Transitions and determinants of ICU-acquired frailty after critical illness: A multicenter cohort study using a multistate Markov model Authors: Tingting Wu, Yueqing Wei, Lan Shi, Rongjin Lin, Hong Li Published: 2026 ABSTRACT BackgroundPreexisting frailty at ICU admission has been well investigated and linked to adverse outcomes, the continuum and determinants of new onset of frailty remain poorly characterized. This study aimed to estimate transition probabilities and intensities across non-frailty, frailty, and death after critical illness, and explore a set of potential variables associated with these transitions.MethodsWe conducted a multicenter prospective cohort study of ICU survivors, followed from ICU discharge to 6 months. Frailty was assessed with the Clinical Frailty Scale (CFS), with scores ≥5 defined as frail and <4 as non-frail. Multistate Markov models were used to estimate transition probabilities and intensities among three states: frailty, non-frailty, and death. Univariate and multivariable Markov models were used to identify predictors of state transitions.ResultsAt follow-up, 66.1% of frail patients reversed to non-frail, 25.7% remained frail, and 8.2% died. Among those non-frails at baseline, 72.9% remained stable, 21.2% developed frailty, and 5.9% died. Recovery from frailty occurred more than three times as often as progression from non-frailty to frailty (transition intensity ratio, 3.12; 95% CI, 2.22–4.38). The transition from frailty to death was over twice as frequent as from non-frailty to death (ratio, 2.52; 95% CI, 0.39–16.37). Sepsis-related complications significantly increased the risk of frailty onset and mortality, whereas higher body mass index, muscle mass, handgrip strength, and phase angle consistently promoted recovery and survival. Older age, female sex, comorbidities, malnutrition, and prolonged ICU or hospital stay impaired recovery.ConclusionsFrailty after critical illness is a dynamic and potentially reversible condition, with transitions shaped by sepsis-related complications, nutritional and muscular status, and patient vulnerability factors. Early post-ICU frailty assessment and targeted interventions addressing infection, nutrition, and physical function may facilitate recovery and improve survival in ICU survivors. Introduction Despite rising age and illness severity among ICU populations, survival rates have steadily improved in recent years [1]. However, survival does not imply full recovery. Many ICU survivors experience prolonged impairments in physical, psychological, and cognitive domains—collectively termed post-intensive care syndrome (PICS) [2]. Frailty has traditionally been conceptualized in geriatric medicine as a multidimensional syndrome characterized by age-related decline in physiological reserve and reduced resilience to stressors [3]. However, in the context of critical illness, frailty-oriented measures may also capture an acquired state of vulnerability in adult ICU patients, including younger adults, suggesting that frailty may not be defined by chronological age alone but may also reflect biological vulnerability [4,5]. Longitudinal studies report high rates of frailty after critical illness, with nearly 70% classified as frail at hospital discharge [6] and approximately 40% developing or sustaining frailty within the first year [7]. While traditionally viewed as a product of aging or chronic disease, frailty may arise acutely during critical illness [3]. Inflammatory responses, malnutrition, immobility, hormonal dysregulation, and mitochondrial dysfunction can rapidly drive muscle wasting, deconditioning, and new-onset frailty [3]. Frailty is associated with adverse outcomes during acute hospitalization and long-term follow-up, including increased mortality, prolonged hospitalization, impaired recovery, and long-term disability [[8], [9], [10], [11], [12]]. Most prior studies, however, have examined preexisting frailty at ICU admission [[8], [9], [10], [11], [12]], whereas new-onset frailty during the ICU stay remains underexplored. Recent research has shifted toward the dynamic and potentially reversible nature of frailty, encompassing both preexisting and new-onset cases [6,7,13,14]. For example, the FORECAST study used alluvial plots to illustrate dynamic frailty transitions across ICU admission, hospital discharge, and 6-month follow-up [6]. The prevalence of frailty rose from 29.8% at ICU admission to 54.6% at discharge, then declined to 34.1% by 6 months [6]. Another study found that approximately 60% of patients without pre-hospital frailty developed frailty by 3 or 12 months after discharge [7]. From baseline to 6 month poster discharge, 54.5% of patients showed improvement in frailty severity, while 45.5% remained unchanged or worsened [6]. These findings underscore the dynamic and potentially reversible nature of frailty but do not distinguish whether changes reflect new-onset frailty or progression of preexisting frailty. Descriptive and visual methods provide limited information on transition probabilities or the factors driving recovery and decline. Given the bidirectional and time-dependent nature of frailty, static models are insufficient to capture its complexity. Multistate Markov models [15] provide a rigorous framework for estimating transition probabilities and intensities among non-frail, frail, and death states, while identifying modifiable risk factors. However, the longitudinal course and drivers of ICU-acquired frailty remain poorly characterized. To address this gap, we conducted a multicenter prospective cohort study using a multistate Markov model to examine frailty transitions among critically ill patients from ICU discharge through 6 months post-discharge. The objectives were to (1) estimate transition probabilities and intensities among non-frailty, frailty, and death states after critical illness, and (2) identify determinants of these transitions. Methods Study design, setting, and participants This multicenter prospective cohort study was conducted in 10 intensive care units (ICUs) across five tertiary grade A hospitals in Fujian Province, China, between February 2022 and August 2023. Inclusion criteria were: (1) age ≥18 years; (2) ICU stay of at least 48 h. Exclusion criteria: (1) preexisting frailty, defined as a Clinical Frailty Scale (CFS) score ≥5 [16], based on functional status two months before ICU admission [7,10,11,17] as determined by patient or surrogate interview, to better reflect baseline status and reduce misclassification of acute illness–related decline as preexisting frailty; (2) physical disability resulting from acute or chronic brain injury, spinal cord injury, bone fractures, neuromuscular disorders, etc.; (3) sedation, coma, or death at ICU discharge; (4) severe cognitive impairment, substance use disorder, or psychotic disorder prior to ICU admission; (5) terminal illness with an expected survival of less than six months. Participants were withdrawn if they (1) completed only one frailty assessment; (2) were lost to follow-up post-discharge; or (3) voluntarily withdrew consent. Exposure variable Frailty was assessed using a modified version of the CFS from the Canadian Study of Health and Aging [16]. The CFS scores range from 1 to 8, 1 = very fit, 2 = well, 3 = managing well, 4 = vulnerable, 5 = mildly frail, 6 = moderately frail, 7 = severely frail, and 8 = very severely frail. For analysis, frailty was dichotomized as frail (CFS 5–8) and non-frail (CFS 1–4) [16]. The CFS has been validated in critically ill patients, showing good inter-rater reliability and strong correlation with other frailty measures [18]. Other covariates Patient characteristics were recorded using a study-specific case report form covering four domains:(1)Demographics: Age, sex, body mass index (BMI), education level, pre-admission employment status, smoking history, and alcohol use.(2)Disease information: Data included primary diagnosis, comorbidities, and clinical conditions during ICU stay. Comorbidity burden was quantified using the Charlson Comorbidity Index (CCI) [19]. Major ICU-related complications included respiratory failure, heart failure, sepsis, and multiple organ dysfunction syndrome. Illness severity and organ dysfunction were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) [20] and the Sequential Organ Failure Assessment (SOFA) [21], respectively. Nutritional risk was evaluated using the Nutritional Risk Screening 2002 (NRS2002) [22] and the Nutrition Risk in Critically Ill (NUTRIC) score [23].(3)Muscle Mass and strength: Body composition was assessed via bioelectrical impedance vector analysis (BIVA) using the NUTRILAB device (AKERN, Italy), which applies a 400 μA, 50 kHz alternating current to estimate fat-free mass (FFM), appendicular skeletal muscle mass (ASMM) and phase angle (PhA) [24]. Muscle strength was assessed using the Medical Research Council (MRC) scale [10] and handgrip strength, measured with a CAMRY EH101 handheld electronic dynamometer.(4)ICU Treatment Variables: Duration of mechanical ventilation, and participation in early rehabilitation (e.g., sitting, walking, resistance training, or bedside cycling during ICU stay), ICU length of stay, hospital length of stay. (1)Demographics: Age, sex, body mass index (BMI), education level, pre-admission employment status, smoking history, and alcohol use.(2)Disease information: Data included primary diagnosis, comorbidities, and clinical conditions during ICU stay. Comorbidity burden was quantified using the Charlson Comorbidity Index (CCI) [19]. Major ICU-related complications included respiratory failure, heart failure, sepsis, and multiple organ dysfunction syndrome. Illness severity and organ dysfunction were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) [20] and the Sequential Organ Failure Assessment (SOFA) [21], respectively. Nutritional risk was evaluated using the Nutritional Risk Screening 2002 (NRS2002) [22] and the Nutrition Risk in Critically Ill (NUTRIC) score [23].(3)Muscle Mass and strength: Body composition was assessed via bioelectrical impedance vector analysis (BIVA) using the NUTRILAB device (AKERN, Italy), which applies a 400 μA, 50 kHz alternating current to estimate fat-free mass (FFM), appendicular skeletal muscle mass (ASMM) and phase angle (PhA) [24]. Muscle strength was assessed using the Medical Research Council (MRC) scale [10] and handgrip strength, measured with a CAMRY EH101 handheld electronic dynamometer.(4)ICU Treatment Variables: Duration of mechanical ventilation, and participation in early rehabilitation (e.g., sitting, walking, resistance training, or bedside cycling during ICU stay), ICU length of stay, hospital length of stay. (1)Demographics: Age, sex, body mass index (BMI), education level, pre-admission employment status, smoking history, and alcohol use.(2)Disease information: Data included primary diagnosis, comorbidities, and clinical conditions during ICU stay. Comorbidity burden was quantified using the Charlson Comorbidity Index (CCI) [19]. Major ICU-related complications included respiratory failure, heart failure, sepsis, and multiple organ dysfunction syndrome. Illness severity and organ dysfunction were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) [20] and the Sequential Organ Failure Assessment (SOFA) [21], respectively. Nutritional risk was evaluated using the Nutritional Risk Screening 2002 (NRS2002) [22] and the Nutrition Risk in Critically Ill (NUTRIC) score [23].(3)Muscle Mass and strength: Body composition was assessed via bioelectrical impedance vector analysis (BIVA) using the NUTRILAB device (AKERN, Italy), which applies a 400 μA, 50 kHz alternating current to estimate fat-free mass (FFM), appendicular skeletal muscle mass (ASMM) and phase angle (PhA) [24]. Muscle strength was assessed using the Medical Research Council (MRC) scale [10] and handgrip strength, measured with a CAMRY EH101 handheld electronic dynamometer.(4)ICU Treatment Variables: Duration of mechanical ventilation, and participation in early rehabilitation (e.g., sitting, walking, resistance training, or bedside cycling during ICU stay), ICU length of stay, hospital length of stay. (1)Demographics: Age, sex, body mass index (BMI), education level, pre-admission employment status, smoking history, and alcohol use. Demographics: Age, sex, body mass index (BMI), education level, pre-admission employment status, smoking history, and alcohol use. (2)Disease information: Data included primary diagnosis, comorbidities, and clinical conditions during ICU stay. Comorbidity burden was quantified using the Charlson Comorbidity Index (CCI) [19]. Major ICU-related complications included respiratory failure, heart failure, sepsis, and multiple organ dysfunction syndrome. Illness severity and organ dysfunction were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) [20] and the Sequential Organ Failure Assessment (SOFA) [21], respectively. Nutritional risk was evaluated using the Nutritional Risk Screening 2002 (NRS2002) [22] and the Nutrition Risk in Critically Ill (NUTRIC) score [23]. Disease information: Data included primary diagnosis, comorbidities, and clinical conditions during ICU stay. Comorbidity burden was quantified using the Charlson Comorbidity Index (CCI) [19]. Major ICU-related complications included respiratory failure, heart failure, sepsis, and multiple organ dysfunction syndrome. Illness severity and organ dysfunction were assessed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) [20] and the Sequential Organ Failure Assessment (SOFA) [21], respectively. Nutritional risk was evaluated using the Nutritional Risk Screening 2002 (NRS2002) [22] and the Nutrition Risk in Critically Ill (NUTRIC) score [23]. (3)Muscle Mass and strength: Body composition was assessed via bioelectrical impedance vector analysis (BIVA) using the NUTRILAB device (AKERN, Italy), which applies a 400 μA, 50 kHz alternating current to estimate fat-free mass (FFM), appendicular skeletal muscle mass (ASMM) and phase angle (PhA) [24]. Muscle strength was assessed using the Medical Research Council (MRC) scale [10] and handgrip strength, measured with a CAMRY EH101 handheld electronic dynamometer. Muscle Mass and strength: Body composition was assessed via bioelectrical impedance vector analysis (BIVA) using the NUTRILAB device (AKERN, Italy), which applies a 400 μA, 50 kHz alternating current to estimate fat-free mass (FFM), appendicular skeletal muscle mass (ASMM) and phase angle (PhA) [24]. Muscle strength was assessed using the Medical Research Council (MRC) scale [10] and handgrip strength, measured with a CAMRY EH101 handheld electronic dynamometer. (4)ICU Treatment Variables: Duration of mechanical ventilation, and participation in early rehabilitation (e.g., sitting, walking, resistance training, or bedside cycling during ICU stay), ICU length of stay, hospital length of stay. ICU Treatment Variables: Duration of mechanical ventilation, and participation in early rehabilitation (e.g., sitting, walking, resistance training, or bedside cycling during ICU stay), ICU length of stay, hospital length of stay. Data measurement Frailty was assessed using the CFS at multiple time points: within 24 h after ICU discharge, and at 1 and 6 months post-discharge. In-hospital assessments were conducted face-to-face by five ICU-specialized nurses, while post-discharge evaluations were completed via telephone follow-up. Premorbid muscle mass was assessed via bioelectrical impedance analysis (BIA) within 24 h of ICU admission. Muscle strength was measured by trained ICU rehabilitation nurses using the MRC scale and handgrip dynamometry within 24 h after ICU discharge. Handgrip strength was tested with the patient seated, elbow at 90°, and forearm in neutral position; the highest value from three trials was recorded [25,26]. Demographic and clinical data, including ICU treatments, were extracted from the electronic medical records at discharge. For variables recorded repeatedly (e.g., APACHE II, SOFA), only values from the first 24 h after ICU admission were analyzed. Daily early rehabilitation activities were documented by bedside nurses. Quality control All participating centers followed standardized protocols for patient screening, data collection, and outcome assessment, applying consistent inclusion and exclusion criteria to reduce inter-site variability. Comprehensive data on potential confounders—including demographics, premorbid muscle mass, disease characteristics, and ICU treatments—were collected to enable appropriate adjustment in statistical analyses. Data collectors received structured training and underwent calibration sessions to ensure consistency. Regular audits and quality assurance measures were conducted to monitor adherence to study protocols and maintain data integrity. Ethics statement The study was approved by the central ethics committee of Fujian Provincial Hospital (Approval No. K2021-04-004), and written informed consent was obtained from all participants or their legal representatives. All procedures complied with the Declaration of Helsinki and national ethical regulations. The manuscript was prepared in accordance with the STROBE reporting guidelines. Statistical analysis Simple size Although no universally accepted method exists for a priori sample size calculation in multistate Markov models, methodological literature recommends that each covariate per transition should be supported by at least 10–20 events to ensure reliable parameter estimation and model stability [27,28]. Our model includes four transitions: from frailty to non-frailty, from non-frailty to frailty, from frailty to death, and from non-frailty to death, with a total of 12 planned covariates. Based on pilot data indicating an average of 1.5 transitions per patient, a sample size of approximately 320–640 patients were considered adequate. Markov multistate model A time-homogeneous Markov multistate model was used to assess the primary outcome, defining three states: frailty, non-frailty, and all-cause death. Frailty and non-frailty were treated as transient states, allowing transitions between them and to death, which was modeled as an absorbing state. The model assumed that transition probabilities depended solely on the time interval, not on calendar time. Based on this structure, cumulative transition probabilities and transition intensities were estimated over a 6-month period from ICU admission. Cumulative transition probabilities represent the likelihood of being in a given state at future time points and are useful for long-term outcome prediction. Transition intensities reflect the instantaneous risk of transitioning between states, capturing short-term dynamics. Subsequently, hazard ratios (HRs) and corresponding 95% confidence intervals for covariates influencing transition intensities were estimated. Missing data were imputed using random forest methods. All analyses were conducted in R (version 4.5.1) using the "msm" package, with model fitting optimized via the BFGS algorithm. Time was modeled in months for clinical interpretability. Baseline characteristics were summarized as mean (SD), median (IQR), or frequency (%), as appropriate. Results Participant characteristics Between February and August 2022, 1,253 patients from five centers were screened. After exclusions, 361 patients were assessed at ICU discharge, 311 at 1 month, and 269 at 6 months; 337 with data from at least two time points were included in the multi-state Markov model analysis. Among them, 29 died within 6 months after discharge. Details are shown in Fig. 1.Fig. 1Flowchart of the participants.Fig. 1 Flowchart of the participants. The mean age was 60.7 ± 13.3 years, and 245 (72.7%) were male. The median Charlson Comorbidity Index was 3.0 (IQR, 2.0–4.0), the APACHE II score was 19.0 (IQR, 13.0–26.0), and the SOFA score was 7.0 (IQR, 4.0–10.0). Postoperative patients comprised 194(57.57%) of the cohort. At baseline, 47 patients (13.9%) were classified as non-frail and 290 (86.1%) as frail (see Table 1).Table 1Baseline characteristics according to frailty status assessed within 24 h of ICU discharge.Table 1VariablesTotal (n = 337)GroupStatisticsP valuesNon-frailty (n = 47)Frailty (n = 290)Age [years, χ¯±s]60.7 ± 13.353.6 ± 13.761.9 ± 12.9t = 4.032<0.001†Male [n (%)]245(72.70)37(78.72)208(71.72)χ2 = 0.9980.318BMI [kg/m2, n (%)] <18.541(17.52)5(16.67)36(17.65)χ2 = 9.8640.041 18.5–23.9170(72.65)20(66.67)150(73.53) ≥24.023(9.83)5(16.67)18(8.82)Marital Status [n (%)] Divorced/Widowed13(3.86)2(5.00)11(3.70)χ2 = 1.3600.506 Unmarried80(23.74)7(17.50)73(24.58) Married239(70.92)31(77.50)208(70.03)Education Level [n (%)] Illiterate/Semi-literate55(16.32)2(4.26)53(18.28)χ2 = 7.1520.067 Primary School110(32.64)16(34.04)94(32.41) Junior to High School141(41.84)22(46.81)119(41.03) College and above31(9.20)7(14.89)24(8.28)Work Type Prior to admission [n (%)] Non-Physical Labor/Unemployed267(79.23)38(80.85)229(78.97)χ2 = 0.0870.768 Physical Labor70(20.77)9(19.15)61(21.03)Smoking [n (%)]122(36.20)17(36.17)105(36.21)χ2 = 0.0000.996Alcohol use [n (%)]110(32.64)19(40.43)91(31.38)χ2 = 1.5050.220Primary Diagnosis [n (%)] Surgery194(57.57)27(57.45)167(57.59)—0.447* Acute Disease18(5.34)5(10.64)13(4.48) Acute Exacerbation of Chronic Disease78(23.15)9(19.15)69(23.79) Sepsis32(9.50)5(10.64)27(9.31) Other15(4.45)1(2.13)14(4.83)Cardiac surgery [n (%)]80(23.74)15(31.91)65(22.41)χ2 = 2.0170.156Non-cardiac surgery [n (%)]116(34.42)12(25.53)104(35.86)χ2 = 1.9120.167ICU Patient Sources [n (%)] Emergency62(18.40)10(21.28)52(17.93)χ2 = 1.3290.722 General ward32(9.50)6(12.77)26(8.97) Post-operation164(48.66)22(46.81)142(48.97) Transferred from another hospital79(23.44)9(19.15)70(24.14)Planed to ICU transfer [n (%)]163(48.37)21(44.68)142(48.97)χ2 = 0.2970.586Complications [n (%)] Respiratory Failure89(26.41)7(14.89)82(28.28)χ2 = 3.7270.054 Heart Failure68(20.18)4(8.51)64(22.07)χ2 = 4.6160.032 Sepsis51(15.13)5(10.64)46(15.86)χ2 = 0.8590.354 MODS20(5.93)0(0.00)20(6.90)χ2 = 3.4460.063Charlson Comorbidity Index [points, M(IQR)]3.00(2.00,4.00)2.00(1.00,3.00)3.00(2.00,4.00)Z = −1.9790.047#Oxygenation index at admission [M(IQR)]287.1(201.3,360.0)271.0(185.7,340.0)287.1(201.3,360.0)Z = −0.1900.849#APACHEII [points, M(IQR)]19.0(13.0,26.0)16.0(11.0,22.0)19.0(14.0,26.0)Z = −2.2160.026#SOFA [points, M(IQR)]7.00(4.00,10.00)6.00(3.00,10.00)7.00(4.00,10.00)Z = −1.6000.109#NRS2002 [points, M(IQR)]3.00(3.00,4.00)3.00(3.00,3.65)3.65(3.00,4.00)Z = −2.8830.003#NUTRIC [points, M(IQR)]5.00(4.00,7.00)5.00(2.00,6.00)5.00(4.00,7.00)Z = −2.6650.007#Length of stay prior to ICU transfer [days, M(IQR)]2.00(0.00,7.00)2.00(0.00,6.00)3.00(0.00,7.00)Z = −0.5240.600#Length of ICU stay [days, M(IQR)]5.00(3.00,8.00)4.00(2.00,6.00)5.00(3.00,9.00)Z = −2.7650.005#Length of hospital stay [days, M(IQR)]19.0(14.0,28.0)16.0(10.0,22.0)20.0(14.0,29.0)Z = −3.316<0.001#Duration of mechanical ventilation [h M(IQR)]15.0(0.0,72.0)8.0(0.0,79.0)15.0(0.0,72.0)Z = −0.9210.356#FFM [kg, χ¯±s]50.6 ± 8.852.2 ± 9.150.3 ± 8.7t = 1.3820.168†ASMM [kg, M(IQR)]19.5(16.1,21.9)19.8(16.7,23.9)19.4(15.8,21.9)Z = 1.6550.097#Handgrip [kg, M(IQR)]18.4(11.4,22.8)21.9(18.4,29.7)16.7(10.5,22.1)Z = 4.425<0.001#MRC [points, M(IQR)]48.0(43.5,56.0)57.0(49.5,60.0)48.0(42.0,54.0)Z = 5.642<0.001#Barthel index [points, M(IQR)]40.0(25.0,55.0)50.0(30.0,80.0)40.0(25.0,55.0)Z = 2.4480.014#Early Rehabilitation [n (%)]121(35.91)25(53.19)96(33.10)χ2 = 7.0920.008Note: #: Wilcoxon rank-sum test; *: Fisher’s exact test; †: Student’s t-test. Bold values indicate statistical significance (P < 0.05).BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale. Table 1Baseline characteristics according to frailty status assessed within 24 h of ICU discharge.Table 1VariablesTotal (n = 337)GroupStatisticsP valuesNon-frailty (n = 47)Frailty (n = 290)Age [years, χ¯±s]60.7 ± 13.353.6 ± 13.761.9 ± 12.9t = 4.032<0.001†Male [n (%)]245(72.70)37(78.72)208(71.72)χ2 = 0.9980.318BMI [kg/m2, n (%)] <18.541(17.52)5(16.67)36(17.65)χ2 = 9.8640.041 18.5–23.9170(72.65)20(66.67)150(73.53) ≥24.023(9.83)5(16.67)18(8.82)Marital Status [n (%)] Divorced/Widowed13(3.86)2(5.00)11(3.70)χ2 = 1.3600.506 Unmarried80(23.74)7(17.50)73(24.58) Married239(70.92)31(77.50)208(70.03)Education Level [n (%)] Illiterate/Semi-literate55(16.32)2(4.26)53(18.28)χ2 = 7.1520.067 Primary School110(32.64)16(34.04)94(32.41) Junior to High School141(41.84)22(46.81)119(41.03) College and above31(9.20)7(14.89)24(8.28)Work Type Prior to admission [n (%)] Non-Physical Labor/Unemployed267(79.23)38(80.85)229(78.97)χ2 = 0.0870.768 Physical Labor70(20.77)9(19.15)61(21.03)Smoking [n (%)]122(36.20)17(36.17)105(36.21)χ2 = 0.0000.996Alcohol use [n (%)]110(32.64)19(40.43)91(31.38)χ2 = 1.5050.220Primary Diagnosis [n (%)] Surgery194(57.57)27(57.45)167(57.59)—0.447* Acute Disease18(5.34)5(10.64)13(4.48) Acute Exacerbation of Chronic Disease78(23.15)9(19.15)69(23.79) Sepsis32(9.50)5(10.64)27(9.31) Other15(4.45)1(2.13)14(4.83)Cardiac surgery [n (%)]80(23.74)15(31.91)65(22.41)χ2 = 2.0170.156Non-cardiac surgery [n (%)]116(34.42)12(25.53)104(35.86)χ2 = 1.9120.167ICU Patient Sources [n (%)] Emergency62(18.40)10(21.28)52(17.93)χ2 = 1.3290.722 General ward32(9.50)6(12.77)26(8.97) Post-operation164(48.66)22(46.81)142(48.97) Transferred from another hospital79(23.44)9(19.15)70(24.14)Planed to ICU transfer [n (%)]163(48.37)21(44.68)142(48.97)χ2 = 0.2970.586Complications [n (%)] Respiratory Failure89(26.41)7(14.89)82(28.28)χ2 = 3.7270.054 Heart Failure68(20.18)4(8.51)64(22.07)χ2 = 4.6160.032 Sepsis51(15.13)5(10.64)46(15.86)χ2 = 0.8590.354 MODS20(5.93)0(0.00)20(6.90)χ2 = 3.4460.063Charlson Comorbidity Index [points, M(IQR)]3.00(2.00,4.00)2.00(1.00,3.00)3.00(2.00,4.00)Z = −1.9790.047#Oxygenation index at admission [M(IQR)]287.1(201.3,360.0)271.0(185.7,340.0)287.1(201.3,360.0)Z = −0.1900.849#APACHEII [points, M(IQR)]19.0(13.0,26.0)16.0(11.0,22.0)19.0(14.0,26.0)Z = −2.2160.026#SOFA [points, M(IQR)]7.00(4.00,10.00)6.00(3.00,10.00)7.00(4.00,10.00)Z = −1.6000.109#NRS2002 [points, M(IQR)]3.00(3.00,4.00)3.00(3.00,3.65)3.65(3.00,4.00)Z = −2.8830.003#NUTRIC [points, M(IQR)]5.00(4.00,7.00)5.00(2.00,6.00)5.00(4.00,7.00)Z = −2.6650.007#Length of stay prior to ICU transfer [days, M(IQR)]2.00(0.00,7.00)2.00(0.00,6.00)3.00(0.00,7.00)Z = −0.5240.600#Length of ICU stay [days, M(IQR)]5.00(3.00,8.00)4.00(2.00,6.00)5.00(3.00,9.00)Z = −2.7650.005#Length of hospital stay [days, M(IQR)]19.0(14.0,28.0)16.0(10.0,22.0)20.0(14.0,29.0)Z = −3.316<0.001#Duration of mechanical ventilation [h M(IQR)]15.0(0.0,72.0)8.0(0.0,79.0)15.0(0.0,72.0)Z = −0.9210.356#FFM [kg, χ¯±s]50.6 ± 8.852.2 ± 9.150.3 ± 8.7t = 1.3820.168†ASMM [kg, M(IQR)]19.5(16.1,21.9)19.8(16.7,23.9)19.4(15.8,21.9)Z = 1.6550.097#Handgrip [kg, M(IQR)]18.4(11.4,22.8)21.9(18.4,29.7)16.7(10.5,22.1)Z = 4.425<0.001#MRC [points, M(IQR)]48.0(43.5,56.0)57.0(49.5,60.0)48.0(42.0,54.0)Z = 5.642<0.001#Barthel index [points, M(IQR)]40.0(25.0,55.0)50.0(30.0,80.0)40.0(25.0,55.0)Z = 2.4480.014#Early Rehabilitation [n (%)]121(35.91)25(53.19)96(33.10)χ2 = 7.0920.008Note: #: Wilcoxon rank-sum test; *: Fisher’s exact test; †: Student’s t-test. Bold values indicate statistical significance (P < 0.05).BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale. Baseline characteristics according to frailty status assessed within 24 h of ICU discharge. Note: #: Wilcoxon rank-sum test; *: Fisher’s exact test; †: Student’s t-test. Bold values indicate statistical significance (P < 0.05). BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale. Transition probability During the 6-month follow-up after ICU discharge, 66.1% of frail patients recovered to non-frailty, 25.7% remained frail, and 8.2% died. Among those initially non-frails, 72.9% remained stable, 21.2% developed frailty, and 5.9% died (Fig. 2). Expected and observed state prevalences were broadly aligned over time, supporting an acceptable overall fit of the multistate Markov model (Figure S1).Fig. 2Transition probabilities six months after ICU discharge.Fig. 2 Transition probabilities six months after ICU discharge. Fig. 3 shows 12-month transition probabilities by baseline frailty status. Among initially frail patients, the probability of transition to non-frailty rose rapidly and plateaued at about 65%, whereas the probability of remaining frail declined from 80% to about 20% by 12 months; the probability of death remained below 15%. Among initially non-frail patients, the probability of remaining non-frail stayed high, the probability of transition to frailty increased modestly to about 20%, and the probability of death remained below 5% (Table 2).Fig. 3Temporal Trends in Transition Probabilities Within 12 Months.Fig. 3Table 2Estimated transition probabilities and intensities in the multistate Markov model.Table 2State(From-To)Transition probability matricesNon-frailtyFrailtyDeathTransition Probability (1st month) Non-frailty0.900(0.847–0.932)0.091(0.059–0.138)0.009(0.004–0.028) Frailty0.284(0.232–0.338)0.697(0.643–0.745)0.019(0.011–0.038) Death–––Transition Probability (6th month) Non-frailty0.729(0.643–0.791)0.212(0.156–0.273)0.059(0.037–0.139) Frailty0.661(0.598–0.698)0.257(0.213–0.306)0.082(0.059–0.144) Death–––Transition Probability (12th month) Non-frailty0.671(0.547–0.736)0.209(0.148–0.263)0.120(0.080–0.275) Frailty0.652(0.547–0.703)0.206(0.152–0.256)0.142(0.105–0.260) Death–––Ratio (form→Death/Other transition)Ratio(95%CI)Frailty→Death vs Frailty→Non-frailty0.057(0.025–0.128)Non-frailty→Death vs Non-frailty→Frailty0.071(0.017–0.287)Frailty→Non-frailty vs Non-frailty→Frailty3.122(2.223,4.382)Frailty →Death vs Non-frailty→Death2.515(0.386,16.367)Non-frailty→Frailty vs Frailty→Death5.628(2.387,13.269)Non-frailty→Frailty vs Non-frailty→Death14.154(3.484,57.494) Temporal Trends in Transition Probabilities Within 12 Months. Table 2Estimated transition probabilities and intensities in the multistate Markov model.Table 2State(From-To)Transition probability matricesNon-frailtyFrailtyDeathTransition Probability (1st month) Non-frailty0.900(0.847–0.932)0.091(0.059–0.138)0.009(0.004–0.028) Frailty0.284(0.232–0.338)0.697(0.643–0.745)0.019(0.011–0.038) Death–––Transition Probability (6th month) Non-frailty0.729(0.643–0.791)0.212(0.156–0.273)0.059(0.037–0.139) Frailty0.661(0.598–0.698)0.257(0.213–0.306)0.082(0.059–0.144) Death–––Transition Probability (12th month) Non-frailty0.671(0.547–0.736)0.209(0.148–0.263)0.120(0.080–0.275) Frailty0.652(0.547–0.703)0.206(0.152–0.256)0.142(0.105–0.260) Death–––Ratio (form→Death/Other transition)Ratio(95%CI)Frailty→Death vs Frailty→Non-frailty0.057(0.025–0.128)Non-frailty→Death vs Non-frailty→Frailty0.071(0.017–0.287)Frailty→Non-frailty vs Non-frailty→Frailty3.122(2.223,4.382)Frailty →Death vs Non-frailty→Death2.515(0.386,16.367)Non-frailty→Frailty vs Frailty→Death5.628(2.387,13.269)Non-frailty→Frailty vs Non-frailty→Death14.154(3.484,57.494) Estimated transition probabilities and intensities in the multistate Markov model. Transition intensity The recovery intensity from frailty to non-frailty was 0.363 (95% CI, 0.285–0.462), more than three times the progression intensity from non-frailty to frailty (0.116; 95% CI, 0.072–0.187), yielding a transition ratio of 3.12 (95% CI, 2.22–4.38). The intensity of death was also higher among frail than non-frail patients (0.021 [95% CI, 0.010–0.044] vs. 0.008 [95% CI, 0.002–0.029]), with a transition ratio of 2.52 (95% CI, 0.39–16.37). Predictors of frailty transitions Independent predictors of each frailty transition are shown in Table 3 and Fig. 4. Sepsis-related complications increased the risk of transition from non-frailty to frailty (HR 4.00, 95% CI 1.17–14.00; Fig. 4A). Higher BMI and greater phase angle were associated with a lower risk of transition from non-frailty to death (Fig. 4B). Recovery from frailty to non-frailty was associated with better nutritional and functional reserve, including higher BMI, greater fat-free mass, appendicular skeletal muscle mass, handgrip strength, MRC score, and phase angle, as well as alcohol use, postoperative cardiac surgery, and planned ICU admission; by contrast, older age, female sex, greater comorbidity burden, malnutrition, and longer ICU and hospital stays reduced the likelihood of recovery (Fig. 4C). Sepsis (HR 6.70, 95% CI 1.02–43.52) and prolonged ICU stay (HR 1.39, 95% CI 1.11–1.74) increased the risk of transition from frailty to death, whereas higher handgrip strength was protective (HR 0.28, 95% CI 0.09–0.87; Fig. 4D).Table 3Coefficient for univariate analysis of multistate Markov model [HR(95%CI)].Table 3VariableNon-frailty → FrailtyNon-frailty → DeathFrailty → Non-frailtyFrailty → DeathAge0.834(0.474–1.467)0.973(0.445–2.128)0.647(0.510–0.822)*1.925(0.718–5.163)Gender MaleReferenceReferenceReferenceReference Female0.361(0.103–1.266)0.000(0.000–23,489.000)0.512(0.324–0.809)*1.486(0.369–5.993)BMI, kg/m2 <18.5RefRefRefRef 18.5–23.91.02(0.63–1.65)0.18(0.03–0.97)*1.24(1.00–1.53)*0.86(0.44–1.65) ≥24.01.06(0.66–1.70)0.15(0.02–0.98)*1.27(1.01–1.58)*0.83(0.42–1.64)Marital status MarriedReferenceReferenceReferenceReference Divorced/Widowed0.804(0.094–6.886)0.007(0.000–18,616.000)0.442(0.174–1.122)2.281(0.567–9.173) Unmarried1.086(0.376–3.131)0.149(0.000–77,220.000)1.446(0.886–2.361)1.897(0.463–7.771)Education level Illiterate/Semi-illiterateReferenceReferenceReferenceReference Primary school0.234(0.060–0.913)*18,967.000(0.000–Inf)0.643(0.310–1.335)1.490(0.200–11.110) Junior to senior high0.500(0.148–1.688)14,653.000(0.000–Inf)1.374(0.673–2.805)2.809(0.490–16.100) College or higher0.000(0.000–Inf)0.002(0.000–Inf)0.549(0.245–1.230)4.347(0.790–23.920)Employment type prior to admission Non-manual labor/UnemployedReferenceReferenceReferenceReference Manual labor1.896(0.687–5.229)0.002(0.000–14,549.000)1.266(0.726–2.207)1.094(0.361–3.318)Smoking NoReferenceReferenceReferenceReference Yes1.346(0.501–3.611)29.330(0.000–25,976.000)1.039(0.642–1.681)0.477(0.059–3.860)Alcohol consumption NoReferenceReferenceReferenceReference Yes1.365(0.516–3.610)1.065(0.031–36.230)1.655(1.027–2.666)*1.514(0.389–5.882)Primary diagnosis Various surgeriesReferenceReferenceReferenceReference Acute disease0.994(0.136–7.294)0.000(0.000–Inf)1.168(0.378–3.616)0.000(0.000–Inf) Acute exacerbation of chronic disease0.141(0.018–1.122)2.172(0.288–16.360)0.408(0.254–0.654)*2.775(0.424–18.150) Sepsis2.606(0.671–10.120)0.000(0.000–Inf)1.476(0.598–3.640)6.851(1.053–44.570)* Other1.262(0.242–6.568)0.000(0.000–Inf)1.034(0.435–2.459)4.359(0.488–38.940)Postoperative status Non-surgicalReferenceReferenceReferenceReference Post-cardiac surgery1.029(0.344–3.078)0.559(0.015–21.170)2.108(1.274–3.489)*0.185(0.007–4.619) Post-non-cardiac surgery1.578(0.588–4.235)3.113(0.066–147.800)1.248(0.757–2.057)0.431(0.044–4.231)ICU patient source In-hospital general wardReferenceReferenceReferenceReference Transferred from other hospitals0.401(0.071–2.277)0.032(0.000–92,489.000)0.688(0.318–1.489)1.656(0.361–7.608) Postoperative transfer0.860(0.278–2.656)16.880(0.000–40,116.000)1.345(0.758–2.388)0.033(0.000–43,510.000) Direct admission or emergency room0.552(0.114–2.676)20.010(0.000–50,033.000)0.665(0.338–1.311)1.573(0.372–6.657)Planned ICU transfer NoReferenceReferenceReferenceReference Yes1.210(0.483–3.033)1.631(0.057–46.750)1.681(1.096–2.577)*0.212(0.020–2.192)Complications NoReferenceReferenceReferenceReference Respiratory failure0.711(0.260–1.945)0.019(0.000–77,349.000)0.406(0.256–0.644)*2.742(0.773–9.734) Heart failure2.350(0.771–7.159)0.009(0.000–13,362.000)0.972(0.529–1.788)1.207(0.401–3.634) Sepsis4.526(1.204–17.000)*11.980(0.133–1079.000)1.587(0.688–3.660)1.368(0.119–15.720) MODS2.844(0.155–52.080)8.365(0.143–488.900)0.956(0.186–4.917)1.153(0.063–20.990)CCI, points1.072(0.730–1.576)1.054(0.320–3.475)0.614(0.495–0.763)*1.126(0.721–1.758)PO2/FiO21.112(0.701–1.765)1.573(0.877–2.821)1.225(0.998–1.503)0.583(0.316–1.074)APACHE II, points1.233(0.784–1.939)1.025(0.330–3.179)0.921(0.736–1.152)0.802(0.483–1.334)SOFA, points1.151(0.729–1.817)2.085(0.867–5.017)1.100(0.871–1.388)0.570(0.219–1.483)NRS2002, points0.736(0.436–1.243)0.686(0.035–13.400)0.674(0.545–0.834)*1.237(0.881–1.738)NUTRIC, points1.247(0.791–1.965)1.098(0.428–2.820)0.911(0.728–1.140)0.669(0.407–1.099)Length of stay prior to ICU transfer, days0.756(0.372–1.539)0.167(0.002–13.160)0.910(0.761–1.088)0.856(0.492–1.490)ICU length of stay, days0.825(0.453–1.500)1.239(0.329–4.660)0.694(0.544–0.884)*1.434(1.137–1.809)*Total hospital length of stay, days0.916(0.449–1.868)0.093(0.003–3.126)0.745(0.576–0.964)*1.230(0.920–1.644)FFM, kg1.091(0.713–1.670)0.975(0.370–2.568)1.288(1.057–1.569)*0.514(0.243–1.088)ASMM, kg1.266(0.785–2.040)1.200(0.646–2.229)1.333(1.054–1.685)*0.532(0.235–1.203)PhA3.476(1.585–7.627)*0.009(0.000–0.186)*2.415(1.214–4.804)*1.133(0.918–1.398)Handgrip strength, kg1.420(0.829–2.434)1.025(0.544–1.931)1.764(1.401–2.221)*0.272(0.087–0.847)*MRC, points1.170(0.785–1.745)1.596(0.497–5.129)1.751(1.398–2.194)*0.796(0.524–1.208)Duration of mechanical ventilation, h0.867(0.421–1.785)0.993(0.183–5.395)0.711(0.548–0.923)*0.871(0.537–1.413)Participation in early rehabilitation NoReferenceReferenceReferenceReference Yes0.558(0.207–1.508)0.569(0.027–11.920)0.923(0.595–1.431)0.324(0.069–1.527)Note: where HR > 1 indicates increased transition risk and HR < 1 indicates decreased transition risk. Asterisks indicate statistical significance: *P < 0.05, **P < 0.01, and ***P < 0.001.BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale.Fig. 4Forest plots for predictors of state transitions in the multistate Markov model. Multivariable analyses are presented for four transition pathways:(a) Non-frailty → Frailty, (b) Non-frailty → Death, (c) Frailty → Non-frailty, and (d) Frailty→Death. Hazard ratios (HRs) and 95% confidence intervals (CIs) are shown for demographic, clinical, and functional variables. Dashed vertical lines indicate HR = 1. BMI: Body Mass Index, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, PhA: Phase Angle, CCI: Charlson Comorbidity Index, NRS 2002: Nutritional Risk Screening 2002, MRC: the Medical Research Council scale.Fig. 4 Table 3Coefficient for univariate analysis of multistate Markov model [HR(95%CI)].Table 3VariableNon-frailty → FrailtyNon-frailty → DeathFrailty → Non-frailtyFrailty → DeathAge0.834(0.474–1.467)0.973(0.445–2.128)0.647(0.510–0.822)*1.925(0.718–5.163)Gender MaleReferenceReferenceReferenceReference Female0.361(0.103–1.266)0.000(0.000–23,489.000)0.512(0.324–0.809)*1.486(0.369–5.993)BMI, kg/m2 <18.5RefRefRefRef 18.5–23.91.02(0.63–1.65)0.18(0.03–0.97)*1.24(1.00–1.53)*0.86(0.44–1.65) ≥24.01.06(0.66–1.70)0.15(0.02–0.98)*1.27(1.01–1.58)*0.83(0.42–1.64)Marital status MarriedReferenceReferenceReferenceReference Divorced/Widowed0.804(0.094–6.886)0.007(0.000–18,616.000)0.442(0.174–1.122)2.281(0.567–9.173) Unmarried1.086(0.376–3.131)0.149(0.000–77,220.000)1.446(0.886–2.361)1.897(0.463–7.771)Education level Illiterate/Semi-illiterateReferenceReferenceReferenceReference Primary school0.234(0.060–0.913)*18,967.000(0.000–Inf)0.643(0.310–1.335)1.490(0.200–11.110) Junior to senior high0.500(0.148–1.688)14,653.000(0.000–Inf)1.374(0.673–2.805)2.809(0.490–16.100) College or higher0.000(0.000–Inf)0.002(0.000–Inf)0.549(0.245–1.230)4.347(0.790–23.920)Employment type prior to admission Non-manual labor/UnemployedReferenceReferenceReferenceReference Manual labor1.896(0.687–5.229)0.002(0.000–14,549.000)1.266(0.726–2.207)1.094(0.361–3.318)Smoking NoReferenceReferenceReferenceReference Yes1.346(0.501–3.611)29.330(0.000–25,976.000)1.039(0.642–1.681)0.477(0.059–3.860)Alcohol consumption NoReferenceReferenceReferenceReference Yes1.365(0.516–3.610)1.065(0.031–36.230)1.655(1.027–2.666)*1.514(0.389–5.882)Primary diagnosis Various surgeriesReferenceReferenceReferenceReference Acute disease0.994(0.136–7.294)0.000(0.000–Inf)1.168(0.378–3.616)0.000(0.000–Inf) Acute exacerbation of chronic disease0.141(0.018–1.122)2.172(0.288–16.360)0.408(0.254–0.654)*2.775(0.424–18.150) Sepsis2.606(0.671–10.120)0.000(0.000–Inf)1.476(0.598–3.640)6.851(1.053–44.570)* Other1.262(0.242–6.568)0.000(0.000–Inf)1.034(0.435–2.459)4.359(0.488–38.940)Postoperative status Non-surgicalReferenceReferenceReferenceReference Post-cardiac surgery1.029(0.344–3.078)0.559(0.015–21.170)2.108(1.274–3.489)*0.185(0.007–4.619) Post-non-cardiac surgery1.578(0.588–4.235)3.113(0.066–147.800)1.248(0.757–2.057)0.431(0.044–4.231)ICU patient source In-hospital general wardReferenceReferenceReferenceReference Transferred from other hospitals0.401(0.071–2.277)0.032(0.000–92,489.000)0.688(0.318–1.489)1.656(0.361–7.608) Postoperative transfer0.860(0.278–2.656)16.880(0.000–40,116.000)1.345(0.758–2.388)0.033(0.000–43,510.000) Direct admission or emergency room0.552(0.114–2.676)20.010(0.000–50,033.000)0.665(0.338–1.311)1.573(0.372–6.657)Planned ICU transfer NoReferenceReferenceReferenceReference Yes1.210(0.483–3.033)1.631(0.057–46.750)1.681(1.096–2.577)*0.212(0.020–2.192)Complications NoReferenceReferenceReferenceReference Respiratory failure0.711(0.260–1.945)0.019(0.000–77,349.000)0.406(0.256–0.644)*2.742(0.773–9.734) Heart failure2.350(0.771–7.159)0.009(0.000–13,362.000)0.972(0.529–1.788)1.207(0.401–3.634) Sepsis4.526(1.204–17.000)*11.980(0.133–1079.000)1.587(0.688–3.660)1.368(0.119–15.720) MODS2.844(0.155–52.080)8.365(0.143–488.900)0.956(0.186–4.917)1.153(0.063–20.990)CCI, points1.072(0.730–1.576)1.054(0.320–3.475)0.614(0.495–0.763)*1.126(0.721–1.758)PO2/FiO21.112(0.701–1.765)1.573(0.877–2.821)1.225(0.998–1.503)0.583(0.316–1.074)APACHE II, points1.233(0.784–1.939)1.025(0.330–3.179)0.921(0.736–1.152)0.802(0.483–1.334)SOFA, points1.151(0.729–1.817)2.085(0.867–5.017)1.100(0.871–1.388)0.570(0.219–1.483)NRS2002, points0.736(0.436–1.243)0.686(0.035–13.400)0.674(0.545–0.834)*1.237(0.881–1.738)NUTRIC, points1.247(0.791–1.965)1.098(0.428–2.820)0.911(0.728–1.140)0.669(0.407–1.099)Length of stay prior to ICU transfer, days0.756(0.372–1.539)0.167(0.002–13.160)0.910(0.761–1.088)0.856(0.492–1.490)ICU length of stay, days0.825(0.453–1.500)1.239(0.329–4.660)0.694(0.544–0.884)*1.434(1.137–1.809)*Total hospital length of stay, days0.916(0.449–1.868)0.093(0.003–3.126)0.745(0.576–0.964)*1.230(0.920–1.644)FFM, kg1.091(0.713–1.670)0.975(0.370–2.568)1.288(1.057–1.569)*0.514(0.243–1.088)ASMM, kg1.266(0.785–2.040)1.200(0.646–2.229)1.333(1.054–1.685)*0.532(0.235–1.203)PhA3.476(1.585–7.627)*0.009(0.000–0.186)*2.415(1.214–4.804)*1.133(0.918–1.398)Handgrip strength, kg1.420(0.829–2.434)1.025(0.544–1.931)1.764(1.401–2.221)*0.272(0.087–0.847)*MRC, points1.170(0.785–1.745)1.596(0.497–5.129)1.751(1.398–2.194)*0.796(0.524–1.208)Duration of mechanical ventilation, h0.867(0.421–1.785)0.993(0.183–5.395)0.711(0.548–0.923)*0.871(0.537–1.413)Participation in early rehabilitation NoReferenceReferenceReferenceReference Yes0.558(0.207–1.508)0.569(0.027–11.920)0.923(0.595–1.431)0.324(0.069–1.527)Note: where HR > 1 indicates increased transition risk and HR < 1 indicates decreased transition risk. Asterisks indicate statistical significance: *P < 0.05, **P < 0.01, and ***P < 0.001.BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale. Coefficient for univariate analysis of multistate Markov model [HR(95%CI)]. Note: where HR > 1 indicates increased transition risk and HR < 1 indicates decreased transition risk. Asterisks indicate statistical significance: *P < 0.05, **P < 0.01, and ***P < 0.001. BMI: Body Mass Index, MODS: Multiple Organ Dysfunction Syndrome, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, SMI: Skeletal Muscle Index, PhA: Phase Angle, CCI: Charlson Comorbidity Index, APACHE II: Acute Physiology and Chronic Health Evaluation II, SOFA: Sequential Organ Failure Assessment scores, NRS 2002: Nutritional Risk Screening 2002, NUTRIC: Nutrition Risk in Critically Ill scores, MRC: the Medical Research Council scale. Forest plots for predictors of state transitions in the multistate Markov model. Multivariable analyses are presented for four transition pathways:(a) Non-frailty → Frailty, (b) Non-frailty → Death, (c) Frailty → Non-frailty, and (d) Frailty→Death. Hazard ratios (HRs) and 95% confidence intervals (CIs) are shown for demographic, clinical, and functional variables. Dashed vertical lines indicate HR = 1. BMI: Body Mass Index, FFM: Fat-Free Mass, ASMM: Appendicular Skeletal Muscle Mass, PhA: Phase Angle, CCI: Charlson Comorbidity Index, NRS 2002: Nutritional Risk Screening 2002, MRC: the Medical Research Council scale. Discussion This multicenter cohort study is the first to characterize transition probabilities, intensities, and determinants of new-onset frailty after critical illness using a multistate Markov model. Three principal findings emerged. First, frailty was highly reversible: two-thirds of frail patients recovered within 6 months, with recovery occurring at more than three times the rate of progression from non-frailty to frailty. Second, frailty was strongly associated with mortality: the transition from frailty to death was more than twice as frequent as that from non-frailty. Third, patient characteristics strongly influenced these trajectories. Recovery was promoted by higher BMI, alcohol use, postoperative cardiac surgery, planned ICU admission, greater muscle mass and strength, and higher phase angle, whereas older age, female sex, greater comorbidity burden, malnutrition, and longer ICU or hospital stays impaired recovery. Notably, sepsis and prolonged ICU stay emerged as major determinants of adverse outcomes, driving both progression to frailty and death. Frailty trajectories after critical illness have also been described in prior cohorts [6,7,29], though findings varied by population and timing of assessment. In COVID-19 survivors, frailty prevalence at discharge was 27.8% [29], while another study reported ICU-acquired frailty rates of 25.7% at 3 months and 17.3% at 12 months [7]. Our study extends this literature in two ways. First, we show that frailty after critical illness is not only common but also highly reversible, with recovery occurring three times more often than progression. Second, by applying a multistate Markov model, we quantified both transition probabilities and intensities, offering a more precise depiction of frailty dynamics than prior descriptive reports. At ICU discharge, 86% of patients in our cohort were frail, a prevalence far exceeding earlier studies [6,7,10,11,17]. Several factors may explain the high frailty prevalence at ICU discharge. Pre-existing frailty was defined using functional status 2 months before ICU admission [7,10,11,17] rather than 2 weeks before admission [6], which may have reduced misclassification of acute illness–related decline as baseline frailty, lowered the apparent prevalence of pre-existing frailty, and contributed to the higher observed prevalence of new-onset frailty at ICU discharge. In addition, more than half of our cohort were postoperative patients, who were more likely to experience transient functional decline followed by recovery than patients with respiratory failure or shock [7,29]. Finally, the frailty captured in our cohort may not be equivalent to classical geriatric frailty. Because assessment occurred during ongoing recovery, the phenotype observed here likely overlapped with the physical domain of PICS and may be better understood as ICU-acquired frailty. Taken together, these factors may have contributed to both the high prevalence of frailty at ICU discharge and the marked recovery observed during follow-up. However, 6-month mortality after ICU discharge was relatively low in our cohort, possibly owing in part to the high proportion of planned ICU admissions. Nevertheless, we also demonstrate that new-onset frailty independently conferred a 2.6-fold higher risk of death, underscoring its prognostic significance beyond preexisting frailty [6,8,10,12,17]. Body composition emerged as a major determinant of frailty transitions. Higher BMI, fat-free mass, and appendicular muscle mass consistently favored recovery, highlighting the protective role of both muscle and adipose tissue. Prior studies similarly demonstrated differential risks across BMI categories: underweight was associated with greater odds of in-hospital death and frailty, whereas overweight and mild obesity were protective [30]. This “obesity paradox” though debated, suggests that moderate adipose reserves may confer metabolic and survival benefits. Mechanistically, muscle and adipose tissue function as active metabolic organs rather than passive energy stores [31]. Muscle catabolism supplies amino acids such as alanine and glutamine, while adipose tissue mobilizes fatty acids, glycerol, and lactate [31]. These substrates are converted in the liver into glucose and ketone bodies, fueling immune responses and tissue repair [31]. Adequate reserves may therefore enhance metabolic resilience, counteract catabolism, and facilitate functional recovery. Consistent with this, studies in sepsis survivors confirmed the obesity paradox but emphasized that muscle mass may be more critical than fat. Patients with a high psoas muscle index and low body fat had the lowest one-year mortality [32], underscoring the central role of muscle in post-ICU recovery. The role of alcohol consumption in frailty transitions remains debated. In our study, alcohol use was associated with greater likelihood of recovery, suggesting a potential protective effect. However, evidence from prior research is inconsistent. Experimental research indicates that alcohol disrupts glycolysis and mitochondrial metabolism, impairing muscle bioenergetics and predisposing to weakness [33]. Conversely, epidemiologic studies in older adults have reported a non-linear association: low-to-moderate consumption (<20 g/day) was linked to a modest reduction in frailty risk, whereas higher intake increased vulnerability [34]. Potential mechanisms underlying the apparent benefit include attenuation of systemic inflammation, improved lipid and glucose metabolism, increased adiponectin and HDL levels, and antioxidant properties of polyphenols present in alcoholic beverages [34]. These biological effects may partially explain why modest alcohol intake could facilitate functional recovery. Nonetheless, causality remains uncertain, and the discrepancy between experimental and clinical data highlights the need for further mechanistic and longitudinal studies. Cardiac surgery emerged as an important predictor of frailty recovery. Although major surgery often causes transient worsening due to perioperative trauma, anesthesia, and systemic inflammation, this decline is typically reversible: about 76% of patients deteriorate at one month but recover by two months [35]. In cardiac surgery, restoration of cardiac performance and circulatory function further accelerates recovery. Frail patients tend to show steady improvement, while non-frail patients often experience a short-term decline followed by recovery, usually returning to baseline within three months [36]. In this study, recovery was significantly greater in cardiac surgery patients than in those undergoing non-cardiac surgery or no surgery. This difference may reflect two factors. First, non-cardiac procedures were highly heterogeneous—including abdominal, thoracic, gynecologic, and urologic operations—limiting detection of consistent patterns. Second, many non-cardiac surgery patients had cancer and required adjuvant therapy, with added risks of recurrence. These factors increased catabolic stress, heightened complications, and, together with the metabolic burden of cancer, constrained functional recovery. By contrast, cardiac surgery directly restored circulatory function, creating a more favorable physiological environment for frailty reversal. Sepsis emerged as a dominant driver of frailty progression. In our multistate model, sepsis increased the risk of transition from non-frailty to frailty by more than fourfold and from frailty to death by nearly sevenfold. Mechanistically, sepsis promotes muscle catabolism through the JAK2/STAT3 pathway [37], impairs repair via PD-1 and ILC2-derived IL-13 signaling [38], and induces mitochondrial dysfunction with oxidative stress, activating proteolytic and autophagic pathways that accelerate muscle loss [39]. These biological mechanisms are supported by large clinical datasets: in a cohort of 8,518 sepsis survivors, 64.1% were frail at discharge and 54.6% experienced worsening frailty compared with baseline [40]. More recently, machine learning models based on routine clinical variables have shown strong predictive performance for post-sepsis frailty, providing a potential tool for early risk stratification and targeted intervention [40]. This study offers several clinical implications. First, ICU-acquired frailty is often reversible, but outcomes are strongly influenced by age, sex, comorbidities, and hospitalization length. Routine frailty screening at ICU discharge, followed by individualized rehabilitation integrating physical training, nutritional support, and psychosocial care, may maximize recovery [41]. Early rehabilitation limited to hospitalization showed little effect, underscoring the need for structured long-term programs after discharge. In addition, nutritional and functional reserves were critical: higher BMI, greater muscle mass, and preserved nutrition favored recovery. Practical interventions to preserve muscle include adequate protein intake, essential amino acids, creatine, vitamin D, β-hydroxy-β-methylbutyrate, and omega-3 fatty acids [42]. Finally, sepsis was a dominant driver of frailty progression. Although sepsis-induced muscle wasting is well recognized [43,44], its mechanisms remain incompletely understood, warranting further research to develop targeted therapies that interrupt the trajectory from sepsis to frailty and death. This study also has several limitations. First, although this was a multicenter study, the sample size was modest, and the CFS was therefore analyzed as a dichotomous rather than ordinal measure. In addition, the limited number of transitions to death may have reduced the precision of the estimates. Larger studies are needed to evaluate frailty across the full range of CFS categories and to better characterize transitions between frailty states. Second, follow-up was discontinued at 6 months because the loss to follow-up exceeded 30%, preventing assessment of long-term frailty trajectories. Third, although the CFS has been validated in older ICU populations [18,45], it remains a screening-based rather than comprehensive frailty measure. Moreover, although it has been applied in broader adult ICU populations [4,5], its psychometric performance in younger adults remains insufficiently established, and its use during acute critical illness and recovery is uncertain. Fourth, cohort heterogeneity should be considered when interpreting the findings. Approximately 60% of patients were admitted after surgery, whereas the remainder were admitted for medical conditions. Frailty transitions may vary between these populations, and larger studies are needed to evaluate them separately. Finally, predictors of frailty transitions were assessed only at baseline, although these factors are likely to evolve over time. Future studies should incorporate time-varying covariates into multistate Markov models to capture dynamic changes in risk and provide more accurate estimates of transition probabilities. Conclusion In this multicenter cohort of ICU survivors, ICU-acquired frailty proved to be a dynamic and largely reversible condition. Two-thirds of frail patients recovered within 6 months, with recovery occurring more than three times as often as progression, whereas frailty carried a 2.5-fold higher risk of death compared with non-frailty. Sepsis-related complications markedly increased the risk of frailty onset and mortality, while greater BMI, muscle mass, handgrip strength, and phase angle consistently favored recovery and survival. Conversely, older age, female sex, comorbidity burden, malnutrition, and prolonged ICU or hospital stay impaired recovery. These findings underscore the importance of routine frailty screening at ICU discharge and highlight the potential benefits of multimodal rehabilitation, optimized nutrition, and targeted infection control strategies to improve long-term outcomes in critically ill patients. CRediT authorship contribution statement TTW: Conceptualization, Methodology, Writing - original draft preparation, Writing - review & editing, Project administration RJL and HL: Conceptualization, Methodology YQW and LS: Investigation TTW: data analysis and interpretation. All authors mentioned above have granted final approval for the submitted version. Ethics approval and consent to participate This study was conducted in accordance with the principles of the Declaration of Helsinki. Prior to enrollment, all participants received a detailed explanation of the study’s purpose, procedures, and content, and provided written informed consent. Participant confidentiality was strictly maintained throughout. Ethical approval was obtained from the Ethics Committee of Fujian Provincial Hospital (Approval No. K2021-04-004). Declaration of Generative AI and AI-assisted technologies in the writing process During the preparation of this manuscript, generative AI (ChatGPT) was used to assist in improving the clarity and linguistic quality of the text (approximately 15%). The author has reviewed, edited, and approved the final manuscript and takes full responsibility for its content. Funding This study was supported by the Joint Funds for the Innovation of Science and Technology, Fujian Province (Grant No. 2023Y9326), the General Program of the Natural Science Foundation of Fujian Provincial Department of Science and Technology (Grant No. 2025J01104), and the Talent Recruitment Scientific Research Project at the First Affiliated Hospital of Fujian Medical University (Grant No. YJRC4219). The funding agencies had no role in the study design, data collection, analysis, manuscript preparation, or the decision to publish. Availability of data and materials The datasets generated and/or analysed during the proposed study will be available from the corresponding author upon reasonable request Declaration of competing interest The authors declare that they have no competing interests.