a b s t r a c t

Background: Sepsis is a systemic inflammatory response syndrome with high mortality. There is an upward trend in sepsis prevalence and mortality worldwide. Early and accurate prediction of outcome in sepsis is important. There remains a great need to improve a reliable prognostic model for sepsis patients with widely available var- iables. The aim of this study was to explore the correlation between serum thyroid hormone levels and prognosis in sepsis patients.

Methods: Septic patients were identified from the Medical Information Mart for Intensive Care (MIMIC)-III data- base. Factors that were found to contribute to the outcome in the uni-variate analysis at P value <0.1 were included in the multivariate. Multivariate analysis was performed by binary logistic regression analysis, which al- lows adjust for confounding factors. We combined an assessment of thyroid hormone and some variables to- gether, which improve the accurate prediction of outcome. The accuracy of the test was assessed measuring the area under the ROC curve (AUROC).

Results: A total of 929 eligible septic patients were included in the data analysis. Seventy hundred and three patients had a good functional outcome, whereas 226 patients had a bad functional outcome. Thyroxin (T₄) level was significantly decreased in patients with an unfavorable functional outcome as compared to pa- tients with a favorable functional outcome (P < 0.01). Binary logistic regression analyses revealed that lower thyroxin concentrations on admission were associated with a risk for poor outcomes (OR 0.556, 95% CI 0.41-0.75; P < 0.01). In addition, in ROC curve analysis, the combined model AUROC was 0.82 for ICU survival, which was significantly higher than the AUROCs of original fT₄ (0.65 and 0.65), T₄ (0.71 and 0.71) and SAPSII (0.70 and 0.72) (all P < 0.05).

Conclusions: Low serum thyroxin levels can be a predictive marker of Short-term outcome after sepsis. A com- bined model (fT₄, T₄ and SAPSII score) can add significant additional predictive information to the Clinical score of the SAPSII.

(C) 2021 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://

creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Sepsis is a systemic inflammatory response syndrome with high mortality that results from severe infection and can lead to secondary organ dysfunction [1]. Sepsis and septic shock are major health care problems, affecting millions of people around the world each year. There is an upward trend in sepsis prevalence and mortality worldwide [2]. Furthermore, sepsis is a crucial issue for policymakers due to the high cost of intensive care unit (ICU) treatment and the significant rehospitalization rate among sepsis survivors [3]. Therefore, early and accurate prediction of outcomes in patients is important.

* Corresponding author.

E-mail address: [email protected] (Z. Lu).

¹ Yiping Wang and Fangyuan Sun contributed equally to this work and should be regarded as co-first authors.

Considering the aforementioned data, precise and reliable predictive factors in sepsis are needed.

Sepsis is a life-threatening disease that can not only cause organ dysfunction but can also cause a series of endocrine and metabolic damage. Among them, the hypothalamus-pituitary-thyroid (HPT) axis endocrine hormone changes are the first measurable [4]. The activation of the HPT axis comprises the hypothalamus release of thyrotropin releasing hormone (TRH), which stimulates the pitui- tary gland to produce Thyroid stimulating hormone [5]. Thy- roxin (T₄) and triiodothyronine (T₃) are produced and released from the thyroid gland in response to TSH stimulation. The primary secretory product of the thyroid gland is the less metabolically ac- tive T₄, which is usually considered a prohormone. The majority of the more metabolically active T₃ is produced from T₄ in extra thy- roidal tissues by 5?-monodeiodination that is carried out by type1 (D₁) and type2 (D₂) deiodinase enzymes. 20% of T₃ is directly

https://doi.org/10.1016/j.ajem.2021.02.014

0735-6757/(C) 2021 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

secreted by the thyroid gland and 80% is transformed from the pe- riphery T₄. T₃ biological activity is 3 to 5 times greater than T₄ and stress reduces hypothalamus thyrotropin releasing hormone and pituitary thyroid stimulating hormone secretion with impaired extra-thyroidal T₄ to T₃ conversion [5].

Acute illness results in a decreased peripheral conversion of free thyroxin (fT₄) to free triiodothyronine (fT₃). Such alterations contribute toward the development of low T₃ syndrome (low serum T₃ concentra- tions with normal TSH concentrations) [6]. In Serious disease, a low serum T₃ level is a compensatory mechanism to save energy and reduce the consumption of protein [7]. The phenomenon is described as non-thyroidal illness syndrome. It is a combination of physiological ad- aptation and pathological response to systemic lesions [8]. According to current knowledge, the mechanism of developing a non-thyroidal dis- ease syndrome is associated with a reduced signaling of TRH production by the hypothalamus [4]. However, the basic reason for these changes has not been clearly explained. The impact of severe systemic diseases on thyroid metabolism is increasingly emphasized by researchers.

Our previous studies revealed an association between thyroid hor- mone levels and short-term outcomes after stroke [5]. Based on the available knowledge, we hypothesized that thyroid function may also be associated with outcomes in patients with sepsis. The purpose of this study was to use a freely available US database to verify this hypothesis.

1. Material and methods
   1. Sources of data

The data of the present study were collected from a large critical care database named Multiparameter Intelligent Monitoring in Intensive Care Database III (MIMIC III) [9]. MIMIC III is a publicly and freely

available database which is well described in previous papers [10]. In brief, MIMIC III database contains ICU patient data from the Beth Israel Deaconess Medical Center (single) between 2001 and 2012. This data- base was approved by the Institutional Review Boards (IRB) of the Mas- sachusetts Institute of Technology (MIT) [11]. After successfully completing the National Institutes of Health (NIH) Web-based training course and the Protecting Human research participants examination (certification number: no.37920768), we were given the permission to extract data from MIMIC III. Data extraction was performed using PostgreSQL (version12.4).

• 1. Date collection and follow-up

There were 50,766 distinct hospital admissions for adult patients (aged 18 years or above) admitted to the ICU during the study period. Infection was identified from the ICD-9 code in the MIMIC-III database. The inclusion criteria were patients with sepsis: the third sepsis defini- tion defined sepsis as a condition with life-threatening organ dysfunc- tion caused by a dysregulated host response to infection. In the present study, we screened patients with documented or suspected in- fection (qSOFA score >= 2 points), and on this basis, we added evidence of organ dysfunction (SOFA score >= 2 points) to ultimately record eligible sepsis patients. Patients with a history of thyroid disease and those with more than 25% missing individual were excluded. For patients who had multiple admissions to the ICU, only the first ICU admission was included for analysis. The design of the study is presented in Fig. 1. The following variables were extracted from the MIMIC-III database for the first day of ICU admission: age at the time of hospital admission, sex, Sequential Organ Failure Assessment score, qSOFA score, and Simplified Acute Physiology Score II (SAPSII). The SOFA score was calcu- lated within the first 24 h after ICU admission. In the present study, all variables had less than 25% missing values. The primary endpoint was

Fig. 1. Flowchart of patient selection.

Baseline characteristics and comparison between survivors and non-survivors.

Demographic characteristics	All	Survivors, n=703	Non-Survivors,n=226	p
Male gender, n (%)	420(45.2%)	304(43.2%)	116(51.3%)	0.024
Age (years)	64.9 +- 14.93	64.3 +- 15.32	66.8 +- 13.50	0.021
SOFA [median (IQR)]	6 (4-8)	5 (3-8)	8 (5-11)	<0.01
qSOFA [median (IQR)]	2 (2-2)	2 (2-2)	2 (2-2)	0.012
SAPSII [median (IQR)]	44 (36-53)	41 (34-50)	51 (43-59)	<0.01
Clinical presentation BMI (kg/m2)	28 (23-33)	28 (23-33)	28 (24-33)	0.310
Systolic blood pressure (mmHg)	110 (101-120)	110 (102-121)	104 (98-114)	<0.01
Heart rate (beats/min1)	91 (79-104)	89 (79-103)	94 (81-107)	0.044
Laboratory findings T4 (nmol/L)	5.4 (3.9-7.1)	6.0 (4.1-7.4)	4.3 (3.0-5.3)	<0.01
fT4 (pmmol/L)	1.1 (0.84-1.3)	1.1 (0.87-1.3)	1.1 (0.80-1.3)	0.072
TSH (mIU/L)	3.0 (0.85-6.7)	3.0 (0.86-6.6)	3.0 (0.79-6.8)	0.797
White blood cells (109/L)	11.23 +- 7.01	11.2 +- 6.29	11.25 +- 8.9	0.497
Hemoglobin (g/dl)	9.6 +- 1.92	9.8 +- 1.97	9.24 +- 1.6	<0.01
Platelets (109/L)	213 +- 126	225.74 +- 124	174.03 +- 126	<0.01
C-reactive protein (mg/L)	74.95 +- 78	76.39 +- 80	70.02 +- 69	0.72
Creatinine (mg/dl)	1.71 +- 1.63	1.60 +- 1.54	2.05 +- 1.83	0.01
Albumin (g/dl)	2.82 +- 0.67	2.87 +- 0.63	2.69 +- 0.74	0.012
Lactate (nmol/L)	2.43 +- 0.67	2.3 +- 1.59	2.71 +- 1.89	0.09

Abbreviations: SOFA, sequential organ failure assessment; SAPSII, Simplified Acute Physiology Score II; BMI, Body Mass Index; T₄, thyroxin; fT₄, free thyroxin; TSH, thyroid-stimulating hormone.

hospital survival, which was defined as the status of patient survival at the time of hospital discharge.

• 1. Statistical analysis

Continuous variables were expressed as mean (standard deviation) or median [inter-quartile range (IQR)] and categorical variables were expressed as the number of percentage as appropriate. Differences in con- tinuous variables were analyzed using independent samples t-test orcom- pared using Mann-Whitney test. Categorical variables were compared with the chi-squared test or Fisher’s exact test as appropriate. Factors that were found to contribute to the outcome in the uni-variate analysis at P value <0.1 were included to the multivariate. Multivariate analysis was performed by binary logistic regression analysis, which allows adjust for confounding factors. For each variable, the odds ratio (OR), and 95% confidence interval (CI) was given. The accuracy of the test was assessed measuring the area under the ROC curve (AUROC). All statistical tests with P value <0.05 were considered statistically significant. Statistical analysis was performed using the statistical program for social science (SPSS) statistical software (version24.0) and MedCalc (version 16.4).

1. Results
   1. Baseline characteristics of study samples

The initial search identified 50,766 ICU admissions from the MIMIC- III database. A total of 8929 patients fulfilled the definition of sepsis. After excluding the patients according to the exclusion criteria, 929 eli- gible patients were enrolled. The comparisons of mortality between 929 eligible sepsis patients and 8000 ineligible sepsis patients amounted to 24.3% (226/929) and 23.9% (1915/8000), respectively (P = 0.82). The mean age of included patients was 64.90 +- 11.30 years old. The pa- tients’ baseline clinical and laboratory data are reported in Table 1.

In our study, 703 patients had good functional outcomes, whereas 226 patients had bad functional outcomes. The mortality in the study group amounted to 24.3% (226/929). Table 1 also shows that patients who had bad functional outcomes after sepsis were older in age (66.8 +- 13.5 vs. 64.3 +- 15.32; P = 0.02) and had lower systolic blood

pressure (104 [98-114] vs. 110 [102-121]; P = 0.01) and lower serum thyroxin concentrations (4.3 [3.0-5.3] vs. 6.0 [4.1-7.4]; P = 0.01).

Fig. 2. A. Comparison of outcomes between different groups based on thyroxin. B. Comparison of outcome between different groups based on SAPSII score.

Fig. 2 (continued).

• 1. Thyroid hormone levels and outcome

Serum T₄ concentration decreased in patients with unfavorable out- come compared to those with favorable outcome (Fig. 2A). Univariate logistic regression analyses revealed that poor functional outcomes

Table 2

Univariate and multivariate between thyroid hormone level and sepsis outcomes.

Factor	Univariate analysis				Multivariate analysis
	Odd ratio (OR)	95% CI	P		Odd ratio (OR)	95% CI	P
Gender	0.722	0.53-0.97	0.034		–	–	–
Age	1.012	1.00-1.02	0.029		–	–	–
SOFA	1.205	1.15-1.26	0.001		–	–	–
qSOFA	1.555	1.10-2.19	0.012		–	–	–
SAPSII	1.055	1.04-1.08	0.001		1.085	1.04-1.13	0.001
BMI	1.00	0.98-1.02	0.908		–	–	–
SBP	0.97	0.96-0.98	0.001		–	–	–
Heart rate	1.00	1.00-1.02	0.050		–	–	–
T4	0.69	0.58-0.83	0.001		0.556	0.41-0.75	0.001
fT4	0.717	0.49-1.06	0.090		4.015	1.35-9.97	0.012
TSH	0.996	0.98-1.01	0.633		–	–	–
WBC	1.000	0.98-1.02	0.964		–	–	–
Hemoglobin	0.850	0.78-0.92	0.001		–	–	–
Platelets	0.996	0.99-1.00	0.001		–	–	–
CRP	0.999	0.99-1.01	0.719		–	–	–
Creatinine	1.163	1.07-1.27	0.001		–	–	–
Albumin	0.663	0.49-0.90	0.007		–	–	–
Lactate	1.130	1.02-1.24	0.012		–	–	–

Abbreviations: SOFA, sequential organ failure assessment; SAPSII, Simplified Acute Physiology Score II; BMI, Body Mass Index; SBP,Systolic blood pressure; T₄, thyroxin; fT₄, free thyroxin; TSH, thyroid-stimulating hormone; WBC, white blood cell; CRP, C-reactive protein.

Table 3 Receive operating characteristics curve analysis.
	Functional outcome at discharge		P
	AUC	95% confidence
MARKER	0.82	0.71-0.84	0.03
SAPSII	0.70	0.64-0.76	0.03
T4	0.71	0.63-0.78	0.04
fT4	0.65	0.58-0.71	0.04

were significantly associated with lower T₄ concentration, fT₄ concen- tration, systolic blood pressure, albumin, hemoglobin and platelets, and higher SOFA score SAPSII score (Fig. 2B), heart rate, creatinine and lactate. We used univariate and multivariate compared the sepsis out- comes and the risk factors as presented in Table 2. The functional out- come of patients at hospital discharge after sepsis coded 0 as survivors and 1 as non-survivors, which was the dependent variable. After adjusting for all other significant outcome predictors, T₄, fT₄ and SAPSII score remain can be seen as independent outcome predictor. Binary logistic regression analyses revealed that poor outcomes were associated with lower T₄ (OR 0.556, 95% CI 0.41-0.75; P < 0.01) and

fT₄ concentrations (OR 4.015, 95% CI 1.35-9.97; P = 0.01) and higher

SAPSII score (OR 1.085, 95% CI 1.04-1.13; P = 0.01) on admission.

• 1. Construction of the combined model

Weperformedreceiveroperatingcharacteristics(ROC) curveanalysis to compare the overall Prognostic accuracy. The area under the receiver operating characteristics curve (AUROC) to predict outcome for T₄ with an AUC of 0.71 (0.63-0.78), the fT₄ with an AUC of 0.65 (0.58-0.71) and the SAPSII score with an AUC of 0.70 (0.64-0.76) (Table 3). At the same time, we combined T₄, fT₄ and SAPSII score together named Combined Model, which could improve the prediction accuracy.

The ability to predict hospital survival and outcome were performed by ROC analyses (Fig. 3). The Combined Model AUROC was 0.82, which were both significantly higher than the AUROCs of original T₄ (0.71), fT₄ (0.65) and SAPSII (0.70) (all P < 0.05). We added T₄ and fT₄ into the SAPSII scoring system to construct the Combined Model. The compari- son is shown in Table 3.

1. Discussion

Sepsis not only can cause damage to multiple organs but can also trig- ger a series of endocrine and metabolic damages [12]. Among the endo- crine and metabolic damages, HPT axis endocrine hormone changes are the first that can be measured. In this study, we first evaluated the mortal- ity of eligible sepsis patients and ineligible sepsis patients. There was no significant difference in mortality between the two groups of patients (Table 4). We then assessed thyroid hormone levels regarding their accu- racy in predicting functional outcome in patients with sepsis. Our main finding is that thyroxin can beseen as an independent short-term prognos- tic marker of outcome in patients even after correcting for possible con- founding factors. For patients who had a good prognosis, the serum thyroxin and free thyroxin concentrations were significantly higher than those in the poor prognosis group (P < 0.01), which showed that a low T₄ level was associated with the short-term prognosis of patients with sep- sis. Moreover, we found that the combined model (thyroxin, free thyroxin and SAPSII score) can improve prediction information accuracy.

Hormones play a pivotal role in the clinical course and outcome. Most of the clinical studies focus on the HPT axis and a related mecha- nism. Various low T₃ states have long been reported in a variety of se- vere acute and chronic diseases [6]. It has been well confirmed that lower serum T₃ concentrations were independently associated with dis- eases, including patients with acute cardiac events [13]; after brain tumor surgery and patients with respiratory failure [14]. It has been suggested that T₃ levels may be a reliable predictor of clinical outcome. Several publications bring up the topic of the role of thyroid hormones in the management of sepsis and their impact on the outcomes [15].

In the retrospective review of 231 patients with surgical sepsis Todd et al. [16] showed results similar to those presented in the current study,

Fig. 3. Comparison of the area under the receiver operating curve.

Table 4

Comparison of mortality between eligible sepsis patients and ineligible patients.

Grouping Survivors Non-Survivors Total

Eligible

Not-eligible

703

6085

226

1915

929

8000

Total 6788 2141 8929

All Eligible

n = 929

Not eligible p

n = 8000

Mortality, n (%) 2141(23.1%) 226 (24.3%) 1915 (23.9%) 0.82

decreased baseline T₃ and T₄ levels were associated with mortality. There are also other reports that, similarly to ours, a study by Maria Foks et al. [8] showed that thyroid hormone levels were significantly lower among patients who died during ICU stay, documented a correla- tion between hypothyroxinaemia and mortality in sepsis. Moreover, a study by Meyer et al. [17] suggested a relationship between serum T₃ and fT₄ level and mortality rate on the last day of hospitalization; how- ever, thyroid hormone levels measured on admission did not differ be- tween survivors and non-survivors. Interestingly, there are also several studies showing an opposite tendency. In the research by Ray et al. [18] TSH, T₃, and fT₄ levels were measured in 180 critically ill patients after 3 h of ICU admission. It did not show any statistically significant differ- ences in T₃ and fT₄ levels between patients who had died and those who had survived. Nevertheless, numerous factors may explain the con- trasting results; for instance, the differences in timing of blood sample gathering, the fraction of thyroid hormones measured, and medical di- agnosis. Furthermore, the research was performed more than 20 years ago, and there was a significant disproportion in the number of patients between the study groups.

However, there are some limitations in this observational study. First, the study was based on electronic health-care records (EHRs) whose data were generated during routine clinical practice [19]. It is possible that the cohort selection was not exactly consistent with the definition of sepsis from guidelines [10]. However, we tried to identify septic patients who were consistent with the third definition of sepsis. Second, the retrospective design of the study made it subject to

confounding by indication, and the result needs to be replicated in more prospective studies to conclude whether they are stable and valid. Prospective studies with longer-term follow-up periods are needed. Third, the database spanned more than 10 years, and clinical practice for the management of sepsis was changed during the study period. The results may not be generalizable to current practice.

1. Conclusion

In conclusion, our study suggested that low serum thyroxin levels can be a predictor marker of short-term outcome after sepsis. Combined model can add significant additional predictive information to the clin- ical. This model may be useful for patient counseling and prognosis evaluation.

Ethics approval and consent to participate

MIMIC III database used in the present study was approved by the Institutional Review Boards (IRB) of the Massachusetts Institute of Technology and does not contain protected health information.

Credit author statement

YPW: Conceptualization, Data curation and Writing – original draft; FYS: Software, Validation; GLH: Conceptualization, Data curation; ZQL: Writing – review & editing and Supervision.

Declaration of Competing Interest

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

All authors declare that there are no competing financial interests. The authors declare the following financial interests/personal rela-

tionships which may be considered as potential competing interests:

Acknowledgements

This work was supported by Research Incubation Project of the First Affiliated Hospital of Wenzhou Medical University.

References

1. Martin GS. Sepsis, severe sepsis and septic shock: changes in incidence, pathogens and outcomes. Exp Rev Anti-infective Ther. 2012 Jun;10(6):701-6.
2. Rhodes A, Evans LE, Alhazzani W, Levy MM, Antonelli M, Ferrer R, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock: 2016. Intens Care Med. 2017 Mar;43(3):304-77.
3. Mayr FB, Yende S, Angus DC. Epidemiology of severe sepsis. Virulence. 2014 Jan 1;5 (1):4-11.
4. Luo B, Yu Z, Li Y. Thyroid hormone disorders and sepsis. Bio-med Mater Eng. 2017; 28(s1):S237-41.
5. Wang Y, Zhou S, Bao J, Pan S, Zhang X. Low T3 levels as a predictor marker predict the prognosis of patients with acute ischemic stroke. Int J Neurosci. 2017 Jul;127 (7):559-66.
6. Bunevicius A, Deltuva V, Tamasauskas S, Tamasauskas A, Laws Jr ER, Bunevicius R. Low triiodothyronine syndrome as a predictor of poor outcomes in patients under- going brain tumor surgery: a pilot study: clinical article. J Neurosurg. 2013 Jun;118 (6):1279-87.
7. Perez AC, Jhund PS, Stott DJ, Gullestad L, Cleland JG, van Veldhuisen DJ, et al. Thyroid-stimulating hormone and clinical outcomes: the CORONA trial

(controlled rosuvastatin multinational study in heart failure). JACC Heart Fail. 2014 Feb;2(1):35-40.

1. Foks M, Dudek A, Polok K, Nowak-Kozka I, Fronczek J, Szczeklik W. Thyroid hor- mones as potential prognostic factors in sepsis. Anaesthesiol Intens Ther. 2019;51 (3):205-9.
2. Zhang Z, Zhu C, Mo L, Hong Y. Effectiveness of sodium bicarbonate infusion on mor- tality in septic patients with metabolic acidosis. Intens Care Med. 2018 Nov;44(11): 1888-95.
3. Li K, Feng C, Jia L, Chen L, Pan F, Li T. Discussion of the implementation of MIMIC da- tabase in emergency medical study. Zhonghua wei zhong bing ji jiu yi xue. 2018 May;30(5):494-6.
4. Johnson AEW, Pollard TJ, Shen L, Lehman L-wh, Feng M, Ghassemi M, et al. MIMIC- III, a freely accessible critical care database. Sci Data. 2016;3(1).
5. Coceani M, Iervasi G, Pingitore A, Carpeggiani C, L’Abbate A. Thyroid hormone and coronary artery disease: from clinical correlations to prognostic implications. Clin Cardiol. 2009 Jul;32(7):380-5.
6. Iervasi G, Pingitore A, Landi P, Raciti M, Ripoli A, Scarlattini M, et al. Low-T3 syn- drome: a strong prognostic predictor of death in patients with heart disease. Circu- lation. 2003 Feb 11;107(5):708-13.
7. Scoscia E, Baglioni S, Eslami A, Iervasi G, Monti S, Todisco T. Low triiodothyronine (T3) state: a predictor of outcome in respiratory failure? Results of a clinical pilot study. Eur J Endocrinol Eur Federat Endocr Soc. 2004 Nov;151(5):557-60.
8. Artero A, Zaragoza R, Camarena JJ, Sancho S, Gonzalez R, Nogueira JM. Prognostic fac- tors of mortality in patients with community-acquired bloodstream infection with severe sepsis and septic shock. J Crit Care. 2010 Jun;25(2):276-81.
9. Todd SR, Sim V, Moore LJ, Turner KL, Sucher JF, Moore FA. The identification of thyroid dysfunction in surgical sepsis. J Trauma Acute Care Surg. 2012 Dec;73(6): 1457-60.
10. Meyer S, Schuetz P, Wieland M, Nusbaumer C, Mueller B, Christ-Crain M. Low triiodothyronine syndrome: a prognostic marker for outcome in sepsis? Endocrine. 2011;39(2):167-74.
11. Ray DC, Drummond GB, Wilkinson E, Beckett GJ. Relationship of admission thyroid function tests to outcome in critical illness. Anaesthesia. 1995;50(12):1022-5.
12. Zhao GJ, Xu C, Ying JC, Lu WB, Hong GL, Li MF, et al. Association between furosemide administration and outcomes in critically ill patients with acute kidney injury. Crit Care. 2020 Mar 4;24(1):75.