Association of synergistic trajectories between early mobilization potential and actual implementation with clinical outcomes among ICU patients with pneumonia: a multivariate latent class mixed model analysis

Abstract

Background

Although early mobilization is a key intervention in the management of patients admitted to the intensive care unit (ICU), its effects on patients remain controversial in previous studies. This study aimed to explore the synergistic trajectories of early mobilization potential and actual implementation in ICU patients with pneumonia and their impacts on clinical outcomes.

Methods

Data were extracted from the Medical Information Mart for Intensive Care (MIMIC) database. Multivariate latent class mixed model (MLCMM) was adopted to identify mobilization trajectories within the first 72 hours of ICU admission among patients with pneumonia and perform subgroup stratification. Linear regression analysis was applied to investigate the association between early mobilization and length of stay (LOS) in the ICU across each subgroup. Subgroup analyses were conducted to explore the heterogeneity of the above association in populations with different characteristics. Kaplan-Meier curves were used to depict survival disparities, and Sankey diagrams were applied to illustrate discharge destinations.

Results

A total of 3,226 patients with pneumonia were included. In the fully adjusted model, Class 2 showed no statistically significant difference in ICU LOS relative to Class 1 [β=−0.04, 95% confidence interval (CI): −0.11 to 0.04, P=0.35]. Class 3 (β=−0.43, 95% CI: −0.51 to −0.35, P<0.001) and Class 4 (β=−0.42, 95% CI: −0.50 to −0.33, P<0.001) were associated with shorter ICU LOS. Sensitivity analyses yielded results consistent with the primary analysis. Significant interaction effects were observed in the subgroups of age (P for interaction =0.01) and mechanical ventilation (P for interaction <0.001) on the above association. Compared with Class 1, the other three subgroups were associated with higher survival probabilities (log-rank P<0.001). The Sankey diagram revealed that Class 1 was characterized by adverse clinical prognoses.

Conclusions

This study demonstrates that low-level stable early mobilization is associated with longer ICU LOS and lower survival rate in patients with pneumonia. These findings suggest that distinct mobilization trajectories may reflect patients’ early recovery status.

Highlight box.

Key findings

• Four distinct early mobilization trajectories were identified among intensive care unit (ICU) patients with pneumonia. Compared with Class 1 (low-level stable pattern), Class 3 (early active maintenance pattern) and Class 4 (rapidly progressive pattern) were associated with shorter ICU length of stay and lower mortality.

What is known and what is new?

• Early mobilization is an important component of ICU care, but evidence regarding dynamic mobilization patterns in patients with pneumonia remains limited.

• This study jointly characterized mobilization potential and actual implementation using repeated assessments and identified distinct early mobilization trajectories associated with clinical outcomes.

What is the implication, and what should change now?

• Early mobilization trajectories may reflect patients’ early mobility and recovery status. Further prospective studies are needed to determine whether trajectory-based assessment can support individualized mobilization strategies in ICU practice.

Introduction

Pneumonia constitutes a prevalent and critical clinical issue among patients admitted to the intensive care unit (ICU) (1). For pneumonia patients receiving ICU care, multiple factors, including acute disease severity, demands for respiratory and circulatory support, sedation and analgesia, delirium, and invasive monitoring devices, frequently lead to varying degrees of immobilization (2). Persistent low physical activity or prolonged bed rest suppresses muscle protein anabolism and amplifies catabolic turnover, which may further trigger ICU-acquired sarcopenia, intensive care unit-acquired weakness, and other complications (3,4). These metabolic derangements are often accompanied by impaired respiratory muscle function and global physical dysfunction, and correlate with prolonged length of stay (LOS), deteriorated long-term quality of life, and delayed rehabilitation progress (5,6). Accordingly, alongside targeted anti-pneumonia therapy, minimizing prolonged bed rest is strongly linked to improved long-term quality of life for these patients.

Early mobilization represents a core intervention in the management of ICU patients (7). Results of multiple randomized controlled trials have shown that early mobilization intervention for ICU patients can improve muscle strength, shorten the duration of mechanical ventilation and LOS, and may reduce functional disability after discharge (8,9). However, one study reported that increased early mobilization did not significantly increase the number of days patients were alive and out of the ICU, and might be associated with more adverse events (10). These results suggest that the clinical benefits of early mobilization still need further research and evaluation. Furthermore, studies investigating early mobilization in critically ill patients with pneumonia have gradually emerged, yet relevant evidence remains scarce. Findings from randomized controlled trials indicate that proactive or combined physical therapy regimens incorporating early mobilization may improve respiratory function in ICU patients with bilateral pneumonia (11). Case series have demonstrated that physical therapy including early mobilization plays a vital role in the management of patients with coronavirus disease 2019 (COVID-19) pneumonia (12). Nevertheless, these investigations primarily focus on the feasibility of implementing early mobilization and are insufficient to establish independent associations between distinct mobilization patterns and patient clinical outcomes. More importantly, there may be a discrepancy between the assessment of patients’ mobilization ability and the actual implementation of mobilization (13). For example, some patients who meet the assessment criteria for mobilization ability fail to receive timely corresponding interventions, while mobilization interventions delivered to other patients may not be based on systematic evaluations of their mobilization capacity and safety, which may affect the efficiency and quality of nursing work.

Therefore, based on the Medical Information Mart for Intensive Care (MIMIC) database, this study explored the association between the synergistic trajectories of early mobilization potential and actual implementation and the LOS and prognosis of ICU patients with pneumonia, aiming to provide a scientific basis for developing safer and more effective nursing protocols for ICU patients with pneumonia. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0939/rc).

Methods

Data source

Data for this study were extracted from MIMIC-IV, a public database developed by the Massachusetts Institute of Technology, containing information on inpatients at the Beth Israel Deaconess Medical Center from 2008 to 2019. All information in the database was de-identified, so no additional ethical approval was required. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

A total of 3,957 patients with pneumonia were initially enrolled. The exclusion criteria were as follows: (I) patients with ICU LOS less than 24 hours (n=687); (II) patients were excluded if they had persistent Braden Activity subscale scores of 1 (bedfast, confined to bed) across multiple time points within the first 72 hours of ICU admission, which precluded analyzable dynamic mobilization potential trajectories (n=19), or if their mobilization/ambulation capacity could not be assessed (n=9); (III) patients who did not undergo mobility assessment within 72 hours of ICU admission (n=16). Finally, 3,226 patients with pneumonia were included in the final analysis (Figure 1).

Figure 1.

Flowchart of patient inclusion and exclusion. ICU, intensive care unit; MIMIC, Medical Information Mart for Intensive Care.

Disease definition

Pneumonia was identified using the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9 and ICD-10). The operational definition was as follows: among adult patients admitted to the ICU for the first time, at least one pneumonia-related ICD code appeared within the top three discharge diagnostic sequences (diagnoses_icd.seq_num ≤3); detailed codes are presented in Table S1 (14). This algorithm relies on diagnostic ranking rather than a 3-day post-admission cutoff and contains no timing information regarding pneumonia onset, making it unable to distinguish community-acquired, hospital-acquired, or ventilator-associated pneumonia.

Exposures and outcomes

All exposure data were extracted from nursing flow sheets within the electronic health records. The Braden Activity subscale and Johns Hopkins Highest Level of Mobility (JH-HLM)/Activity records were selected because both contain repeated structured documentation in MIMIC-IV with complementary interpretations: the former reflects mobilization potential, whereas the latter captures actual mobilization delivery. No other ICU mobility scales included repeated longitudinal records covering the full cohort within the dataset locked for this analysis. (I) Mobilization potential assessment (itemid =224057) utilized the Activity subscale of the Braden Pressure Injury Risk Scale, coded on a 1–4 ordinal scale, where itemid denotes unique measurement identifiers in the MIMIC database. (II) Actual mobilization levels were prioritized using the 1–8 scoring system of JH-HLM (itemid =229321). Missing JH-HLM values were mapped from contemporaneous Activity records (itemid =224084) following pre-specified rules. Every entry contained an exact charttime timestamp, which was converted to hours elapsed since ICU admission while retaining the original recorded time points without alignment to fixed time windows. This approach incorporated all available records generated within the first 72 hours post-ICU admission; however, documentation frequency and timing were subject to fluctuations driven by patient clinical status and routine nursing workflows.

The primary outcome was ICU LOS. Secondary outcomes encompassed ICU mortality, in-hospital mortality, and 28-day mortality.

Covariates

Covariates were selected based on clinical relevance, temporal sequence, and their potential confounding effects on the trajectory-outcome association, rather than univariate analyses or P values for between-group differences. Model 2 was adjusted for age, sex, and race/ethnicity; Model 3 additionally incorporated the Sequential Organ Failure Assessment (SOFA) score, Charlson Comorbidity Index, mechanical ventilation, vasopressor administration, and delirium (Table 1).

Table 1. Baseline characteristics of patients across different early activity trajectory subgroups.

Characteristic	Overall (N=3,226)	Class 1 (N=1,777)	Class 2 (N=533)	Class 3 (N=505)	Class 4 (N=411)	P value
Age, years	66.4±16.6	67.7±16.3	67.9±15.9	62.7±17.0	63.3±17.4	<0.001
Sex (male)	1,739 [54]	950 [53]	283 [53]	277 [55]	229 [56]	0.80
Race						<0.001
Black	321 [10.0]	168 [9.5]	52 [9.8]	59 [12]	42 [10]
Other	736 [23]	459 [26]	107 [20]	99 [20]	71 [17]
White	2,169 [67]	1,150 [65]	374 [70]	347 [69]	298 [73]
Surgical patients	1,333 [41]	731 [41]	199 [37]	218 [43]	185 [45]	0.09
Mechanical ventilation	1,086 [34]	825 [46]	133 [25]	46 [9.1]	82 [20]	<0.001
Vasopressor drug	611 [19]	415 [23]	90 [17]	62 [12]	44 [11]	<0.001
Sedative drug	1,623 [50]	1,042 [59]	259 [49]	159 [31]	163 [40]	<0.001
Analgesic drug	1,527 [47]	1,047 [59]	228 [43]	122 [24]	130 [32]	<0.001
Restraint drug	2,931 [91]	1,592 [90]	492 [92]	471 [93]	376 [91]	0.04
CCI	5 [3, 8]	5 [3, 8]	6 [4, 8]	5 [3, 8]	5 [3, 7]	0.02
SOFA	3 [1, 5]	4 [2, 6]	3 [1, 4]	2 [1, 4]	2 [1, 4]	<0.001
APS III	47.2±19.1	51.2±21.3	45.6±16.2	40.4±13.7	40.3±13.5	<0.001
GCS	15.0 [14.0, 15.0]	15.0 [13.0, 15.0]	15.0 [14.0, 15.0]	15.0 [14.0, 15.0]	15.0 [14.0, 15.0]	<0.001
Heart rate, beats/min	95.4±21.2	93.4±21.2	97.2±21.2	97.8±21.2	97.4±20.3	0.008
SBP, mmHg	126.2±24.1	124.9±24.5	124.7±23.0	130.6±24.4	128.1±23.6	0.20
DBP, mmHg	68.9±17.9	68.2±18.2	68.5±17.0	71.9±16.7	69.1±19.3	0.20
Resp rate, breaths/min	21.6±6.3	21.2±6.2	21.6±6.2	22.7±6.5	22.5±6.4	<0.001
Temperature, ℃	36.9±0.8	36.8±0.9	36.9±0.6	37.0±0.7	36.9±0.8	0.10
SpO2, %	95.3±4.3	95.6±4.5	94.9±4.6	94.9±3.9	95.2±3.9	0.004
Delirium	256 [7.9]	177 [10.0]	43 [8.1]	26 [5.1]	10 [2.4]	<0.001

Data are presented as mean ± standard deviation, n [%], or median [interquartile range]. Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. APS III, Acute Physiology Score III; CCI, Charlson Comorbidity Index; DBP, diastolic blood pressure; GCS, Glasgow Coma Scale; SBP, systolic blood pressure; SOFA, Sequential Organ Failure Assessment; SpO₂, saturation of peripheral oxygen.

Statistical analysis

A multivariate latent class mixed model (MLCMM) was constructed via the latent process approach to characterize dynamic changes in mobilization status within the first 72 hours of ICU admission (15). The model incorporated all available mobilization records and fitted nonlinear trends using second-order time polynomials. Indicator-specific random effects were applied to address heterogeneity in baseline levels and measurement errors across distinct assessment scales. The derived data-driven trajectory classes do not represent distinct clinical phenotypes and may simultaneously reflect disease severity and early recovery status. The MLCMM was built using mobilization levels and Braden subscale scores. Model fit metrics demonstrated that the four-class solution yielded the lowest Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values. While the two-class and three-class models failed to converge, the four-class model converged smoothly after 94 iterations with superior numerical stability. In contrast, the five-class model exhibited elevated BIC values and non-convergence, indicating a high risk of overfitting (Table S2). The four-class model was therefore selected as the optimal classification scheme. Between-group comparisons of baseline continuous variables were performed using analysis of variance or the Kruskal-Wallis test according to data distribution, whereas categorical variables were compared via the χ² test.

Given the right-skewed distribution of ICU LOS, linear regression models were constructed with log(ICU LOS) as the dependent variable. Model 2 was adjusted for age, sex, and race/ethnicity; Model 3 further incorporated the SOFA score, Charlson Comorbidity Index, mechanical ventilation, vasopressor use, and delirium. All pairwise comparisons across trajectory classes were conducted based on Model 3, with multiple testing corrected by the Holm procedure. Subgroup analyses were exploratory in nature. P values for interaction terms were adjusted using the Benjamini-Hochberg method. Subgroup modifiers were selected based on clinical relevance and their potential to modify early mobilization delivery and clinical outcomes, including sex, age, SOFA score, mechanical ventilation, and delirium. Additional sensitivity analyses included landmark analyses restricted to patients who remained in the ICU for at least 72 hours after admission, with adjustment for the timing of initial Braden and JH-HLM/Activity assessments and a constraint that both baseline assessments were completed within the first 24 hours post-admission. Kaplan-Meier curves and log-rank tests were used to visualize and compare survival disparities. Sankey diagrams were generated to illustrate frequency distributions linking mobilization trajectory classes to discharge destinations.

All analyses were performed using R software (Version 4.5.2), and a P value <0.05 was considered statistically significant.

Results

Cohort characteristics of all included participants

A total of 3,226 patients were ultimately included in this study, with a mean age of 66.4±16.6 years; 1,739 (53.9%) were male. The numbers and proportions of patients assigned to Class 1, Class 2, Class 3, and Class 4 were 1,777 (55.1%), 533 (16.5%), 505 (15.7%), and 411 (12.7%), respectively (Table 1).

Latent class trajectory changes of early mobilization level in ICU patients with pneumonia

Four distinct evolution patterns of early mobilization were identified based on the dual-trajectory model. In terms of Braden Activity, Class 1 showed a low-level stable pattern with almost no improvement within 72 hours of ICU admission. Class 2 corresponds to the slowly rising trajectory subgroup, characterized by a slight increase in activity level over time. Class 3 maintained a moderate-to-high activity level from an early stage, defined as the early active maintenance pattern, indicating the initiation of relatively proactive activity management immediately after ICU admission. Class 4 was categorized as the rapidly progressive pattern; although its activity level decreased slightly within 0–24 hours, it remained consistently higher than that of Class 1 and Class 2, and increased significantly during 48–72 hours, ultimately reaching the highest level (Figure 2A).

Figure 2.

Latent class trajectory changes of early activity levels within 72 hours after ICU admission. (A) Braden Activity scores (1–4 points). (B) JH-HLM scores (1–8 points). The x-axis represents time after ICU admission (hours), and the y-axis represents the corresponding scale scores. Different colors indicate distinct latent classes (Class 1–4). The shaded areas represent the 95% confidence intervals of the model estimates. Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. ICU, intensive care unit; JH-HLM, Johns Hopkins Highest Level of Mobility.

The trajectory trend of JH-HLM was basically consistent with that of Braden. Class 3 and Class 4 exhibited relatively high or continuously rising functional status, among which the mobilization level of Class 4 at 48 hours was significantly better than that of the other groups (Figure 2B).

Baseline characteristics and distribution of initial mobilization grades

Baseline characteristic comparisons revealed that patients in Class 1 exhibited higher SOFA and Acute Physiology Score III (APS III) scores, alongside greater proportions receiving mechanical ventilation, vasopressors, sedatives, and analgesics, as well as the highest prevalence of delirium (all P<0.001). Additionally, Class 4 had a higher rate of mechanical ventilation than Class 3 (20.0% vs. 9.1%), with correspondingly higher utilization of sedative and analgesic agents (Table 1).

Records of mobilization assessments demonstrated that within the first 72 hours after ICU admission, each patient had a median of 8 Braden Activity subscale assessments [interquartile range (IQR), 6–11] and 25 JH-HLM/Activity entries (IQR, 16–31), amounting to a combined median of 35 total mobilization records (IQR, 23–42) distributed across a median of 30 distinct time points (IQR, 20–36). The median timing of the initial Braden and JH-HLM/Activity assessments was 1.05 hours (IQR, 0.50–2.64 hours) and 0.90 hours (IQR, 0.42–1.97 hours) post-admission, respectively. Initial assessments for both tools were completed within the first 24 hours of ICU admission in 99.8% of patients (Table S3).

In the initial Braden Activity subscale assessment, Class 1 was mainly rated as “bedfast” (level 1); Class 2 was still dominated by bedrest, but the proportion of “activities near wheelchairs” increased; Class 3 and Class 4 had a higher proportion of higher-level mobilization such as “occasional walk” or “frequent walks” in the first assessment, with the highest proportion of high-level mobilization in Class 4 (Figure 3A). Findings from the first actual mobilization record (JH-HLM) were basically consistent with the Braden Activity subscale assessments (Figure 3B).

Figure 3.

Distribution of Braden assessment grades at ICU admission and first activity records across different early activity trajectory subgroups. (A) Distribution of the first Braden Activity assessment grades after ICU admission (1= bedrest, 2= activities near wheelchairs, 3= occasional walking, 4= frequent walking). (B) Distribution of the first recorded JH-HLM activity grades (1= bedrest, 2= in-bed activity, 3= sitting at the bedside, 4= sitting in a chair, 5= standing for >1 min, 6= walking for >10 steps, 7= walking for >25 feet, 8= walking for >250 feet). Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. ICU, intensive care unit; JH-HLM, Johns Hopkins Highest Level of Mobility.

Association between different early mobilization trajectories and ICU LOS

Multivariate linear regression analysis

Multivariate linear regression analyses revealed that in Model 3, Class 2 showed no statistically significant difference in log(ICU LOS) relative to Class 1 (P=0.35). By contrast, Class 3 [β=−0.43, 95% confidence interval (CI): −0.51 to −0.35, P<0.001] and Class 4 (β=−0.42, 95% CI: −0.50 to −0.33, P<0.001) were independently associated with shorter ICU LOS (Table 2). Pairwise post-hoc comparisons with Holm correction demonstrated that Class 2 had longer log(ICU LOS) than both Class 3 and Class 4 (all Holm-adjusted P<0.001), whereas no significant difference was detected between Class 3 and Class 4 (Holm-adjusted P=0.75) (Table S4).

Table 2. Multivariable linear regression analysis of early activity trajectory subgroups and ICU length of stay [log(ICU LOS)].

Class	Model 1		Model 2		Model 3
	β (95% CI)	P value	β (95% CI)	P value	β (95% CI)	P value
Class 1	Reference		Reference		Reference
Class 2	−0.16 (−0.23, −0.08)	<0.001	−0.15 (−0.23, −0.07)	<0.001	−0.04 (−0.11, 0.04)	0.35
Class 3	−0.62 (−0.70, −0.54)	<0.001	−0.64 (−0.72, −0.56)	<0.001	−0.43 (−0.51, −0.35)	<0.001
Class 4	−0.57 (−0.65, −0.48)	<0.001	−0.58 (−0.67, −0.49)	<0.001	−0.42 (−0.50, −0.33)	<0.001

Results are presented as regression coefficients β (95% CI), with Class 1 serving as the reference group. The dependent variable is the natural logarithm of ICU LOS [log(ICU LOS)]. Model 1 is the unadjusted model; Model 2 adjusts for age, sex, and race on the basis of Model 1; Model 3 further adjusts for disease severity and clinical intervention factors, including SOFA score, CCI, mechanical ventilation, vasopressor use, and delirium status. All P values correspond to two-sided tests, with the statistical significance level set at 0.05. Multicollinearity diagnostics were performed for all covariates prior to inclusion in the final model; all VIFs were less than 5, indicating no substantial multicollinearity (the number of adjusted covariates was minimized due to small sample sizes in several subgroups). Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. CCI, Charlson Comorbidity Index; CI, confidence interval; ICU, intensive care unit; LOS, length of stay; SOFA, Sequential Organ Failure Assessment; VIF, variance inflation factor.

Sensitivity analyses yielded consistent findings with the primary regression model after additional adjustment for the timing of initial Braden Activity and JH-HLM/Activity assessments, and after restricting the cohort to patients with both baseline assessments completed within the first 24 hours of ICU admission. A 72-hour landmark analysis limited to patients remaining hospitalized in the ICU at the 72-hour time point further confirmed that Class 3 and Class 4 were associated with shorter ICU LOS compared with Class 1 (P<0.05) (Table S5).

Subgroup analysis

Subgroup analysis showed that the association between early mobilization trajectory classes and log(ICU LOS) had significant interaction effects in the subgroups of age (P for interaction =0.01) and mechanical ventilation (P for interaction <0.001) (Table 3).

Table 3. Association analysis between early activity trajectory and ICU length of stay [log(ICU LOS)] in different subgroups.

Subgroup	Term	Estimate	P value	P for interaction
Sex				0.51
Female	Class 1	Reference
	Class 2	0.00 (−0.11, 0.10)	0.94
	Class 3	−0.44 (−0.55, −0.32)	<0.001
	Class 4	−0.40 (−0.52, −0.27)	<0.001
Male	Class 1	Reference
	Class 2	−0.07 (−0.17, 0.04)	0.21
	Class 3	−0.43 (−0.54, −0.32)	<0.001
	Class 4	−0.43 (−0.55, −0.32)	<0.001
Age				0.01
≥60 years	Class 1	Reference
	Class 2	−0.01 (−0.10, 0.08)	0.83
	Class 3	−0.34 (−0.44, −0.24)	<0.001
	Class 4	−0.38 (−0.49, −0.27)	<0.001
<60 years	Class 1	Reference
	Class 2	−0.09 (−0.23, 0.05)	0.19
	Class 3	−0.57 (−0.71, −0.44)	<0.001
	Class 4	−0.47 (−0.62, −0.33)	<0.001
Delirium				0.08
No	Class 1	Reference
	Class 2	−0.05 (−0.13, 0.03)	0.22
	Class 3	−0.42 (−0.50, −0.34)	<0.001
	Class 4	−0.40 (−0.49, −0.32)	<0.001
Yes	Class 1	Reference
	Class 2	0.09 (−0.17, 0.35)	0.51
	Class 3	−0.65 (−0.98, −0.32)	<0.001
	Class 4	−0.90 (−1.40, −0.40)	<0.001
Mechanical ventilation				<0.001
No	Class 1	Reference
	Class 2	0.10 (0.02, 0.19)	0.02
	Class 3	−0.30 (−0.39, −0.22)	<0.001
	Class 4	−0.27 (−0.36, −0.18)	<0.001
Yes	Class 1	Reference
	Class 2	−0.32 (−0.47, −0.17)	<0.001
	Class 3	−0.91 (−1.15, −0.67)	<0.001
	Class 4	−0.71 (−0.90, −0.53)	<0.001
SOFA				0.15
>3	Class 1	Reference
	Class 2	−0.31 (−0.42, −0.20)	<0.001
	Class 3	−0.45 (−0.57, −0.33)	<0.001
	Class 4	−0.53 (−0.66, −0.40)	<0.001
<3	Class 1	Reference
	Class 2	−0.02 (−0.12, 0.09)	0.72
	Class 3	−0.39 (−0.50, −0.29)	<0.001
	Class 4	−0.05 (−0.15, 0.06)	0.39

Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. ICU, intensive care unit; LOS, length of stay; SOFA, Sequential Organ Failure Assessment.

In both patients with pneumonia aged <60 years and ≥60 years, compared with Class 1, Class 3 (β_<60=−0.57, P<0.001; β_≥60=−0.34, P<0.001) and Class 4 (β_<60 =−0.47, P<0.001; β_≥60 =−0.38, P<0.001) were associated with shorter ICU LOS, with a higher effect size in patients aged <60 years. Regardless of mechanical ventilation status, compared with Class 1, Class 3 (β_{non-mechanical ventilation} =−0.30, P<0.001; β_{mechanical ventilation} =−0.91, P<0.001) and Class 4 (β_{non-mechanical ventilation} =−0.27, P<0.001; β_{mechanical ventilation} =−0.71, P<0.001) were associated with shorter ICU LOS, with a higher effect size in the subgroup receiving mechanical ventilation. In addition, in the subgroup receiving mechanical ventilation, Class 2 was also associated with shorter ICU LOS compared with Class 1 (β_{mechanical ventilation} =−0.32, P<0.001) (Table 3).

Association between different early mobilization trajectories and mortality

Survival analysis showed that there were significant differences in the 7-day survival probabilities of each mobilization trajectory subgroup in the ICU (P-log-rank <0.001). Among them, Class 1 had the lowest survival probability, while Class 3 and Class 4 maintained a high survival probability throughout the period (Figure S1).

In terms of ICU mortality, Class 1 had the highest ICU mortality (15%); Class 2 was 6.8%; Class 3 and Class 4 were even lower, at 2.2% and 2.4%, respectively (P<0.001). Hospital mortality and 28-day mortality showed similar trends (all P<0.001) (Figure 4).

Figure 4.

Comparison of ICU mortality, hospital mortality, and 28-day mortality among different early activity trajectory subgroups. The bar chart shows the incidence rates of different mortality outcomes in the four early activity trajectory subgroups (Class 1–4). The y-axis represents the mortality percentage (%). Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. ICU, intensive care unit.

Evolution of final clinical outcomes of early mobilization trajectories

The Sankey diagram intuitively showed the heterogeneous pathways of patients’ evolution from early mobilization trajectories to final clinical outcomes. In Class 1, 387 patients (21.8%) had an outcome of in-hospital death and 68 (3.8%) were discharged to hospice care; these proportions were higher than those observed in the other trajectory subgroups. Class 2 exhibited the most diverse discharge pathways: aside from 60 patients (11.3%) who died, 256 patients (48.0%) were transferred to skilled nursing or rehabilitation facilities. Class 3 and Class 4 demonstrated superior rehabilitation outcomes, with 65.1% of Class 3 and 66.7% of Class 4 patients discharged home (Figure 5, Table S6).

Figure 5.

Sankey diagram of early ICU activity trajectory classes and the evolution of patient clinical outcomes. The Sankey diagram adheres to the principle of flow balance, i.e., the sum of the input ends of each node is consistent with the sum of its output ends, and the proportional width of the links intuitively reflects the frequency distribution of transitions between categories. The colored blocks on the left represent the four early activity trajectory classes (Class 1–4) identified by the LCMM; the colored blocks on the right represent the final destinations of patients after ICU discharge. The width of the flow links is proportional to the patient proportion. Class 1, low-level stable pattern; Class 2, slowly rising pattern; Class 3, early active maintenance pattern; Class 4, rapidly progressive pattern. ICU, intensive care unit; LCMM, latent class mixed model; LTAC, long-term acute care; SNF, skilled nursing facility.

Discussion

This study found that compared with Class 1, patients with pneumonia in Class 3 and Class 4 were associated with shorter ICU LOS and lower mortality. These findings illustrate substantial heterogeneity in early mobilization patterns and corresponding clinical outcome distributions among pneumonia patients admitted to the ICU.

Over half of the patients in this study (55%) were categorized into Class 1, indicating that persistent hypoactivity in the early phase following ICU admission is prevalent among patients with pneumonia. Moreover, patients in this class exhibited greater baseline disease severity and higher requirements for organ support. Compared with patients in other classes, Class 1 patients had higher SOFA and APS III scores, greater reliance on mechanical ventilation, vasoactive agents and sedatives, alongside a higher incidence of delirium. These factors may directly restrict patients’ mobility potential and the delivery of mobilization interventions (7,16,17); they are strongly correlated with prolonged ICU LOS and elevated mortality risk (18,19). Accordingly, the stable low-mobility trajectory of Class 1 and its associated unfavorable clinical outcomes may be jointly driven by more severe illness, greater organ support requirements, and slower early recovery, rather than simply attributable to reduced mobility levels alone.

In addition, Class 3 and Class 4 were associated with more favorable prognoses. Class 3 was characterized by sustained moderate mobility with intermittent ambulation, whereas Class 4 presented an initial decline or persistently low mobility followed by a rapid rebound, representing a state of swift mobility improvement in affected patients. Previous studies have similarly demonstrated that such patterns correlate with shorter ICU LOS and lower mortality rates (20,21). Nevertheless, the mobility trajectories identified in the present study do not constitute validated rehabilitation phenotypes or strata defined by mobilization intervention intensity. As such, these trajectories may merely serve as markers of milder disease severity, faster recovery, and greater opportunities for mobilization, rather than reflecting clinical benefits conferred by mobilization itself. Notably, Class 4 patients demonstrated higher rates of mechanical ventilation, sedative, and analgesic administration relative to Class 3 patients. This suggests that Class 4 patients may experience more mobility limitations stemming from respiratory support, sedation-analgesia regimens, and clinical care workflows in the early ICU period, which could partially account for their trajectory featuring an initial drop or sustained low mobility followed by a rapid increase. This remains a hypothesis proposed based on clinical characteristics, however, causality between mechanical ventilation or sedation-analgesia and the observed trajectory shifts cannot be established from these data.

Prior literature may offer partial biological mechanistic explanations for the association between early mobility trajectories and clinical outcomes observed among ICU-admitted pneumonia patients in this investigation. At the respiratory and pulmonary function level, early mobilization can enhance the strength and coordination of respiratory muscles (22,23), thereby directly counteracting the pathological process of pneumonia, reducing the risk of atelectasis, and accelerating the control and repair of pulmonary infection. At the systemic metabolic and inflammatory regulation level, early mobilization can stimulate skeletal muscle to release anti-inflammatory myokines (e.g., interleukin-6), downregulate excessive systemic inflammatory responses, and thus alleviate pneumonia-related immune pathological damage (24). At the neuromuscular and functional maintenance level, early mobilization maintains muscle mass and strength through mechanical load and neural stimulation, laying a foundation for early weaning and recovery of daily functions (25).

Exploratory stratified analyses revealed divergent associations between distinct early mobility trajectories and ICU LOS across age and mechanical ventilation subgroups. Among pneumonia patients younger than 60 years and those aged 60 years or older, both Class 3 and Class 4 were correlated with shorter ICU LOS relative to Class 1; however, the absolute magnitude of the estimated association was larger in patients aged <60 years. This discrepancy may stem from superior cardiopulmonary function, greater muscle reserve, and enhanced anabolic capacity in younger individuals (26), physiological features linked to accelerated physical recovery (27). Nevertheless, this study did not directly quantify muscle reserve, metabolic status, or functional improvements following mobilization; thus, the above interpretations remain speculative physiological hypotheses. Class 3 and Class 4 were associated with abbreviated ICU LOS regardless of mechanical ventilation status, with stronger effect estimates observed in mechanically ventilated patients. Within the mechanically ventilated subgroup, Class 2 also correlated with shorter ICU hospitalization duration. Given the exploratory nature of subgroup analyses, these findings constitute preliminary effect modification analyses only. Prospective studies specifically enrolling patients with ventilator-associated pneumonia are warranted to validate these observations.

In summary, this study identified pronounced dynamic heterogeneity in early mobility patterns among pneumonia patients admitted to the ICU. Mobility trajectories constructed from repeated assessments may complement single-timepoint mobility scores. Nonetheless, these trajectories merely reflect patients’ early mobility status and recovery course; prospective investigations are required to verify whether they can guide individualized clinical interventions.

This work has several limitations. First, the mobility trajectories identified may partially reflect baseline disease severity and organ support requirements rather than independent rehabilitation phenotypes. As a retrospective observational analysis, this study only establishes associations between mobility trajectories and clinical outcomes and cannot infer causal relationships or intervention efficacy. Second, although multivariate models were adjusted for multiple markers of illness severity and therapeutic exposures, mobility trajectories were built using assessments collected within the first 72 hours of ICU admission. Patients who died early or were transferred out of the ICU rapidly had fewer mobility evaluations and limited opportunities for mobilization. Accordingly, residual confounding (e.g., respiratory muscle training), reverse causation, time-varying confounding, and guarantee-time bias could not be fully eliminated. Third, the frequency and timing of mobility documentation were subject to patient clinical status and nursing workflows, introducing information bias. Additionally, the underlying reasons for missing mobility assessments in 16 patients and persistently minimal mobility scores in another 19 patients were unavailable, potentially introducing selection bias. Fourth, selection of outcome measures was constrained by data availability within the MIMIC-IV database. Standardized, consistent structured measurements of ICU-acquired weakness, pressure injuries, and post-discharge quality of life were absent across the full cohort, precluding robust quantitative analyses. Delirium, which correlates with mobility capacity, disease severity, and clinical endpoints, was incorporated as a potential confounder in regression adjustment rather than analyzed as a standalone outcome.

Conclusions

Compared with Class 1 (low-level stable pattern), ICU patients with pneumonia in Class 3 (early active maintenance pattern) and Class 4 (rapidly progressive pattern) were associated with shorter ICU LOS and lower mortality. While these trajectories mirror patients’ early mobility profiles and recovery trajectories, further prospective research is needed to confirm their utility for nurses delivering personalized bedside mobilization interventions.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.