A dysregulated host response to infection is a cornerstone of the pathophysiology of sepsis [1, 2]. A dysregulated response alters vital signs such as body temperature (BT), heart rate (HR), blood pressure (BP), and respiratory rate (RR) [3, 4]. Alterations such as hypothermia, fever, tachycardia, hypotension, or tachypnea tend to be associated with altered clinical outcomes [5,6,7,8,9].

Elderly patients generally have a blunted host inflammatory response [10,11,12], which could contribute to different associations between the vital signs and mortality between non-elderly and elderly patients, as observed in a variety of pathological settings [13,14,15]. In sepsis, a single-center study with a limited sample size demonstrated that BT, as a component of the vital signs, is an independent predictor of mortality only in elderly patients [13]. However, whether the relationships between altered vital signs and clinical outcomes differ between elderly and non-elderly patients remains unclear. Investigating and clarifying these differences might contribute to improving the quality of sepsis care, thus reducing mortality rates in critically ill septic patients.

Thus, we hypothesized that associations between initial vital signs and mortality are different for elderly and non-elderly patients with sepsis. Three large sepsis cohorts were investigated to assess the primary outcome of 90-day in-hospital mortality.

Materials and methods

Study cohorts

The current observational study was retrospectively conducted using three sepsis cohorts. Written informed consent was waived because of the study design.

Discovery cohort: FORECAST cohort

FORECAST (Focused Outcomes Research in Emergency Care in Acute Respiratory Distress Syndrome, Sepsis, and Trauma) was a multicenter (ICUs, n = 59, Japan), prospective observational study conducted by the Japanese Association for Acute Medicine (JAAM) from January 2016 to March 2017 [16]. Of the 1184 patients with sepsis, 1148 patients who had records of 90-day in-hospital mortality were included in the analyses. The Institutional Review Board of Chiba University Graduate School of Medicine has approved the study (approval number, 3407).

Validation cohort 1: JAAMSR cohort

JAAMSR (Japanese Association for Acute Medicine Sepsis Registry) was a multicenter (ICUs, n = 15, Japan), prospective observational study conducted by the JAAM from June 2010 to May 2011 [17]. Some of the clinical centers were included in both JAAMSR and FORECAST cohorts; however, not all were included in both of these cohorts. All 624 patients screened with sepsis were included in the analyses. The Institutional Review Board of Chiba University Graduate School of Medicine has approved the study (approval number, 3407).

Validation cohort 2: SPH cohort

Patients admitted to the ICU at St. Paul’s Hospital (SPH) in Vancouver, Canada, between July 2000 and January 2004 (n = 1337) were screened [18]. Of these, 1037 patients met the definition of sepsis on ICU admission. Of these, 1004 patients had records of 90-day in-hospital mortality and were included in the analyses. The Institutional Review Board of SPH has approved the study (approval number, H02-50076).

Data collection and definitions

Sepsis and septic shock were defined based on the sepsis-2 criteria [19]. According to the current standard [1], severe sepsis according to the sepsis-2 criteria was assumed as sepsis. Patients with age ≥ 75 years were defined as elderly and < 75 years as non-elderly, based on previous reports [20,21,22]. In this study, the vital signs included BT, HR, mean arterial pressure (MAP), systolic blood pressure (SBP), and RR. The worst vital sign values corresponding to the APACHE II scoring were retrieved within 24 h of the diagnosis of sepsis in all cohorts. Each vital sign was divided into two groups based on the cutoff values from the sepsis-2 and sepsis-3 criteria [1, 19] (additional details are provided in Additional file 1: Table S1). The definition of the suspected site of infection was applied as described elsewhere [16].

Statistical analysis

Our primary analysis used Cox regression to test for hazard of death over 90 days by the vital signs in the elderly and non-elderly groups. We selected this approach to adjust for potential baseline imbalances, including age (per year), sex, chronic steroid use, and APACHE II score (Additional file 1: Table S2, correlation analysis) based on previous reports [18, 23]. Significant discovery results were tested for replication and generalizability in patients of the same ancestry in validation cohort 1 and patients of different ancestry in validation cohort 2. We tested for homogeneity of results across cohorts using a meta-analytic technique. Sensitivity analysis using the Cox regression was conducted by adding parameters with significant difference between the non-elderly and elderly patients including comorbidities as covariates. Sensitivity analyses using age cutoffs of 65, 70, and 80 years old for significant discovery results were conducted [24,25,26,27,28]. Interactions between age and BT for mortality were tested using Cox regression analysis. In addition, the Cox regression analysis for 28-day in-hospital mortality was conducted.

Univariate analysis was performed using Pearson’s chi-square test or a Mann-Whitney U test. Two-tailed P values < 0.05 were considered to be significant. Analyses were performed using either the SPSS software version 24.0 (IBM Corporation, Armonk, NY, USA) or the Review Manager (RevMan) version 5.3 software (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark).


There were 628 non-elderly and 520 elderly patients in the discovery FORECAST cohort (Table 1 and Additional file 1: Table S3). Of the vital sign variables on day 1, elderly patients had significantly lower BT and HR compared to non-elderly patients. Elderly patients also had significantly higher 90-day mortality compared to non-elderly patients (Table 1).

Table 1 Baseline characteristics and clinical outcomes in the derivation cohort (FORECAST cohort)

Discovery analysis for all vital signs (FORECAST cohort)

In the primary analysis of the five vital signs using Cox regression, non-elderly patients with BT < 36.0 °C had a significantly increased hazard of death over 90 days (P = 0.025, adjusted hazard ratio 1.70, 95% CI 1.07–2.71) (Fig. 1, Fig. 2a). In contrast, elderly patients with hypothermia did not show differences in 90-day mortality (P = 0.16) (Fig. 1, Fig. 2a). In elderly patients, HR > 90 beats/min and RR > 30 breaths/min were associated with increased mortality (HR, P = 0.040, adjusted hazard ratio 1.72, 95% CI 1.03–2.88; RR, P = 0.044, adjusted hazard ratio 1.49, 95% CI 1.01–2.19) (Fig. 1).

Fig. 1
figure 1

Associations between the vital signs and 90-day in-hospital mortality in discovery cohort (FORECAST cohort). Non-elderly patients with BT < 36.0 °C showed a significant increase for hazard of death over 90-day period. The hypothermia was not associated with mortality in elderly patients. Heart rate > 90 beats/min and RR > 30 breaths/min had a significant association for increased mortality in elderly patients. CI, confidence interval. The adjusted hazard ratio was calculated using potentially confounding factors such as the age, sex, chronic steroid use, and acute physiology and chronic health evaluation (APACHE) II score

Fig. 2
figure 2

Probability of mortality corresponding to the body temperature category of < 36.0 °C. a Discovery cohort (FORECAST cohort). b Validation cohort 1 (JAAMSR cohort). c Validation cohort 2 (SPH cohort). Non-elderly patients with hypothermia had higher mortality in all three cohorts

Validation of significant discovery results (JAAMSR, SPH cohorts)

Non-elderly patients with BT < 36.0 °C showed significantly increased mortality in validation cohort 1 (Japan) (JAAMSR, P = 0.0024, adjusted hazard ratio 2.05, 95% CI 1.29–3.26) (Fig. 2b) (Additional file 1: Table S4, Baseline characteristics). The same effect was replicated in the patients of different ancestry in validation cohort 2 (Canada) (SPH, P = 0.029, adjusted hazard ratio 1.36, 95% CI 1.03–1.80) (Fig. 2c) (Additional file 1: Table S5, Baseline characteristics). Similar to the discovery cohort, there were no significant associations between hypothermia and mortality in elderly patients (JAAMSR, P = 0.34; SPH, P = 0.81) (Fig. 2b, c). Test for homogeneity of the effect of BT < 36.0 °C on mortality using a meta-analytic technique revealed no significant differences across all three cohorts (I2 = 17% in non-elderly patients) (Fig. 3).

Fig. 3
figure 3

Meta-analysis to test for the homogeneity between each study for association of body temperature < 36.0 °C and 90-day in-hospital mortality. Non-elderly patients revealed no significant differences in effect between all three cohorts (I2 = 17%) and a significant combined effect (P = 0.0002, hazard ratio 1.58, 95% CI 1.24–2.01). JP, Japan; CA, Canada; CI, confidence interval. The hazard ratio from each cohorts was calculated with adjusting the potentially confounding factors such as the age, sex, chronic steroid use, and acute physiology and chronic health evaluation (APACHE) II score

The effects of HR > 90 beats/min and RR > 30 breaths/min on mortality observed in the elderly patients of the discovery cohort were not replicated in validation cohort 1 (HR > 90 beats/min, P = 0.82; RR > 30 breaths/min, P = 0.96).

Analysis including additional covariates

Repeating the Cox regression analysis including age, sex, chronic steroid use, APACHE II score, and significantly different baseline parameters between non-elderly and elderly patients (FORECAST suspected site of infection, body mass index, Charlson comorbidity index; JAAMSR suspected site of infection, body mass index, comorbidities [stroke, heart failure, chronic lung disease]; SPH comorbidities [chronic hepatic disease, chronic lung disease] (Table 1 and Additional file 1: Table S4 and Table S5)) yielded the same results (Additional file 2: Figure S1).

Analysis using different age cutoffs

We performed the sensitivity analysis for the association of BT < 36.0 °C and mortality using different age cutoff values of 65, 70, and 80 years. The age cutoffs of 70 and 80 years revealed similar results to those observed in the primary analysis. However, the effect of hypothermia was eliminated using a cutoff of 65 years (Additional file 3: Figure S2).

Analysis of fever, the interaction of age and BT, and 28-day in-hospital mortality

Because hypothermia had effects on mortality across all three cohorts, we tested for the effect of fever in all three cohorts in the secondary analysis. In non-elderly patients, a trend towards reduced hazard of death was observed when BT > 38.3 °C in the discovery cohort (P = 0.11, adjusted hazard ratio 0.71, 95% CI 0.47–1.08) (Fig. 1 and Additional file 4: Figure S3a). This potential effect was highly significant in both validation cohorts (JAAMSR, P = 0.0043, adjusted hazard ratio 0.51, 95% CI 0.32–0.81; SPH, P = 0.018, adjusted hazard ratio 0.76, 95% CI 0.61–0.95) (Additional file 4: Figure S3b-c). A meta-analytic test for heterogeneity revealed no significant difference in effect across cohorts (I2 = 14%) and a significant combined effect (P = 0.0006, adjusted hazard ratio 0.70, 95% CI 0.57–0.86) (Additional file 5: Figure S4). However, no significant effect of fever was observed in elderly patients in any cohort (FORECAST, P = 0.71; JAAMSR, P = 0.50; SPH, P = 0.72) (Additional file 4: Figure S3a-c).

The interaction of age (non-elderly or elderly) and hypothermia (BT < 36.0 °C or ≥ 36.0 °C) for mortality was statistically significant across the three cohorts (FORECAST, P = 0.0059; JAAMSR, P = 0.15; SPH, P = 0.012; three cohorts combined, P = 0.0001). The interaction term between the age and fever (BT ≤ 38.3 °C or > 38.3 °C) for mortality was also significant across the three cohorts (FORECAST, P = 0.23; JAAMSR, P = 0.13; SPH, P = 0.033; three cohorts combined, P = 0.005). These results indicate that the BT response (both low and high) has a significant impact on mortality in non-elderly patients but does not appear to have an impact on elderly patients.

Repeating analysis using 28-day mortality yielded similar, but partly stronger relationships between vital signs and mortality as those observed in the primary analysis (Additional file 6: Figure S5).


In the present study, non-elderly sepsis patients with hypothermia (BT < 36.0 °C) had significantly increased mortality, and those with fever (BT ≥ 38.3 °C) had decreased mortality. In contrast, hypothermia and fever were not associated with altered mortality in elderly patients. Interestingly, the other vital sign values were not consistently associated with differences in the outcome in patients of any age. Thus, BT was distinct among the vital signs for its association with the clinical outcomes in the non-elderly, but not elderly, patients with sepsis.

The key finding of our analysis is that the association of BT alterations on mortality was only observed in the younger septic population as opposed to elderly patients. The meta-analytic technique showed that there was low heterogeneity (I2 = 17%) in the association between hypothermia and mortality in non-elderly patients in the three cohorts (Fig. 3). In a secondary analysis, the opposite effect of fever was also only observed in non-elderly sepsis patients (Additional file 4: Figure S3 and Additional file 5: Figure S4). The changes in the association between the vital signs and mortality based on age have been investigated in several categories of critically ill patients. In blunt trauma, the presenting vital signs are less predictive of mortality in elderly patients as compared to the non-elderly [14]. Additionally, vital signs are less predictive of in-hospital cardiac arrest in elderly patients as compared to the non-elderly [15]. Since the significant associations of tachycardia and tachypnea on mortality in elderly patients were not replicated in the validation cohorts, we confirmed that BT alterations are uniquely predictive of mortality among the vital signs in non-elderly patients with sepsis but not in the elderly patients. Significant interactions with mortality between hypothermia or fever and age groups reinforced these findings.

In the current understanding of pathomechanism of sepsis-related BT changes, fever and hypothermia both have adaptive biological value for the host. In this context, fever is considered as an indicator of active (disease-fighting) strategy which may be beneficial in diseases of mild-to-moderate severity in a previously healthy host while hypothermia is a passive (disease-tolerating) strategy that is advantageous in severe forms of the disease, especially in the presence of comorbidities [29,30,31]. Nevertheless, BT is not the cause but rather the indicator of the severity, thus consequently of the outcome of the disease. In our previous study employing the FORECAST and JAAMSR cohorts, with no distinction between the non-elderly and elderly sepsis patients, hypothermia was associated with poor clinical outcomes [32, 33]. In another study of adult sepsis patients, hypothermia (< 36.0 °C) within 24 h from sepsis diagnosis was associated with a higher 28-day and 1-year mortality [34]. Oppositely, increased BT in sepsis/septic shock patients admitted to an ICU within 24 h of hospital arrival was associated with decreased mortality [35]. These reports were in line with our results in non-elderly patients.

Several factors could explain the difference in the relationships between the hypothermia and clinical outcome in elderly and non-elderly patients with sepsis. The physiology of fever generation and development of hypothermia is complicated; however, reduced production of pyrogenic cytokines such as IL-6 and TNF-a may be related to the lower incidence of fever seen in elderly septic patients [36, 37]. Comorbidities such as stroke, which was a higher prevalence in the elderly group in this study, may comprise neurological deficits that could impair thermoregulatory/inflammatory reflexes [38, 39]. Decreased heat production due to reduced muscle mass and increased heat radiation due to a reduced fat mass and blunted peripheral vasoconstriction capacity might be related to the lower BTs [40]. Observational studies have indeed reported that higher age populations have lower BTs and smaller diurnal BT variations [41, 42], which is consistent with our results. Lower BT and altered BT response in elderly septic patients as the result of age-related physical and functional deterioration may mitigate the hypothermic/febrile effects on clinical outcomes.

There were several limitations to the present study. First was the descriptive nature of the observational study. The primary findings were validated using two independent cohorts, which included multiple centers and patients of different ancestry; however, these do not prove a causal link. Second limitation of the study was the absence of data on variables that may have confounded BT measurement, including temperature measurement site and whether patients received antipyretics or targeted temperature management. Third was on the retrieval of vital signs. The vital signs corresponding to the APACHE II score were analyzed in the present study. However, according to the APACHE II scoring, a temperature as lower than 34 °C receives 1 point, whereas fever higher than 38.9 °C receives 3 points. Therefore, patients presenting both hypothermia and fever were more likely to be categorized in the fever group, which may have influenced the results. In addition, since the analyzed data were only in the first 24 h after the diagnosis of sepsis, it is unclear whether the duration of hypothermia or temperature changes over time affects the outcome. Another limitation could be the absence of established criteria for defining elderly patients. In this study, we applied the currently accepted standards in literature [20,21,22]. The analyses in the age cutoff of 70 and 80 years yielded the same conclusion, which could help strengthen the results of this study.


We investigated the difference in the associations of the vital signs with mortality between non-elderly and elderly patients with sepsis. Among the vital signs, only BT showed significantly different effects on mortality between non-elderly and elderly patients. Hypothermia is associated with increased mortality and fever with decreased mortality in non-elderly septic patients. In contrast, hypothermia and fever did not impact mortality in elderly patients.