Distinct Health‐Related Quality of Life Recovery Trajectories Among ICU Survivors: A Multicentre Prospective Cohort Study Using the EQ‐5D‐5L Authors: Jiyeon Kang, Yujin Jeong Published: 2026 ABSTRACT ABSTRACTBackgroundICU survivors frequently experience reduced health‐related quality of life (HRQoL); however, recovery trajectories vary considerably, underscoring the need to identify distinct subgroups for targeted follow‐up.AimsTo identify distinct HRQoL trajectory groups among ICU survivors, compare EQ‐5D‐5L profiles across groups and explore associated factors.Study DesignThis multicentre prospective study included 19 ICUs in four South Korean university hospitals. Adult survivors with admissions ≥ 24 h were assessed using the EQ‐5D‐5L over 24 months post‐discharge. Trajectory groups were identified via latent‐class mixed modelling, and associated factors via multinomial logistic regression.ResultsAmong 609 survivors, three groups emerged: normative stable (58.3%), suboptimal (31.7%) and persistently low (10.0%). The normative stable group maintained HRQoL near the Korean population norm. The suboptimal group showed persistently reduced HRQoL, with moderate‐to‐extreme problems in mobility and pain/discomfort at 24 months. The persistently low group exhibited profound, sustained impairment, particularly in mobility, usual activities and pain/discomfort. Older age (aOR 1.72, 95% CI 1.07–2.75), lower educational attainment (aOR 2.11, 95% CI 1.34–3.32) and ICU delirium (aOR 2.14, 95% CI 1.10–4.15) were associated with suboptimal membership, whereas surgical admission showed lower odds (aOR 0.49, 95% CI 0.26–0.94). Multimorbidity (aOR 3.36, 95% CI 1.69–6.69) and discharge to an extended care facility (aOR 4.97, 95% CI 2.40–10.27) were associated with persistently low membership.ConclusionsThree distinct HRQoL recovery trajectories emerged over 24 months post‐discharge. Although most survivors maintained near‐normal HRQoL, unfavourable groups experienced sustained impairment in mobility and pain/discomfort. These findings support trajectory‐based approaches for identifying high‐risk phenotypes and guiding individualised post‐ICU care.Relevance to Clinical PracticeThe persistent mobility and pain/discomfort problems highlight the importance of repeated assessment during post‐ICU follow‐up. Critical care nurses may consider them alongside ICU delirium history, multimorbidity and discharge destination; however, prospective validation is required before these trajectories or factors can identify high‐risk survivors.Trial RegistrationThis study was prospectively registered with the Korean Clinical Research Information Service (KCT0004045). Registered 11 June 2019. First recruitment 11 June 2019Keywords: cohort study, critical care, mobility limitation, quality of life, survivorship Introduction Intensive care unit (ICU) survivors frequently experience persistent physical, cognitive and psychological impairments after hospital discharge, substantially impairing their health‐related quality of life (HRQoL) [1]. A systematic review reported that although HRQoL tends to improve during the first year after ICU discharge, it generally remains lower than population norms [2]. For many survivors, recovery is incomplete, and HRQoL may not return to pre‐illness levels [3], with impairments persisting for 5–10 years following critical illness [4, 5]. These long‐term sequelae highlight the importance of patient‐centred outcomes and survivorship research in critical care populations. Understanding longitudinal HRQoL trajectories among ICU survivors may help identify recovery phenotypes associated with poor outcomes and facilitate risk stratification, targeted interventions and tailored follow‐up care. Previous studies have demonstrated substantial heterogeneity in post‐ICU recovery patterns. Soliman et al. [6] reported that HRQoL outcomes at 1 year differed according to diagnostic categories, whereas a more recent study showed that even within the same diagnostic group, some survivors improved while others deteriorated over time [7]. Similarly, studies of sepsis survivors identified distinct functional recovery trajectories, including groups with complete recovery and persistent functional decline [8]. Among COVID‐19 survivors who underwent mechanical ventilation, HRQoL recovery was also classified into multiple distinct profiles [9]. Collectively, these findings suggest that recovery after critical illness is heterogeneous and cannot be adequately explained by a single recovery pathway, and that a trajectory‐based approach may provide a more nuanced understanding of post‐ICU recovery. Because post‐ICU recovery is multidimensional, often involving concurrent changes across physical, psychological and social domains, HRQoL offers a global patient‐reported measure of overall recovery that complements assessments of individual domains [1]. The EuroQol 5‐dimension 5‐level (EQ‐5D‐5L) assesses HRQoL across five dimensions: mobility, self‐care, usual activities, pain/discomfort and anxiety/depression. Responses can be converted into a utility‐based index score, enabling quantitative comparisons over time and facilitating longitudinal analyses [10, 11]. Its brevity and utility‐based scoring make it particularly suitable for longitudinal follow‐up in critically ill populations. The EQ‐5D‐5L was recently recommended as a core outcome measure for HRQoL assessment in ICU survivors in an international consensus study [12] and is increasingly used in post‐ICU survivorship research. Despite growing interest in post‐ICU recovery, most EQ‐5D studies have focused on mean HRQoL at specific time points, potentially obscuring heterogeneity in longitudinal recovery patterns. Prior studies suggest that recovery trajectories differ not only between individuals but also across HRQoL dimensions, with physical domains improving more consistently than psychological ones [13, 14]. However, studies examining longitudinal HRQoL trajectory groups among ICU survivors and characterising dimension‐specific impairment patterns using the EQ‐5D‐5L remain limited, and little is known about factors associated with unfavourable recovery phenotypes. Trajectory modelling enables identification of clinically meaningful subgroups with distinct longitudinal recovery patterns that may not be apparent in conventional analyses based on group averages, potentially facilitating early identification of high‐risk survivors and supporting individualised follow‐up strategies after critical illness. Aims This study aimed to identify distinct HRQoL trajectory groups among ICU survivors, compare EQ‐5D‐5L dimensional profiles across trajectory groups and examine factors associated with each trajectory. Design and Methods Design This multicentre prospective cohort study investigated longitudinal HRQoL trajectories among ICU survivors and was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline. Setting and Sample This study was conducted between June 2019 and July 2022 across 19 ICUs in four university hospitals located in Busan, South Korea. The participating hospitals had between 812 and 1205 beds, including 56 to 99 adult ICU beds, and nurse‐to‐patient staffing ratios ranging from 1:1 to 1:4. Eligible participants were adults aged ≥ 18 years who had been admitted to the ICU for at least 24 h. Patients were excluded if they (1) had been admitted to an ICU within the previous year, (2) had been diagnosed with a physical disability or mental health problem prior to ICU admission or (3) were unable to complete the survey. Of the 891 patients enrolled during ICU admission, 609 completed at least one EQ‐5D‐5L assessment at 3, 6, 12 or 24 months after discharge and were included in the analysis. The required sample size was estimated for logistic regression analysis using G*Power version 3.1.9.7, as no universally accepted formula exists for sample size estimation in latent‐class mixed modelling (LCMM). Assuming a significance level α = 0.05, 1 − β = 0.80, OR = 2.0, event rate = 0.20, R 2 other X = 0.20 and π = 0.50, the minimum required sample size was 434. The final sample of 609 participants exceeded this requirement. Measurements Participant Characteristics The following demographic and clinical characteristics were collected: sex, age, education, employment status, household income, caregiver status, body mass index (BMI), alcohol consumption, smoking status, Functional Comorbidity Index (FCI), reason for ICU admission, route of ICU admission, illness severity, surgery, mechanical ventilation, cardiopulmonary resuscitation, medication, organ support therapies, delirium, ICU length of stay and discharge destination. Comorbidity was assessed using the Functional Comorbidity Index (FCI), an 18‐item index developed to predict physical function rather than mortality, in which each condition present contributes one point to a total score. A higher score indicates a greater comorbidity burden; in this study, an FCI score of ≥ 2, indicating the presence of two or more of the included conditions, was used to denote a higher comorbidity burden. Health‐Related Quality of Life HRQoL was assessed using the EQ‐5D‐5L instrument. The EQ‐5D‐5L evaluates health status across five dimensions: mobility, self‐care, usual activities, pain/discomfort and anxiety/depression. Each dimension is rated on a five‐level scale ranging from no problems (Level 1) to extreme problems (Level 5). The EQ‐5D‐5L results can be expressed both as a descriptive health profile and as a single utility index score. The utility index score is derived using country‐specific value sets that reflect population preferences for health states. Scores generally range from 1 (perfect health) to 0 (death‐equivalent health state), with negative values indicating health states perceived as worse than death [11, 15]. In this study, EQ‐5D‐5L index scores were calculated using the Korean value set [16]. HRQoL outcomes were interpreted in reference to the Korean population norm for individuals aged 55–64 years (EQ‐5D‐5L index score: 0.909) [17]. This age group was selected as the reference because it corresponds most closely to the mean age of the study sample (60.0 years). Procedures Participants were recruited through notices posted in the participating ICUs and study information leaflets made available to patients and caregivers at the nursing stations. Because recruitment relied on posted notices and information leaflets, screening was conducted among patients or caregivers who expressed interest in participating, rather than through consecutive screening of all ICU admissions. Written informed consent was obtained before enrolment from participants or, when they were unable to provide consent during their ICU stay, from their legal representatives. Demographic and clinical data were extracted from electronic medical records. HRQoL assessments were conducted at 3, 6, 12 and 24 months after hospital discharge. The study was originally designed to conduct standardised face‐to‐face interviews during scheduled outpatient follow‐up visits. Prior to the COVID‐19 pandemic, assessments were performed through face‐to‐face interviews during outpatient follow‐up visits. Following the onset of the pandemic in March 2020, follow‐up assessments were conducted via telephone interviews to minimise the risk of infection. A previous study demonstrated acceptable agreement between telephone‐based and face‐to‐face administration of the EQ‐5D, supporting the reliability of telephone‐based assessments [18]. Data Analysis Statistical analyses were performed using IBM SPSS Statistics version 28.0 and the lcmm package in R version 4.5.0. Participant characteristics were summarised using descriptive statistics, including frequencies, percentages, means and standard deviations. Because different illness severity scoring systems were used across participating hospitals, APACHE II and SAPS 3 scores were standardised (z‐scores) before inclusion in the analyses. To identify distinct HRQoL recovery trajectories among ICU survivors, LCMMs were fitted with a linear function of time using EQ‐5D‐5L index scores as the dependent variable. Models with two to four latent classes were compared using the Akaike information criterion (AIC), Bayesian information criterion (BIC), entropy, average posterior probabilities, solution stability and clinical interpretability, with the final model selected based on these criteria collectively (Table S1) [19, 20]. To complement the trajectory analysis, longitudinal changes across trajectory groups were examined using repeated‐measures ANOVA. Changes from 3 months were also classified using an MCID of 0.08 [6] as improved, no clinically important change, or deteriorated and displayed in Sankey diagrams overall and by trajectory group. For dimensional analyses, responses in each EQ‐5D‐5L dimension were dichotomised into no/slight problems (Levels 1–2) and moderate‐to‐extreme problems (Levels 3–5). Dimensional profiles were presented as the proportions of participants reporting moderate‐to‐extreme problems at each follow‐up time point. Two alternative thresholds—Level 1 versus Levels 2–5 and Levels 1–3 versus Levels 4–5—were also examined. Differences in participant characteristics and EQ‐5D‐5L dimensional profiles across trajectory groups were analysed using one‐way ANOVA or chi‐square tests. Multinomial logistic regression analysis was performed to identify factors associated with trajectory group membership. Additionally, binary logistic regression was used to identify factors associated with non‐normative HRQoL trajectory membership, defined as membership in either the suboptimal or persistently low group, with the normative stable group as the reference. Baseline characteristics were compared according to EQ‐5D‐5L assessment availability at each follow‐up time point (Table S2). Missing data were handled using multiple imputation with chained equations, generating 50 datasets under the missing‐at‐random assumption. Age, sex, Functional Comorbidity Index and standardised illness severity score were included in the imputation model. To assess the potential influence of imputation, observed and imputed EQ‐5D‐5L index and dimension scores were compared using absolute mean differences, with Cohen's d additionally calculated for index scores (Tables S3 and S4). A two‐sided p‐value < 0.05 was considered statistically significant. Ethical Considerations This study was approved by the Institutional Review Board of Dong‐A University (approval date: 15 May 2019; IRB No. 2‐104 709‐AB‐N‐01‐201 904‐HR‐021‐14). Written informed consent was obtained prior to enrolment. All study procedures were conducted in accordance with the principles of the Declaration of Helsinki and relevant institutional regulations. Clinical Trial Registration This study was prospectively registered with the Korean Clinical Research Information Service (KCT0004045). Registered 11 June 2019. First recruitment 11 June 2019. Results Participant Characteristics A total of 609 ICU survivors who responded to the EQ‐5D‐5L assessment at least once were included in the study (Figure S1). The mean age was 60.0 ± 13.3 years, and 59.6% were male. More than half of the participants were unemployed (53.4%), 64.4% had at least a high school education, and 69.8% reported a monthly household income of <. Family caregivers were present for 84.4% of participants. Approximately one‐quarter of the participants (24.3%) had an FCI score ≥ 2. The most common reasons for ICU admission were surgical conditions (28.7%) and neurological diseases (27.3%). Approximately half of the participants (52.4%) were admitted through the emergency department. Mechanical ventilation was administered in 18.9% of participants, and organ support therapies were used in 5.6%. The mean ICU length of stay was 4.62 ± 7.78 days, and most participants (83.4%) were discharged home (Table 1). TABLE 1Characteristics of the participants (N = 609).CharacteristicCategory n %Mean ± SDMedian (IQR)Age (year)< 6026643.760.02 ± 13.2861.0 (18.0)≥ 6034356.3SexMale36359.6Female24640.4EmploymentYes28346.5No32553.4Education< High school21435.1≥ High school39264.4Monthly household income (million KRW)< 342569.8≥ 318129.7Family caregiverYes51484.4No9515.6BMI< 2538663.424.17 ± 4.2423.82 (4.42)≥ 2522336.6Current alcohol useYes23839.1No37160.9SmokingYes17328.4No43671.6Functional comorbidity index< 246175.70.92 ± 1.02≥ 214824.3Reasons for ICU admissionMedical10417.1Cardiac11619.0Neurological16627.3Surgical17528.7Trauma487.9ICU admission routeEmergency department31952.4Others29047.6Illness severity at admissionAPACHE II38111.42 ± 6.2410.0 (8.0)SAPS 316633.75 ± 15.7833.5 (21.0)SedativesYes18630.5No42369.5AnalgesicsYes43070.6No17929.4Organ support therapiesYes345.6No57594.4Mechanical ventilationYes11518.9No49481.1CPRYes122.0No59798.0SurgeryYes33254.5No27745.5Delirium in the ICUYes8313.6No52686.4ICU days< 751384.24.62 ± 7.783.0 (2.0)≥ 79615.8Discharge destinationHome50883.4Extended care facilities10116.6 Note: Missing data: Employment, n = 1; Education, n = 3; Monthly household income, n = 3; Illness severity at admission, n = 62. Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index.Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ICU, intensive care unit. Characteristics of the participants (N = 609). Note: Missing data: Employment, n = 1; Education, n = 3; Monthly household income, n = 3; Illness severity at admission, n = 62. Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index. Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ICU, intensive care unit. Longitudinal Trajectories of HRQoL The mean EQ‐5D‐5L index scores for the overall cohort were 0.80 ± 0.18 at 3 months, 0.82 ± 0.17 at 6 months, 0.81 ± 0.18 at 12 months and 0.83 ± 0.16 at 24 months after hospital discharge. LCMM analysis identified three distinct HRQoL trajectory groups over the 24‐month follow‐up period. The normative stable, suboptimal and persistently low groups comprised 58.3% (n = 355), 31.7% (n = 193) and 10.0% (n = 61) of participants, respectively. The normative stable group demonstrated consistently high HRQoL scores that remained close to the Korean population norm (0.909) throughout follow‐up. Mean EQ‐5D‐5L index scores in this group ranged from 0.90 to 0.91 across all time points. The suboptimal group showed persistently reduced HRQoL, with mean EQ‐5D‐5L index scores ranging from 0.74 to 0.78, consistently below the population norm throughout follow‐up. The persistently low group exhibited markedly impaired HRQoL over the entire follow‐up period. Mean EQ‐5D‐5L index scores ranged from 0.42 to 0.52 and showed only minimal recovery over 24 months (Figure 1). EQ‐5D‐5L index scores differed significantly across trajectory groups and over time, with a significant time‐by‐group interaction (Table S5). Although the group‐level trajectories were relatively stable in their ordering, individual‐level change from the 3‐month score was considerable: across the cohort, 32.0% improved, 21.8% deteriorated and 46.1% showed no clinically important change at 6 months, with broadly similar distributions thereafter (Figure S2). Change patterns differed by group. The normative stable group was the most stable, with roughly half showing no clinically important change at each time point. In contrast, the persistently low group showed the highest proportion of improvement (57.4% at both 6 and 24 months), most plausibly reflecting partial recovery from a low 3‐month baseline (a floor effect) rather than a shift out of the impaired range, as this group remained the most impaired throughout. Longitudinal trajectories of EQ‐5D‐5L index scores over 24 months after hospital discharge among ICU survivors (N = 609). (A) Mean EQ‐5D‐5L index scores of the overall cohort at 3, 6, 12 and 24 months after discharge. (B) Predicted longitudinal trajectories derived from latent‐class mixed modelling. The dashed line indicates the Korean population norm value (0.909). (C) Longitudinal changes in EQ‐5D‐5L index scores according to trajectory group. Although the two‐class model had lower AIC and BIC values, the three‐class model had higher entropy, acceptable average posterior probabilities and a stable solution across random starts, whereas the four‐class model showed substantially poorer classification quality. Considering these findings collectively, the three‐class model was selected as the final solution (Table S1). EQ‐5D‐5L Dimensional Health Profiles According to Trajectory Groups Figure 2 presents the proportions of participants reporting moderate‐to‐extreme problems (Levels 3–5) across the five EQ‐5D‐5L dimensions during follow‐up. The normative stable group consistently demonstrated low proportions of moderate‐to‐extreme problems across all dimensions throughout the study period. Although pain/discomfort was the most frequently reported problem within this group, the proportion remained relatively low (5.4%–12.7%). Proportions of participants reporting moderate‐to‐extreme problems (Levels 3–5) across EQ‐5D‐5L dimensions at 3, 6, 12 and 24 months after discharge by trajectory group. All between‐group comparisons at each time point were statistically significant (chi‐square test, all p < 0.001). In the suboptimal group, problems related to mobility and pain/discomfort persisted throughout follow‐up. Moderate‐to‐extreme mobility problems were reported by 23.3%–28.0% of participants across all time points. Pain/discomfort peaked at 50.8% at 12 months before declining to 28.5% at 24 months. Limitations in usual activities remained above 20% between 3 and 12 months and decreased to 14.0% at 24 months. Anxiety/depression gradually improved over time, from 24.4% at 6 months to 7.3% at 24 months, whereas self‐care problems remained relatively uncommon (below 10% at all time points). The persistently low group demonstrated the highest burden of impairment across all EQ‐5D‐5L dimensions throughout follow‐up. Notably, substantial proportions of participants continued to report moderate‐to‐extreme problems at 24 months in mobility (88.5%), usual activities (73.8%) and pain/discomfort (77.0%). Although self‐care and anxiety/depression problems showed some improvement over time, impairment levels remained considerably higher than those observed in the other trajectory groups. The proportions of moderate‐to‐extreme problems across all five EQ‐5D‐5L dimensions differed significantly among the three trajectory groups at each follow‐up time point (Table S6). The corresponding profiles using the two alternative thresholds are presented in Tables S7 and S8. Comparison of Participant Characteristics Across Trajectory Groups Significant differences were observed among the three trajectory groups with respect to age, employment status, education level, household income, alcohol consumption, FCI score, reason for ICU admission, sedative use, organ support therapies, mechanical ventilation, ICU delirium, ICU length of stay and discharge destination (Table 2). TABLE 2Characteristics of the HRQoL trajectory groups.VariableCategoryNormative stableSuboptimalPersistently lowχ2 (N = 609) p n (%) n (%) n (%)Age (year)< 60184 (51.8)60 (31.1)22 (36.1)23.465< 0.001≥ 60171 (48.2)133 (68.9)39 (63.9)SexMale225 (63.4)105 (54.4)33 (54.1)5.0380.081Female130 (36.6)88 (45.6)28 (45.9)EmploymentYes185 (52.1)79 (40.9)19 (31.1)12.9030.002No169 (47.6)114 (59.1)42 (68.9)Education< High school93 (26.2)95 (49.2)26 (42.6)30.463< 0.001≥ High school260 (73.2)98 (50.8)34 (55.7)Monthly household income (million KRW)< 3232 (65.4)143 (74.1)50 (82.0)9.6940.008≥ 3122 (34.4)48 (24.9)11 (18.0)Family caregiverYes302 (85.1)161 (83.4)51 (83.6)0.2910.864No53 (14.9)32 (16.6)10 (16.4)BMI< 25225 (63.4)122 (63.2)39 (63.9)0.0100.995≥ 25130 (36.6)71 (36.8)22 (36.1)Current alcohol useYes151 (42.5)71 (36.8)16 (26.2)6.4370.040No204 (57.5)122 (63.2)45 (73.8)SmokingYes105 (29.6)52 (26.9)16 (26.2)0.5850.746No250 (70.4)141 (73.1)45 (73.8)Functional comorbidity index< 2289 (81.4)138 (71.5)34 (55.7)21.350< 0.001≥ 266 (18.6)55 (28.5)27 (44.3)Reasons for ICU admissionMedical52 (14.6)34 (17.6)18 (29.5)19.3960.013Cardiac66 (18.6)40 (20.7)10 (16.4)Neurological98 (27.6)50 (25.9)18 (29.5)Surgical118 (33.2)47 (24.4)10 (16.4)Trauma21 (5.9)22 (11.4)5 (8.2)ICU admission routeEmergency department185 (52.1)94 (48.7)40 (65.6)5.3120.070Others170 (47.9)99 (51.3)21 (34.4)Standardised illness severity score< 0189 (53.2)93 (48.2)29 (47.5)1.1620.446≥ 0166 (46.8)100 (51.8)32 (52.5)SedativesYes100 (28.2)59 (30.6)27 (44.3)6.3550.042No255 (71.8)134 (69.4)34 (55.7)AnalgesicsYes251 (70.7)135 (69.9)44 (72.1)0.1100.946No104 (29.3)58 (30.1)17 (27.9)Organ support therapiesYes17 (4.8)9 (4.7)8 (13.1)7.2990.026No338 (95.2)184 (95.3)53 (86.9)Mechanical ventilationYes58 (16.3)37 (19.2)20 (32.8)9.2100.010No297 (83.7)156 (80.8)41 (67.2)CPRYes9 (2.5)2 (1.0)1 (1.6)1.4930.474No346 (97.5)191 (99.0)60 (98.4)SurgeryYes196 (55.2)103 (53.4)33 (54.1)0.1760.916No159 (44.8)90 (46.6)28 (45.9)Delirium in the ICUYes32 (9.0)33 (17.1)18 (29.5)21.463< 0.001No323 (91.0)160 (82.9)43 (70.5)ICU length of stay (days)< 7306 (86.2)165 (85.5)42 (68.9)12.1290.002≥ 749 (13.8)28 (14.5)19 (31.1)Discharge destinationHome315 (88.7)157 (81.3)36 (59.0)34.101< 0.001Extended facilities40 (11.3)36 (18.7)25 (41.0) Note: Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index.Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ICU, intensive care unit. Characteristics of the HRQoL trajectory groups. Note: Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index. Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ICU, intensive care unit. Compared with the normative stable group, multinomial logistic regression analysis demonstrated that age ≥ 60 years (aOR = 1.72, 95% CI 1.07–2.75), education below high school level (aOR = 2.11, 95% CI 1.34–3.32) and ICU delirium (aOR = 2.14, 95% CI 1.10–4.15) were independently associated with membership in the suboptimal trajectory group. Surgical admission was associated with a lower likelihood of belonging to the suboptimal group compared with medical admission (aOR = 0.49, 95% CI 0.26–0.94). For the persistently low trajectory group, FCI score ≥ 2 (aOR = 3.36, 95% CI 1.69–6.69) and discharge to an extended care facility (ECF; aOR = 4.97, 95% CI 2.40–10.27) were identified as independent associated factors (Table 3). TABLE 3Associated factors of HRQoL trajectory group (ref: Normative stable group).VariableSuboptimalPersistently lowBOR (95% CI) p BOR (95% CI) p Age ≥ 60 years0.541.72 (1.07–2.75)0.0250.341.41 (0.67–2.97)0.368Female sex0.361.43 (0.91–2.25)0.1170.221.24 (0.61–2.54)0.556Unemployed0.081.08 (0.70–1.68)0.7290.491.62 (0.77–3.43)0.202Less than high school graduation0.752.11 (1.34–3.32)0.0010.371.44 (0.70–2.99)0.321Monthly household income < 3 million KRW−0.110.90 (0.56–1.45)0.6660.191.21 (0.52–2.82)0.661No family caregiver0.181.20 (0.71–2.01)0.5000.151.16 (0.51–2.62)0.724BMI ≥ 250.111.11 (0.74–1.68)0.6050.141.15 (0.59–2.25)0.678Current alcohol use0.701.03 (0.65–1.64)0.904−0.700.50 (0.22–1.12)0.090Current smoker0.511.06 (0.64–1.77)0.8190.511.67 (0.73–3.84)0.226FCI ≥ 21.211.38 (0.87–2.19)0.1721.213.36 (1.69–6.69)0.001Reasons for ICU admission (ref: medical)Cardiac0.011.01 (0.51–2.00)0.982−0.580.56 (0.20–1.56)0.267Neurological−0.020.98 (0.51–1.91)0.964−0.120.89 (0.34–2.33)0.806Surgical−0.700.49 (0.26–0.94)0.032−0.890.41 (0.15–1.16)0.094Trauma0.782.17 (0.91–5.17)0.080−0.240.78 (0.21–2.99)0.720ED ICU admission route−0.410.66 (0.42–1.05)0.0800.261.30 (0.61–2.74)0.497Standardised illness severity score ≥ 00.181.20 (0.80–1.79)0.3830.001.00 (0.52–1.92)0.989Use of sedatives−0.010.99 (0.57–1.74)0.9850.151.16 (0.48–2.83)0.740Use of analgesics−0.020.98 (0.61–1.57)0.9390.001.00 (0.47–2.13)0.995Organ support therapies−0.130.88 (0.34–2.26)0.7850.862.36 (0.80–6.99)0.121Mechanical ventilation0.091.09 (0.55–2.16)0.7960.531.69 (0.61–4.70)0.311CPR−1.160.31 (0.06–1.80)0.194−1.120.33 (0.03–3.54)0.358Delirium in the ICU0.762.14 (1.10–4.15)0.0250.521.69 (0.71–4.01)0.238ICU length of stay ≥ 7 days0.321.38 (0.72–2.64)0.3370.151.17 (0.48–2.86)0.729Discharge to extended care facilities0.551.74 (0.99–3.05)0.0531.604.97 (2.40–10.27)< 0.001χ2 (p) = 135.86 (< 0.001), R 2 = 0.241 Note: Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index.Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ED, emergency department; FCI, functional comorbidity index. Associated factors of HRQoL trajectory group (ref: Normative stable group). Note: Functional Comorbidity Index ≥ 2 indicates the presence of two or more comorbid conditions included in the index. Abbreviations: BMI, body mass index; CPR, cardiopulmonary resuscitation; ED, emergency department; FCI, functional comorbidity index. In the binary logistic regression analysis, age ≥ 60 years, education below high school level, an FCI score ≥ 2, ICU delirium and discharge to an extended care facility were independently associated with non‐normative trajectory membership, whereas neurological admission (versus medical admission) was associated with a lower likelihood (Table S9). Discussion In this multicentre prospective cohort study, longitudinal HRQoL recovery among ICU survivors followed three distinct trajectories over 24 months after hospital discharge. These recovery phenotypes differed not only in overall HRQoL levels but also in dimensional burden patterns and associated clinical and sociodemographic factors. Compared with the normative stable group, older age, lower educational attainment and ICU delirium were associated with suboptimal trajectory membership, whereas surgical admission was associated with a lower likelihood of membership in this group. Multimorbidity and discharge to ECFs were associated with persistently low trajectory membership. The normative stable group, comprising more than half of the cohort, maintained HRQoL levels close to the Korean population norm throughout follow‐up with minimal burden across all EQ‐5D‐5L dimensions. Problems related to mobility, self‐care, usual activities and anxiety/depression were uncommon, whereas pain/discomfort, though the most frequently reported problem in this group, remained low throughout follow‐up. Similar findings were reported in the Dutch MONITOR‐IC cohort, in which nearly half of ICU survivors demonstrated favourable long‐term HRQoL profiles with EQ‐5D‐5L index scores of 0.84–0.88 at 1 year after ICU discharge [21], somewhat lower than those in the normative stable group of the present study, possibly reflecting differences in case mix or healthcare context. Collectively, these findings suggest that a substantial proportion of ICU survivors may achieve long‐term recovery approaching general population norms. However, the comparatively large normative stable group should be interpreted in light of the broad eligibility criteria, relatively low rates of mechanical ventilation and organ support, short ICU length of stay and high proportion of home discharge. Together with the substantial proportion of neurological admissions, these characteristics may have contributed to the comparatively large proportion of survivors following the normative stable trajectory, and the resulting trajectory distribution may therefore differ from that of cohorts enriched for severe critical illness such as sepsis or ARDS. At the same time, the broad eligibility criteria enabled the inclusion of a heterogeneous population of general adult ICU survivors encountered in routine clinical practice, supporting the relevance of these findings to unselected ICU populations. The suboptimal trajectory group was characterised by persistently reduced HRQoL, with the group‐mean trajectory showing little change and remaining below the population norm throughout the 24‐month follow‐up. Because pre‐ICU HRQoL was not assessed, we could not determine the extent to which this pattern reflected post‐ICU impairment or pre‐existing low HRQoL. The association of this trajectory with older age, lower educational attainment and ICU delirium suggests that a chronic health burden predating ICU admission may have contributed, although this remains a hypothesis that future studies incorporating baseline HRQoL should test. Factors associated with this trajectory—older age, lower educational attainment and ICU delirium—have also been linked to unfavourable post‐ICU outcomes in previous studies [22, 23, 24, 25], suggesting that suboptimal recovery may reflect the combined influence of ageing, socioeconomic vulnerability and acute neurological stress during critical illness rather than illness severity alone. Notably, illness severity was not independently associated with suboptimal trajectory membership, further supporting this interpretation. Although surgical admission was associated with a lower likelihood of suboptimal group membership, a previous study demonstrated substantial heterogeneity in long‐term outcomes among surgical ICU populations [26], and further studies are needed to clarify trajectory differences across specific surgical subgroups. Among participants in the suboptimal trajectory group, problems related to mobility and pain/discomfort persisted throughout follow‐up, possibly reflecting the interplay between chronic pain burden and impaired physical functioning during recovery. Devine et al. [27] reported that many ICU survivors experience chronic pain interfering with walking, daily activities and overall functioning after discharge. Previous studies have demonstrated that pain interference is associated with impaired physical HRQoL and the pain/discomfort dimension of the EQ‐5D‐5L in ICU survivors [28, 29]. These findings suggest that pain assessment and mobility‐focused rehabilitation may represent important targets for survivors with persistently suboptimal recovery. Although the persistently low trajectory group represented only 10% of the cohort, this subgroup demonstrated profoundly impaired HRQoL throughout the 24‐month follow‐up period. EQ‐5D‐5L index scores in this group ranged from 0.42 to 0.52, which is substantially lower than values previously reported in ICU survivorship cohorts from Korea, the United Kingdom and the Netherlands [4, 7, 30]. Importantly, severe impairments persisted across multiple EQ‐5D‐5L dimensions, particularly mobility, usual activities and pain/discomfort. Self‐care and anxiety/depression demonstrated partial improvement over time but remained substantially impaired at 24 months. The usual activities dimension reflects the ability to maintain work, household responsibilities, family roles and leisure activities. The sustained impairment observed in this dimension suggests that survivors in the persistently low group experienced not only prolonged physical disability but also substantial disruption of daily functioning and social participation over the long term. Membership in the persistently low trajectory group was associated with multimorbidity and discharge to ECFs. These findings are consistent with previous studies demonstrating that patients with greater comorbidity burden, frailty or pre‐existing functional vulnerability tend to experience poorer long‐term recovery after critical illness [31, 32, 33]. Importantly, discharge to an ECF is unlikely to represent a direct cause of poor HRQoL; rather, it may function as a marker of severe dependency and vulnerability at hospital discharge. A previous study reported high rates of cognitive impairment and dependency in activities of daily living among critically ill survivors discharged to ECFs [34]. Together, these findings suggest that survivors discharged to ECFs likely require more intensive longitudinal rehabilitation and follow‐up support. Although the group‐level trajectories remained distinct and stable in their ordering over follow‐up, individual‐level change was considerable, and the two constructs should not be conflated. Notably, the persistently low group showed the highest proportion of improvement from the 3‐month score, which most plausibly reflects partial recovery from a very low baseline rather than a shift out of the impaired range, as this group remained the most impaired at every time point. These patterns indicate that impairment in the suboptimal and persistently low groups represents a durable burden that does not resolve spontaneously over 2 years, underscoring the importance of early identification and continued monitoring. Limitations This study has several limitations. First, because of the observational design, causal relationships between associated factors and HRQoL trajectories cannot be established. Second, baseline HRQoL before ICU admission was unavailable; therefore, the observed trajectories describe the prevalence of post‐discharge HRQoL impairment rather than the incidence of newly developed impairment. Whether the suboptimal and persistently low trajectories reflect newly acquired impairment, pre‐existing vulnerability or both cannot be determined. The relative stability of the group‐mean trajectories over follow‐up is consistent with the possibility that some of the observed impairment predated ICU admission, and previous work reporting pre‐admission EQ‐5D‐5L scores overlapping with those observed in our cohort supports this interpretation [7], although direct comparison is constrained by differences in case mix, value sets and assessment methods. This distinction is clinically important because pre‐existing impairments may be less amenable to interventions initiated after ICU discharge. Future studies incorporating reliable pre‐ICU HRQoL assessments are needed to distinguish these possibilities. Third, the findings may have been affected by selection bias at both enrolment and follow‐up. Because recruitment relied on posted notices and leaflets, screening was limited to patients who expressed interest rather than conducted consecutively across all eligible admissions, so the enrolled sample may have over‐represented less severely affected or more engaged survivors. During follow‐up, only survivors who completed at least one assessment were included, and those assessed at later time points tended to be younger with a lower comorbidity burden (Table S2). Non‐participation may have been more common among severely impaired survivors, for whom attending assessments—particularly in‐person visits—is more difficult; the shift to telephone assessment during the pandemic may have partly offset this, though its net effect is uncertain. More vulnerable survivors may therefore have been under‐represented, likely leading to an underestimation of the burden of post‐discharge HRQoL impairment. Fourth, although the same structured questionnaire was used throughout the study, the mode of follow‐up assessment changed from face‐to‐face to telephone interviews during the COVID‐19 pandemic, which may have introduced mode‐related variability in response patterns or follow‐up participation. However, previous studies have reported acceptable agreement between administration modes of the EQ‐5D, suggesting that the impact of this change was likely limited [18]. Fifth, although the EQ‐5D‐5L is useful for evaluating multidimensional HRQoL, it does not directly assess cognitive function, a major component of post‐intensive care syndrome. Finally, because the study was conducted in university hospital ICUs in one region of South Korea, caution is warranted when generalising the findings to other settings and populations. Implications for Clinical Practice The persistence of mobility limitations and pain/discomfort through 24 months in the suboptimal and persistently low profiles suggests that these domains warrant particular attention during post‐ICU follow‐up. Practically, brief patient‐reported measures such as the EQ‐5D‐5L could be administered repeatedly beyond the early recovery period, rather than at a single time point, to detect problems that persist or emerge later. Because multimorbidity and discharge to an extended care facility were most strongly associated with the persistently low trajectory, survivors with these characteristics may benefit from earlier planning of rehabilitation and pain management at the point of discharge. These suggestions reflect associations observed in this cohort rather than validated decision rules; because the trajectory groups were identified retrospectively, prospective studies are required to establish whether this approach can guide risk‐stratified follow‐up. Conclusion In this multicentre prospective cohort study, HRQoL recovery among ICU survivors followed three distinct trajectories over 24 months after hospital discharge. While the normative stable group maintained HRQoL levels close to population norms, the suboptimal and persistently low groups demonstrated sustained impairment with distinct dimensional burden patterns and associated factors. Problems related to mobility and pain/discomfort persisted in both unfavourable trajectory groups, whereas the persistently low group additionally demonstrated substantial long‐term impairment in usual activities, self‐care and anxiety/depression. These findings support the need for trajectory‐based post‐ICU follow‐up strategies focused on early identification of high‐risk recovery phenotypes, risk‐stratified rehabilitation, pain management and continuity of care across the post‐acute recovery continuum. Funding This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (nos. NRF2022R1A2C1011917, RS‐2026‐25484185). This funding source had no role in the study design, analysis, data interpretation or decision to submit for publication. Ethics Statement This study was approved by the Institutional Review Board of Dong‐A University (approval date: 15 May 2019; IRB No. 2‐104 709‐AB‐N‐01‐201 904‐HR‐021‐14). All study procedures were conducted in accordance with the principles of the Declaration of Helsinki and relevant institutional regulations. Consent Written informed consent was obtained prior to enrolment. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: Study flowchart. ‘Not responded’ indicates that a participant could not be reached at that time point. Participants who did not explicitly withdraw remained eligible and were contacted again at subsequent time points; consequently, the number of respondents at each time point does not decrease monotonically over follow‐up. Figure S2: Changes in EQ‐5D‐5L index scores relative to 3 months in the overall cohort and by HRQoL trajectory group. Changes at 6, 12 and 24 months were classified relative to each participant's 3‐month EQ‐5D‐5L index score. Improved, no clinically important change and deteriorated were defined as changes of ≥ 0.08, > −0.08 to < 0.08 and ≤ −0.08, respectively. The width of each flow represents the number of participants transitioning between change categories across follow‐up assessments. Multiple‐imputation data were used. Table S1: Model fit indices for latent‐class mixed models. Table S2: Baseline characteristics of the overall cohort and respondents at each follow‐up assessment. Table S3: Comparison of observed and imputed EQ‐5D‐5L index scores across follow‐up time points. Table S4: Comparison of observed‐case and imputed dimension scores across follow‐up time points. Table S5: Longitudinal changes in EQ index score by trajectory group and repeated‐measures ANOVA results. Table S6: Comparison of the proportions of participants reporting moderate‐to‐extreme problems (Levels 3–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S7: Comparison of the proportions of participants reporting any problems (Levels 2–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S8: Comparison of the proportions of participants reporting severe‐to‐extreme problems (Levels 4–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S9: Associated factors of non‐normative HRQoL trajectory membership (ref: normative stable group). Acknowledgements The authors have nothing to report. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Figure S1: Study flowchart. ‘Not responded’ indicates that a participant could not be reached at that time point. Participants who did not explicitly withdraw remained eligible and were contacted again at subsequent time points; consequently, the number of respondents at each time point does not decrease monotonically over follow‐up. Figure S2: Changes in EQ‐5D‐5L index scores relative to 3 months in the overall cohort and by HRQoL trajectory group. Changes at 6, 12 and 24 months were classified relative to each participant's 3‐month EQ‐5D‐5L index score. Improved, no clinically important change and deteriorated were defined as changes of ≥ 0.08, > −0.08 to < 0.08 and ≤ −0.08, respectively. The width of each flow represents the number of participants transitioning between change categories across follow‐up assessments. Multiple‐imputation data were used. Table S1: Model fit indices for latent‐class mixed models. Table S2: Baseline characteristics of the overall cohort and respondents at each follow‐up assessment. Table S3: Comparison of observed and imputed EQ‐5D‐5L index scores across follow‐up time points. Table S4: Comparison of observed‐case and imputed dimension scores across follow‐up time points. Table S5: Longitudinal changes in EQ index score by trajectory group and repeated‐measures ANOVA results. Table S6: Comparison of the proportions of participants reporting moderate‐to‐extreme problems (Levels 3–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S7: Comparison of the proportions of participants reporting any problems (Levels 2–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S8: Comparison of the proportions of participants reporting severe‐to‐extreme problems (Levels 4–5) across HRQoL trajectory groups by EQ‐5D‐5L dimension and follow‐up time point. Table S9: Associated factors of non‐normative HRQoL trajectory membership (ref: normative stable group). Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.