Background

Accidental hypothermia is defined as a drop in core temperature below 35°C. If untreated, it impairs metabolic processes, induces respiratory and circulatory depression, and may lead to life-threatening arrhythmias and cardiac arrest. It may develop despite intact thermoregulation due to cold exposure or pathological thermoregulatory failure [1]. In winter and early spring, the incidence increases [2,3]. As global climate change reduces extended cold periods, strong, brief cold spells may increase, placing older, chronically ill, and socially marginalized people in danger [4,5].

Two tools are generally used for hypothermia staging, when direct core temperature is not available (Table 1). The Swiss staging model combines physiological responses and symptomatology with predefined core temperature ranges [6]. A revised version estimates circulatory-arrest risk when core temperature is not available [7]. The field-use Wilderness Medical Society classification even offers guidance on rewarming techniques for each severity level [8].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Hypothermia severity classification systems [6–8].

https://doi.org/10.1371/journal.pone.0334328.t001

Emergency triage systems assess hypothermia in different ways, prioritizing patients by evaluating the urgency of their condition, frequently employing five-category scales [9]. Despite this common foundation, these systems differ in the detailed criteria they use for patient assessment and coexisting specific medical conditions, also in terms of hypothermia (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Five-level emergency triage systems [9–13].

https://doi.org/10.1371/journal.pone.0334328.t002

In the Manchester Triage System (MTS), a core temperature below 35°C is a modifier for category 2, but the final decision is made with additional symptoms and clinical factors considered. [10]. Hungary used a modified version of the widely used, multidimensional Canadian Triage and Acuity Scale (CTAS) [11]. In the CTAS-based Hungarian Emergency Triage System (MSTR), a core temperature between 32°C and 35°C is sufficient for classification into category 3, whereas a core temperature less than 32°C is an independent modifier for assigning patients to category 2 (S1–S2 Tables), indicating the need for immediate intervention [11, 12]. Croatia is among the countries that have implemented the Australasian Triage Scale, despite its rarity in Europe. It places life-threatening conditions above the severity of hypothermia on its own. [13–15]. When core temperature is a unique modifier, triage system criteria may affect categorization and even treatment [9]. Demographics and critical outcomes may serve as a foundation for local procedure adaptation [16].

In real-world emergency care, the predictive accuracy and clinical relevance of specialized hypothermia classification systems—along with conventional triage protocols applied to hypothermic patients—remain insufficiently documented. Furthermore, epidemiological data on hypothermia in Hungary, including seasonal variation, patient demographics, and emergency department outcomes, are notably scarce. To address these gaps, we conducted a retrospective study of hypothermic patients treated at the Department of Emergency Medicine, Semmelweis University, between 2020 and 2024.

Our objectives were as follows:

• Analyze the demographic characteristics and seasonal trends of hypothermia cases.
• Evaluate the predictive performance of the Hungarian Emergency Triage System – as it is the nationally mandated, CTAS-based protocol in use across Hungary – and of the severity categories defined by measured tympanic temperature thresholds in the Swiss staging model and Wilderness Medical Society guidelines for identifying patients at risk of critical outcomes.
• Assessing whether incorporating the admission tympanic temperature into the MSTR improves risk stratification.
• Investigate the impact of outdoor temperature on hypothermic patient outcomes

Methods

A retrospective cohort analysis was conducted to identify patients presenting with hypothermia to the Department of Emergency Medicine, Semmelweis University, between January 1, 2020, and December 31, 2024. During this period, a total of 197 751 emergency department visits occurred (31 264 in 2020; 41 571 in 2021; 38 777 in 2022; 39 771 in 2023; and 46 368 in 2024). Hungarian emergency departments are required to operate at least one ICU-equipped resuscitation bay and maintain 4–8 monitored beds, including at least one capable of intensive-level care [17]. The inclusion criteria were ICD-10 code T68 and a tympanic temperature below 35.0°C at ED admission, which was manually retrieved from the EHR on January 5, 2025 (eMedSolution system database version 2024/Q4/1). Patients with a tympanic temperature of 35.0°C or higher and those with incomplete data were excluded. (Fig 1). Ethical approval for the study was granted by the Semmelweis University Regional and Institutional Review Board (SE RKEB 274/2024). Only fully de-identified, retrospectively collected clinical records were accessed and analyzed. Consequently, individual informed consent was waived.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Patient selection flowchart.

https://doi.org/10.1371/journal.pone.0334328.g001

Demographic data (age, sex, residence) and clinical characteristics (mode of arrival, MSTR triage category, hypothermia severity) were recorded. Categorical variables are reported as frequencies and percentages, age as the median (IQR), and tympanic temperature as the median (IQR). Hypothermia severity was categorized via the Swiss staging model and the Wilderness Medical Society classification based on standardized measurements of tympanic temperature ranges at admission. Admission tympanic temperature used as a pragmatic proxy for core temperature was assessed using Braun ThermoScan PRO 6000 tympanic thermometer, which uses infrared technology (Hillrom Inc.) [18]. The instrument displays values between 20 °C and 42.2 °C, clinical accuracy is ± 0.2 °C within the 35.0-42.0 °C range and ±0.3 °C outside this interval. These limits conform to ASTM E 1965−98 requirements for infrared thermometers (S3 Table). Hypothermia severity was then categorized according to the ranges specified in the Swiss Staging System and the Wilderness Medical Society classification. Ambient temperature was exported from the monthly mean temperature data from the local meteorological station based on the registered residence of the patients (S4 Table) [19]. Data normality was evaluated via the Kolmogorov‒Smirnov and Shapiro‒Wilk tests.

Critical outcomes were defined as mortality during emergency care or admission to the ICU to capture all critical events requiring intensive care during the ED stay and to reflect the clinical relevance of both early mortality and the need for high-level intervention in a setting with limited intensive care availability. Relationships among categorical variables were assessed via the chi-square test. The effect size was quantified via Cramer’s V. The Mann‒Whitney U test and rank‒biserial correlation coefficients were implemented for continuous data that were not normally distributed. Pearson’s correlation coefficient was used to assess the association between normally distributed data. Univariate logistic regression models were implemented to evaluate the correlations of the triage category, admission tympanic temperature, WMS, Swiss staging, and ambient temperature with the primary outcome. A multivariate logistic regression model was employed to calculate odds ratios with 95% confidence intervals, p values, and Wald statistics for components with low collinearity, as determined by the variance inflation factor (VIF). We further assessed potential collinearity between triage category (MSTR) and admission tympanic temperature using Pearson’s and Spearman’s correlations in combined models (S7 Table). In the multivariate models, we evaluated the predictive accuracy and clinical relevance of combinations of predictors that were independently significant for the critical outcomes. Variables without significant univariate association were not included in the multivariate models.

We used multiple model validity measures to conduct a thorough assessment of the predictive performance and model fit. A high-risk indicator for critical outcomes was determined by combining the triage category and tympanic admission temperature in a logistic regression model. Calibration was evaluated using the Brier score, intersection point, slope, and visual inspection with decile grouping and a locally weighted scatterplot smoothing (LOWESS) curve. The probability threshold was set based on the Youden index. We report odds ratios with 95% confidence intervals and Nagelkerke R² from logistic models, and PPV, NPV, sensitivity, specificity, LR + , LR − , and AUC with 95% CIs at the Youden-optimal threshold. Analyses were performed in SPSS 28.0 and R 4.2.0 with p < 0.05.

All relevant de-identified data that support the findings of this study are publicly available via the Open Science Framework at https://doi.org/10.17605/OSF.IO/5URSV. The public dataset removes all direct and indirect identifiers in accordance with PLOS ONE’s data policy on human participant data.

Results

Study population and demographics

Of the 131 patients, n = 88 (67.2%) were male. The median age was 67.0 years (IQR: 59.0–75.0). A total of 74 (56.5%) patients were aged ≥65 years (18–44: 4 [3.1%]; 45–64: 53 [40.5%]) (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Demographics, clinical characteristics, and outcomes of emergency care.

https://doi.org/10.1371/journal.pone.0334328.t003

Seasonal variation

When the monthly mean temperature dropped below 7°C in winter, the incidence of hypothermia increased across all severity stages. All Swiss staging categories (I–IV) occurred at temperatures below 10°C, with more than half of Stage II–IV cases occurring in months with a mean temperature less than 7°C. Stage IV hypothermia (<24°C) was the most temperature dependent, with 90.0% (9/10) of the cases occurring below 10°C and none occurring above 18°C (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Seasonal distribution of hypothermia cases and ambient temperatures (2020–2024). Bars indicate monthly case counts stratified by Swiss staging (I: 35–32 °C; II: 32–28 °C; III: 28–24 °C; IV: < 24 °C); line shows the monthly mean outdoor temperature for patients’ registered area. Sample size: n = 131. Meteorological source: monthly means from the local station (S4 Table).

https://doi.org/10.1371/journal.pone.0334328.g002

Total ED visits rose from 31 264 in 2020–46 368 in 2024. Mean monthly ambient temperatures varied between 0.9 °C and 26.6 °C over the same period (S4 Table).

Predictors and univariate correlations with severe outcomes

The triage category was the strongest predictor (p < 0.001), indicating a medium–strong association (Cramer’s V = 0.349) with the primary outcome. The temperature ranges from the Swiss Staging System also showed a significant, but moderate-strength association with the primary outcome (p = 0.032; Cramer’s V = 0.259). There was also a moderate but significant association for the WMS classification (p = 0.017; Cramer’s V = 0.249). The initial tympanic temperature was negatively correlated (p < 0.001; Pearson’s r = −0.294) with ICU admission and ED death. In contrast, ambient external temperature was not significantly associated with our outcome measure (p = 0.694; r = −0.034) (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Association between predictors and critical outcomes (ICU admission and ED mortality).

https://doi.org/10.1371/journal.pone.0334328.t004

Predictive performance

The diagnostic accuracy of predicting critical outcomes was assessed via receiver operating characteristic (ROC) curves. Among the triage category (AUC = 0.683, 95% CI: 0.598–0.767), initial tympanic temperature (AUC = 0.666, 95% CI: 0.566–0.764), Swiss- (AUC = 0.644, 95% CI: 0.545–0.736) and WMS classification (AUC = 0.637, 95% CI: 0.536–0.730), the triage category was found to be the strongest independent predictor (Fig 3A), but the combination of triage classification and tympanic temperature (AUC = 0.740, 95% CI: 0.644–0.829) provided the best predictive performance for critical outcomes (Fig 3B). At the Youden‑optimal threshold, the combined triage + temperature model yielded sensitivity 0.795 and specificity 0.621, corresponding to LR+ = 2.10 and LR− = 0.33. The model showed good overall accuracy (Brier 0.188 vs. null 0.223) and near-ideal calibration (intercept ~ 0, slope ~ 1). At the Youden-optimal threshold (p ≥ 0.282), this is a simple, high-risk signal (S1 Fig; S6 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. ROC curves for predictors of the composite critical outcome (ICU admission or ED mortality).

Curves are shown for MSTR triage category, admission tympanic (*core‑proxy) temperature (°C), Swiss staging, Wilderness Medical Society (WMS) classification, and the combined triage+temperature model. AUC values showing overall discrimination (higher is better). Abbreviations: AUC, area under the ROC curve; MSTR, Hungarian Emergency Triage System; WMS, Wilderness Medical Society.

https://doi.org/10.1371/journal.pone.0334328.g003

Discussion

Consistent with earlier epidemiological findings, our data confirm that hypothermia cases in emergency care follow a seasonal trend, reaching their peak in winter [4]. Hypothermia cases were at their highest in the winter months. While the number of patients treated in the ED also increased throughout the years of the study, our dataset does not allow us to analyze if the increase in patient throughput had an independent effect on the incidence of hypothermia. Hence, we choose not to make any causal statements. The greater proportion of male patients is also consistent with prior research, which links this trend to differences in exposure and behavioral factors [16]. The highest occurrence was seen during the cold and winter period. The severity of the cases could not be predicted by the average monthly temperature, probably indicating that patients were exposed to different environmental conditions: frailty, comorbidity, intoxication, indoor occurrence, and pre-hospital times. All of these factors influence outcomes far beyond the ambient temperature. Preventive measures should be continual, ensuring adequate readiness even before the cold season begins.

The age distribution of the patients indicated that middle-aged individuals were also prominently represented, whereas hypothermia primarily affected the older age group, as previously reported in large-scale studies [20]. Although comorbidities and age-related decreases in thermoregulatory ability are significant variables, hypothermia may also be caused by non-physiological factors [5]. Recent findings show distinct comorbidity and cognitive decline patterns between indoor and outdoor hypothermia mortality. [21].

The significance of prehospital detection and timely intervention is well documented, particularly in the context of hypothermia, where rapid recognition and management can enhance outcomes [22]. The early recognition of hemodynamic instability and the use of advanced rewarming methods improves outcomes [23]. Even the patients in unwitnessed arrest, asystole, or at old age may benefit from extracorporeal support [24]. Most of our patients arrived via ambulances, representing the critical importance of emergency medical services in hypothermia. There were no hypothermic arrests on arrival, suggesting that local EMS already transports these patients to ECMO centers, in line with European guidelines [25]. However, prehospital destination policies were not captured in the study. Emergency triage systems are tools to prioritize urgency, and they are not meant to predict outcomes. The CTAS-based Hungarian Triage System (MSTR) is, however, multimodal. It synthesizes vital-sign thresholds, mental status and signs of shock, and treats core temperature as a secondary modifier (<32 °C → MSTR II; 32–35 °C → MSTR III). Admission temperature might capture exposure and when compared to triage, it is only modestly correlated, even in hypothermic patients. Adding the two together, may serve as complementary information. The predictive function of decision-making systems in emergency treatment was substantiated by the strong association between critical outcomes and triage categorization. The MSTR exhibited greater accuracy for the primary outcome than hypothermia-specific categorization methods did. Beyond discriminatory power, calibration analysis also showed that the triage + temperature model appears to be reliable for estimating the risk of critical outcomes. This correlation could serve as the basis for developing a simple signaling system to aid early escalation decisions in the ED.

Although their predictive power was not as strong as that of the triage classification, the temperature ranges of Wilderness Medical Society classification and the Swiss staging model for hypothermia both correlated with critical outcomes. Earlier symptom-based reports found only about 50% match between clinical stage and centrally measured core temperature [26], while a later study with higher case count showed 61% correct classification by Swiss temperature thresholds (18% overestimation; 21% underestimation) [27]. Based on this, the Swiss staging model and other hypothermia classification systems are good for determining severity, the involvement of additional clinical parameters within triage systems may aid in risk stratification and guide treatment in emergency environments [8].

Critical outcomes were associated with lower tympanic temperatures. Myocardial irritability below 28°C is a known key driver of rescue collapse and mortality in severe hypothermia [28]. Only 12.2% of patients were admitted to the ICU, despite a 21% in-ED mortality rate. Limited ICU capacity may shift critical care to the ED and make the combined outcome a better reflection of real-world severity. In isolation however, the tympanic temperature showed only moderate standalone discrimination. The absence of a significant correlation between the primary outcome and outside temperature aligns with the hypothesis that the severity of hypothermia is influenced by exposure duration, comorbidities, and other environmental variables, in addition to ambient temperature. Combining triage with the admission tympanic temperature results in a more accurate prediction than does depending on only a few factors. Our results imply that advanced risk assessment models in emergency care may perform better than single clinical indicators do and may improve patient management and resource allocation [29]. Frailty, acidosis, hyperkalemia, and indoor occurrence are known predictors of poor outcomes in hypothermia [30], which may be integrated into future severity scales.

Limitations

The retrospective, single-center design may limit generalizability. Our reliance on routinely documented ED indicators potentially introduced inconsistency and residual confounding, in addition to the lack of comorbidity profiles and psychosocial variables. Detailed information on exposure setting, comprehensive vital signs, and metabolic state were not systematically addressed. We only assessed short-term outcomes, and the limited sample size may also affect the statistical reliability. Furthermore, we did not capture in-hospital mortality beyond the ED nor detailed causes of death. Frailty, comorbidities, potential intoxicants, specific details of exposure, and the workflow of the local emergency department/ intensive care unit capacity can all change the direction of correlations. Such factors can not only impact the way triage or temperature are modulated but also influence how a model can be generalized and calibrated from one system to another. Even though these parameters were not measured, they are extremely important in both mortality and intensive care unit admission assessment. The triage categorization of patients with a tympanic temperature of less than 32°C may have introduced collinearity, regardless of our statistical effort.

The use of monthly average temperatures may have resulted in oversimplified weather patterns, potentially resulting in the absence of finer meteorological effects. Despite standardization, tympanic thermometry may underestimate true core temperature in profoundly hypothermic patients compared with central measurements, although the thermometer used fulfils current ASTM accuracy criteria. Of the 131 readings, 14 lay within 0.3 °C of the 24/28/32/35 °C boundaries, and 13 within 20–24 °C. The thermometer we used has a declared accuracy of 0.3 °C below 35 °C, these cases lie exactly within the device’s tolerance range. Because of this, patients may have been classified into adjacent Swiss or WMS threshold categories purely due to measurement inaccuracy, which should be considered when interpreting analyses. While we acknowledge the use of tympanic temperature measurement as a limitation of our study—given that it may have influenced the classification of some patients between the Swiss and WMS categories—we emphasize that our study was designed to reflect real-world triage conditions, where tympanic thermometry is the most used method in emergency departments and general clinical practice. Tympanic temperature was measured directly instead of applying the full Swiss or WMS criteria to estimate core temperature and evaluate the scales’ utility. We relied solely on the temperature ranges specified by those scales, and thus we cannot draw definitive conclusions about the usefulness of core temperature estimation systems. We only analyzed the Hungarian Emergency Triage System, which may limit broader comparability. We did not assess additional ED workflow factors, such as waiting times or resource utilization. Generalizability may vary with triage protocols and ICU capacity, which influence admission thresholds and the composite outcome. External validation across health systems is required.

Conclusion

Accidental hypothermia continues to pose a substantial clinical challenge in Hungary and other temperate regions, necessitating the implementation of specialized prehospital and in-hospital care. Although the temperature ranges of Swiss staging model and the Wilderness Medical Society guidelines can be beneficial in the assessment of severity, our results suggest that triage alone offers moderate discrimination for early critical outcomes. In terms of critical outcomes, deeper integration of admission temperature on arrival into triage decisions can serve as a simple, accurately calibrated risk indicator and prepare for early escalation decisions. However, external validation and local recalibration are required before routine use. In future studies, severity scales can be fine-tuned by adding dynamic temperature thresholds.

Supporting information