Demographic, perioperative, biochemical and illness‐severity predictors of thirst in critically ill patients: a systematic review and meta‐analysis

Abstract

Thirst is a prevalent and distressing symptom among critically ill and surgical patients, yet its predictors remain poorly characterised. We aimed to quantify the associations between patient demographics, treatment modalities, biochemical markers and illness severity with self‐reported thirst in critical settings. We conducted a systematic review and meta‐analysis of studies published through April 2025 in PubMed, Embase, Web of Science, Cochrane Library and Ovid MEDLINE. Eligible observational studies in adult intensive care or perioperative cohorts reported associations between thirst and at least one of six factors: sex, anaesthetic technique, mechanical ventilation, blood urea nitrogen (BUN), serum osmolality or Acute Physiology and Chronic Health Evaluation (APACHE‐II) score. Data were pooled using DerSimonian–Laird random‐effects models to estimate odds ratios (ORs) for dichotomous predictors and weighted or standardised mean differences (WMDs/SMDs) for continuous variables. Heterogeneity was assessed by I ². Fifteen studies (n = 11 657) met inclusion criteria. Sex (OR = 1.04 (95% confidence interval (CI): 0.95–1.14) I ² = 0%) and mechanical ventilation (OR = 1.36 (95% CI: 0.91–2.05) I ² = 63%) were not significantly associated with thirst. Anaesthetic technique (OR = 1.43 (95% CI: 0.97–1.89); P < 0.001; I ² = 57%). APACHE‐II score (WMD = 0.24 points (95% CI: −0.55 to 1.04); I ² = 26%) and BUN (SMD = 0.27 (95% CI: −0.11 to 0.65); I ² = 84%) showed no significant effects. Serum osmolality exhibited the strongest relationship, with a mean elevation of 3.04 mOsm/kg in thirsty patients (95% CI: 0.15–5.93 (P = 0.039); I ² = 85%). Among the factors examined, elevated serum osmolality most reliably predicts patient‐reported thirst. Incorporating osmolality‐guided monitoring and tailored oral‐care protocols into critical‐care pathways may help pre‐empt and alleviate thirst, enhancing patient comfort in high‐acuity environments.

Introduction

Critically ill patients frequently experience intense sensations of thirst during their hospital stay, yet this distressing symptom remains under‐recognised and inadequately managed. ¹ Thirst in the intensive care unit (ICU) and perioperative settings arises from a complex interplay of demographic variables, treatment modalities, biochemical derangements and overall illness severity. ² Although often dismissed as a minor discomfort, unrelieved thirst can contribute to agitation, delirium and decreased patient satisfaction, and may even impair adherence to necessary therapies, such as fluid restrictions or mechanical ventilation protocols. ³ Understanding which patients are at greatest risk for severe thirst is therefore essential for developing targeted prevention and treatment strategies. ⁴

Demographic factors, such as sex, appear to modulate thirst perception and reporting. ⁵ Several observational studies have suggested that women may report higher thirst intensity scores than men under comparable clinical circumstances, possibly due to differences in osmoreceptor sensitivity or variations in oropharyngeal sensation and oral mucosal hydration. ⁶ Clarifying the influence of sex on thirst among critically ill cohorts will help clinicians anticipate which patients may require closer monitoring or earlier intervention. ⁷

Perioperative and ICU treatment factors also play pivotal roles in thirst genesis. ⁸ Anaesthetic technique, including the choice of general anaesthetics and regional blocks, can impair salivary secretion and alter central thirst pathways. ⁹ Patients undergoing prolonged mechanical ventilation are particularly vulnerable, as endotracheal tubes bypass normal oral sensory feedback and promote mucosal dryness, while ventilator humidification device inefficiencies can exacerbate airway and oropharyngeal desiccation. ¹⁰ Sedation practices, especially deep sedation, further blunt behavioural responses to thirst, delaying both patient‐reported symptoms and caregiver recognition. ¹¹ By categorising anaesthetic and ventilatory variables, we can identify modifiable care elements to mitigate thirst intensity without compromising essential life‐support measures. ¹²

Biochemical parameters provide objective markers that correlate with subjective thirst. Blood urea nitrogen (BUN) and serum osmolality, classic indices of hydration status, reflect the balance between fluid intake, renal concentration ability and insensible losses. ¹³ Elevated BUN and increased serum osmolality stimulate osmoreceptors in the hypothalamus, triggering vasopressin release and the behavioural drive to drink. ¹⁴ In critically ill patients, however, these homeostatic signals may be dysregulated by factors such as renal impairment, hypermetabolism and systemic inflammatory responses. ¹⁵ Precisely quantifying the relationship between serum solute markers and thirst severity can inform whether routine laboratory monitoring may serve as an early warning system for impending thirst episodes and whether pre‐emptive fluid adjustment could ameliorate discomfort. ¹⁶

Finally, overall illness severity as measured by validated scoring systems such as the Acute Physiology and Chronic Health Evaluation (APACHE‐II) captures the cumulative burden of organ dysfunction, hemodynamic instability and metabolic derangement, all of which influence thirst. ¹⁷ Patients with higher APACHE‐II scores often require more aggressive fluid management strategies, vasoactive support and mechanical ventilation, each of which can independently exacerbate thirst. ¹⁸ Moreover, severe illness can blunt neuroendocrine thirst regulation through hypothalamic suppression, resulting in paradoxical combinations of physiological dehydration and diminished thirst awareness. ¹⁹ Elucidating how global severity indices interact with thirst experience is critical for tailoring symptom management to patients' overall clinical trajectories, rather than relying on a one‐size‐fits‐all approach. ²⁰

Despite the high prevalence of thirst and its clear impact on patient comfort and outcomes, existing literature has yet to converge on which factors most reliably predict severe thirst in ICU populations. By synthesising evidence among demographic, perioperative, biochemical and illness‐severity domains, we aim to provide clinicians with a comprehensive risk‐stratification framework. Hence, this review was performed to determine the demographic, perioperative, biochemical and illness‐severity predictors of thirst in critically ill patients.

Methods

This systematic review and meta‐analysis was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) 2020 checklist.

Eligibility criteria

We included original studies of adult patients in high‐acuity hospital settings, including ICU cohorts and peri‐operative/post‐anaesthesia care cohorts, which quantitatively evaluated at least one pre‐specified factor in relation to patient‐reported thirst. This broader inclusion was chosen because thirst in ICU and peri‐operative care involves overlapping, clinically modifiable mechanisms (e.g. fasting/NPO status, airway/oxygen delivery and mouth‐breathing, fluid balance and osmoregulatory shifts), while also allowing us to examine peri‐operative/anaesthetic predictors that are not typically captured in ICU‐only cohorts.

The factors eligible for inclusion are as follows: (i) demographic variables (sex); (ii) perioperative/anaesthetic techniques (general anaesthesia, regional anaesthesia); (iii) mechanical ventilation status; (iv) biochemical markers (BUN, serum osmolality); and (v) illness‐severity scores (APACHE‐II). Eligible study designs comprised randomised controlled trials, cohort studies and case‐control studies. We excluded studies in paediatric populations (younger than 18 years), those without extractable effect estimates or measures of association, qualitative reports and conference abstracts lacking full data.

Search strategy

We searched MEDLINE (via PubMed and Ovid), Embase, Web of Science and Cochrane Library from inception through April 2025. In addition, we hand‐searched reference lists of included articles and relevant reviews. No language restrictions were applied. A medical librarian–assisted search combined controlled vocabulary (e.g. ‘thirst’, ‘xerostomia’, ‘critical illness’, ‘surgical patient’) and free‐text terms for each risk factor (e.g. ‘sex’, ‘mechanical ventilation’, ‘BUN’, ‘osmolality’, ‘APACHE‐II’). We used similar syntax adapted to each database. Search strategy for all the databases are provided in Appendix S1.

Selection process

After de‐duplication in EndNote 20, two reviewers independently screened titles and abstracts for relevance. Full texts of potentially eligible articles were retrieved and assessed against inclusion criteria. Disagreements at either stage were resolved through discussion or, if necessary, adjudication by a third reviewer. A PRISMA flow diagram (Fig. 1) depicts study selection. ²¹

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta‐Analyses flowchart.

Data collection process

Using a pilot‐tested, standardised form, two reviewers independently extracted study‐level data on author, year, country, study design, sample size, patient population, risk factor definitions, outcome measures (thirst intensity scales, prevalence), effect estimates (odds ratios (ORs), mean differences) and adjustment covariates. A data‐audit log recorded all resolution decisions. For missing or unclear data, we contacted corresponding authors up to two times; unresolved items were handled via sensitivity analyses. Where reported, we extracted respiratory support as described in each primary study, including invasive mechanical ventilation (IMV), non‐invasive ventilation (NIV) and oxygen delivery modalities (e.g. high‐flow nasal cannula (HFNC) or conventional oxygen therapy). Because reporting was heterogeneous, we harmonised these data at the level most consistently available among studies (typically ‘mechanical ventilation/intubation: yes/no’ or ‘any ventilatory support’). In addition, we summarised study‐specific operational definitions and the available distribution/range of key clinical variables (e.g. ventilation modality where reported, illness‐severity scores and biochemical markers) in Table S1 to facilitate clinical interpretability and to transparently indicate where primary studies did not report these details.

Data items

Primary data items included: (i) sex (male vs female), (ii) anaesthetic technique (general vs regional), (iii) mechanical ventilation (yes/no), (iv) BUN (mg/dL, continuous), (v) serum osmolality (mOsm/kg, continuous) and (vi) APACHE‐II score (continuous). We also extracted study characteristics for subgroup and meta‐regression analyses, including geographic region, publication year and study quality score.

For outcome timing, we extracted (when reported) the index timepoint for thirst assessment and any reported timing linkage between thirst measurement and exposure ascertainment. When studies did not report timing sufficiently to align exposures (e.g. urea/osmolality) with thirst assessment, this was recorded as not reported and is addressed as a limitation.

Selection of predictors

Candidate predictors were pre‐specified a priori based on: (i) biological plausibility and established mechanisms contributing to thirst (e.g. fluid restriction/NPO status, respiratory support, medications affecting fluid balance and biochemical indices related to osmotic regulation), and (ii) feasibility of extraction among studies (i.e., variables commonly defined and reported in primary studies). We did not restrict selection to a single prior systematic or scoping review; rather, we synthesised mechanistic rationale with the reporting patterns observed in the included literature to enable consistent extraction and pooling where possible.

Operational definitions of predictors

Sex was defined as biological sex (male vs female) as reported in the primary studies. Anaesthetic technique was defined as exposure to general anaesthesia versus regional anaesthesia/neuraxial techniques, where this was the anaesthetic exposure evaluated. Mechanical ventilation was extracted as a binary exposure (yes/no) as reported in each study; where studies explicitly distinguished modalities, we extracted invasive mechanical ventilation (endotracheal tube or tracheostomy ventilation) and non‐invasive ventilation separately, but where studies combined ventilatory support categories (e.g. mechanical ventilation alongside other respiratory supports such as HFNC or facemask/nasal cannula), we retained the study's original categorisation to avoid misclassification. BUN was defined as serum/plasma BUN (mg/dL) and serum osmolality as measured or calculated osmolality (mOsm/kg) as reported. APACHE‐II was extracted as the APACHE‐II score at or nearest to the time of thirst assessment.

Risk‐of‐bias assessment

The Newcastle–Ottawa Scale (NOS) was used to assess risk of bias in cohort, case–control or cross‐sectional analytical studies across selection, comparability and outcome/exposure domains. ²² Two reviewers rated each study independently; discrepancies were reconciled by consensus. Studies scoring ≥7 were deemed ‘high quality’, 4 to 6 ‘moderate’ and ≤3 ‘low’.

Statistical analysis

For dichotomous predictors (s, mechanical ventilation, anaesthetic technique), we either extracted number of events and participants with and without thirst outcome or point estimate (ORs with confidence interval (CI)) to calculate pooled ORs with 95% CIs. For continuous predictors (BUN, osmolality, APACHE‐II), we extracted mean, standard deviation (SD) with sample size to estimate pooled weighted mean difference (WMD) or standardised mean difference (SMD). We conducted random‐effects meta‐analyses using the DerSimonian–Laird method in STATA version 19 (StataCorp LLC). Heterogeneity was assessed with the Cochran Q statistic and quantified by I ² statistic. An I ² >75% indicates substantial heterogeneity.

Given potential baseline‐risk differences between ICU and peri‐operative cohorts, we pre‐specified an exploratory sensitivity approach stratifying analyses by clinical setting (ICU vs peri‐operative) where sufficient studies were available. However, for most predictors, the number of studies within each setting was limited and reporting definitions varied, which constrained the statistical power and interpretability of setting‐restricted pooled estimates; therefore, we present setting‐restricted findings narratively and emphasise this as a limitation. Sensitivity analyses excluded studies one at a time to perform leave‐one‐out analysis to assess the small study effects. For any predictor with ≥10 studies, we evaluated small‐study effects via funnel plots and Egger regression (two‐sided P < 0.10). ²³

Results

PRISMA results

As shown in Figure 1, our comprehensive database search identified 2777 unique records (1273 from PubMed, 632 from Ovid MEDLINE, 348 from EMBASE, 412 from Web of Science and 112 from the Cochrane Library). After removal of 872 duplicates, 1905 titles and abstracts were screened and 1801 were excluded for clearly not meeting our exposure, population or outcome criteria. We then retrieved and reviewed the full text of the remaining 104 articles (none of which were unavailable), excluding 89: 51 studies employed a different exposure definition, 23 did not enrol critically ill patients and 15 reported non‐thirst outcomes. The final quantitative and qualitative synthesis therefore comprised 15 studies that met all pre‐specified inclusion criteria. ⁵ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷

Characteristics of the included studies

Fifteen studies across nine countries were included, comprising nearly 14 000 participants (per‐study sample sizes ranged from 42 to 7134). Twelve studies employed a cross‐sectional design, one was a prospective cohort, one was retrospective and one was a randomised controlled trial. Overall, the evidence base was predominantly observational (14 of 15 studies), with only one randomised controlled trial (n = 61), representing approximately 0.5% of the total pooled cohort (61 of 11 657). The mean ages in the thirst and no‐thirst groups were similar, spanning approximately 51.5 to 68.7 years versus 52.1 to 68.2 years respectively. Thirst was most frequently measured using a Numerical Rating Scale (0–10) in eight studies and visual analog scale (VAS; 0–100 or 0–10) in three, with the remainder using self‐report or persistence scores. Sample sizes for those reporting thirst varied from 12 to 2776, while no‐thirst groups ranged from nine to 4358. Index timepoints of each of the included studies were widely varied. ICU assessments after 24‐h admission and specific clock times were explicitly reported in one study, while thirst was assessed after endotracheal extubation or daily ICU symptom assessments over 7 days, with clinical variables abstracted and medications administered within 6 h before assessment reported. Quality appraisal rated six studies as low risk of bias, four as moderate and five as high (Table 1).

Table 1.

General characteristics of the study details (n = 15)

Study (year)	Region	Study design	Characteristics of the participants	Thirst assessment tool	Sample size (thirst/no‐thirst)	Mean age (years) – thirst and no‐thirst group	Risk‐of‐bias assessment
Belete (2022)	Ethiopia	Cross‐sectional study	Adult patients who had undergone surgery and were admitted to PACU	Assessed based on NRS scale 0–10	252/172	Not reported	Moderate
Doi (2021)	Japan	Cross‐sectional study	Patients aged ≥18 years admitted in mixed medical‐surgical ICU	Assessed based on NRS	40/9	67.25 (5.4), 64.5 (4.6)	Moderate
Eng (2020)	Spain	Cross‐sectional study	Age ≥18 years, diagnosed HF with reduced or preserved ejection fraction.	Measured using VAS (0–100)	143/159	67 (13), 67 (12)	Low
Fan (2013)	China	Cross‐sectional study	≥5 months on haemodialysis, ≥18 years of age, mentally and physically being able to participate	Measured using VAS	12/30	65.2 (10.9), 63.5 (9.2)	Low
Lee (2020)	China	Cross‐sectional study	Patients ≥20 years in PACU	Assessed based on NRS scale 0–10	675/536	56.8 (15.7), 57.16 (16.16)	Low
Lin (2022)	China	Cross‐sectional study	Patients aged ≥18 years and ICU stay for >24 h	Self‐reported by patients	210/90	Not reported	Moderate
Ning (2024)	China	Cross‐sectional study	Age ≥18 years, patients who underwent orthopaedic spinal surgery	Measured using VAS	277/26	Not reported	Low
Puntillo (2014)	USA	Cross‐sectional study	Patients ≥18 years and ICU stay for >24 h	Assessed based on NRS scale	127/125	54.5 (14), 55.5 (15)	High
Saltnes (2023)	Norway	Prospective cohort study	Patients aged ≥18 years of age ICU stay for >24 h	Self‐reported by patients	225/128	61.7 (7.2), 60.5 (6.8)	Low
Sato (2019)	Japan	Retrospective study	Patients aged ≥18 years RASS of −1 to +1	Measured using thirst persistence score ≥8	65/261	68.7 (4), 68.2 (3.7)	High
Seyhan (2023)	Turkey	Cross‐sectional study	Patients ≥18 years of age RASS of −1 to +1	Self‐reported by patients	177/177	Not reported	High
Stotts (2015)	US	Cross‐sectional study	Patients >18 years of age and ICU stay for >24 h and RASS of −1 to +1	Assessed based on NRS scale 0–10	252/101	55 (14.4), 57.4 (14.5)	Moderate
Walker (2016)	UK	Cross‐sectional study	Patients with non‐obstetric surgery requiring anaesthesia care over 48 h period and within 24 h of surgery	Self‐reported by patients	2776/4358	Not reported	High
Zeng (2024)	China	Cross‐sectional study	Patients in PACU recovering from general anaesthesia without history of oral or head and neck radiotherapy	Measured using VAS ranged from 0 to 10	1371/1324	51.5 (16), 52.1 (15.5)	Low
Zhang (2022)	China	RCT	≥18 years of age and ICU stay for >24 h	Assessed based on NRS scale ranged from 0 to 10	31/30	63.1 (17), 65.2 (15.9)	High

APACHE‐II, Acute Physiology and Chronic Health Evaluation II; BUN, blood urea nitrogen; HF, heart failure; ICU, intensive care unit; NRS, Numerical Rating Scale; PACU, post‐anaesthesia care unit; RASS, Richmond Agitation–Sedation Scale; RCT, randomised controlled trial; VAS, visual analog scale.

Sex and thirst

Twelve studies encompassing 9380 patients were pooled using DerSimonian–Laird random‐effects inverse‐variance model to assess the association between sex and thirst. The overall OR was 1.04 (95% CI: 0.95–1.14; z = 0.80; P = 0.424), indicating no significant association between the two sex groups (Fig. 2). Between‐study heterogeneity was negligible (Cochran Q = 5.45, df = 11, P = 0.907; I ² = 0.0%). Publication bias assessment revealed a symmetrical plot (Fig. S1) with an Egger test P value of 0.657. Sensitivity analysis (Fig. S2) did not reveal any single study effects.

Figure 2.

Forest plot showing the association between sex and thirst. Weights are from a random‐effects model.

Mechanical ventilation and thirst

Six studies including 2354 patients were pooled to assess the association between mechanical ventilation and thirst. The overall OR was 1.36 (95% CI: 0.91–2.05; z = 1.49, P = 0.137), indicating no statistically significant association between mechanical ventilation and thirst (Fig. 3). Moderate heterogeneity was observed (Cochran Q = 13.44 (df = 5, P = 0.020); I ² = 62.8%), suggesting variability in effect estimates among studies. Sensitivity analysis (Fig. S3) did not reveal any single study effects.

Figure 3.

Forest plot showing the association between mechanical ventilation and thirst. Weights are from a random‐effects model.

Anaesthetic technique and thirst

Five studies were pooled to assess the association between anaesthetic technique and thirst, yielding a significant overall effect of 1.43 (95% CI: 0.97–1.89; z = 6.07, P < 0.001) (Fig. 4). Between‐study heterogeneity was moderate, with Cochran Q = 9.33 (df = 4, P = 0.053) and I ² = 57.1%, indicating some variability in effect sizes among studies. Sensitivity analysis (Fig. S4) did not reveal any single study effects.

Figure 4.

Forest plot showing the association between anaesthetic technique and thirst. Weights are from a random‐effects model.

APACHE‐II score and thirst

Six studies (n = 1342) were pooled using a DerSimonian–Laird random‐effects inverse‐variance model to assess the difference in APACHE‐II scores between patients with and without significant thirst. The overall WMD was 0.24 points (95% CI: –0.55 to 1.04; z = 0.60, P = 0.546), indicating no significant association (Fig. 5). Heterogeneity was low (Cochran = 6.76 (df = 5, P = 0.239); I ² = 26.1%), suggesting consistency of findings among studies. Sensitivity analysis (Fig. S5) did not reveal any single study effects.

Figure 5.

Forest plot showing the association between APACHE‐II and thirst. Weights are from a random‐effects model. APACHE‐II, Acute Physiology and Chronic Health Evaluation.

BUN and thirst

Five studies comprising 1041 patients were pooled to examine the association between BUN levels and reported thirst. The overall SMD was 0.27 (95% CI: −0.11 to 0.65; z = 1.42, P = 0.157), indicating a small, non‐significant trend toward higher BUN in patients experiencing thirst (Fig. 6). However, there was substantial between‐study heterogeneity (Cochran Q = 25.33 (df = 4, P < 0.001); I ² = 84.2%), suggesting marked variability in effect estimates among the included studies. Sensitivity analysis (Fig. S6) did not reveal any single study effects.

Figure 6.

Forest plot showing the association between blood urea nitrogen and thirst. Weights are from a random‐effects model. CI, confidence interval; SMD, standardised mean difference.

Osmolality and thirst

Five studies including 1281 patients were pooled using a DerSimonian–Laird random‐effects inverse‐variance model to compare serum osmolality between patients with and without thirst. Thirsting patients had a significantly higher mean osmolality by 3.04 mOsm/kg (95% CI: 0.15–5.93; z = 2.06, P = 0.039) (Fig. 7). However, between‐study heterogeneity was substantial (Cochran Q = 27.51 (df = 4, P < 0.001); I ² = 85.5%), indicating considerable variability among studies. Sensitivity analysis (Fig. S7) revealed that exclusion of any studies except Puntillo et al. ³¹ changes the association from a significant to a non‐significant association.

Figure 7.

Forest plot showing the association between osmolality and thirst. Weights are from a random‐effects model. CI, confidence interval; WMD, weighted mean difference.

Discussion

In this comprehensive meta‐analysis of critically ill patients, we evaluated six domains such as sex, mechanical ventilation, anaesthetic technique, illness severity (APACHE‐II), BUN and serum osmolality as predictors of patient‐reported thirst. Among 15 studies, nearly 14 000 participants were included. Sex did not emerge as a significant risk factor; pooled estimates showed virtually identical odds of thirst in men and women, and between‐study heterogeneity was negligible. Similarly, mechanical ventilation, a factor long assumed to promote oropharyngeal dryness, was not significantly associated with higher thirst, although moderate variability existed among studies. The choice of anaesthetic technique demonstrated no relationship with thirst, and this analysis was marked by moderate heterogeneity. Illness severity, as measured by APACHE‐II scores in six trials, bore no consistent association with thirst intensity, nor did BUN levels in five studies despite the theoretical link between azotaemia and osmotic drive. By far the strongest and only statistically robust association was observed for serum osmolality: patients reporting thirst had, on average, a modest but significant elevation in serum osmolality compared with those without thirst. Taken together, these findings suggest that, among the factors evaluated, biochemical cues of hyperosmolarity and specific pharmacologic interventions hold greater predictive value for thirst in high‐acuity settings than do demographic characteristics or global illness scores. ³⁸

Our results both confirm and nuance earlier observations in the critical‐care literature. ¹ Prior single‐centre and observational studies have variably reported higher thirst intensity among women, theorising sex‐linked differences in osmoreceptor sensitivity or oral mucosal innervation; our pooled analysis, however, found no meaningful sex disparity. ³⁹ This discrepancy may reflect the broader, more diverse patient populations captured among multiple international cohorts, or it may be attributed to inconsistent adjustment for confounders such as baseline hydration status and comorbidity profiles. Likewise, mechanical ventilation has been widely implicated in oropharyngeal desiccation and consequent thirst through mechanisms of mucosal drying and suppression of normal swallowing reflexes. ⁴⁰ Yet, our meta‐analysis did not corroborate a clear ventilator‐associated thirst signal, suggesting that modern humidification technologies and routine oropharyngeal care protocols may mitigate what was once a pervasive discomfort.

Older investigations into opioid‐induced xerostomia, for example, reported transient reductions in unstimulated salivary flow and subjective dry‐mouth scores, findings that resonate with our aggregate estimate of heightened thirst odds under specific anaesthetic regimens. ⁴⁰ Our work extends these observations by quantifying the effect size among varied ICU populations, although the moderate heterogeneity underscores the influence of procedural complexity, adjunctive medications and perioperative fluid strategies.

The absence of an association between APACHE‐II score and thirst diverges from a limited body of research that has hinted at greater discomfort among more severely ill patients, perhaps due to higher doses of vasopressors or sedatives. Our standardised, multi‐study approach suggests that global severity indices alone lack the granularity to capture symptom‐specific distress and that patient‐level factors such as individual neuroendocrine sensitivity and direct mucosal insult may play more central roles. ⁴¹

Finally, our finding that elevated serum osmolality significantly predicts thirst consolidates long‐standing physiologic principles and empirical observations. Classical human‐laboratory studies have shown that even small increases in plasma osmolality robustly trigger thirst via osmoreceptor activation in the hypothalamus. Smaller clinical cohorts have reported positive correlations between measured plasma osmolality and subjective thirst scores in ICU patients, but these studies were underpowered and lacked uniform measurement timing. ⁶ By pooling nearly 1300 patients, our analysis provides stronger quantitative support for osmolality as a key driver of thirst in critical‐care contexts, and it suggests a potential threshold above which patients may experience clinically meaningful thirst that warrants pre‐emptive management.

The divergent associations observed among risk domains reflect the multifactorial nature of thirst regulation in high‐acuity settings. The robust link between serum osmolality and thirst aligns with classical osmoreceptor physiology: even modest elevations in plasma tonicity trigger hypothalamic osmoreceptors, which in turn drive vasopressin release and conscious thirst sensation. ⁴² In critical illness, homeostatic control of water balance is often impaired by factors such as vasopressor‐mediated renal hypoperfusion, insensible fluid losses, and inflammatory‐induced capillary leak; these perturbations can elevate osmolality beyond the normal compensatory range, eliciting stronger thirst signals than would be predicted by fluid balance alone. ⁴³ Conversely, the absence of a clear relationship between BUN and thirst may indicate that urea concentration, while an important marker of azotaemia, does not independently activate osmoregulatory pathways with the same sensitivity as sodium and other osmolytes. ⁴⁴ Many of the included studies measured BUN in isolation, without concurrent assessment of total serum osmolality, limiting the ability to disentangle uremic effects from broader osmotic stimuli. ²⁸ , ²⁹

Although, intuitively, mechanical ventilation would exacerbate mucosal dryness through bypassed oral humidification and endotracheal tube–mediated airflow disruption, modern ventilator humidifiers and routine oropharyngeal moisturising protocols appear to have mitigated this effect, as evidenced by the non‐significant pooled association. Similarly, sex differences in thirst have been postulated on the basis of sex hormone–mediated variations in osmoreceptor sensitivity, yet our aggregate data suggest that any such differences are clinically negligible when averaged among diverse ICU populations.

The lack of association between APACHE‐II score and thirst points to a disconnect between global illness severity and symptom‐specific distress. APACHE‐II captures acute physiologic aberrations and chronic comorbidity burden but does not account for steroid or diuretic use, the specific fluid‐management strategies or individual neuroendocrine responsiveness that more directly influence thirst perception. This disparity underscores the limitation of using broad severity indices as proxies for symptom risk and highlights the need for targeted physiologic or pharmacologic markers when predicting symptom burden. ⁴⁵

This meta‐analysis has several key strengths. We adhered to a PRISMA‐compliant protocol, applied an exhaustive, multi‐database search strategy without language restrictions and included both ICU and perioperative cohorts to maximise generalisability. Our dual‐reviewer data extraction with adjudication minimised bias, and the comprehensive risk‐of‐bias assessments informed sensitivity analyses. We also formally tested for publication bias among adequately powered domains, wherever possible. However, certain limitations warrant consideration. First, the majority of included studies were observational, with retrospective designs that are subject to residual confounding and variable adjustment for key covariates. Second, outcome measures varied among studies: thirst was assessed using different scales and at varying time points, introducing measurement heterogeneity. Importantly, many observational studies did not report sufficiently granular timing to ensure close temporal alignment between exposure measurement and thirst assessment (e.g. the interval between laboratory measures such as osmolality/BUN and the thirst rating, or the time elapsed between extubation and thirst evaluation), raising the possibility of non‐differential exposure misclassification that would tend to bias associations toward the null. Third, thirst is an inherently subjective outcome and may be susceptible to measurement bias, including framing/leading effects and variable assessor scripts; while several studies used standardised numeric rating/VAS prompts for symptom self‐report, heterogeneity in phrasing and recall context remains unavoidable and may contribute to between‐study variability.

Fourth, the review protocol was not prospectively registered (e.g. PROSPERO) because registration could not be completed at the time of initiation. Although we followed an a priori protocol with pre‐specified methods before screening and extraction, the absence of prospective registration may limit external verification of protocol adherence. Fifth, Although we included both ICU and peri‐operative cohorts to enhance clinical breadth, these populations can differ meaningfully in baseline thirst risk and in upstream determinants (e.g. elective fasting practices, peri‐anaesthetic drugs, postoperative timing of assessment vs prolonged critical illness with complex fluid/renal trajectories). Because many predictors were reported in only a small number of studies within each setting, we were not consistently able to generate robust setting‐restricted pooled estimates (ICU only vs peri‐operative only) for all predictors; thus, our pooled effects should be interpreted as overall associations among high‐acuity settings, and future primary studies and syntheses should support stratified estimates using standardised thirst definitions and timepoints. Sixth, the predictors assessed in this review should be interpreted as a non‐exhaustive set of clinically plausible factors that were also sufficiently and consistently reported for extraction among the included studies. Other potentially relevant predictors have been reported in the wider literature but could not be evaluated here due to absent or inconsistent reporting in our included studies (e.g. American Society of Anesthesiologists physical status classification, which has been associated with thirst in latest similar meta‐analysis). ⁴⁶

Additionally, the included evidence was overwhelmingly observational (14 of 15 studies), which limits causal inference and leaves open the possibility of residual confounding and reverse causation. Only one randomised controlled trial contributed to this review (n = 61; ~0.5% of the pooled sample), and its small size limits the precision and generalisability of intervention effects; therefore, our pooled associations should be interpreted primarily as hypothesis‐generating and supportive of the need for adequately powered trials. Finally, respiratory support exposure was not uniformly characterised a studies. Although some studies distinguished IMV from NIV or described specific oxygen delivery approaches, most studies reported respiratory support in broader terms (e.g. ‘mechanical ventilation/intubation’ or ‘oxygen therapy’), limiting our ability to conduct robust stratified analyses by device type (e.g. HFNC vs conventional oxygen, humidified vs dry gas) or by duration of ventilatory support.

Our findings have direct relevance for patient comfort and safety protocols in critical‐care settings. Measurement of serum osmolality at the time of routine laboratory draws offers a simple, objective indicator of thirst risk; a threshold‐based alert, akin to hypernatremia notifications, could prompt proactive oral‐care interventions such as chilled swabs, oral moisturisers and judicious fluid administration when clinically permissible. Given the non‐significant association with mechanical ventilation, current humidification practices appear largely effective, but regular auditing of humidifier performance and oropharyngeal moisturising schedules remains important.

Moreover, the disconnect between APACHE‐II scores and thirst underscores that global severity indices are insufficient for symptom prediction; instead, multidisciplinary rounds should include standardised thirst assessments, ideally using a brief, validated scale to identify and address thirst as part of comprehensive symptom management. Integrating thirst evaluation into nursing flow sheets and electronic medical records will elevate its visibility alongside pain and delirium, reinforcing the notion that thirst is a meaningful patient‐reported outcome rather than a trivial discomfort.

Further investigation is needed to refine our understanding of thirst pathophysiology and management. Prospective cohort studies and clinical trials should standardise thirst measurement tools and harmonise timing of assessments to reduce measurement heterogeneity. Mechanistic studies employing continuous osmolality monitoring could clarify dynamic relationships between fluid shifts, vasopressor use and thirst onset. Randomised trials comparing oral‐care bundles (e.g. swab protocols, gel applications) against usual care in patients stratified by osmolality could establish efficacy thresholds and optimise resource allocation. Additionally, patient‐level meta‐analyses incorporating granular covariates such as fluid balance charts, medication dosing and detailed ventilator settings would allow more modelling of combined risk factors. Finally, exploration of novel humidification technologies and salivary gland–protective pharmacotherapies may yield targeted interventions to pre‐empt or ameliorate thirst without compromising other aspects of critical care.

Conclusions

This review identifies serum osmolality as the most reliable predictors of thirst among critically ill patients, while demographic factors, mechanical ventilation, illness‐severity scores, anaesthetic technique and BUN levels appear less informative. These findings emphasise the primacy of biochemical homeostasis and pharmacologic influences in driving thirst, and they call for integration of osmolality‐guided protocols and standardised thirst assessments into routine care. Addressing thirst proactively through objective monitoring and evidence‐based interventions will enhance patient comfort, potentially reduce delirium and agitation and foster a more patient‐centred approach in high‐acuity environments.

Supporting information

Figure S1. Funnel plot for the association between sex and thirst.

Figure S2. Sensitivity analysis for the association between sex and thirst.

Figure S3. Sensitivity analysis for the association between mechanical ventilation and thirst.

Figure S4. Sensitivity analysis for the association between anaesthetic technique and thirst.

Figure S5. Sensitivity analysis for the association between APACHE‐II and thirst.

Figure S6. Sensitivity analysis for the association between blood urea nitrogen and thirst.

Figure S7. Sensitivity analysis for the association between osmolality and thirst.

Table S1. Study‐level operational definitions and available distributions of key predictors and clinical context variables (all included studies).

Appendix S1. Supporting Information.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.