Thirst level trajectories and influencing factors in critically ill patients based on a group-based trajectory model: A prospective observational study

Abstract

Background

Thirst is widely recognized as a prevalent, severe, and often under-managed symptom in critically ill patients. Despite increased recognition of its importance in Intensive Care Unit (ICU) practice, the current management status of thirst has not substantially improved. Therefore, an urgent need exists to understand the short-term dynamics and influencing factors of thirst in this population.

Objective

This study aimed to identify distinct trajectories and predictors of trajectory subgroups in critically ill patients within the initial 24 hours of ICU admission.

Methods

A prospective observational study was conducted between November 2024 and February 2025 at a tertiary hospital in Chengdu, China. Data were collected using a unified electronic case report form comprising three sections: general demographics, clinical characteristics, and the Critically Ill Patient’s Thirst Assessment Scale. Thirst levels were assessed at five time points: upon ICU admission, and at 6, 12, 18, and 24 hours post-admission. Group-based trajectory modeling was employed to identify distinct thirst trajectories, and unordered multivariate logistic regression was used to determine the predictors of trajectory subgroups.

Results

A total of 209 eligible participants were included in the study. Three distinct thirst trajectories were identified: the low-level sustained group (41.8%), the moderate-level decline group (41.9%), and the high-level fluctuating group (16.3%). Key predictors associated with a higher likelihood of belonging to the moderate to high thirst trajectory groups included serum potassium and ionized calcium levels, oxygenation index, and antibiotic use.

Conclusions

This study enhances the understanding of distinct thirst trajectories and the predictors associated with the moderate to high thirst trajectory subgroups. Early identification and stratification of thirst levels upon ICU admission are crucial for developing targeted management strategies to alleviate thirst distress in critically ill patients.

What is already known.

• •Thirst is a prevalent, severe, and often under-managed symptom among critically ill patients.
• •Thirst not only causes distress recollections for patients but is also associated with an increased risk of potential complications, including aspiration, delirium and infection.

What this paper adds.

• •Thirst exhibited a dynamic trend over time.
• •ICU patients displayed significant heterogeneity in thirst symptom presentation.
• •Key predictors associated with a higher likelihood of belonging to the moderate to high thirst trajectory subgroups included serum potassium and ionized calcium levels, oxygenation index, and antibiotic use.

1. Introduction

Thirst is widely recognized as a prevalent, severe, and commonly under-managed symptom in the Intensive Care Unit (ICU) , with reported incidence rates ranging from 43.8% to 90% (Çelik et al., 2023; Saltnes-Lillegård et al., 2023). Critically ill patients frequently undergo fluid restriction and receive therapeutic interventions (e.g., mechanical ventilation support, anesthetic agents, anticholinergics, antihypertensive agents, and diuretics) that directly contribute to xerostomia. Furthermore, even with intravenous hydration, patients experiencing hypovolemia or underlying cardiac or renal dysfunction often report thirst due to their inability to self-administer oral fluids (Negro et al., 2022). The impact of this symptom is profound, with some survivors reporting enduring, intense recollections of thirst post-discharge —— akin being in the desert (Flim et al., 2022). Moreover, persistent thirst is associated with an elevated risk of potentially fatal complications, including aspiration, delirium, oral candidiasis, and systemic infections (Cannon et al., 2023; Mhatre et al., 2024; Sato et al., 2019). These complications substantially worsen patient outcomes, increase the demand for longer-term care, and contribute to significant economic burden.

Unrelieved thirst is a significant source of patient distress that caregivers can and should endeavor to alleviate (Lin et al., 2023). Despite increased recognition of thirst management in ICU practice, recent studies indicate that the management status of this symptom has not substantially improved (Lee et al., 2024; Stoopler et al., 2024). In clinical practice, considerable emphasis is often placed on correcting osmotic and electrolyte parameters—such as serum sodium abnormalities or clinical and ultrasonographic indicators—leading to a management focus predominantly on physiological balance. Therefore, in the high-acuity ICU environment, nurses frequently prioritize acute physiological changes and treatment regimens over subjective symptoms like thirst (Doherty et al., 2024). A prevailing belief that thirst is transient and not immediately life-threatening may contribute to insufficient assessment and intervention. Communication barriers, notably in intubated patients, further complicate the expression and recognition of thirst needs, creating a critical disconnect between patient experience and nursing action. Drawing upon established research on thirst-quenching mechanisms, thirst can be alleviated through both pre-absorptive and post-absorptive pathways. Pre-absorptive mechanisms, such as the application of low-temperature and menthol, engage the transient receptor potential melastatin subtype 8 (TRPM8) channel located in cold-sensitive receptors of the trigeminal and glossopharyngeal nerve endings. Post-absorptive effects, achieved by restoring plasma osmotic pressure balance after substantial fluid intake. Given this, a low-volume yet effective pre-absorptive mechanism may be particularly beneficial for alleviating thirst in ICU patients, especially those who undergo frequent fasting or require strict fluid volume management (Armstrong and Kavouras, 2019; Nakaya et al., 2024). Thus, addressing these clinical priorities—specifically, identifying patients at high risk for thirst, establishing optimal intervention timing, and developing personalized management strategies—is essential to inform and improve clinical practice.

Previous surveys examining the association between thirst intensity and risk factors have largely employed cross-sectional methodologies, often limiting assessment to conscious patients (Belete et al., 2022; Ho et al., 2021; Lee et al., 2020). Longitudinal data tracking short-term thirst fluctuation in the ICU population are notably limited. While a recent observational study detailed temporal thirst trends (Zhao et al., 2024), its failure to address inherent population heterogeneity may result in generalized conclusions regarding dynamic trends. This methodological constraint warrants further consideration.

Trajectory analysis is a longitudinal methodology that employs repeated measurements to characterize the temporal progression of indicators such as behavior, clinical status, symptoms, or biomarkers. The group-based trajectory model (GBTM), proposed by Nagin in 1999, is a widely utilized technique within clinical psychology and medical research for identifying and analyzing these trajectories. The primary objective of GBTM is to identify distinct subgroups exhibiting similar longitudinal profiles for a single indicator (Nagin et al., 2018).

Guided by the Symptom Management Dynamic Model (Brant et al., 2010), this study aimed to explore the dynamic nature of thirst and the complexities of its management beyond simple identification. This model framework comprises five core dimensions: influencing factors, symptom experience, symptom change trajectories, intervention strategies, and outcomes. The model emphasizes that the temporal evolution of symptoms and the precision of targeted interventions are pivotal. Based on this theoretical foundation, we hypothesized that thirst levels fluctuate over time, and that critically ill patients exhibit significant heterogeneity in their symptom presentation.

Therefore, this study employed GBTM to identify distinct trajectory subgroups and predictors of thirst levels in critically ill patients within the initial 24 hours of ICU admission, thereby providing evidence for informed, targeted thirst management.

2. Methods

2.1. Study design and setting

This prospective and observational study was conducted at a tertiary hospital hospital in Chengdu, China, from November 2024 to February 2025. This facility includes two ICUs followed similar operational protocols. The doctor-to-nurse ratio is 1:3.44, and the nurse-to-patient ratio is 1:3.05. The approximate annual admission rate for the ICUs was about 3,000 patients. Building upon previous findings that identified thirst as the most prevalent and severe symptom on the first day of ICU admission among 353 patients (66%) (Saltnes-Lillegård et al., 2023). This study assessed thirst levels at five key time points: upon ICU admission (T0) and at 6 hours (T1), 12 hours (T2), 18 hours (T3), and 24 hours (T4) post-admission. The study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (see Supplementary Appendix 1).

2.2. Participants and sample

Based on the population availability or clinical feasibility, a convenience sampling method was employed to recruit the necessary sample size. The required number of participants was determined using the sample size estimation formula appropriate for a one-way repeated measures design. (Chang and Chang, 2012): $M=2[1+(K-1)\rho ]\frac{{\sigma }^2{(Z_{1-\alpha /2}+Z_{1-\beta })}^2}{K{\delta }^2}$, where M represents the sample size; $\sigma$ denotes the standard deviation of thirst intensity at admission (T0) ($\sigma$= 0.546); K is the number of repeated measurements per subject (K = 5); $\rho$ is correlation coefficient between measurements ($\rho$ = 0.5); $\delta$ is the smallest difference detectable between groups ($\delta$ = 0.25$\sigma$) (Klukowska et al., 2024); Z_1-α/2 and Z_1-β are critical value from the standard normal distribution (α = 0.05, 1- β = 0.8, Z_1-α/2 = 1.96, Z_1-β = 0.84). Accounting for an anticipated 10%-20% attrition rate, the minimum required sample size for this study was calculated to be 166. Given the absence of a universally accepted standard for determining sample size and specific calculation methods for Group-Based Trajectory Modelling (GBTM), a minimum sample size of 200 is recommended when utilizing the Bayesian Information Criterion (BIC) as the primary model selection criterion to ensure robust outcomes (Wu et al., 2025). A final sample of 209 patients was included in the analysis.

Inclusion criteria were: (1) age ≥18 years old; (2) ICU admission for ≥ 24 hours; Exclusion criteria based on a review of the medical records were: (1) diagnosis of Sjögren’s syndrome or uremia; (2) history of psychological disorders (e.g., schizophrenia, anxiety or depression) or cognitive impairment (e.g., dementia); (3) patients with oral cancer or those unable to have their oral condition assessed post-oral surgery. Withdrawal criteria were: (1) ICU readmission within 72 hours; (2) death, transfer, or discharge within 24 hours of ICU admission; (3) voluntary withdrawal at any point during the study.

2.3. Measurement

2.3.1. The critically ill patient’s thirst assessment scale

Currently, no gold standard exists for thirst diagnosis. Commonly used tools such as the Numeric Rating Scale (NRS) or Visual Analogue Scale (VAS) (Faiz, 2014) are not feasible for non-conscious or sedated ICU patients. To address this limitation, the Critically Ill Patient’s Thirst Assessment Scale (CIP-TAS) (Liao et al., 2024) was employed to assess thirst levels (Supplementary Appendix 2). Developed by Liao (Yunnan Cancer Hospital, Department of Critical Care Medicine, 2024) (Liao et al., 2024), the CIP-TAS objectively evaluates indicators including lip moisture/dryness, oral mucosal condition, tongue texture and coating, sputum viscosity, saliva volume and viscosity, dry mouth symptoms, and behavioral signs. The scale demonstrated good internal consistency, with an overall Cronbach's α coefficient of 0.827. Thirst severity was categorized as follows: no thirst (0 scores); mild thirst (1–3 scores); moderate thirst (4–6 scores); severe thirst (≥ 7 scores). Given that thirst can be subjective, the Intraclass Correlation Coefficient (ICC) was calculated to compare CIP-TAS scores with self-reported NRS scores in conscious patients (Koo and Li, 2016). The resulting ICC of 0.588 (P < 0.05) indicated a medium-intensity consistency between the two tools (Supplementary Appendix 3), thereby validating the CIP-TAS for measuring thirst intensity in this population.

2.3.2. The Mini-Cog test

Upon regain consciousness following ICU admission, patients’ cognitive function was assessed using the Mini-Cog test. The Mini-Cog test is a brief screening tool for cognitive impairment, comprising a three-item recall and a clock-drawing test (CDT). Scoring was performed as follows: (1) Word recall: 1 point was awarded for each correctly recalled word (maximum of 3 points); (2) Clock drawing test: scored as normal or abnormal. A normal CDT indicated the correct placement of numbers and hands. Scores were interpreted as follows, 0 point indicated a positive screen for cognitive impairment; 1-2 points with abnormal CDT indicated a positive screen for cognitive impairment; 1-2 points with normal CDT indicated a negative screen; and 3 points indicated a negative screen for dementia (in which case the CDT was not required) (Borson et al., 2003, 2000).

2.3.3. Other covariates

A custom-designed general information questionnaire was utilized to gather comprehensive data from ICU patients. Guided by the Symptom Management Dynamic Model framework, informed by previous literature (Wu et al., 2022; Xie et al., 2025) and clinical experience, the following variables were selected for investigation: (1) general demographics: gender, age, reason for ICU admission, marital status, smoking history (yes/no), alcohol taking history (yes/no), and presence of full denture (yes/no); (2) clinical characteristics: mechanical ventilation (MV) status (yes/no), nothing by mouth (NPO) status (yes/no), fluid input and output balance, comorbidities, medications, medical history, urine specific gravity and colour, total ICU length of stay, 24-hour fluid input and output volumes, body temperature, respiration rate, blood pressure, Body Mass Index (BMI), Acute Physiology and Chronic Health Evaluation II (APACHE II) scores, venous thromboembolism (VTE) risk assessment, D-dimer levels, blood glucose, serum sodium, serum potassium, ionized calcium levels, serum osmolality, C-reactive protein (CRP) levels, creatinine clearance (Ccr), oxygenation index, duration of mechanical ventilation, and duration of NPO status.

2.4. Data collection

A unified electronic case report form (eCRF) was developed for centralized data management, facilitating efficient data entry, retrieval, and subsequent analysis. All research personnel and nurses received comprehensive training on study procedures. To minimize variability, strict adherence to standardized protocols for patient screening, data collection, and evaluation, along with uniform inclusion and exclusion criteria, was maintained throughout the study. The team leader of each shift was responsible for data collection, with one team member assisting in its collection and verification. During the data collection period, patients received routine care in the ICU. To avoid potential bias, no specific interventions to alleviate thirst were provided to participants. Routine oral care was provided every 6 hours. For severe thirst, a damp gauze pad was offered for oral moistening and relief. For patients who had undergone endotracheal intubation, a 6-hour NPO period was maintained following extubation. After this 6-hour period, patients who successfully passed the Wakada Drinking Water Test were permitted to attempt oral fluid intake, which typically occurred approximately 24 hours after ICU admission for these intubated patients (Faleiros et al., 2025). The ICU environment’s temperature (18-22°C) and humidity (50%-60%) were monitored daily. All collected data were subjected to a rigorous double-checking process to ensure accuracy.

2.5. Statistical analysis

This study employed the group-based trajectory model (GBTM) to yield two key outputs: trajectory shapes, typically represented by polynomial functions of time, and individual probabilities of group membership (Nagin et al., 2024, 2018). The selection of an optimal GBTM followed a two-step procedure (Wang et al., 2024). Step 1: Determination of the optimal number of trajectory groups. This process begins by fitting models with a single group and incrementally increasing the number of groups. To ensure parsimony and prevent overfitting, the maximum polynomial order for the trajectories was fixed at a cubic term. The optimal number of groups was selected based on the following criteria: (1) a Bayesian Information Criterion (BIC) value closest to 0; (2) an Average Posterior Probability of Assignment (AvePP) > 0.7 for each group; (3) an Odds of Correct Classification (OCC) > 5.0 for each group; and (4) a minimum group membership of 5% of the participants. Step 2: Determination of trajectory shapes. Once the optimal number of groups was established, the shape of each trajectory was determined by fitting models, starting with the highest-order cubic term (Hyun et al., 2023; Wu et al., 2025). The outcome variable was assumed to follow a normal distribution in our GBTM model. The general functional form for the trajectory of variable Y over time ‘t’ was defined as: Y(t) = β0 + β1·t + β2·t² + β3·t³ (Nagin et al., 2024; Nagin and Odgers, 2010). The optimal model was identified based on the minimum Bayesian principle, with consideration for the model’s clinical interpretability and reliability. The GBTM model was implemented using the gbmt package in R (version 4.4.2).

Normality of continuous variables was assessed using the Shapiro-Wilk test. Categorical variables were presented as frequencies and proportions. Normally distributed continuous variables were reported as mean ± standard deviation ($\overline{X}$ ±$S$), and non-normally distributed continuous variables as medians (interquartile range: Q1, Q3) [M (Q1, Q3)]. Group differences for categorical variables were analyzed with the χ² test or Fisher’s exact test. Continuous variables were analyzed using the Kruskal-Wallis test or One-way Analysis of Variance (ANOVA), according to their distribution. Subsequently, the trajectory subgroups categorization derived from the model served as the dependent variable in an unordered multivariate logistic regression analysis to identify potential influencing factors (Hyun et al., 2023; Raymond and Darsaut, 2025; Wu et al., 2025). A P-value < 0.05 was considered statistically significant. Missing data were imputed using maximum likelihood (ML) estimation. To mitigated the risk of model overfitting, which can arise from an excessive number of variables, a collinearity diagnosis was performed. Variables with a Variance Inflation Factor (VIF) greater than 10 were excluded. All analyses were conducted using IBM SPSS (version 25.0).

2.6. Ethical consideration

This study adhered to the principles of the Declaration of Helsinki and received ethical approval from the Medical Ethics Committee and Clinical Trial Review Committee of the Sichuan Cancer Hospital (SCCHEC-02-2024-229). Written informed consent was obtained from all participants or their families.

3. Results

3.1. Participant characteristics

A total of 323 patients were recruited from November 2024 to February 2025. During eligibility screening, 76 patients were excluded: 5 due to being under 18 years of age, 10 for refusing to participate, and 61 for having an observation period shorter than 24 hours. Following a review of medical records and on-site assessments, an additional 38 patients were excluded: 11 due to a history of psychological disorders, 12 with cognitive impairment from traumatic brain injury, and 15 who had been readmitted within 72 hours. Finally, 209 met the eligibility criteria into the GBTM analysis (see Fig. 1). Participant characteristics are presented in Table 1A, Table 1B, Table 1C. The participant cohort was predominantly male (68.4%), and aged 40-79 years (87.1%). Over half of the participants underwent major surgery (53.6%). Furthermore, 60.8% of patients received mechanical ventilation (MV), and 82.3% of patients were in NPO status. The median ICU length of stay was 48 (24.0, 106.5) hours.

Fig. 1.

Flowchart of the data collection procedure.

Table 1A.

General demographics for the total cohort and each trajectory subgroup (n=209).

Characteristics	Total (n=209)	Trajectory subgroups			Statistics	P
		1 (n=86)	2 (n=35)	3 (n=88)
Gender (n=209)
Male	143 (68.4)	59 (68.6)	22 (62.9)	62 (70.5)	$\chi 2$=0.671	0.715
Female	66 (31.6)	27 (31.4)	13 (37.1)	26 (29.5)
Age, years (n=209)
18-39	16 (7.7)	9 (10.5)	1 (2.9)	6 (6.8)	$\chi 2$=6.834	0.323*
40-59	78 (37.3)	29 (33.7)	14 (40)	35 (39.8)
60-79	104 (49.8)	46 (53.5)	19 (54.3)	39 (44.3)
≥80	11 (5.3)	2 (2.3)	1 (2.9)	8 (9.1)
Marital status (n=209)
Single	6 (2.9)	3 (3.5)	1 (2.9)	2 (2.3)	$\chi 2$=3.789	0.704*
Married	187 (89.5)	75 (87.2)	33 (94.3)	79 (89.8)
Divorced	9 (4.3)	6 (7.0)	0 (0.0)	3 (3.4)
Widowed	7 (3.3)	2 (2.3)	1 (2.9)	4 (4.5)
Smoking historya(n=209)	45 (21.5)	20 (23.3)	10 (28.6)	15 (17)	$\chi 2$=2.226	0.329
Alcohol taking historya(n=209)	52 (24.9)	25 (29.1)	11 (31.4)	16 (18.2)	$\chi 2$=3.723	0.155
Full denture (n=209)	7 (3.3)	3 (3.5)	0 (0.0)	4 (4.5)	$\chi 2$=1.183	0.687*

^aNote: 1 week before admission to the ICU

^⁎indicates Fisher’s exact test.

Trajectory 1: low-level sustained group; Trajectory 2: high-level fluctuating group; Trajectory 3: moderate-level decline group;

Data are reported as $\overline{\mathrm{X}}$±$\mathrm{S}$ or median (Q1, Q3) for continuous variables and number (percentage) for categorical variables.

Table 1B.

Clinical characteristics for the total cohort and each trajectory subgroup (n=209).

Characteristics	Total (n=209)	Trajectory subgroups			Statistics	P
		1 (n=86)	2 (n=35)	3 (n=88)
Reason for ICU admission (n=209)
Major surgery	112 (53.6)	42 (48.8)	20 (57.1)	50 (56.8)	$\chi 2$=10.002	0.425*
Sepsis	12 (5.7)	6 (7.0)	1 (2.9)	5 (5.7)
Respiratory failure	15 (7.2)	3 (3.5)	3 (8.6)	9 (10.2)
Acute Exacerbation of Chronic Disease	31 (14.8)	15 (17.4)	6 (17.1)	10 (11.4)
Acute Disease	28 (13.4)	15 (17.4)	5 (14.3)	8 (9.1)
Other	11 (5.3)	5 (5.8)	0 (0.0)	6 (6.8)
Mechanical ventilation (MV) (n=209)	127 (60.8)	49 (57)	21 (60)	57 (64.8)	$\chi 2$=1.119	0.571
NPO (n=209)	172 (82.3)	73 (84.9)	26 (74.3)	73 (83.0)	$\chi 2$=1.963	0.375
Fluid input and output balanceb(n=209)
Negative balance	45 (21.5)	22 (25.6)	6 (17.1)	17 (19.3)	$\chi 2$=1.489	0.475
Positive balance	164 (78.5)	64 (74.4)	29 (82.9)	71 (80.7)
Comorbidities (n=209)
Renal	8 (3.8)	6 (7.0)	0 (0.0)	2 (2.3)	$\chi 2$=3.343	0.156*
Cardiovascular	58 (27.8)	21 (24.4)	9 (25.7)	28 (31.8)	$\chi 2$=1.275	0.529
Diabetes	27 (12.9)	10 (11.6)	6 (17.1)	11 (12.5)	$\chi 2$=0.696	0.706
Medications (n=209)
Anestheticsc	111 (53.1)	43 (50.0)	18 (51.4)	50 (56.8)	$\chi 2$=0.860	0.651
Sedativesc	121 (57.9)	49 (57.0)	19 (54.3)	53 (60.2)	$\chi 2$=0.413	0.813
Opioidsc	118 (56.5)	45 (52.3)	19 (54.3)	54 (61.4)	$\chi 2$=1.526	0.466
Diureticsc	31 (14.8)	12 (14.0)	6 (17.1)	13 (14.8)	$\chi 2$=0.201	0.905
Vasopressor drugsc	63 (30.1)	29 (33.7)	10 (28.6)	24 (27.3)	$\chi 2$=0.908	0.635
Proton pump inhibitorsc	141 (67.5)	61 (70.9)	24 (68.6)	56 (63.6)	$\chi 2$=1.078	0.583
Antibioticsc	130 (62.2)	50 (58.1)	26 (74.3)	54 (61.4)	$\chi 2$=2.804	0.246
Total ICU length of stay, h (n=209)	48.0, (24.0, 106.5)	48.0, (24.0, 112.0)	48.0, (24.0, 96.0)	39.5, (24.0, 107.3)	H=0.553	0.758
24-hour fluid input and output volumesb(n=209)	948.0, (100.0, 1640.5)	1012.5, (-94.0, 1786.3)	726.0, (100.0, 1414.0)	1006.0, (334.3, 1672.5)	H=0.990	0.610
Temperatured(n=209)	36.7, (36.5, 36.9)	36.8, (36.5, 37.0)	36.6, (36.30, 36.8)	36.7, (36.50, 36.9)	H=3.024	0.220
Respiration ratee(n=209)	18.0, (15.0, 21.0)	19.0, (15.0, 21.0)	19.0, (15.0, 22.0)	18.0, (15.0, 21.0)	H=0.854	0.652
Systolic blood pressure, mmHge(n=209)	121.7 ± 19.0	120.1 ± 18.1	122.5 ± 18.6	123.0 ± 20.2	F=0.558	0.573
Diastolic blood pressure, mmHge(n=209)	72.8 ± 11.6	71.3 ± 10.7	71.9 ± 13.8	74.7 ± 11.4	F=1.973	0.142
BMI, kg/m2 (n=209)	22.4, (19.9, 24.9)	22.3, (19.9, 24.9)	22.5, (19.4, 24.7)	22.2, (19.8, 25.3)	H=0.020	0.990
APACHE Ⅱ (n=209)	12.0, (9.0, 16.0)	12.0, (9.0, 17.0)	11.0, (8.0, 17.0)	11.0, (8.3, 16.0)	H=1.573	0.455
VTE (n=209)	6.0, (5.0, 8.0)	7.0, (5.0, 8.0)	7.0, (5.0, 8.0)	6.0, (5.0, 8.0)	H=2.339	0.311
Duration of MV, h (n=209)	4.0, (0.0, 9.0)	3.0, (0.0, 8.3)	4.0, (0.0, 9.0)	5.0, (0.0, 9.0)	H=0.983	0.612
Duration of NPO, h (n=209)	24.0, (16.5, 48.0)	24.0, (16.5, 36.0)	32.0, (0.0, 72.0)	24.0, (24.0, 48.0)	H=0.867	0.648

Note: NPO, nothing by mouth; BMI, body mass index; APACHE Ⅱ, acute physiology and chronic health evaluation II; VTE, venous thromboembolism; MV, mechanical ventilation, including endotracheal intubation, tracheotomy, or mask oxygen inhalation.

Trajectory 1: low-level sustained group; Trajectory 2: high-level fluctuating group; Trajectory 3: moderate-level decline group;

Data are reported as $\overline{\mathrm{X}}$±$\mathrm{S}$ or median (Q1, Q3) for continuous variables and number (percentage) for categorical variables;

^⁎indicates Fisher’s exact test

^bthe first 24 hours in ICU

^c12 hours before admission to ICU to 24 hours within ICU stay

^dthe maximum body temperature within 24 hours of ICU stay

^ewhen admitted to the ICU.

Table 1C.

Laboratory parameters for the total cohort and each trajectory subgroup (n=209).

Characteristics	Total (n=209)	Trajectory subgroups			Statistics	P
		1 (n=86)	2 (n=35)	3 (n=88)
Urine specific gravity (n=162)
< 1.003	10 (4.8)	7 (8.1)	1 (2.9)	2 (2.3)	$\chi 2$=8.314	0.048*
1.003-1.030	146 (69.9)	56 (65.1)	27 (77.1)	63 (71.6)
> 1.030	6 (2.9)	5 (5.8)	1 (2.9)	0 (0.0)
Urine colour§(n=162)
Level 1	24 (11.5)	11 (12.8)	6 (17.1)	7 (8.0)	$\chi 2$=10.529	0.272*
Level 2	126 (60.3)	54 (62.8)	19 (54.3)	53 (60.2)
Level 3	5 (2.4)	1 (1.2)	1 (2.9)	3 (3.4)
Level 4	3 (1.4)	0 (0.0)	2 (5.7)	1 (1.1)
Level 5	2 (1.0)	1 (1.2)	1 (2.9)	0 (0.0)
Level 6	2 (1.0)	1 (1.2)	0 (0.0)	1 (1.1)
D-dimer, μg/mL (n=199)	3.0, (1.2, 6.1)	3.5, (1.2, 6.8)	3.4, (0.8, 5.8)	2.8, (1.4, 5.6)	H=0.868	0.648
Blood glucose, mmol/Le(n=209)	7.6, (6.2, 9.3)	7.4, (6.0, 9.0)	7.1, (6.3, 9.3)	8.1, (6.2, 9.4)	H=0.712	0.700
Sodium, mmol/Lb(n=209)	139.9, (135.2, 143.0)	139.9, (133.7, 143.4)	138.0, (135.8, 143.0)	140.2, (135.6, 142.6)	H=0.515	0.773
Potassium, mmol/Lb(n=209)	3.9, (3.6, 4.3)	3.9, (3.7, 4.4)	3.8, (3.6, 4.2)	3.9, (3.7, 4.2)	H=2.189	0.335
Ionized calcium, mmol/Lb(n=209)	1.13, (1.09, 1.17)	1.12, (1.09, 1.16)	1.13, (1.10, 1.20)	1.13, (1.09, 1.18)	H=1.923	0.382
Osmolality, mmol/Lb(n=209)	301.9, (293.9, 308.7)	303.3, (293.5, 310.4)	299.9, (293.9, 310.5)	301.1, (296.1, 307.8)	H=0.057	0.972
CRP, mg/L (n=184)	60.4, (9.1, 136.8)	66.6, (9.8, 161.1)	62.5, (17.6, 123.4)	54.1, (5.6, 122.3)	H=1.137	0.566
Ccr, ml/min/1.73m2 (n=209)	96.1, (75.1, 106.2)	93.4, (73.0, 107.1)	92.3, (73.4, 107.8)	98.6, (78.9, 105.2)	H=1.942	0.379
Oxygenation index, mmHg, (n=209)	370.5, (269.6, 486.8)	320.3, (243.0, 436.4)	385.6, (350.4, 491.0)	392.3, (284.0, 502.8)	H=5.730	0.057

Note: CRP, C-reactive protein; Ccr, creatinine clearance;

Data are reported as $\overline{\mathrm{X}}$±$\mathrm{S}$ or median (Q1, Q3) for continuous variables and number (percentage) for categorical variables;

Bold indicates statistical significance

^⁎indicates Fisher’s exact test

Trajectory 1: low-level sustained group; Trajectory 2: high-level fluctuating group; Trajectory 3: moderate-level decline group;

^bthe first 24 hours in ICU

^ewhen admitted to the ICU;

^§from the urine routine report. Level 1: colorless; Level 2: light yellow or red; Level 3: light brown; Level 4: yellow; Level 5: red; Level 6: brown.

3.2. The trajectories of thirst level

Fig. 2 displays an overall downward trend in thirst levels across all investigated patients. Model selection was based on clinical interpretability and goodness-of-fit metrics, primarily the Bayesian Information Criterion (BIC). Although the four-group trajectory model (C=4) exhibited the lowest BIC score, the small sample size within this fourth group rendered it clinically impractical (Nagin et al., 2024). Consequently, considering all fitting metrics and clinical relevance, a three-group trajectory model (C=3) was selected for subsequent analysis (Table 2).

Fig. 2.

The overall trend of all the investigated patients. The time of upon ICU admission (T0) and at 6 hours (T1), 12 hours (T2), 18 hours (T3), and 24 hours (T4) post-admission.

Table 2.

Trajectory evaluation metrics.

Tracks	AIC	BIC	AvePP	OCC	Probability
1	4693.442	4708.298	1	-	1
2	4356.752	4406.269	0.943/0.923	11.1/18	0.599/0.401
3	4193.022	4267.298	0.939/0.923/0.905	21.3/61.5/13.3	0.418/0.163/0.419
4	4135.249	4234.284	0.923/0.929/0.896/0.907	15.2/77/16.8/117.7	0.440/0.146/0.338/0.076

Note: Bold values indicate the selected class. AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; AvePP, Average posterior probability; OCC, Odds of Correct Classification.

The identified trajectories were: Trajectory 1 (the low-level sustained group, 41.8%, 2 < thirst levels < 4 throughout 24 hours), Trajectory 2 (the high-level fluctuating group, 16.3%, 6 < thirst levels < 8 throughout 24 hours), and Trajectory 3 (the moderate-level decline group, 41.9%, 4 < thirst levels < 6 throughout 24 hours) (Fig. 3). To accurately characterize the trajectory shapes of each group, a time-series polynomial regression model was fitted to each group separately. Trajectory 1 followed a cubic function (with significant coefficients for β1, β2, and β3; P < 0.05) and an intercept of 4.009. This trajectory exhibited a complex pattern characterized by a rapid initial decline followed by a slower rate of decline. Trajectory 2 also followed a cubic function (with significant coefficients for β1, β2, and β3; P < 0.05) and had an intercept of 7.578. However, the overall magnitude of change for Trajectory 2 was smaller compared to Trajectory 1. Trajectory 3 followed a linear function (with a significant t coefficient; P < 0.05), and had an intercept of 5.480, showing a stable linear decline over time (Table 3). Fig. 3 presents that apart from the high-level fluctuating group, which showed an upward trend from 6 hours post-ICU admission, the thirst levels of the other two groups peaked upon admission and subsequently decreased to varying degrees.

Fig. 3.

Trajectories of thirst level within the initial 24 hours of ICU admission. The Y-axis indicates the outcome of the Critically Ill Patient’s Thirst Assessment Scale, and the X-axis represents the time of upon ICU admission, and at 6 hours, 12 hours, 18 hours, and 24 hours post-admission.

Table 3.

Functional form for each trajectory.

Tracks	Parameters	Estimate	Standard error	P
1	Intercept	4.009	0.140	<0.000
	β1	-0.357	0.054	<0.000
	β2	0.025	0.005	<0.000
	β3	-0.000	0.000	<0.000
2	Intercept	7.578	0.183	<0.000
	β1	-0.133	0.071	0.011
	β2	0.017	0.007	<0.050
	β3	-0.000	0.000	<0.010
3	Intercept	5.480	0.074	<0.000
	β1	-0.055	0.005	<0.000

3.3. Univariate analysis of thirst level trajectory subgroups

Univariate analysis indicated no significant group differences for most variables, with the exception of urine specific gravity (P = 0.048) (Table 1C). Despite the absence of significant differences in most variables, descriptive observations revealed that major surgery was the most common reason for ICU admission across all groups (53.6%). When comparing specific groups, the high-level fluctuating group exhibited a higher proportion of admissions for major surgery (57.1%) compared to the other groups. This group also had a higher proportion of patients with a history of smoking (28.6%) and alcohol consumption (31.4%). Antibiotic use was also higher in the high-level fluctuating group (74.3%) compared to the other groups (58.1% and 61.4%). Furthermore, the duration of NPO was longer (32 hours vs 24 hours) and serum potassium levels were lower (3.8 mmol/L vs 3.9 mmol/L) in the high-level fluctuating group. Patients admitted for respiratory failure, those receiving mechanical ventilation, and the duration of MV was comparatively higher in the moderate-level decline group than in the other two groups (respiratory failure: 10.2% vs 3.5% and 8.6%; MV receipt: 64.8% vs 57% and 60%; MV duration: 5 hours vs 3 hours and 4 hours, respectively). The oxygenation index was also higher in the moderate-level decline group (392.3 mmHg vs 385.6 mmHg and 320.3 mmHg, respectively). Furthermore, ionized calcium levels were elevate in both the moderate-level decline and the high-level fluctuating groups compared to the low-level sustained group. Table 4 summarizes the representative phenotype of each trajectory subgroup.

Table 4.

Summary of the demographics of the trajectory subgroups and the trajectory ranks for characteristics.

Items	Trajectory 1	Trajectory 2	Trajectory 3
Demographics
Gender	Male	Male	Male
Age	40-79	40-79	40-79
Reason for ICU admission	Major surgery (third)	Major surgery (first)	Major surgery (second)
Ranks for characteristics
Smoking history	Second	First	Third
Alcohol taking history	Second	First	Third
Mechanical ventilation	Third	Second	First
Antibiotic use	Third	First	Second
Potassium	First	Third	Second
Ionized calcium	Second	First	Third
Oxygenation index	Third	Second	First
Duration of MV	Third	Second	First
Duration of NPO	Third	First	Second

Note: MV, mechanical ventilation; NPO, nothing by mouth.

Trajectory 1: low-level sustained group; Trajectory 2: high-level fluctuating group; Trajectory 3: moderate-level decline group.

3.4. Influencing factors of thirst level trajectory subgroups

Drawing upon clinical experience and existing literature, a comprehensive set of variables was initially selected for inclusion in a multivariate logistic regression model. These variables encompassed the reason for ICU admission, smoking and alcohol taking history, denture, mechanical ventilation status (yes/no), NPO status (yes/no), fluid input and output balance, comorbidities, medications, urine specific gravity, urine colour, total ICU length of stay, temperature, BMI, APACHE Ⅱ scores, VTE, D-dimer, blood glucose, sodium, potassium, ionized calcium, osmolality, CRP, Ccr, oxygenation index, and the duration of MV and NPO were also considered. Collinearity diagnostics were performed. Variables exhibiting a Variance Inflation Factor (VIF) > 10, such as anesthetics, sedatives, opioids, sodium, and osmolality, were subsequently excluded from the model (see Supplementary Appendix 4). The goodness-of-fit for the regression mode was assessed and adjusted following the candidacy assessment of remaining variables after removing those with weaker clinical relevance. The final model’s fit result was χ² =30.805 (P=0.014). The overall regression model was statistically significant, identifying potassium, ionized calcium, antibiotic use, and oxygenation index as independent predictors within the three subgroups (Fayers and King, 2008) (see Supplementary Appendix 5 and 6).

Taking the low-level sustained group as the reference category, multivariate logistic regression analysis identified several independent predictors. Patients with low potassium levels exhibited a significantly increased likelihood of classification into either the moderate-level decline group (Odds Ratio [OR] = 0.629, 95% Confidence Interval [CI]: 0.398-0.993) or the high-level fluctuating group (OR = 0.454, 95% CI: 0.223-0.922); A higher oxygenation index was associated with a significantly increased likelihood of belonging to the moderate-level decline group (OR = 1.003, 95% CI: 1.000-1.005). Conversely, higher ionized calcium levels (OR = 3.797, 95% CI: 1.080-13.354) and antibiotic use (OR = 0.381, 95% CI: 0.150-0.969) were associated with a significantly increased likelihood of classification into the high-level fluctuating group (Table 5).

Table 5.

Multivariate logistic regression analysis for three trajectory subgroups.

Group comparison	predictors	Coefficient	Standard error	Warl$\mathbf{d}\chi 2$	P	OR	95% CI
Moderate-level decline groupa	Potassium	-0.464	0.233	3.962	0.047	0.629	0.398 - 0.993
	Oxygenation index	0.003	0.001	5.410	0.020	1.003	1.000 - 1.005
High-level fluctuating groupa	Potassium	-0.790	0.362	4.768	0.029	0.454	0.223 - 0.922
	Ionized calcium	1.334	0.642	4.324	0.038	3.797	1.080 - 13.354
	Antibiotic1	-0.965	0.476	4.103	0.043	0.381	0.150 - 0.969

^aNote: With the low-level sustained group as the reference category

¹With the antibiotic use as the reference category; VS, versus.

4. Discussion

This is the first observational study using an objective tool to investigate thirst levels in critically ill patients during their initial 24 hours of ICU stay. Furthermore, we utilized GBTM to capture distinct dynamic changes in thirst levels within this population. A key finding was the identification of three distinct trajectory subgroups: the low-level sustained group (41.8%), the moderate-level decline group (41.9%), and the high-level fluctuating group (16.3%). This finding highlights the heterogeneity of the thirst experience among ICU patients. Several significant predictor factors, including serum potassium, ionized calcium, oxygenation index, and antibiotic use, were found to be associated with these trajectory subgroups. The identification of these predictors provides a valuable foundation for the early identification of individuals at high risk for thirst by clinical medical staff, thereby informing assessment and intervention strategies for thirst management in ICU patients.

Overall, thirst levels among all investigated patients in our study demonstrated a downward trend, consistent with previous observational studies using subjective assessment tools (Waldréus et al., 2018; Zhao et al., 2024), which indicate a general decline in overall thirst levels among their study populations. However, analysis of patient distribution within each trajectory subgroup revealed that 58.2% experienced moderate to high thirst levels. This prevalence aligns with an observational study of 353 ICU patients, which reported thirst as the most prevalent symptom on the first day of ICU admission (66%) and the most severe symptom throughout a seven-day observation period (64%) (Saltnes-Lillegård et al., 2023). Therefore, while the overall thirst levels may be decreasing, the persistent issue of thirst symptoms in patients within the moderate to high thirst groups warrants continued attention.

Further detailed analysis of our study revealed distinct shapes and characteristics of thirst progression within the three identified subgroups. Specifically, Trajectory 2 (the high-level fluctuating group) exhibited an upward trend in thirst beginning approximately 6 hours post-ICU admission. Patients in Trajectory 2 experienced the most severe thirst, with thirst levels reaching as high as 6 to 8. This group commonly reported a history of smoking and alcohol use, received antibiotic therapy, and had prolonged NPO status. Furthermore, they exhibited lower serum potassium levels and higher ionized calcium levels. Due to treatment protocols, ICU patients frequently undergo fluid and nutritional restrictions. As most patients in the moderate to high thirst level groups had undergone surgery and were NPO prior to ICU admission, they may have developed electrolyte imbalances, such as hyponatremia and hypokalemia. Trajectory 2 indicated an upward trend in thirst levels approximately 6 hours after ICU admission. Given that the majority of patients in this subgroup were postoperative, we hypothesize that this upward trend may be attributable to the metabolism of anesthetic sedatives and anticholinergic agents, coupled with the planned tracheal extubation within 24 hours post-surgery (Faleiros et al., 2025). Following surgery, patients transferred to the ICU with stable vital signs and approaching extubation criteria often regained consciousness and expressed a desire for oral fluids. However, healthcare providers frequently prioritize postoperative fluid and electrolyte balance and safety concerns over the patient’s reported thirst. Therefore, patients typically maintain NPO status. Furthermore, the ICU environment may induce anxiety and tension in alert patients, potentially exacerbating thirst (Belete et al., 2022). Thirst is regulated by both homeostatic and non-homeostatic mechanisms. Homeostatic regulation is influenced by osmotic and electrolyte parameters, which may serve as predictors of thirst severity. Therefore, clinical caregivers should prioritize monitoring osmotic and electrolyte balance in patients undergoing prolonged fasting and adjust fluid volume accordingly. For patients with prolonged NPO status, integrating mindfulness, meditation techniques, or virtual reality (VR) may help alleviate thirst distress (Javvaji et al., 2024; Sezer and Sacchet, 2025). In critically ill patients, non-homeostatic pathways, such as oral cavity stimuli, habits, and interoception, may trigger thirst, even with strict monitoring of fluid and electrolyte balance (Nakaya et al., 2024). Given these insights, enhanced attention is warranted for the diagnosis of thirst and the implementation of safe management strategies (Nascimento et al., 2023). Further research is recommended to explore the efficacy of low-dose pre-absorptive agents (e.g., low-temperature and menthol) in ICU settings (Nascimento et al., 2014, 2023; Shu et al., 2026).

We observed that patients receiving antibiotics were more likely to be classified into the high-level fluctuating group. It is hypothesized that this association may be explained by antibiotic-induced gastrointestinal disturbances, such as diarrhea or dyspepsia. These disturbances can lead to fluid loss, subsequently triggering thirst as a rehydration signal (Kollef and Micek, 2014). Routine antibiotic use in the ICU for infection prevention may contribute to this mechanism by altering gut microbiota, potentially resulting in mild gastrointestinal symptoms and subsequent fluid depletion. While the relationship between antibiotic use and thirst has not been previously reported in cross-sectional studies, the growing concern regarding antibiotic resistance within the ICU setting necessitates further research to confirm this novel finding.

In contrast, both Trajectory 1 (the low-level sustained group) and Trajectory 3 (the moderate-level decline group) peaked upon admission, followed by a gradual descending trend over time to varying degrees. Patients in Trajectory 1 experienced the mildest thirst intensity (ranging from 2 to 4). This group also demonstrated a lower proportion of MV and antibiotic use, as well as shorter duration of MV and NPO status. Conversely, patients in Trajectory 3 experienced relatively high thirst intensity, ranging from 4 to 6, with a descending trend observed over time. Although Trajectory 2 and 3 shared some similar characteristics and did not exhibit statistically significant differences, notable findings emerged. Specifically, a greater proportion of patients in Trajectory 3 were admitted to the ICU for respiratory failure. This subgroup also presented with a higher prevalence of MV treatment and longer MV duration, accompanied by a higher oxygenation index. MV support is a routine practice in ICU settings; however, prolonged exposure to iatrogenic hyperoxia and high arterial oxygen tension is common (Sarıtaş et al., 2023). This may explain the elevated oxygenation index observed in patients receiving mechanical ventilation. Previous research suggests that MV can lead to mouth breathing, resulting in dry mouth and subsequent thirst. Patients with tracheal intubation may also experience complications such as bronchial mucosal injury due to overly deep intubation, which can exacerbate discomfort (Wu et al., 2022). Therefore, while ensuring patient safety, maintaining an appropriate depth of tracheal intubation, keeping a reasonable oxygenation index, and facilitating early extubation may effectively alleviate thirst distress in these patients.

The Symptom Management Dynamic Model posits that symptoms change dynamically over time, underscoring the necessity of personalized interventions aligned with these temporal manifestations to improve patient outcomes (Brant et al., 2010). Early risk stratification upon admission is also crucial to enable targeted interventions for patients identified in the moderate to high thirst trajectory groups. Healthcare providers should prioritize interventions addressing factors associated with prolonged NPO or MV status and monitor for antibiotic resistance. Implementing non-pharmacological strategies for thirst relief is recommended. Effective measures include a structured “thirst bundle” (Doherty et al., 2024), “optimized oral hygiene care” (Serra et al., 2023), and “the use of saliva substitutes and mucosal moistening agents” (Clark and Archer, 2022; Hong et al., 2024) may provide reference. Future efforts should focus on improving healthcare provider training in recognizing and addressing patient thirst needs, establishing a gold standard for thirst assessment, implementing communication aids to overcome barriers in nurse-patient dialogue, and the efficacy of low-dose pre-absorptive agents for ICU settings, ultimately enhancing the patient experience during ICU hospitalization (Dind et al., 2021).

5. Strengths and limitations

By longitudinally tracking changes in thirst severity among patients within the initial 24 hours of ICU admission, this study identified three heterogeneous subgroups and their associated influencing factors. These findings will assist clinical staff in accurately screening for patients at high risk of severe thirst early in the ICU admission process, thereby enabling optimal intervention timing and facilitating a proactive rather than reactive approach to thirst management. While this study has several limitations. Firstly, the generalizability of our findings may be limited due to recruitment from a single center and the use of convenience sampling. Although acknowledging the potential impact of convenience sampling on generalizability, efforts were made to mitigate selection bias by collecting data across various times. Additionally, the investigation period was limited to the initial 24 hours following ICU admission to understand the short-term, dynamic nature of thirst. This constraint may have influenced the observed statistical effect sizes in the subgroups. Future research should aim to incorporate longer observation periods and larger sample sizes to enhance the accuracy of findings. Secondly, although the objective thirst assessment scale utilized in this study possessed verified reliability and enabled assessment prior to patient regaining consciousness, its clinical utility remains limited. While the moisture degree of the oral mucosa can serve as an indicator of dehydration (Armstrong et al., 2016), this finding alone is insufficient. Further validation across a broader range of settings is therefore required. Moreover, a moderate ICC (0.588) was calculated between the CIP-TAS and NRS. Given the inherent complexity of thirst assessment and the absence of a universally accepted gold standard, further investigation into the consistency between subjective and objective thirst evaluations is necessary. Inter-rater reliability has not been reported, posing a risk of measurement bias. Finally, observational designs are limited in establishing causality, and residual confounding is an inherent risk. Furthermore, the multinomial regression model, which incorporated numerous variables relative, carried a risk of overfitting. Addressing these limitations through further research will provide more comprehensive and robust guidance for thirst management in ICU settings.

6. Conclusions

This study employed GBTM to identify three distinct thirst level trajectory subgroups in critically ill patients: the low-level sustained group, the moderate-level decline group, and the high-level fluctuating group. Serum potassium, ionized calcium, oxygenation index, and antibiotic use emerged as key indicators associated with an increased likelihood of belonging to the moderate to high thirst level groups. Consequently, early identification and stratification of thirst levels within the first 24 hours of ICU admission are crucial for developing timely, targeted management strategies to alleviate thirst distress in this population.

Data availability statement

Raw data are available on request from the corresponding author, M. X.

CRediT authorship contribution statement

Xing Shu: Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Shuang Yang: Writing – review & editing, Validation, Supervision, Project administration, Methodology, Funding acquisition. Aiping Hu: Writing – review & editing, Validation, Investigation, Conceptualization. Jiang Yuan: Visualization, Software, Methodology. Xin Liu: Supervision, Resources, Project administration, Investigation, Data curation. Zewen Pan: Supervision, Resources, Project administration, Data curation. Zhongjun Cao: Supervision, Resources, Investigation, Data curation. Mingfang Xiang: Writing – review & editing, Validation, Supervision, Resources, Project administration, Conceptualization.

Declaration of competing interest

All authors declare no conflicts of interest in this study.