The impact of temperature on non-typhoidal Salmonella and Campylobacter infections: an updated systematic review and meta-analysis of epidemiological evidence

Summary

Background

As temperatures rise, the transmission and incidence of enteric infections such as those caused by Salmonella and Campylobacter increase. This study aimed to review and synthesise the available evidence on the effects of exposure to ambient temperatures on non-typhoidal Salmonella and Campylobacter infections.

Methods

A systematic search was conducted for peer-reviewed epidemiological studies published between January 1990 and March 2024, in PubMed, Scopus, Embase, and Web of Science databases. Original observational studies using ecological time-series, case-crossover or case-series study designs reporting the association between ambient temperature and non-typhoidal Salmonella and Campylobacter infections in the general population were included. A random-effects meta-analysis was performed to pool the relative risks (RRs) per 1 °C temperature increase, and further meta regression, and subgroup analyses by climate zone, temperature metrics, temporal resolution, lag period, and continent were conducted. The Navigation Guide systematic review methodology framework was used to assess the quality and strength of evidence. The study protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO).

Findings

Out of 3472 results, 44 studies were included in this systematic review encompassing over one million cases each of Salmonella and Campylobacter infections. Geographically, the 44 studies covered 27 countries across five continents and most of the studies were from high income countries. The meta-analysis incorporated 23 Salmonella studies (65 effect estimates) and 15 Campylobacter studies (24 effect estimates). For each 1 °C rise in temperature, the risk of non-typhoidal Salmonella and Campylobacter infections increased by 5% (RR: 1.05, 95% CI: 1.04–1.06), and 5% (RR: 1.05, 95% CI: 1.04–1.07%), respectively, with varying risks across different climate zones. The overall evidence was evaluated as being of “high” quality, and the strength of the evidence was determined to be “sufficient” for both infections.

Interpretation

These findings emphasise the relationship between temperature and the incidence of Salmonella and Campylobacter infections. It is crucial to exercise caution when generalising these findings, given the limited number of studies conducted in low and middle-income countries. Nevertheless, the results demonstrate the importance of implementing focused interventions and adaptive measures, such as the establishment of localised early warning systems and preventive strategies that account for climatic fluctuations. Furthermore, our research emphasises the ongoing need for surveillance and research efforts to monitor and understand the changing dynamics of temperature-related enteric infections in the context of climate change.

Funding

Australian Research Council Discovery Projects grant (ARC DP200102571) Program.

Research in context.

Evidence before this study

Systematic searches of major databases including PubMed, Embase, Scopus, and Web of Science were conducted to identify existing systematic reviews and meta-analyses focused on the association between temperature exposure and Salmonella and Campylobacter infections. Terms related to enteric infections, including “diarrhoea” or “gastroenteritis” or “foodborne illnesses” or “Salmonella” or “Salmonellosis” or “Campylobacter” or “Campylobacteriosis” were combined with meteorological terms such as “temperature” or “heat” or “heatwave” or “weather” or “meteorological factor” or “climate” or “climatic factor” or “climate change” or “global warming”. While four reviews were identified, they often lacked pathogen-specific analyses and did not comprehensively assess evidence quality and strength. Notably, Salmonella studies published before and after 2019 were largely overlooked, leaving a gap in understanding the latest evidence.

Added value of this study

Our research aimed to address these gaps by conducting a systematic review and meta-analysis to comprehensively evaluate the relationship between ambient temperature and Salmonella and Campylobacter infections. Different from previous reviews, our study has conducted a comprehensive literature search and includes previously overlooked studies, particularly on Salmonella that were published before 2019. Moreover, by applying rigorous methods, including the Navigation Guide framework for systematic reviews, we not only synthesised the available evidence but also assessed its quality and strength. This allowed for a more robust understanding of the temperature-enteric infection relationship, thereby providing valuable insights for public health decision-making.

Implications of all the available evidence

Our systematic review and meta-analysis contribute to the growing body of evidence that emphasises the link between temperature exposure and Salmonella and Campylobacter infections. The findings showed a 5% increase in risk of both Salmonella and Campylobacter infections per 1 °C temperature increase. By clarifying the quality and strength of the evidence, our study offers valuable insights for policymakers and public health officials. These findings highlight the importance of implementing specific interventions and preventive measures to reduce the impact of climate change on enteric infections. Furthermore, our research emphasises the ongoing need for surveillance and research efforts to monitor and understand the changing dynamics of temperature-related enteric infections in the context of climate change. Policymakers can use this evidence to implement targeted public health interventions, such as the development of localised early warning systems and preventive strategies that consider socio-economic conditions and climatic variations.

Introduction

There is a growing body of evidence indicating the negative impact of climate change on human health, including the effect on the transmission of infectious diseases.1, 2, 3, 4 Climate change can directly affect the survival and transmission of pathogens through water, food, and the environment.5, 6, 7 Among these infectious diseases, enteric infections, which result from consuming contaminated food or water are among the leading causes of morbidity.⁸

Salmonella and Campylobacter infections are among the most frequently reported enteric infections and account for a substantial percentage of food-borne diseases annually.⁹ For example, non-typhoidal Salmonella causes more than 80.3 million cases and around 155,000 deaths per year worldwide.¹⁰ Meanwhile, Campylobacter, which is one of the main causes of foodborne illnesses in humans, is responsible for approximately 96 million cases of infection and over 20,000 deaths annually in both developed and developing countries.¹⁰ However, studies have indicated that the actual number may exceed as Campylobacter incidence is under reported.¹¹

Epidemiological studies have quantified the relationships between Salmonella and Campylobacter infections and climate variability including temperature, humidity, and precipitation in different regions with varying climatic characteristics.12, 13, 14, 15, 16

Rising temperature facilitates the transmission of pathogens responsible for enteric infections, such as Salmonella and Campylobacter, as temperature plays a crucial role in their life cycles.17, 18, 19, 20 Both pathogens exhibit increased transmission rates as temperatures rise, although the specific temperature range and environmental conditions that promote their spread vary.

Salmonella exhibits optimal multiplication within a temperature range of 35 °C–37 °C, because this range closely matches the normal body temperature of warm-blooded hosts, including humans. However, Salmonella can survive at lower temperatures, leading to a lagged increase in infection rates as temperatures rise.^{14,18,19,21,22}

Campylobacter species thrive best at a temperature of 30 °C–42 °C, closely mirroring the body temperature of birds, a principal reservoir.²³ However, Campylobacter is less resilient than Salmonella, requiring moist environments, for their survival outside the host.²⁴

Warm summer days also lead to an increase in outdoor activities, such as picnics and barbecues, where people prepare and consume food outside. In these situations, proper cooking and safe food handling practices are often overlooked because of the casual setting and limited access to proper food preparation facilities.²⁵ Additionally, there can be higher consumption of high-risk food items, such as dairy products, eggs, poultry and vegetables, which may not be cooked thoroughly or left at ambient temperatures, whereby bacteria can multiply and increase the risk of food spoilage.²⁶ The combination of these factors raises the risk of food contamination and, as a result, the likelihood of foodborne illnesses like Salmonella and Campylobacter infections.^16,17,27

Although the impact of temperature on enteric infection transmission is apparent, the specific temperature thresholds at which the incidence of Salmonella and Campylobacter increases vary across different climate zones.^16,18,19,28 In temperate climates, a modest increase in temperature can significantly elevate the susceptibility to Salmonella and Campylobacter infections.²⁰ For example, when the temperature rises from 15 °C to 25 °C, the rates of infection escalate, particularly if the temperature exceeds 20 °C.^20,29 In tropical and subtropical climates, where higher temperatures are more common, sustained temperatures above 30 °C are typically necessary to augment bacterial proliferation and provoke an upsurge in infection rates. Nevertheless, the relative surge in infections in these regions may not be as conspicuous as in temperate regions due to the naturally elevated baseline temperatures.^20,26,30 Thus, several epidemiological studies reporting varying risk increase in Salmonella and Campylobacter infection per 1 °C in temperature.

Moreover, lag periods, which are delays between the occurrence of enteric infection, diagnostic testing, and reporting of results, play a crucial role in providing insights into the temporal relationship between temperature and enteric infections outcomes.^31,32 By considering these delays, studies can more accurately attribute changes in infection rates to specific temperature exposures across different studies.

Although there have been four previous systematic reviews assessing the impact of temperature on enteric infections,33, 34, 35, 36 three of them only included studies published until 2019.34, 35, 36 The fourth systematic review specifically focused on the association between weather variability and Campylobacter infection, included studies conducted until 2022.³³ However, among the systematic review and meta-analysis studies, only the meta-analysis conducted by Chua et al.³⁶ provided pathogen-specific effect estimates for Salmonella and Campylobacter. The review by Chua et al.³⁶ covered studies published between 2000 and 2019 and found that the risk of Salmonella and Campylobacter infections increased by 5% and 2%, respectively, for every 1 °C rise in temperature. While the review assessed the risk of bias in individual studies, it did not assess the quality and strength of the synthesised evidence. Moreover, several Salmonella studies conducted before 2019 were not included in the review, and numerous relevant studies have been published since 2019.

To address these gaps, we conducted a systematic review and quantitative synthesis of the available epidemiological evidence aimed to answer the question “In the general population, what is the change in risk of non-typhoidal Salmonella and Campylobacter infections per 1 °C change in exposure to ambient temperature, observed in human epidemiological studies?”

Methods

Overview and study search strategy

This review followed the Navigation Guide systematic review methodology in environmental health as a guiding framework.^37,38 The Preferred Reporting Items for Systematic Review and Meta-analysis statement (PRISMA) was adhered to report the findings.³⁹ A detailed description is provided in the Supplementary material (Supplementary Table S1). The review protocol was registered in PROSPERO (CRD42022323608) and can be accessed online.⁴⁰

Based on the guidelines and framework for systematic review, we formulated the objective of the review using the Participants-Exposure-Comparisons-Outcome-Studies (PECOS) framework⁴¹ as follows:

• •Population: The general population.
• •Exposure: Ambient temperatures measured in different temporal resolutions (daily/weekly/monthly) and temperature metrics (mean, minimum and maximum).
• •Comparator: The risk of non-typhoidal Salmonella and Campylobacter infection case counts, incidence rates, morbidity, hospital admissions, or mortality that were estimated based on reference temperatures (optimal temperature, or lower level of exposure) or per 1 °C increase above a certain threshold as identified by the study authors.
• •Outcome: Non-typhoidal Salmonella and Campylobacter infection case counts or incidence rates
• •Study design: Quantitative observational human studies

In accordance with the stated objective, a comprehensive literature search strategy was developed (Supplementary Table S2) to identify relevant studies in the PubMed, Embase, Scopus, and Web of Science databases from January 1, 1990 to December 17, 2021 and was subsequently updated on March 1, 2024. The search utilised specific keywords related to health outcomes, including “diarrhoea” or “gastroenteritis” or “foodborne illnesses” or “Salmonella” or “Salmonellosis” or “Campylobacter” or “Campylobacteriosis.” These terms were combined with meteorological terms such as “temperature” or “heat” or “heatwave” or “weather” or “meteorological factor” or “climate” or “climatic factor” or “climate change” or “global warming”. In addition, a manual search was conducted by referring to previous review articles and the reference lists of included studies.

The articles obtained from all the above databases were imported into EndNote, and duplicates were removed. The shortlisted articles were then exported to Rayyan QCRI, the Systematic Reviews web application,⁴² for independent blind screening by two reviewers (YTD and MT). The two reviewers (YTD and MT) worked together to screen titles, abstracts, and full texts of potentially relevant records by applying the inclusion criteria. In case of any disagreements between the two reviewers, a third reviewer (BMV) was consulted, and the disagreement was resolved through discussion. The study selection process was documented in a flow chart following the PRISMA guidelines³⁹ (Fig. 1).

Fig. 1.

PRISMA flow diagram for literature search and study selection process of the systematic review and meta-analysis.

Inclusion and exclusion criteria

Studies were included in the review if they assessed the association between temperature and non-typhoidal Salmonella or Campylobacter infections and met the following criteria:

(1) Original, peer-reviewed literature published in English and reporting on an independent study population which refers to a dataset that is not overlapping or repeated in terms of spatial or temporal scale in relation to outcome and exposure measurements compared to other studies. (2) Exposure included high ambient temperatures, heatwaves, or hot weather, measured using any temperature metrics (mean, minimum, or maximum) with no restrictions on temporal resolution; and (3) Observational ecological time-series, case-crossover or case-series reporting comparative risks associated with different resolutions, including daily, weekly, monthly, or annual data on infection cases, morbidity, hospital admissions, mortality or incidence rates in the general population.

Studies were excluded if: (1) The enteric infection was not caused by Salmonella or Campylobacter (2) They only examined seasonal effects without explicitly considering the temperature metrics; (3) They investigated only outbreaks, presented only descriptive analyses without the exposure of interest; and (4) They were case reports, clinical trials, non-human studies, or were derived from conference abstracts, editorials, reviews, books, posters, or commentaries.

To ensure consistency and clarity in our analysis, we included studies from the same population and study period if they met the following criteria: (a) the studies provided location-specific estimates within a state or across different climate zones, (b) the studies employed different temporal resolutions for measuring exposure and outcome, such as daily, weekly, or monthly, (c) the studies utilized different temperature metrics, including minimum, mean, or maximum, (d) the studies employed varied lag period structures, such as single or cumulative lags. In cases where not all these criteria were met, we prioritized more recent studies and those with the longest study duration.

Data extraction

A standardised data extraction form was developed according to the inclusion criteria and based on the Cochrane Effective Practice and Organization of Care (EPOC) data collection resource guideline.^43,44 One of the authors (YTD) retrieved and extracted the following study characteristics: the last name of the first author, year of publication, study location (Continent, country, city, province, region or state), study period, sample size and sources of outcome data, assessment of exposure, assessment of outcome, confounding and effect modifiers, study design and statistical models, temperature metrics (mean/minimum/maximum) and temporal resolution (daily/weekly/monthly/annually), lag structure (single and cumulative lag: where the single lag approach focuses on examining the impact of temperature on health outcomes at a specific time lag after exposure and, the cumulative lag approach takes into account the delayed effects across a range of time periods), key findings, and effect estimates with corresponding 95% confidence interval (CI) and standard error (SE). Additionally, for studies with location-specific estimates (city, province, region or state), climate zones data were extracted using the Köppen–Geiger climate classification map⁴⁵ and climate data website.⁴⁶ In cases where studies had multiple lags, exposure metrics, and climate zones, the effect estimates, and all other characteristics were recorded separately.

Effect estimates, such as Incidence Rate Ratios (IRR) and Relative Risks (RRs), were extracted from text descriptions, tables, or figures in the main article or Supplementary materials. To ensure consistency, regression coefficients (β), Odds Ratios (OR), and percent change (PC) were all converted to RRs using established methods.47, 48, 49 In cases where effect estimates were presented solely in figures, the WebPlotDigitizer tool⁵⁰ was used to extract the necessary data. Lastly, for studies that only reported the effect size and p-value, we calculated 95% confidence intervals (CI) using the method of Altman and Bland.⁵¹

If a study reported multiple estimates for a pathogen serotypes, the original study’s pooled effect estimate was used if provided. If not provided, we conducted a fixed effect meta-analysis and used a pooled effect estimate.

For the studies that reported estimates from different models, we selected the model that included the relevant confounders (seasonality and time trend) and covariates (rainfall, public holidays and relative humidity).⁵² If a study provided both crude and adjusted estimates, we prioritised the adjusted estimates. When a study presented estimates derived from multiple temperature metrics, lag periods and lag structure we included all of them to calculate the overall pooled estimate. Later, we conducted subgroup analyses based on different categories. If a study reported estimates for multiple locations, estimates from all locations were included in the main pooled effect estimations. For non-linear model estimates which showed effect at both ends of temperature extremes, we chose the RRs for heat effects for the overall effect estimate to align with the comparator presented in the PECOS. For heatwave studies, we incorporated studies that used a reference temperature above which a heatwave was defined, or utilized lower and upper percentiles of temperature, such as the 90th, 95th, or 99th percentiles, relative to the reference temperature to define heatwaves. We calculated the relative risk (RR) increases between these percentiles and then determined the change in RR per 1 °C increase in temperature, considering the temperature difference between the two percentile ranges.

Finally, all effect estimates were standardised to account for a 1 °C increase in temperature, enabling the pooling of effect estimates. In cases where studies reported the increase in effect estimates per X degree Celsius rise in temperature (where X differs from 1), the log-linear of the effect estimate was divided by X and then standardised for a 1 °C increase in temperature.⁴⁷

Data synthesis and meta-analysis

The study findings were synthesised both quantitatively and narratively. Quantitatively, we conducted a separate random-effects meta-analysis for both enteric infections. For the studies that did not meet the criteria for meta-analysis, the key findings were summarized and synthesised narratively.

Given the diverse nature of the included studies, we calculated the pooled effect estimates and corresponding confidence intervals using the DerSimonian and Laird procedure.⁵³ Forest plots were utilised to present the findings of each individual study, as well as the combined results. We have quantified the amount of statistical heterogeneity among the studies using the Higgins I² statistics which measures the proportion of observed variability not caused by sampling error.^49,52 The heterogeneity was categorised as low (I² ≤ 25%), moderate (25% < I² < 75%), or high (I² ≥ 75%).

Moreover, we estimated prediction intervals (PIs) for the effect size measure in the population. The PI determines where the effect estimate of new observations from future studies across different settings would fall based on the existing evidence.^54,55 Severe heterogeneity was suspected when the 80% PI included the null effect and was more than twice as wide as the 95% CI.⁵⁶

After assessing the degree of heterogeneity, subgroup analyses were conducted based on various factors including climate zones,^45,46 temperature metrics, temporal resolution (daily, weekly, monthly, or annually), lag structure, lag period for single lag studies, administrative divisions, continents, and World Bank income categories. These analyses were performed when the total number of effect estimates was two or more (i.e., k ≥ 2), and location (city, district, province or state level) effect estimates were used in the climate zone subgroup analysis. Climate zone subgroup analysis was conducted exclusively for studies carried out at the city or state level, with a focus on a specific climate zone. In cases where location-specific data were available, climate zones were determined by extracting information from the Köppen-Geiger climate classification map. Nationwide (except countries which have only one climate zone) and multinational studies were excluded from the climate zone subgroup analysis.

We also conducted meta-regression using location-specific potential confounders that could explain the relationship between the effect estimates of Campylobacter and Salmonella infections and various factors associated with the study locations. These factors include latitude, longitude, mean temperature, gross domestic product (GDP) per capita of countries, continent where the study was conducted, Köppen–Geiger climate zones, and the administrative level of the study location. Factors such as a country’s gross domestic product (GDP) per capita were considered because many studies were conducted at a nationwide level or covered most states within a single country. Thus, it was assumed that the study samples were representative of their respective countries or regions.

Sensitivity analyses were conducted in multiple steps to assess the robustness of the pooled effect estimate when certain assumptions or study characteristics are varied. Firstly, following the Navigation Guideline systematic review methodology, studies with a high risk of bias were excluded, and the pooled effect estimate was reanalyzed.^38,57 Secondly, leave-one-out analysis and outlier detection were performed to identify the influence of individual estimates.⁵⁸

Thirdly, the studies were analysed separately based on their adjustment for primary confounders, covariates, or both. This process involved categorising the studies into subgroups according to whether they adjusted for primary confounders, covariates, or both primary confounders and covariates, followed by the recalculation of the effect estimates within each subgroup.

Fourthly, sensitivity analyses were conducted by categorising studies into linear and non-linear modelling techniques to evaluate the extent to which the choice of modelling approach influences the effect estimate.

Finally, to address the dependency of multiple effect estimates derived from a single study, robust variance estimation (RVE) was employed. RVE facilitates the inclusion of all estimates from a study while accounting for the correlations among them. Importantly, it does not necessitate the specification of the exact correlation structure among the multiple estimates, thereby rendering it a flexible tool for managing complex data dependencies.^59,60

Publication bias was assessed visually through the use of funnel plots and Egger linear regression tests to detect any asymmetry.⁶¹ The trim-and-fill method was then utilised to adjust the overall estimates, considering the number of studies that may be missing due to publication bias.⁶²

Statistical analysis was performed using the R software version 4.0.2⁶³ with the packages meta,⁶⁴ metafor,⁶⁵ forester⁶⁶ and dmetar.⁶⁷ A p-value of <0.05 was considered statistically significant unless otherwise reported.

Evidence synthesis

The modified version of the Navigation Guide framework for systematic reviews in environmental health, which was developed based on the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) methods, was employed to assess the quality and strength of evidence presented in the included studies.^37,38,68 Details for each step are presented in the Supplementary material (SM-pp.6).

A three-step evidence assessment was applied: First, an assessment of the risk of bias in individual studies was conducted based on a modified version of the Office of Health Assessment and Translation (OHAT) risk of bias rating tool for human and animal studies.^44,57,69 Each study was evaluated against six domains of the risk of bias and the overall rating of an individual study was determined by a combination of risk of bias ratings in three key components of bias (outcome assessment, exposure assessment and confounding assessment)⁷⁰ (Supplementary Tables S3 and S4).

Second, the quality of the evidence was assessed across the studies. According to the Navigation Guide for observational studies, we initially rated the quality of evidence as “moderate”. The final overall quality of evidence was changed to “high”, “moderate”, or “low” by upgrading or downgrading the “moderate” rating based on downgrading and upgrading factors as detailed in the Supplementary material (Supplementary Table S5).

Third, we assessed the strength or certainty of the evidence across studies by combining the overall quality of evidence rating with the direction of effect estimates and other data attributes that may affect certainty. The overall strength or certainty of the evidence was rated as “sufficient evidence,” “limited evidence,” or “inadequate evidence,” following the guidelines in the Navigation Guide framework (Supplementary Table S6). Two authors (YTD and MT) performed the risk of bias assessment and evaluated the body of evidence.

Role of funders

The funder of the study had no role in study design, data collection, data analyses, data interpretation, or writing of the report.

Results

Selection of studies

Search queries in the four databases initially yielded 3472 studies, of which 1009 were duplicates and were hence removed. The remaining 2463 articles underwent title and abstract screening, resulting in the exclusion of 2349 studies. We conducted a full-text assessment of the remaining 114 articles and an additional three articles that were identified from the reference lists of studies. In total, 44 studies were included in the review, of which 22 examined Salmonella infection, 15 investigated Campylobacter infection, and seven studies^13,14,28,71, 72, 73, 74 assessed both infections. The detailed study selection process is presented in the flow diagram (Fig. 1).

Description of selected studies

A detailed description and summarised key findings of the included studies are presented in the Supplementary material (SM-pp. 22 and Supplementary Tables S7 and S8). Geographically, the 44 studies covered 27 countries across five continents. Fourteen studies were nationwide,^22,29,72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83 two were multi-nation studies,^15,19 and the rest were conducted at various administration divisions such as district, city, or province/state level.

The study areas were located across eight climate zones as classified by Koppen-Geiger classification,⁴⁵ with 64.4% of studies conducted within three climate zones (Humid subtropical climate, Warm humid continental climate and Oceanic climate). The highest proportion of studies (24.4%) were conducted in the Humid subtropical climate zone (Fig. 2).

Fig. 2.

Location of studies based on Köppen-Geiger climate zone classification across continents. Details for Köppen-Geiger climate zone abbreviations are described in the Supplementary material (SM-pp. 23).

National notifiable diseases surveillance systems and regional health databases from different countries were the main sources of outcome data in most studies (n = 42) and hospital records were used in two studies.^84,85 Exposure data were collected from national meteorological agencies in 40 studies. The remaining four studies used remotely sensed and validated temperature data.^19,22,75,79 Studies used different temperature metrics, with the mean temperature being widely used among them (n = 33).

The weekly temporal resolution was the most frequently used exposure measurement (n = 20), followed by monthly resolution (n = 16).^{12,22,74,75,77,}84, 85, 86, 87, 88, 89, 90, 91, 92 Daily and annual temporal resolution was also used in four^20,30,93,94 and two^13,78 studies, respectively. Two studies^15,95 applied more than one temporal resolution.

Except for one study,⁹⁶ which used a case-crossover study design, the remaining studies employed a time-series study design. Except for four studies,^13,78,85,90 all studies adjusted for the primary confounders: seasonality and time trend. Most studies also adjusted for other relevant covariates, including rainfall (n = 33), relative humidity (n = 12) and public holidays (n = 12). The studies reported effect estimates in terms of IRR, RR, OR percent change and regression coefficients.

The overall risk of bias assessment is presented in Fig. 3, and detailed domain specific ratings for each study are available in Supplementary Tables S9 and S10. Out of 44 studies, 40 studies received an overall rating of “Definitely Low” or “Probably Low” risk of bias, while three studies had “Probably High” risk of bias as they did not adjust for primary confounders. One study was classified as having an overall “Definitely High” risk of bias.

Fig. 3.

Percentage distribution of overall risk of bias ratings within each bias domain for all included studies. (a) non-typhoidal Salmonella infection studies (n = 29), (b) Campylobacter infection studies (n = 22). The plots were created using the Risk of Bias Visualization Tool (ROBVIS).⁹⁷

Meta-analysis results

The pooled effect estimates from random-effects meta-analyses showed that the risk of infection increased by 5% for both non-typhoidal Salmonellosis (RR: 1.05, 95% CI: 1.04–1.06, I² = 95.5%) and Campylobacteriosis (RR: 1.05, 95% CI: 1.04–1.07, I² = 98.0%) with each 1 °C increase in temperature (Supplementary Figs. S3 and S4).

A subgroup analysis for both enteric infections is summarized in Fig. 4, Fig. 5. Regarding Salmonella infection, statistically significant differences were observed between groups except for temperature metrics and temporal resolution subgroups (Fig. 4 and Supplementary Figs. S5–S11). Compared to the overall pooled effect estimate, higher RR values were found in the Humid subtropical climate zone (RR 1.10; 95% CI: 1.06–1.15), Oceania Continent (RR 1.10; 95% CI: 1.06–1.13), and Cumulative lag (RR 1.09; 95% CI: 1.06–1.11).

Fig. 4.

Random-effects meta-analysis findings for Salmonella infection studies by subgroup showing change in RR and 95% CIs per 1 °C increase in temperature (k = number of effect estimates).

Fig. 5.

Random-effects meta-analysis findings for Campylobacter infection studies by subgroup showing change in RR and 95% CIs per 1 °C increase in temperature (k = number of effect estimates).

Meanwhile, for Campylobacter infection studies, statistically significant differences were observed between groups except for the subgroup Continent (Fig. 5 and Supplementary Figs. S12–S18). A higher RR was observed in the hot-summer Mediterranean climate zones (RR 1.09; 95% CI: 0.96–1.25), despite a wider confidence interval. Similar results were found in multi-nation studies (RR 1.11; 95% CI: 1.03–1.18) and lag 1 for single lag studies (RR 1.07; 95% CI: 1.02–1.12). In the other subgroups, the RR in each group was closer to the pooled effect estimate.

We performed a meta-regression analysis using location specific potential confounders to examine how various study-level characteristics related to the observed effect estimates. For Salmonella infection studies, we observed a positive association between the annual mean temperature of the study locations (Wald test, p = 0.01) and the effect estimates obtained from the Oceania continent (Wald test, p = 0.02), as compared to the estimates obtained from Asia. However, we did not find any association between latitude, longitude, and the GDP per capita of countries with effect estimates for Salmonella infection studies (Supplementary Table S11).

Meanwhile, there was a significant positive association (Wald test, p < 0.01) between the GDP per capita of countries and effect estimates for Campylobacter infection studies. Additionally, it was observed that effect estimates derived from the Mediterranean warm summer climate zone exhibited a positive association compared to the Warm-summer continental climate zone (Wald test, p < 0.01). However, we did not observe any associations between effect estimates for Campylobacter infection studies and continent, latitude, longitude, or annual mean temperature of the study locations (Supplementary Table S12).

Sensitivity analysis and publication bias

The leave-one-out analysis and outlier detection analyses for both enteric infections did not yield a different effect estimate from the original findings (Supplementary Figs. S19 and S20 and Tables S13 and S14) Removing studies with a high or definitely high risk of bias did not alter the overall pooled effect estimate for both infections. The sensitivity analysis comparing studies that adjusted for primary confounders, covariates, or both did not reveal any statistically significant differences from the main pooled effect estimate for either enteric infection. However, studies that adjusted only for confounders (RR = 1.08 95% CI: 1.06–1.10) and the groups that did not include covariates (RR = 1.08 95% CI: 1.06–1.09) exhibited higher effect estimates for Salmonella infection. Likewise, the robust variance estimation also produced similar results to the pooled effect estimates (Supplementary Tables S13 and S14).

For non-typhoidal Salmonellosis the funnel plot showed asymmetry (Supplementary Fig. S21) and Egger’s test for a regression intercept gave a p-value of 0.005. As a result of the observed asymmetry of the funnel plot an adjustment of the effect estimate using the trim and fill method was performed (Supplementary Fig. S22) and resulted in a pooled effect (RR: 1.04; 95% CI 1.02 to 1.05) which was not far from the overall pooled estimate (RR: 1.05, 95% CI: 1.04–1.06). However, For Campylobacteriosis the funnel plot and Egger’s test for a regression intercept show no evidence of publication bias (Supplementary Fig. S23).

Narrative synthesis

A narrative synthesis was conducted for 11 studies, which consisted of five Campylobacter infection and four Salmonella infection studies, and two studies that examined both infections. These studies did not meet the criteria to be included in the quantitative meta-analysis (Supplementary Tables S7 and S8). As for Salmonella infection studies included in the narrative synthesis, all^{12,28,74,76,77,98} reported a positive association between higher temperature and infections at different lag periods. Among the studies included in the qualitative analysis, one study found no evidence of a link between temperature and Campylobacter infection.⁷⁵ The rest of the studies^{28,74,79,85,90,91} found a positive association.

Discussion

The synthesis of the findings from the reviewed studies suggests an association between temperature exposure and both Salmonella and Campylobacter infections. The pooled effect estimates per 1 °C increase in temperature showed that the overall risk of both Salmonella and Campylobacter infections increased by 5%. The direction of effect estimates for both enteric infections is consistent with previous reviews that demonstrated positive associations between ambient temperature and all-cause and pathogen specific enteric infections caused by bacteria including Salmonella and Campylobacter.^35,36 The magnitude of the pooled effect estimate from this study is similar in the case of Salmonella infection. However, the effect estimate for Campylobacter infection was higher compared to the previous finding,³⁶ which showed a 2% increase per 1 °C rise in temperature. The narrow confidence interval suggests that, despite the significant heterogeneity among studies, the overall pooled effect estimate is consistent and robust for both enteric infections. Moreover, the prediction interval did not include null for Salmonella infection, suggesting that future studies are unlikely to overturn the observed effect. Therefore, this strengthens the certainty of the available evidence indicating a strong association between increasing temperature and the risk of Salmonella infection.

The observed positive association between enteric infections and temperature can be elucidated through multifaceted ecological processes. Elevated temperatures directly influence pathogen growth and replication, while also indirectly impacting food storage, dietary habits (such as consuming raw or undercooked foods), and food handling practices during warmer weather.^99,100

For Salmonella, temperature plays a pivotal role in pathogen proliferation. Salmonella exhibits optimal multiplication within a temperature range of 35 °C–37 °C, because this range closely matches the normal body temperature of warm-blooded hosts, including humans. Conversely, when temperatures drop below 15 °C, the proliferation of Salmonella is significantly reduced.^14,22 Higher temperature also coincides with food contamination during warmer seasons underscoring the impact of ambient temperature on bacterial load in food, thereby elevating the risk of food-borne infections. Unlike other infectious diseases, such as vector-borne infections, which exhibit bell-shaped associations with temperature,101, 102, 103 Salmonella infections consistently increase with in the above mentioned temperature ranges.^15,17,27,77

In the case of Campylobacter species, the mechanism of temperature that influences incidence of Campylobacteriosis is not only directly linked to the relationship between temperature and the pathogen, but rather its impact on the entire infection chain. Studies have identified major risk factors for Campylobacter infection, including chicken consumption, contact with raw meat, and ingestion of unpasteurized milk or barbecued beef, all of which are influenced by changes in ambient temperature.^{26,99,100,104} Warmer temperatures also increase the abundance of flies that act as mechanical vectors in Campylobacter transmission to humans.¹⁰⁵

The subgroup analysis at the climate zone level for both enteric infections revealed statistically significant differences between groups. For Salmonella infection, higher relative risks were observed in humid subtropical climate zones (RR = 1.10, 95% CI: 1.06–1.15) compared to the other subgroups and the overall pooled estimate. Conversely, hot humid continental and warm humid continental climate zones exhibited identical but lower relative risks (RR = 1.03, 95% CI: 1.02–1.04) compared to the overall pooled estimate. This difference could be due to the consistently warm and moist weather conditions found in humid subtropical climate zones throughout the year. These conditions may support the growth and replication of the Salmonella pathogen in the reservoirs and within the food chain.^104,106 In contrast, continental climate zones generally experience low mean temperatures for most of the season, which may inhibit the growth and replication of the Salmonella pathogen in reservoirs and within the food chain.

Similarly, subgroup analysis at Continent level indicated higher relative risks (RR = 1.10, 95% CI: 1.06–1.13) in the continent of Oceania compared to other regions. This finding may be linked to the predominance of tropical climate zones within this continent. However, the absence or limited number of studies from other regions, particularly Africa and Asia, makes it difficult to draw definitive conclusions and more studies in these regions are encouraged.

Meanwhile, subgroup analysis for Campylobacter infection studies showed significant association in warm humid continental climate zones but insignificant RR in hot-summer Mediterranean climate zones (RR = 1.09, 95% CI: 0.96–1.25). This finding is contradictory, given the limited growth of the Campylobacter pathogen in regions with colder climates. Generally, the occurrence of Campylobacteriosis decreases as temperatures drop below the optimal temperature for Campylobacter proliferation. On the other hand, during the warmer months when temperatures rise within the optimal range, cases of Campylobacteriosis tend to be more prevalent. However, other environmental factors, such as humidity, water sources, and agricultural practices, can significantly impact the occurrence and transmission of Campylobacter. In the regions with warm, humid, continental climates, these conditions are particularly favourable for the growth of the bacteria. Even a slight rise in temperature can significantly increase the risk of Campylobacter infections, enabling the bacteria to thrive and spread through water and food sources.^14,27

We also conducted subgroup analysis for single-lag studies for both enteric infections. Our analysis revealed consistent estimates across various lag periods. These findings strongly support the notion that temperature influences the occurrence of Salmonella and Campylobacter infections through all stages of the food supply chain.

The significant heterogeneity observed in both enteric infection studies aligns with previous reviews on enteric infections.^35,36 Several factors could contribute to this heterogeneity. Firstly, outcome measurement and data sources could be sources of variability. Gastroenteritis infection reporting is frequently subject to underreporting and exhibits variability across data sources in passive surveillance systems. Consequently, this can lead to substantial heterogeneity between studies that have relied on different data sources. Additionally, cases requiring medical attention may not be reported promptly, causing a temporal mismatch between exposure and case reporting.

Secondly, substantial heterogeneity could be attributed to unaccounted socio-demographic factors. The occurrence of enteric infections differs based on age, with young children and older individuals being more susceptible due to weaker immune systems and inadequate hygiene practices. However, there have been limited studies exploring the incidence of these infections in relation to sex, age, and other socioeconomic factors. This limitation hinders our ability to evaluate their impact and, as a result, they were not considered in the meta-regression analysis.

The third aspect to consider is the definition of effect estimates associated with a 1 °C increase in temperature above the reference point. These estimates vary by location, even within similar climate zones with different temperature ranges. This creates inconsistency, as the optimal reference temperature in one climate zone may be the lowest or highest threshold in another.

Finally, other factors, such as the modelling approach, adjustment for primary confounders and covariates could contribute to the observed heterogeneity. It is worth noting that there could also be unidentified factors which were not acknowledged in the analysed studies but contribute to the observed level of heterogeneity.

We found high quality of evidence on the associations between temperature and both enteric infections (Table 1). Moreover, the overall pooled estimate also showed an increase in both enteric infections with increasing temperature, and sensitivity analyses confirmed the robustness of the result. Thus, the strength of evidence was “sufficient” to show the effect of temperature on increased enteric infection for both pathogens (Table 1). Moreover, the inclusion of large number of studies with definitely low and probably low risk of bias that adjusted for relevant confounding factors makes the estimate worth consideration in terms of preventive health interventions in the context of climate change.

Table 1.

Summary of the body of evidence on temperature as a risk factor for Salmonella and Campylobacter infections.

Salmonella infection studies (n = 29)			Campylobacter infection studies (n = 22)
Quality of evidence assessment
	Rating	Rationale and summary of evidence	Rating	Rationale and summary of evidence
Down grading factors
Risk of bias	0	Except for two studies that were rated as “definitely high” or “probably high” risk of bias, the remaining studies (27/29) were rated as “definitely low” or “probably low” risk of bias. Removing studies with probably high or definitely high risk of bias did not change the overall pooled effect estimate. Therefore, we had no significant concerns regarding bias in the body of evidence.	0	There is no substantial risk of bias since only three studies were rated as “definitely high” or “probably high”. The remaining studies (19/22) were rated as “definitely low” or “probably low” risk of bias. Removing two studies with probably high or definitely high risk of bias did not change the overall pooled effect estimate.
Indirectness	0	We did not have any major concerns about the definition of the outcome as Salmonella cases included in all studies were either laboratory confirmed or clinically diagnosed. Additionally, most of the studies used direct measurements of exposure.	0	Campylobacter cases included in most studies were identified either through laboratory confirmation or clinical diagnosis. Furthermore, most of the studies employed direct measurements of exposure.
Inconsistency	0	A significant level of heterogeneity was observed in the overall pooled estimate, with an I2 value of 95.5%. However, it is worth noting that the 80% prediction interval (PI) did not encompass unity (Supplementary Fig. S3). Furthermore, “leave-one-out” analysis, did not reveal significant variations in the overall pooled effect estimates (RR = 1.05 to 1.05). Additionally, there were no changes in statistical significance.	0	There was substantial heterogeneity (I2 = 99.2%, p < 0.0001) even after conducting subgroup analyses. The 80% prediction intervals include unity and is wide (Supplementary Fig. S4). However, the sensitivity analysis using the “leave-one-out” approach did not identify any significant deviations from the overall pooled effect estimates (RR: 1.05 to 1.06) and there was no change in the statistical significance.
Imprecision	0	Except for five studies, the confidence intervals of the remaining studies were notably narrow (95% CIs). Furthermore, we observed that the majority of the studies included a representative proportion of the population of interest over multiple years.	0	Only three effect estimates (3/22) possessed wider confidence intervals. Furthermore, we observed that the majority of the studies included a representative proportion of the population of interest over multiple years.
Publication bias	0	Despite the observed asymmetrical distribution of studies in the funnel plot and Egger’s tests (Supplementary Figs. S21 snd S22), the pooled effect estimate adjusted after including imputed studies using the Trim and Fill method (RR 1.05; 95% CI: 1.04 to 1.05) did not show difference in statistical significance.	0	Egger’s test (p = 0.156) and the funnel plot did not show publication bias (Supplementary Fig. S23).
Upgrading factors
Large magnitude of effect	0	Effect sizes were small in most of the studies, with the pooled effect estimate being below 2.	0	The effect sizes in the majority of the studies were small, with the pooled effect estimate being less than 2.
Dose response pattern	+1	The quality of the evidence was upgraded because the majority of the studies (27/29) indicated a statistically significant increase in the risk of Salmonella infection with an increase in temperatures beyond reference values.	+1	The quality of the evidence was upgraded, as the majority of the studies (20/22) demonstrated a statistically significant association between an increase in temperatures above reference values and an increased risk of Campylobacter infection.
Confounding minimizes effect	0	Except for one study, all the other studies considered primary confounders. Additionally, most studies considered potential confounders through modelling, and we do not expect any confounding factors that would underestimate the effect estimate.	0	Four studies (two of them included in the meta-analysis) did not adjust for primary confounders. However, the sensitivity analysis did not show evidence to suggest that possible residual confounders would shift the effect to null.
Summary of the quality assessment
Quality of evidence	High	Moderate + (0) +(0) +(0) + (1) = 1, resulting the overall quality of evidence rated as “High”.	High	Moderate + (0) +(0) +(0) + (1) = 1. After down grading two quality factors, the overall quality of evidence was rated as “low”
Summary Findings	n/a	Overall, the studies included in the meta-analysis and narrative synthesis indicated a “high” quality of evidence linking rising temperatures to an increase in Salmonella cases.	n/a	The studies, both included in the quantitative analysis and the narrative synthesis, provided a “high” quality of evidence that links rising temperatures to an increase in Campylobacter cases.
Strength of evidence assessment
Quality of evidence	n/a	High	n/a	High
Direction of effect Estimates	n/a	The direction of effect estimates largely showed an increasing trend in Salmonella infection with increasing in temperatures beyond the reference values.	n/a	The direction of effect estimates largely showed an increasing trend in Campylobacter infection associated with temperature, except one study. The confidence intervals were narrow making the result certain.
Confidence in effect estimate	n/a	The studies have shown a positive association between Salmonella infection and temperature. The PI also showed a positive association in future studies. The meta-analysis includes a sufficient number of studies, providing confidence in the findings for the future.	n/a	Although an increasing trend was observed in the overall random effect the confidence and prediction intervals contain the null effect. Moreover, the number of studies included in the meta-analysis was small and made the result uncertainty.
Other characteristics of the data that may influence certainty	n/a	The sensitivity analysis, which considered various factors, did not reveal significant variation from the pooled overall effect estimate. This analysis further enhances the certainty of the evidence (Supplementary Table S13).	n/a	The sensitivity analysis, which considered various factors, did not reveal significant variation from the pooled overall effect estimate. This analysis further enhances the certainty of the evidence (Supplementary Table S14).
Overall strength of evidence	Sufficient	The quality of the evidence was rated as “High”. The prediction interval suggested that future studies are unlikely to overturn the observed effect, and various sensitivity analyses consistently yielded similar overall estimates. Consequently, we have reached the conclusion that the available evidence is sufficiently robust to demonstrate the impact of temperature on Salmonella infection.	Sufficient	The available evidence consists of the results from one or more meticulously designed and executed studies. The quality of the evidence was rated as “High”. So, the evidence is likely to hold up in possible future studies and we have concluded that the evidence is sufficient to demonstrate the impact of temperature on Campylobacter infection.

This systematic review and meta-analysis have several strengths. Firstly, it was based on a pre-registered protocol and employed a rigorous, transparent, and previously established method to evaluate and synthesise evidence from observational environmental health studies.^38,40,57 Secondly, while previous reviews have evaluated the influence of temperature on enteric infections such as Salmonella and Campylobacter, our comprehensive search approach yielded a greater number of studies. In addition to recent studies published after 2019, we included studies (n = 18) overlooked in previous reviews.28, 29, 30^,74, 75, 76^{,78,80,82,83,85,}88, 89, 90^,94,98,107 Unlike prior studies, we did not exclude studies that only provided effect estimates and p-values without confidence intervals; instead, we used appropriate methods to calculate confidence intervals and incorporated them into our analysis.⁵¹ The larger number of studies included resulted in a narrower and significant prediction interval, providing stronger evidence for the association between temperature and Salmonella infection, which was uncertain in previous review.³⁶ Moreover, the inclusion of additional new studies could potentially explain the higher pooled effect estimate (5% increase per 1 °C rise in temperature) for Campylobacter infection, as compared to the previous review.³⁶ Contrary to prior studies that included only specific lag periods, temperature metrics, and the largest effect estimates, we included multiple effect estimates stratified by factors such as climatic zones, lag period and temperature metrics. The inclusion of multiple effect estimates is more likely to capture non-significant results, helping to mitigate publication bias and provide more robust effect estimates.³⁵ Meanwhile, we also checked the possible dependency from multiple effect estimates from a single study and found no difference from the overall pooled estimates. Finally, we used the established Navigation Guide framework^38,57 to evaluate the quality and strength of evidence, thus enhancing the accuracy and confidence of the systematic review findings on this topic.

This study has three limitations. Firstly, despite our efforts and comprehensive search, we may still have overlooked eligible studies published in other languages and data from grey literature. Moreover, the data extraction was performed by a single individual, despite being verified by a second person. To improve the accuracy and reliability of the data, it would be ideal to duplicate the data extraction process. Future research should aim to include multiple reviewers in the data extraction process to minimize this limitation.

Secondly, although our findings show a positive association between temperature and enteric infections, the lack of relevant studies from low- and middle-income countries (LMICs), particularly in regions like Africa and Asia, challenges the generalisability of these pooled estimates. Most of the reviewed studies originated from high-income countries, despite a higher incidence of enteric infections in low- and middle-income countries (LMICs), particularly in Africa and Asia.^108,109 Climate change significantly impacts LMICs’ vulnerability to enteric infections, exacerbated by inadequate sanitation, poor water quality, and limited healthcare access. Rising temperatures combined with these challenges can lead to higher gastroenteritis incidence. Therefore, more research is needed to understand the effects of temperature on enteric infections in these areas.^110,111

Thirdly, the association between temperature and enteric infections may be influenced by socio-demographic, behavioural, and other environmental factors as indicated in previous studies.^13,85,112 However, few published studies included in this review have examined these factors, and we were unable to explore them further due to a lack of available data in the included studies. Future studies should investigate the influence of socio-demographic and other environmental factors to clarify local and regional vulnerabilities. We also acknowledge the number of Campylobacter studies included in the meta-analysis was relatively small (n = 15). Thus, the quantitative association needs to be interpreted cautiously.

The existing body of evidence from epidemiological studies suggests that high temperatures increase the risk of enteric infections. The association between temperature and enteric infections varies across different climate zones, continents, and income levels. However, the limited number of studies from low- and middle-income countries suggest that more research is needed to understand the effects of temperature on enteric infections in Asia and Africa. Nevertheless, the overall results demonstrate the importance of implementing focused interventions and adaptive measures, such as the establishment of localised early warning systems and preventive strategies that account for climatic variations. These findings could aid in health service planning and inform the development of tailored preventive measures and optimizing resource allocation. Furthermore, our research emphasises the ongoing need for surveillance and research efforts to monitor and understand the changing dynamics of temperature-related enteric infections in the context of climate change.

Contributors

YTD was involved in the literature search, screening the studies, assessing the risk of bias, study design, data collection, data analysis, data interpretation, figure creation, and writing of the manuscript. MT was involved in the literature search, screening the studies, assessing the risk of bias, data collection, data interpretation, reviewing and revising the manuscript. BMV and OA oversaw the meta-analysis design, execution, statistical analysis, data interpretation, and was involved in reviewing and revising the manuscript. PB was involved in all stages of the project, including conception, design, data analysis, data presentation, interpretation, reviewing, and revising the manuscript. AH, KD, YZ, TD, TC were involved in conception, design of the project, reviewing and revising the manuscript. All authors made substantial contributions to each part of the content. YTD and MT have accessed and verified the underlying data. All authors read and approved the final version of the manuscript and approved submission for publication.

Data sharing statement

This review utilised publicly available data; therefore, no original data are available for sharing. However, the data extracted for meta-analysis have been deposited in figshare can be accessed online at https://figshare.com/s/57a4fdc98e8c7b8704c6.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.