), followed by North America (n = 17, 19.3 %). Outside of a few Asian countries and Australia, the rest of the Global South (e.g., Latin America, Africa) was underrepresented (n = 3, 3.4 %). In terms of climate zone, most studies focused on temperate and dry climates (n = 74, 84.1 %); only a few from Ecuador (Zapata, 2021), Brazil (da Silva et al., 2020), Taiwan (Sung et al., 2011; Sung et al., 2013), northern Australia (Xu et al., 2019; Jahan and Wraith, 2021) and Vietnam (Trang et al., 2016a) featured tropical climates, and none polar climates. (See Fig. 2.)

Fig. 2.

Geographic and climate zone distribution of studies included in the review (n = 88).

3.2.2. Types of climatic and meteorological exposures reported

Frequently-investigated climatic and meteorological factors included air/ambient temperature, relative humidity, precipitation, sunlight/solar radiation, and barometric pressure (Fig. 3, left). At least two factors were evaluated in 74 studies (84.1 %), and just one in the remainder (15.9 %). Most studies (n = 74, 84.1 %) included air temperature and temperature-based metrics (e.g., heatwave occurrence), chiefly mean daily temperature, mean of daily maximum or minimum temperature, and daily temperature range. Heat and cold were also commonly considered, typically defined with either a) an absolute threshold based on local climate norms (Hansen et al., 2008; Bulbena et al., 2009; Trang et al., 2016a; Liu et al., 2019) or b) a particular relative percentile (e.g., 95 %, 99 %) of the temperature distribution (Liu et al., 2020; Sherbakov et al., 2018; Khalaj et al., 2010). Likewise, heatwaves or extreme heat events were defined as maximum temperatures exceeding an absolute or relative threshold for consecutive days.

Fig. 3.

Types of meteorological exposures and mental health outcomes (n = 88).

In addition to temperature, 30 studies (34.1 %) examined the effect of solar radiation in terms of sunlight radiation (MJ/m²), sunshine duration (h), day length (h), cloud cover (Okta) (National Weather Service, n.d.), and global radiation (MJ/m²); 25 studies (28.4 %) evaluated precipitation-related factors, including rainfall, snowfall, and the number of rainy days; 23 studies (26.1 %) examined humidity, measured as relative humidity or dew point temperature; 19 (21.6 %) investigated barometric pressure; and nine (10.2 %) studied the impact of wind speed/velocity and direction. Several studies also accounted for horizontal visibility (Yi-Fan et al., 2016; Sarran et al., 2017; Tapak et al., 2018), mist (Sarran et al., 2017; Tapak et al., 2018), and the number of dusty and foggy days (Carney et al., 1988; Noelke et al., 2016).

In most studies, different climatic factors were considered as separate explanatory variables; however, some applied biometeorological models to synthesize human thermal comfort or stress scores (n = 13, 14.8 %). Nine articles (10.2 %) used apparent temperature as the heat indicator (Khalaj et al., 2010; Wilson et al., 2013; Basu et al., 2018; Min et al., 2019; Yi et al., 2019; Niu et al., 2020; Zapata, 2021; Oudin Åström et al., 2015); this thermal comfort index integrates ambient temperature, relative humidity, and wind velocity (Steadman, 1984). Others used the body discomfort index (Shiloh et al., 2005) or physiologically equivalent temperature (PET) (Shiue et al., 2016). Finally, one study developed a novel heat index by generating a cumulative temperature measure reflective of hourly extreme-heat exposure within a day (Tang et al., 2021

3.2.3. Types of mental health outcomes examined

The majority of studies conducted before 1990 used self-reported psychological conditions, while most recent studies utilized hospital admission/utilization records or clinical diagnosis as outcome variables [e.g.17, 23, 65, 76] (n = 65, 73.9 %). In addition, a few studies used drug dispensation (Cornali et al., 2004; Hartig et al., 2007) (n = 2, 2.3 %), symptom worsening or development of a new episode (Molin et al., 1996; Cornali et al., 2004; Bullock et al., 2017; Sarran et al., 2017) (n = 4, 4.5 %). Of those drawing upon hospital records (Fig. 3, right), 83 (94.3 %) examined mental and behavioral morbidity (Obradovich et al., 2018b; Shiloh et al., 2005; Carney et al., 1988; Obrien et al., 2014) and six (6.8 %) used mental health related cause-specific mortality (Liu et al., 2020; Gasparrini et al., 2012; Kim et al., 2015; Ho and Wong, 2019; Oudin Åström et al., 2015; Rocklöv et al., 2014). Most studies extracted MBD status from hospital records based on the ICD-9 or ICD-10 [e.g.,23, 38] (n = 55, 62.5 %), although some used clinical diagnosis by a trained psychiatrist based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) (Lee et al., 2002; Huibers et al., 2010).

Among studies utilizing ICD standards, the top five most examined outcome categories were schizophrenia (n = 25, 45.5 %), mood disorders (n = 20, 36.4 %), organic mental disorders such as dementia (n = 15, 27.3 %), neurotic disorders such as anxiety and depression (n = 13, 23.6 %), and psychoactive substance use (n = 10, 18.2 %). A few studies investigated intellectual disabilities (n = 5, 9.1 %), behavioral syndromes (n = 3, 5.5 %), personality disorder (n = 2, 3.6 %), developmental disorders (n = 2, 3.6 %), and childhood behavioral disorders (n = 2, 3.6 %). Finally, 31 studies (56.4 %) examined MBD outcomes that included multiple categories.

3.2.4. Study type, spatiotemporal scale, and statistical analysis

The reviewed studies most commonly used a time series design (n = 51, 57.9.2 %), followed by cohort (n = 17, 19.3 %), case-crossover/casecontrol (n = 9, 10.2 %), cross-sectional data or a single wave from longitudinal data (n = 5, 5.6 %), and other designs (n = 6, 6.8 %). Time-series studies mostly reported a large number of records (median = 34,000), while cohort studies showed smaller sample sizes (median = 480). Fig. 4 illustrates the spatial and temporal resolutions of the studies. Concerning spatial resolution, most studies were conducted at macroor meso-scale due to the resolution of the data source; for example, data on hospitalization and emergency room visits were often reported at aggregated levels (Bulbena et al., 2005; Bulbena et al., 2009; da Silva et al., 2020) ranging from a hospital catchment or neighborhood (n = 13, 14.8 %) to an entire nation (n = 19, 21.6 %). Regarding temporal resolution, most studies (n = 71, 80.7 %) examined daily climatic factors, with a few considering monthly patterns (10.2 %).

Fig. 4.

Spatiotemporal resolution of the studies (n = 88).

When estimating exposure-response associations, the most widely used statistical technique, especially in earlier studies, was linear modeling (including generalized linear and linear mixed modeling) (n = 49, 55.7 %). Poisson or negative binomial link functions were frequently employed to account for MBD incidence count or prevalence data; linear mixed models were applied for the nested data structure of repeated measurements from individual participants (Bullock et al., 2017; Sarran et al., 2017). Recent studies favored the technique non-linear distributed lag modeling (DLNM, n = 22, 25.0 %), which can simultaneously fit non-linear exposure-response relationships and non-linear delayed effects using a bidimensional matrix (i.e., cross-basis) (Gasparrini and Armstrong, 2010; Gasparrini, 2011). Included studies often compared multiple lag values of the effects, including cumulative lag effects, in the main analyses or sensitivity analyses. The maximum lag values were often determined based on the literature, model fit statistics, or RR-lag plots. Aside from lag 0, the most frequently used lag value was 0–7 (n = 15, 68.2 %), followed by 0–3 (n = 11, 50 %), 0–2 (n = 8, 36.4 %), and 0–21 (n = 6, 27.3 %). Many studies considered seven days as the delay period for short-term exposure to ambient conditions (Almendra et al., 2019), and 21 days for medium-long exposure (Eun-hye et al., 2021). However, findings related to appropriate lag values varied. Table 2 provides details about the curve functions, lag values, and single-day versus cumulative effect estimates. In addition, other less frequently used methods included Pearson and Spearman rank correlation, autoregressive integrated moving average (ARIMA), Cox regression, and generalized estimation equation (GEE) methods.

Table 2.

Studies using distributed lag models and selection of lag and cumulative effects (N = 22)

Author, year	Linearity	Statistical analysis	Immediate or cumulative effect	Lag time/Exposure prior to outcome	Temporal unit
Wilson et al., 2013	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 0–2	Daily
Wang et al., 2014	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 0–7	Daily
Peng et al., 2017	Non-linear	Distributed lag non-linear model	Cumulative	0, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6, 0–7, 0–14, 0–21	Daily
Chan et al., 2018	Non-linear	Distributed lag non-linear model	Cumulative	0–2, 0–8	Daily
Lee et al., 2018	Non-linear	Distributed lag non-linear model	Cumulative	0–7	Daily
Sherbakov et al., 2018	Non-linear	Distributed lag non-linear model	Cumulative	0–3	Daily
Wang et al., 2018	Non-linear	Distributed lag non-linear model	Cumulative	0, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6	Daily
Almendra et al., 2019	Non-linear	Distributed lag non-linear model	Cumulative	0, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6, 0–7	Daily
Gu et al., 2019	Non-linear	Distributed lag non-linear model	Cumulative	0–7, 0–14, 0–21	Daily
Min et al., 2019	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6, 0–7, 0–8, 0–9, 0–10, 0–11, 0–12, 0–13, 0–14, 0–15, 0–16, 0–17, 0–18, 0–19, 0–20, 0–21	Daily
Pan et al., 2019	Non-linear	Distributed lag non-linear model	Immediate	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21	Daily
Xu et al., 2019	Non-linear	Distributed lag non-linear model	Immediate	0, 1, 2	Daily
Yi et al., 2019	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6, 0–7, 0–8, 0–9, 0–10, 0–11, 0–12, 0–1, 0–14	Daily, 0–14 cumulative days
da Silva et al., 2020	Non-linear	Distributed lag non-linear model	Cumulative	0, 0–6, 0–7	Daily
Liu et al., 2020	Non-linear	Distributed lag non-linear model	Cumulative	0, 1–5, 6–21, 0–21	Daily
Niu et al., 2020	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 4, 5, 6, 7, 0–1, 0–2, 0–3, 0–4, 0–5, 0–6, 0–7	Daily, 0–7 cumulative days
Zhang et al., 2020	Non-linear	Distributed lag non-linear model	Immediate & cumulative	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 0–1, 0–3, 0–5, 0–7, 0–9	Daily
Bundo et al., 2021	Linear	Distributed lag linear model	Cumulative	0–3, 0–7	Daily
Eun-hye et al., 2021	Non-linear	Distributed lag non-linear model	Cumulative	0, 0–3, 0–7, 0–14, 0–21	Daily
Jahan & Wraith, 2021	Non-linear	Distributed lag non-linear model	Cumulative	0–3, 0–7, 0–10, 0–15, 0–21, −0–28, 0–30	Daily
Yoo et al., 2021	Non-linear	Distributed lag non-linear model	Cumulative	0–7	Daily

3.3. Meta-analysis of the relationship between climate characteristics and mental health

3.3.1. Climatic factors and MBD

Of the 76 studies in the systematic review set, 42 were included in the meta-analysis, which provided estimated reported comparable effect sizes that could be converted to relative risk. As different studies examined different types of MBD, and some only reported cases of MBD without differentiating subclasses (e.g., schizophrenia, mood disorder), we first estimated random-effect models by combining all subclasses of MBD and considering MBD risk as a single outcome variable (Fig. 5). When multiple outcomes or models were reported, we adhered to the rules detailed in Section 2.5 to select only one effect size. As the majority of the reviewed studies reported morbidity and only a few investigated mortality risk, we report the findings related to morbidity first, followed by mortality. Pooled effects were created for each exposure, such as thermal index, hot/cold air temperature, and sunshine duration. When less than three studies were available for a certain exposure-outcome pair, we presented the effect sizes from the individual studies for the sake of transparency and completeness but did not include them in meta-analysis models. The exposure variables omitted from the meta-analysis were wind direction/speed, barometric pressure, humidity, and precipitation.

Fig. 5.

Forest plot of climate/meteorological factors and risk of mental and behavioral disorders (* denotes outliers; RE Model).

3.3.1.1. Thermal comfort index and MBD.

Apparent temperature is a summary index that considers positive contributions to the human energy budget from air temperature, relative humidity, and radiation, along with the negative contribution of wind speed. Drawing on three studies that examined apparent temperature, our meta-analysis results suggested a heightened risk of MBD during energy budget overload but not energy loss conditions. Namely, when apparent temperature was at the 90th percentile (heat overload), MBD risk was elevated (pooled RR = 1.08, 95 % CI = 1.03, 1.12) compared to median or minimum risk temperatures. Heterogeneity across the studies was small (I² = 5.3 %). Meanwhile, apparent temperature in the 10th percentile (heat loss) was not associated with MBD (pooled RR = 1.02, 95 % CI = 0.99, 1.05) and heterogeneity was high (I² = 80.4 %).

3.3.1.2. Air temperature and MBD.

The meta-analysis results suggested that heat conditions exceeding certain local thresholds (e.g., consecutive daily temperatures of 35 °C or exceeding the 97.5th or 99th local percentile) were consistently associated with increased risk of MBD: heatwave, pooled RR = 1.05, 95 % CI = 1.02, 1.08; 97.5th percentile air temperature, pooled RR = 1.18, 95 % CI = 1.07, 1.30; 99th percentile air temperature, pooled RR = 1.18, 95 % CI = 1.08, 1.29. In contrast, the relationship of cold weather and MBD was less consistent across measures; temperatures in the 1st percentile (pooled RR = 0.97, 95 % CI = 0.86, 1.09) and 2.5th percentile (pooled RR = 1.14, 95 % CI = 0.94, 1.35) were not associated with MBD. Pooling of findings from studies that assumed a log-linear monotonal relationship between 1 °C increase in temperature and MBD risk also yielded insignificant results (pooled RR = 1.01, 95 % CI = 0.99, 1.03), although results with a linearity assumption needed to be interpreted with caution (see Section 4.3). Furthermore, Four studies among them considered particular seasons (Kim et al., 2021; Huibers et al., 2010; Trang et al., 2016b; Oh et al., 2020), (e.g., May to September in the Northern Hemisphere or October to March in the Southern Hemisphere) (Williams et al., 2012; Vida et al., 2012; Bundo et al., 2021), while the other five used data year-round. Finally, while heterogeneity was low for heatwave – MBD (I² = 17.2 %), it was high (I² > 80 %) for all other models, likely due to the vast differences in populations, climate conditions, and high temperature exposure variables.

Studies on cause-specific mortality related to MBD mostly focused on air temperature metrics. Pooling the effects of the three studies that assumed log-linearity, we found a 1 °C increase in temperature to be associated with 3 % increased risk of MBD-related mortality (pooled RR = 1.03, 95 % CI = 1.02, 1.04), and low heterogeneity (I² = 0.0 %).

3.3.1.3. Solar radiation/sunshine, barometric pressure, precipitation, and humidity, and MBD.

Among the various exposure measures related to sunshine and solar radiation, only sunshine duration by hour was used by three studies and therefore amenable to risk estimation using a random effects meta-analysis model. This estimation did not show an association (pooled RR = 0.98, 95 % CI = 0.93, 1.03). Regarding barometric pressure, only two studies were available. Likewise, for precipitation and relative humidity, a variety of variables were examined, each in a single study; these included rain volume (mm or percentile threshold), rain/snow/foggy day, and relative humidity (%). Thus, meta-analyses could not be performed for these exposures.

3.3.2. Outcome-specific analysis

To understand the direction and magnitude of the associations between climate factors and specific subcategories of MBD, we tallied effect sizes according to the specific MBD subclass reported and performed a meta-analysis on each subclass. After sorting the climate factor-mental disorder pairs, only three outcome categories were represented in at least three studies with the same exposure-outcome pairs—schizophrenia (F20–F29, n = 4), mood disorders (F30–F39, n = 7), and neurotic disorders (F40–F49, n = 6) —and thus able to be pooled using random-effect models. Other outcomes such as dementia and behavioral syndromes were not included in the meta-analysis due to the fact that they were featured in less than three studies. We present forest plots showing the pooled effects of the three analyzable exposure-outcome pairs in Fig. 6, while the complete set of individual effect sizes are presented in Supplementary material Figs. S4–1 to S4–6.

Fig. 6.

Forest plot of climate/meteorological factors and schizophrenia, mood disorders, and neurotic disorders. (* denotes outliers; RE Model).

3.3.2.1. Schizophrenia.

Four studies fitted models to compare 99th percentile and median/minimum air temperatures for relation to schizophrenia. Based on these data, we found a 99th percentile high temperature to be associated with higher risk of schizophrenia (pooled RR = 1.07, 95 % CI = 1.01, 1.12), with low to moderate heterogeneity (I² = 40.4 %).

3.3.2.2. Mood disorders.

Pooled results from four studies estimating the effect of a 1 °C increase in temperature yielded insignificant results (pooled RR = 1.00, 95 % CI = 0.97, 1.03), while the pooled association of 99th percentile temperature and elevated risk of mood disorder approached statistical significance (pooled RR = 1.23, 95 % CI = 1.00, 1.51). These models featured high heterogeneity scores (I² > 80 %) due to geographical and population differences, along with variation in outcome classification: three studies defined their outcome variable using ICD-9 or 10, where the fourth used the DSM-IV. Moreover, Kim et al. (2015) only considered ICD codes F31–33, while the other two considered F31–40.

3.3.2.3. Neurotic disorders.

Pooled results related to neurotic disorders (e.g., depression) were generally not significant. The association of 1 °C increase in temperature with neurotic disorders did approach significance (pooled RR = 1.02, 95 % CI = 1.00, 1.04), but that of 99th percentile temperature did not (pooled RR = 1.11, 95 % CI = 0.92, 1.33). Heterogeneity scores were high for these models (I² > 75 %) for reasons similar to those described above.

3.4. Risk of bias and certainty of evidence

Fig. 7 shows the results of the risk of bias assessment by study and assessment category. The individual scores for each item evaluated are given in Supplementary material S5 Fig. S5–1. Across all studies, the major bias concerns sampling selection, outcome measurements, and confounding factors (Supplementary material S5 Fig. S5–2). Studies that utilized records from a few hospitals without explaining the sampling method or their representativeness of the population were considered to have elevated bias. A few early individual-level studies utilized convenient samples from students, hospitals, or online panels (Barnston, 1988), which might be subject to volunteer bias and hence were considered probably high bias. Regarding outcome measurement, studies that extracted outcomes from hospital records using standard classification systems, such as ICD-9 or 10 or DSM-IV, were considered to have low risk of bias, as were those involving clinical diagnosis by a trained psychiatrist or self-report based on validated diagnostic/screening instruments. Studies that used measures not validated for the specific construct to be examined were graded as having probably high bias. In addition, if the outcome was clinically diagnosed, we considered whether the assessors were blinded to exposure conditions; outcomes measured by self-reported scale were uniformly considered to have elevated bias. For confounding factors, the majority of populationand individual-level studies included time-invariant factors such as age, sex, socioeconomic conditions, and pre-existing health conditions, time-variant factors such as season, day of week, time of day, and other exposure factors such as air pollution and noise level. Several studies that referenced morbidity records used auxiliary place-based datasets that provided population-level demographic conditions. Studies that only adjusted for temporal factors (season, day of week, diurnal) but did not include study population/individualor environment-related confounding factors were graded as having higher risk of bias, while those that did not consider or adjust for any confounding factors in their exposure-response estimates were considered to have probably high risk of bias. Finally, selective reporting of results was observed in some studies when multiple exposure, outcome, or statistical models were mentioned in the planned analyses, but only those with significant results were presented and discussed.

Fig. 7.

Risk of bias assessment results.

Using the GRADE criteria, we assessed the certainty of evidence for each category where pooled effects were produced. We assigned a baseline of moderate certainty based on the study design, followed by decisions to downgrade or upgrade for each of the GRADE domains. Effect estimates for which more than half of the contributing studies exhibited greater than low risk of bias in at least one domain were downgraded by one level, such as the associations between MBD and heatwave and sunshine duration. All studies met the specified desired population, exposure, comparator, and outcome criteria and were not downgraded based on those categories. For example, no study used as outcome a surrogate of a clinical endpoint. When we identified inconsistencies, such as a prediction interval more than twice the confidence interval and studies disagreeing in the direction of the association (i.e., positive and negative association for the same exposure-outcome pair), we downgraded the evidence; a case in point is the association of high air temperature and MBD. For the number of participants and specifically person-time, all but one study used a time-series design or featured a large cohort. The only study that did not report the number of participants used hospital admissions records during a 16-year span and was thus considered to have adequate sample size. The relative risks yielded by the random effects meta-analysis models were small, and therefore no evidence was upgraded for large magnitude of effect. As meta-regression was not possible due to the small number of studies, there was insufficient evidence for a dose-response gradient. Confounders may either have positive or negative impacts on the associations, and therefore no upgrading was performed for the criteria regarding confounding.

Ultimately, based on the outcomes of the structured GRADE process, there is a moderate level of evidence supporting that MBD is elevated when the thermal index (apparent temperature) increases. Similarly, the association between risk of schizophrenia and high air temperature has moderate certainty. Other identified associations such as between heatwave or high temperature and MBD still require more attention due to having only low certainty of evidence. Table 3 presents the details of the evaluation outcome.

Table 3.

Evaluation of certainty of evidence

Rating factors	D1	D2	D3	D4	D5	U1	U2	U3	Final rating
Apparent temperature (heat) – MBD	0	0	0	0	0	0	0	0	Moderate (+++)
Apparent temperature (cold) – MBD	0	0	−1	0	0	0	0	0	Low (++)
Heatwave – MBD	−1	0	0	0	0	0	0	0	Low (++)
Air temperature (high temperature) – MBD	0	0	−1	0	0	0	0	0	Low (++)
Air temperature (low temperature) – MBD	0	0	−1	0	0	0	0	0	Low (++)
Air temperature (linear increase) – MBD	−1	0	−1	0	0	0	0	0	Very low (+)
Sunshine duration – MBD	−1	0	−1	0	0	0	0	0	Very low (+)
Air temperature (high temperature) – Schizophrenia	0	0	0	0	0	0	0	0	Moderate (+++)
Air temperature (high temperature) – mood disorders	0	0	−1	0	0	0	0	0	Low (++)
Air temperature (linear increase) – mood disorders	−1	0	−1	0	0	0	0	0	Very low (+)
Air temperature (high temperature) – neurotic disorders	0	0	−1	0	0	0	0	0	Low (++)
Air temperature (linear increase) – neurotic disorders	−1	0	−1	0	0	0	0	0	Very low (+)

4. Discussion

4.1. Main findings

In this study, we performed a systematic review and meta-analysis to examine the associations between climatic and meteorological factors and mental health. The results revealed uneven geographical and climate zone distributions and uncovered underrepresented climates. Temporally, earlier studies focused on psychological states using smaller samples, recent studies emphasized psychiatric morbidity with longitudinal/time-series surveillance data. Extreme heat based on heat thresholds and high thermal index integrating temperature, humidity, and wind velocity were identified as risk factors for MBD, while extreme cold was not. Evidence was too limited to generate pooled effects sizes for humidity, wind, and solar radiation. Regarding different subclasses of MBD, schizophrenia risk increased when temperature rose above the 99th percentile, and mood and neurotic disorder risks were approaching significance with temperature increase.

Our meta-analysis revealed that, in general, heat extremes and temperatures exceeding local thresholds were related to mental and behavioral disorders; specifically, these associations were identified among studies comparing heatwave with non-heatwave days and also among studies estimating MBD risk in 99th or 97.5th percentile temperatures relative to median or minimum temperatures. These results are consistent with findings from prior systematic reviews and meta-analyses, which indicated high temperatures and heat waves to be risk factors for mental disorders (Thompson et al., 2018; Liu et al., 2021). Nevertheless, the certainty of evidence is considered low or very low for most exposure-outcome pairs due to risk of bias issues and high heterogeneity. Thompson et al. (Thompson et al., 2018) reported a similar finding between heat and mental health outcomes, although a meta-analysis was not conducted due to heterogeneity. Liu et al. (Liu et al., 2021) obtained pooled effect sizes (RR) for various definitions of heatwave, and reported pooled RRs between 1.048 and 1.753. Our effect estimate overlap with the lower end of the RR range, for estimates including heatwave (RR = 1.05, 95 % CI = 1.02–1.08), 97.5th percentile temperature (RR = 1.18, 95 % CI = 1.07–1.30), and 99th percentile temperature (RR = 1.18, 95 % CI = 1.08–1.29). The difference on the upper end of the RRs betwween the two studies was largely attributable to the fact that the present study did not consider studies focusing on suicide ideation or attempts while Liu et al. did. Cold extremes, on the other hand, were not significantly associated with MBD risk. No previous studies are available on this relationship to compare the effect estimate.

Beyond temperature metrics alone, we also identified a thermal index, which combines temperature, humidity, wind speed, and radiation, to be factor for elevated risks of MBD. Temperature is the top climatic factor contributing to thermal stress, but not the only one; humidity and radiation variations can have as large effects as temperature change. While popular metrics using 99th or 97.5th percentile temperatures were significant but low-certainty correlates of MBD, thermal index at the 90th percentile yielded results with moderate certainty. This indicates the importance of considering multiple sources of heat strain in the course of human-environment heat exchange. Studies on sport and occupational medicine using established thermal models such as physiological equivalent temperature (PET), the Comfort Formula (COMFA), and wet-bulb globe temperature (WBGT) have demonstrated these models to have strong explanatory power as summaries of thermal stress (Budd, 2008). Meanwhile, we found no significant relationships when considering sunshine duration alone, likely due to the small number of studies that used this metric. Other climate exposures, such as humidity, wind speed, barometric pressure, and solar radiation, were measured heterogeneously and could not be pooled, thus the results were inconclusive.

When it comes to specific mental disorders, in general, studies using homogeneous climate/meteorological metrics were lacking. Previous systematic reviews that provided narrative summaries of patterns mostly noted that high temperatures might be a trigger of schizophrenia, bipolar disorder, and dementia (Thompson et al., 2018; Montes et al., 2021; Bongioanni et al., 2021). To our knowledge, our study is among the first to conduct a meta-analysis on various mental health outcomes, especially for each type of MBD based on ICD classifications and other self-reported psychological states. The only significant pooled effect identified was an association of 99th percentile temperature with schizophrenia, which also showed moderate certainty of evidence. The relationships of temperature with mood and neurotic disorders approached but did not achieve statistical significance, and the confidence in the body of evidence was assessed as low at best.

The heterogeneity of effect estimates was high on most exposure-outcome pairs, likely due to differences in climate conditions, population characteristics, and heat adaptation and mitigation capacities. It has previously been noted that general population studies may yield heterogeneity values in the range of 75 % and above, higher than those reported in randomized controlled trials or other smaller studies (Chen and Hoek, 2020).

4.2. Mechanisms of the observed associations between climate conditions and mental health

The mechanisms underlying the observed associations between climate/meteorological factors and mental health may be multifaceted and complex. First, the associations between heat and mental health may be explained by dysregulation of the serotonin system (5-hydroxytryptamine, 5-HT), which is involved in mood and cognition modulation. Studies have shown that heat and humidity may cause water deprivation and dehydration, which lead to a decrease in serotonin; such deficiency is related to increased mood disorders, depression, anxiety disorders, schizophrenia, and other MBDs (Lin et al., 2014). It has also been suggested that, as serotonin plays an essential role in thermal regulation, acute ambient temperature changes may cause system dysregulation, thereby aggravating mental health symptoms (Crane et al., 2015). On the other hand, many antipsychotic and psychotropic medications have thermoregulatory side effects, and intake of such medications may be related to compromised thermoregulation and perspiration (Zammit et al., 2021). Second, recent research has indicated that heat stress can induce neurodegeneration; in particular, heat stroke can cause excitotoxicity, necrosis, and apoptotic death of neuronal cells (Gong et al., 2022; Zammit et al., 2021; Kourtis et al., 2012). Clinical studies suggest that neuronal cell and neural network alterations contribute to the pathology of mental disorders (Quach et al., 2016). Finally, in addition to biological mechanisms, there may be a socioecological explanation of the link: extended periods of heat may influence individual and family daily routines and social networks, presenting challenges for commute and childcare routines, and also incur emotional strain related to family members with pre-existing health conditions; all of these are external stressors to mental health. Adverse microclimate conditions also limit opportunities to engage in outdoor activities that offer restorative effects (e.g., visiting a park or a beach) (Hartig et al., 2007; Li et al., 2019). Finally, disaster situations such as heat and cold storms often also come with disruption of transportation and energy/water infrastructures (Li et al., 2022b), exacerbating the psychosocial stress experienced by individuals.

One potential pathway that was not substantiated in the current study is human sunlight exposure and the circadian clock. Specifically, research on seasonal affective disorder and depression has outlined a circadian timing mechanism via the hypothalamic suprachiasmatic nucleus that is dependent on ambient conditions, with the light-dark cycle being considered the most relevant synchronizer for the body’s circadian timing (e.g., sleepwake cycle, locomotor activity, hormones); disturbance of that timing is related to depression and distress (Mendoza, 2019).

4.3. Future directions and limitations

While there has been a recent surge of research interest, given the broad array of climatic factors, the various types of mental health conditions, and the insufficient evidence of association determined in this study for most exposure-response pairs, more research is warranted regarding each exposure-response pair and also the joint effects of climatic exposures. Here, we discuss potential future directions for filling critical knowledge gaps and addressing the risk of bias and other methodological concerns.

4.3.1. Standardizing meteorological metrics and thresholds

The wide variety of meteorological metrics used in included studies reduces the ability to perform meta-analysis and promotes high heterogeneity. For example, measures used to represent sunshine and solar radiation included daily sunshine availability, duration, solar radiation, day length, cloud cover, and direct and global radiation. Likewise, for air temperature, studies estimated the effects of 1 °C increase in daily mean or maximum temperature, 1 °C increase in daily temperature range, various percentile thresholds (e.g., 99, 97.5, 95, 90, 75), and various definitions of a heat wave. We recommend that future studies select metrics that provide adequate granularity and reflect biometeorological mechanisms. Regarding temperature, for instance, daily mean or maximum temperature may fail to capture the critical impacts of nighttime temperature on mental health, especially during multi-day heat events. The literature also indicates that high nighttime temperature, especially if persistent, is related to sleeping disturbances (Haskell et al., 1981; Libert and Bach, 2005), which could trigger episodes of mental disorders (Liew and Aung, 2021). It also remains necessary to consider the joint effect of multiple meteorological conditions. As thermal stress is related to most recognized biological neurodegeneration pathways involved in mental health, future studies may consider using established thermal indices that consider metabolic, conductive, convective, and radiative heat flux, either as the main exposure variable or in the sensitivity analysis.

4.3.2. Addressing scale mismatches and attention to microclimate exposure

Although the majority of studies included in this review utilized ground-measured meteorological conditions, spatial mismatches often exist between the measurements taken at weather stations and the microenvironments experienced by individual participants. Studies that consider climate rather than microclimate may be subject to the uncertain geographic context problem (UGCoP): namely that areal measures do not match the environmental exposure that exerts impacts on health and behavior, a problem that has been discussed in geography and spatial epidemiology (Kwan, 2012; Jia et al., 2020). Nevertheless, studies on human exposure to heat, wind, and solar radiation reviewed here still mostly used areal-based climate measures derived from sparsely-located weather stations which may be too coarse to establish exposure-response relationships. In fact, microclimate conditions can be modified by urban heat island effects, site-level structural elements and vegetation, water bodies, and the albedo of materials. All of these factors directly impact human physiology, thermal sensation, and behavioral adaptation at a given location. Future studies are warranted that precisely measure the microclimates to which individuals are exposed in residential neighborhoods and third places, such as through technologies like downscaling and/or microclimate logger networks (George et al., 2015).

4.3.3. Studying disease-specific exposures and pathways

More studies are needed on specific mental disorders (e.g., schizophrenia, depression, anxiety, dementia). In the set of studies reviewed here, many utilized combined disease prevalence from multiple MBDs that may reflect different risk pathways, leading to results that are leveled out and challenging to interpret. For studies that considered specific mental outcomes, ICD and DSM codes were typically used; however, ICD classifications depend on good clinical judgment and can be less accurate, while the DSM is less commonly used outside the U.S. In addition, admission/discharge records were often aggregated without distinction of clinical stages. As the impacts of risk factors and interventions may differ over the disease course, it is critical to consider disease trajectory and climate at different stages such as in prodromal, first episode, persistent, and remitted cases. Additionally, more studies that address neurotransmitters and other mechanisms would help advance knowledge and policy.

4.3.4. Expanding outcomes from morbidity to well-being

Public health policies across the world, such as the Healthy People 2030 framework of the U.S., have been advocating for not only disease prevention/treatment but also for promoting people to achieve their full potential. Although this review set out to frame mental health in a broad definition that encompasses happiness and well-being, the number of studies employing such constructs was limited. Therefore, the quantitative synthesis mostly addressed MBD; obtaining pooled effects for affect and other positive emotions was not possible due to the small number of studies and high heterogeneity. Future studies are warranted that use valid instruments to examine whether climate-related risks impact subjective well-being. In addition to identifying modifiable risk factors, there is value in adopting a salutogenic view that determines the ambient conditions that provide for a quality experience and support subjective well-being.

4.3.5. Modeling technique that accounts for non-linear and delayed dependencies

The two main factors that pertained to statistical model selection for estimating climate and mental health relationships were non-linearity and lagged and cumulative effects. Recent evidence suggests that the relationships between continuous climatic and meteorological factors and mental health outcomes are not linear or monotonically increasing/decreasing (Almendra et al., 2019). In fact, human physiology suggests that for climatic conditions within a reasonable range, the energy budget is maintained, meaning that extreme values at both ends may have negative impacts (Eun-hye et al., 2021). To account for the nonlinearity, studies can use heat and cold thresholds or only records from warm seasons. Additionally, recent studies that considered nonlinear exposure-response relationships have often fitted models with spline functions (Yoo et al., 2021; Lee et al., 2018b; Bundo et al., 2021). Of all studies that considered non-linear exposure-response curves using distributed lag models, only one found a linear relationship; the remainder presented curvilinear functions for both the predictor and lags. This raises a question as to the validity of the findings of studies that assume linearity/log-linearity without appropriate theoretical grounds or statistical tests. In the absence of specified geographic/seasonal conditions, interpretations of MBD risk as increasing with every 1 °C increase in air temperature can be misleading. Future meta-analysis articles should also use caution when assuming linear/loglinear or monotonic relationships. Standardization and methodological comparisons across model specifications and single-day and cumulative lag values may be informative. As more studies accumulate that employ a two-stage time-series design to estimate location-specific exposure-response relationships, advanced multivariate meta-analysis methods should be considered for future reviews (Gasparrini and Armstrong, 2013).

Limitations of this review also need to be noted. First, one major limitation is the small number of studies for each exposure-outcome pair. Often only three studies were available, rendering it impossible to perform sensitivity analysis, exclude outliers and studies based on specific considerations, and examine the impacts of such exclusions on the meta-analysis results. Second, our review did not include experimental and quasiexperimental studies, and hence the findings, depending on the confidence of evidence, contribute to but cannot directly establish causality. Third, we did not include outcomes related to suicide ideation and attempts. Although previously suicide has been considered a symptom associated with other psychiatric disorders, the DSM-5 positioned suicidal behavioral disorder as an independent condition for further study (DSM-IV-TR, 2000). Fourth, we excluded studies considering maternal exposure and child mental health conditions, because there may be additional biophysiological mechanisms involved. Nevertheless, we recognize the importance of studying reproductive health and linked lives in climate exposure. Fifth, although the GRADE tool provides a great framework for evaluating the certainty of evidence (Balshem et al., 2011; Schünemann et al., 2019a), it has limitations in assessing evidence from nonrandomized/observational research (Schünemann et al., 2019b). Finally, in light of our research questions, we conducted outcome-specific analysis, but could not perform subgroup analysis based on age, gender, or geography due to the high heterogeneity and small number of studies available for each exposure-outcome pair. Narrative review evidence, such as Sharpe and Davison (Sharpe and Davison, 2021), highlighted the links between climate-related disasters and mental disorders specifically for lowand middle-income countries. As the evidence base continues to grow, future studies should focus on health disparities across population groups, especially vulnerable groups such as older adults, low-income groups, and socially isolated populations, which may offer scientific insights and inform targeted policy interventions.

5. Conclusion

As the impact of climate change on mental health gains increasing recognition, a comprehensive investigation of the relationship between various climatic and meteorological factors and mental health can reveal gaps in knowledge and inform policy related to public health and climate adaptation. Accordingly, utilizing the PRISMA framework, this meta-analysis aimed to identify factors associated with increased risk of mental and behavioral disorders. Our random-effects meta-analysis models revealed that higher thermal index values, heatwaves, and extreme temperatures surpassing certain thresholds were linked to elevated risks of MBD. Furthermore, when considering specific subtypes of MBD, high temperatures were found to be associated with an increased risk of schizophrenia. Notably, the results underscored the heterogeneity of exposure measures and a scarcity of evidence regarding the effects of climate factors other than air temperature, such as humidity, wind, solar radiation, and barometric pressure. The findings also highlighted the importance of investigating the synergistic effects of multiple climate factors using thermophysiological models, non-linear exposure-outcome relationships, and cumulative and lagged effects of climate exposure.

HIGHLIGHTS.

• There is a recent surge of studies on the impact of various climate factors on mental health

• Earlier evidence focused on climate and self-reported psychology while recent studies focused on morbidity

• Extreme thermal index and heatwave were associated with elevated risks of mental disorders

• Extreme high temperature was associated with high risks of schizophrenia

• Methodological gaps include heterogeneity in climate measures, scale mismatch, and non-linear effects

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.scitotenv.2023.164435.

Data availability

Data are shared in the supplementary documents