Post-operative multi-organ dysfunction following heart transplantation and left ventricular assist device implantation: phenotypes, trajectories, and prognostic implications—a retrospective study

Abstract

Background

Multi-organ dysfunction syndrome (MODS) is a leading cause of mortality after heart transplantation (HTx) and left ventricular assist device (LVAD), yet its phenotypic heterogeneity remains poorly characterised. This study intends to identify and clinically validate distinct multi-organ dysfunction phenotypes following HTx and LVAD implantation, and to analyse their early prognosis using longitudinal biomarker trajectories and independent clinical adjudication.

Methods

This single-center retrospective study included 68 consecutive adult patients who underwent HTx or LVAD implantation in the Cardiothoracic and Vascular Surgery Department of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, between January 2020 and June 2025. The Institutional Review Board of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology approved this with a waiver of informed consent. We collected demographic, preoperative, and operative data from medical records. Postoperative MODS was defined by Sequential Organ Failure Assessment (SOFA) criteria as dysfunction with ≥2 organ systems. Patients were followed up for inpatient, 1-, and 2-year mortality.

Results

MODS occurred in 48 of 68 patients (70.6%). It was more frequent after HTx than LVAD (77.6% vs. 52.6%, P=0.08). The most common organ dysfunctions were cardiovascular (70.6%), hepatic (60.3%), and respiratory (50.0%). Patients with MODS had significantly longer cardiopulmonary bypass (CPB) duration than those without MODS (183.11±57.42 vs. 138.28±29.83 min; P=0.003) and higher rates of respiratory (68.8% vs. 5.0%; P<0.001), cardiovascular (83.3% vs. 40.0%; P=0.001), and renal dysfunction (50.0% vs. 5.0%; P=0.001). After adjustment, MODS was independently associated with 2-year mortality [hazard ratio (HR) =5.47, 95% confidence interval (CI): 1.07–27.9; P=0.04]. Latent class analysis (LCA) identified 2 MODS phenotypes: a high-severity/multi-system phenotype with high probabilities of renal (100%), hepatic (78%), and respiratory (72%) dysfunction, and a lower-severity/selective phenotype. The high-severity phenotype had significantly worse 2-year survival than the lower-severity phenotype (44.8% vs. 74.4%; log-rank P=0.005).

Conclusions

With regard to post-cardiac surgery MODS, we report a novel trajectory-based approach that holds promise for improving early risk stratification and guiding targeted management by distinguishing high-severity multisystem from low-severity selective trajectories following HTx and LVAD implantation.

Highlight box.

Key findings

• Compared to left ventricular assist device (LVAD) implantation, heart transplantation (HTx) was associated with predominantly cardiovascular, respiratory, and renal dysfunction, while LVAD was linked with predominant hepatic and coagulatory dysfunction.

• Two distinct post-operative multi-organ dysfunction syndrome (MODS) phenotypes were identified: a high-severity multi-system phenotype and a low-severity selective phenotype, with significantly different survival outcomes.

What is known and what is new?

• Previous studies have described MODS after advanced heart failure surgery, but direct comparisons of organ dysfunction patterns between HTx and LVAD recipients are limited.

• This study provides evidence from a direct comparison of post-operative organ failure profiles between the two modalities. This is also the first study to use latent class analysis to define distinct MODS phenotypes following HTx and LVAD, and to demonstrate that phenotype class is a strong independent predictor of early mortality.

What is the implication, and what should change now?

• Phenotype-based risk stratification is an important consideration in the postoperative management of HTx and LVAD implantation. The results of this study suggest that early identification of patients with a high-severity multi-system phenotype can help implement intensive monitoring and interventions to improve outcomes.

Introduction

Multi-organ dysfunction syndrome (MODS) remains a primary contributor to adverse outcomes following cardiac surgery with cardiopulmonary bypass (CPB). CPB-induced systemic inflammatory response is a key driver, with elevated levels of interleukin-8 (IL-8), interleukin-18 (IL-18), and procalcitonin observed in patients who develop MODS compared to those with Systemic Inflammatory Response Syndrome (SIRS) alone. Contemporary Society of Thoracic Surgeons Adult Cardiac Surgery Database confirms that mortality and major morbidity rise incrementally as the number of affected organ systems increases, particularly in cases of renal failure and prolonged ventilation (1,2).

In the field of heart failure, heart transplantation (HTx) is currently the only surgical therapy that offers both improved quality of life and extended survival (3,4). However, the limited availability of donor organs results in prolonged waiting time, during which patients experience progressive hemodynamic deterioration, and mortality remains high despite pharmacological treatment. In the COPERNICUS trial, the annual mortality rate for patients with chronic heart failure was 11%, despite the use of angiotensin-converting enzyme (ACE) inhibitors and β-blockers (5).

Consequently, a large number of patients undergo left ventricular assist device (LVAD) implantation as a bridge to transplantation. Despite advances in surgical technique and perioperative care, one of the main causes of early mortality after HTx and LVAD implantation is MODS (6). Multisystem organ failure is a significant contributor to early mortality following LVAD implantation, as demonstrated with the large INTERMACS registry data (7). Sepsis-associated organ dysfunction is driven by an abnormal and poor host response to infection, in which inflammatory and immune activation, metabolic disturbances, and impaired microcirculation interact to compromise organs (8). In cardiac surgery-associated MODS, organ dysfunction arises through a distinct set of mechanisms compared to infection-related sepsis. CPB initiates a systemic inflammatory response, often leading to end-organ injury (8). Additionally, ischemia-reperfusion events contribute further tissue and myocardial damage. Right ventricular (RV) dysfunction may exacerbate this cascade by raising systemic venous congestion, thereby impairing renal and other organ function (9). These interconnected mechanisms collectively account for the varied patterns of organ dysfunction frequently observed after cardiac surgery.

Importantly, MODS is a heterogeneous syndrome rather than a single, uniform disease process. In sepsis and acute respiratory distress syndrome (ARDS), distinct clinical phenotypes have been identified, each linked to different inflammatory profiles, disease severity, treatment responses, and outcomes. In contrast, most prior studies in cardiac surgery have treated MODS as a single entity, and a phenotyping approach has not been established for patients undergoing HTx or LVAD implantation (10,11). Without phenotype-specific stratification, management remains nonspecific, and prognosis remains poor. For example, some patients may benefit from early initiation of renal replacement therapy, while others may require targeted pulmonary vasodilators for right ventricular dysfunction, or personalised adjustments to anticoagulation in the presence of a coagulopathic phenotype.

We hypothesised that distinct MODS phenotypes exist in this population and are associated with differing early outcomes. Therefore, the aim of this retrospective study was to identify and clinically validate distinct multiple organ dysfunction (MOD) phenotypes following HTx and LVAD implantation, and to analyse their association with early prognosis using longitudinal biomarker trajectories and independent clinical adjudication. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1983/rc).

Methods

Study design and participants

We conducted a single-center retrospective observational study in the Cardiothoracic and Vascular Surgery Department of Huazhong University of Science and Technology of Tongji Medical College, Tongji Hospital, between January 2020 and June 2025. The study included 68 consecutive adult patients (≥18 years old) who underwent primary HTx or LVAD implantation during the study period. We collected demographic information, preoperative baseline characteristics, and operative details. Patients were divided into two groups by surgery type: HTx and LVAD. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study received approval from the Institutional Review Board of Huazhong University of Science and Technology, Tongji Medical College, Tongji Hospital, with a waiver of informed consent due to the retrospective nature of the research. Eligible participants were identified from the institutional surgical database and hospital electronic medical record system. For LVAD patients, we included only China-made devices: CH-VAD, CORHEART6, EVAHEART, and HEARTCON. We further stratified patient by device type and bridging strategy. Exclusion criteria included death before ICU admission, active infection or sepsis at the time of surgery, concomitant major non-cardiac surgery, and history of prior HTx or LVAD. Missing data were handled using an available-case approach, meaning patients were not excluded for missing individual variables. Instead, analyses were conducted using all available data for each variable, maximising sample size for each analysis. Two investigators independently reviewed all records to confirm eligibility, and a third senior surgeon resolved disagreements. Data were collected from the electronic medical record of Huazhong University of Science and Technology of Tongji Medical College, Tongji Hospital, including demographic information, preoperative baseline characteristics, and operative details. To assess organ function, daily morning laboratory values and clinical data were obtained throughout the intensive care unit (ICU) stay. Renal function was evaluated by serum creatine, blood urea nitrogen, and 24-hour urine output; hepatic function by total bilirubin, international normalized ratio (INR), aspartate aminotransferase (AST), and alanine aminotransferase (ALT); coagulation by platelet count and fibrinogen; Pulmonary function by ventilation status and partial pressure of arterial oxygen to fraction of inspired oxygen (PaO₂/FiO₂) ratio; cardiovascular support by vasopressor equivalents and inotropic score; neurological status score by Glasgow Coma Scale with sedation assessment by the Richmond Agitation-sedation Scale; and right ventricular function by fractional area change (FAC), tricuspid annular plane systolic excursion (TAPSE), and qualitative grading. Organ dysfunction was defined using modified Sequential Organ Failure Assessment (SOFA) criteria adapted for cardiac surgery with specific thresholds for each organ system. Patients were categorized into two groups according to surgical type: transplantation or LVAD implantation. All patients were followed from the date of surgery to hospital discharge, and subsequently through telephone interviews for up to 2 years. Patients lost to follow-up were censored at the date of last contact. The primary outcome was 2-year all-cause mortality. Secondary outcomes included MODS progression, ICU length of stay, and duration of mechanical ventilation. All patients were followed for up to 2 years.

Statistical analysis

Post-operative organ dysfunction was classified according to SOFA-aligned criteria across six systems: respiratory, cardiovascular, renal, hepatic, neurological, and coagulation (Table 1). MODS was defined as dysfunction of ≥2 organ systems. Patients were further stratified into three groups: no dysfunction (0 systems affected), single-organ dysfunction (1 system), and MODS (≥2 systems).

Table 1. MODS definition—SOFA-aligned cut-offs.

System	Variables used	Criterion
Respiratory	Ventilation post-surgery	Mechanical ventilation >24 h
Cardiovascular	Vasopressors, intra-aortic balloon pump, extracorporeal membrane oxygenation, post-cardiotomy shock	Vasopressor use >24 h
Renal	Renal_failure, acute_renal_failure	Failure
Hepatic	Liver_failure, hyperbilirubinemia_post-surgery, bilirubin_post	Failure or bilirubin >2.0 mg/dL
Neurological	Ischemic_stroke, hemorrhagic_stroke	Clinical stroke (any type)
Coagulation	Disseminated intravascular coagulation, thrombocytopenia, _blood_disorder	Disseminated intravascular coagulation or thrombocytopenia

MODS, multi-organ dysfunction syndrome; SOFA, Sequential Organ Failure Assessment.

Patient characteristics stratified by MODS status (present vs. absent) were compared using the Mann-Whitney U test for continuous variables and Chi-squared/Fisher’s exact test for categorical variables. Organ dysfunction group comparisons (no organ dysfunction, single-organ dysfunction, and multi-organ dysfunction) for continuous clinical variables used the Kruskal-Wallis test followed by pairwise Wilcoxon signed-rank tests for post-hoc analysis. Categorical outcomes across these groups were compared with Chi-squared tests.

Principal component analysis (PCA) was used to visualise the distribution of organ dysfunction variables.

Latent class analysis (LCA) was performed on the six binary organ dysfunction indicators to identify latent subgroups or Phenotypes with distinct profiles. Model selection was guided by the Bayesian Information Criterion (BIC), and a two-class solution best represented the data. Each patient was assigned to the most probable class based on posterior probabilities.

Class conditional probabilities were plotted as a heatmap to characterise the organ dysfunction signature of each class.

We compared inpatient mortality and 2-year mortality between LCA classes using Chi-squared tests. Kaplan-Meier survival curves were plotted for each LCA class and compared using the log-rank test.

A multivariate Cox proportional hazards model was then fitted, adjusting for LCA class, surgical group (HTx vs. LVAD), and age, to evaluate the independent prognostic contribution of the identified phenotypes.

We applied a Cox proportional hazards regression model to identify independent predictors of 2-year mortality. Covariates included surgical modality (LVAD vs. HTx), presence of MODS (≥2 organ systems), individual organ dysfunction domains, and age. Results are presented as hazard ratios (HRs) with 95% confidence intervals (CIs) and associated P values. Statistical significance was set at P<0.05.

Results

Between January 2020 and June 2025, patients underwent HTx and LVAD implantation at Huazhong University of Science and Technology of Tongji Medical College, Tongji Hospital. We assessed the prevalence of organ dysfunction across six primary organ systems and compared patients who underwent HTx and LVAD surgery. Figure 1 illustrates the comparative findings.

Figure 1.

Prevalence of each organ dysfunction domain by surgical group. HTX, heart transplantation; LVAD, left ventricular assist device.

Cardiovascular and respiratory systems: the HTx group showed higher rates of dysfunction in these areas. Cardiovascular dysfunction was the most prevalent complication in the HTx group at 84%, compared to 37% in the LVAD group.

Similarly, respiratory dysfunction was reported in 61% of HTx patients, nearly triple the prevalence seen in the LVAD group (21%).

Hepatic and coagulatory systems: conversely, the LVAD group showed a higher prevalence of hepatic and coagulatory complications. Hepatic dysfunction was observed in 79% of LVAD patients compared to 53% for HTX. Notably, coagulatory dysfunction was present in 42% of the LVAD group, while no instances (0%) were reported in the HTX group.

Renal and neurological systems: renal dysfunction was more common in the HTX group (48%) than in the LVAD group (16%). Neurological complications were the least prevalent across both modalities, affecting 6% of HTX patients and 0% of the LVAD cohort (Figure 1).

MODS prevalence by surgical group

The prevalence of MOD varied between the two surgical groups. In the HTx group, 38 of 49 patients (77.6%) presented with MOD, compared with 10 of 19 patients (52.6%) in the LVAD group. In contrast, absence of MOD was observed in 11 patients (22.4%) in the HTX group and 9 patients (47.4%) in the LVAD group (Table 2).

Table 2. MODS prevalence by group.

Group	Overall	HTx (n=49)	LVAD (n=19)	P value
No-MODS	20	11 (22.4)	9 (47.4)	0.08
MODS	48	38 (77.6)	10 (52.6)

Data are presented as n (%) or n. HTX, heart transplant; LVAD, left ventricular assist device; MODS, multi-organ dysfunction syndrome.

Organ dysfunction co-occurrence heatmap

Figure 2 presents the co-occurrence of post-operative organ dysfunction across the six studied systems. This heatmap illustrates how often multiple organ systems failed simultaneously within the total patient cohort (n=68).

Figure 2.

Co-occurrence heatmap of organ dysfunction domains.

Primary co-occurrence patterns

Cardiovascular and respiratory systems: the strongest association was observed between cardiovascular and respiratory dysfunction, occurring simultaneously in 44.1% of the patients.

Cardiovascular and hepatic/renal systems: cardiovascular dysfunction also showed high rates of co-occurrence with both hepatic and renal systems, with each pairing appearing in 38.2% of cases.

Hepatic and renal systems: the concurrent failure of the hepatic and renal systems was observed in 25.0% of the population.

Respiratory and renal systems: respiratory dysfunction frequently overlapped with renal (26.5%) and hepatic (27.9%) complications.

Low frequency associations

Neurological and coagulatory systems: neurological and coagulatory dysfunctions showed the lowest rates of co-occurrence with other organ systems. Neurological complications overlapped with respiratory and cardiovascular systems in only 4.4% of cases (Figure 2).

Patient characteristics by MODS status

A total of 68 patients were included in the analysis, of whom 48 (70.6%) had MODS and 20 (29.4%) did not. There were no statistically significant differences between the MODS and NO-MODS groups in age, sex, BMI, diabetes mellitus, hypertension, or ischemic heart disease (IHD) (all P>0.05). Similarly, pre-operative cardiac markers, including N-terminal pro-B-type natriuretic peptide (NT-proBNP), estimated glomerular filtration rate (eGFR), creatinine, and left ventricular end-diastolic diameter (LVEDD), did not differ significantly between the two groups.

However, CPB duration was significantly longer in patients with MODS than in those without MODS (183.11±57.42 vs. 138.28±29.83 min; P=0.003).

As expected, the MODS group had significantly higher frequencies of individual organ dysfunctions.

Respiratory dysfunction occurred in the MODS group (P<0.001). Cardiovascular dysfunction was also more frequent in the MODS group (83.3% vs. 40.0%; P=0.001). Similarly, renal dysfunction occurred in 50.0% of patients with MODS compared with 5.0% in the NO-MODS group (P=0.001). Hepatic dysfunction was more frequent among patients with MODS (68.8% vs. 40.0%), although the difference was not statistically significant (P=0.053). Neurological and coagulation dysfunction were uncommon and did not differ significantly between groups.

The mean MOD score was significantly higher among patients classified as having MODS (2.92±0.90 vs. 0.95±0.22; P<0.001).

Although mortality was consistently higher among patients with MODS, the differences in the unadjusted analyses were not statistically significant. Inpatient mortality was 12.5% in the MODS group compared with 0% in the NO-MODS group (P=0.24). Similarly, 6-month mortality was 35.4% versus 15.0% (P=0.16), 1-year mortality was 37.5% versus 15.0% (P=0.12), and 2-year mortality was 45.8% versus 20.0% (P=0.09) in the MODS and NO-MODS groups, respectively.

ICU stay (8.34±7.60 vs. 6.82±2.58 days; P=0.43) and total hospital stay (36.35±18.32 vs. 41.16±14.11 days; P=0.31) did not differ significantly between the groups (Table 3).

Table 3. Patient characteristics by MODS status.

Variable	Total (n=68)	No-MODS (n=20)	MODS (n=48)	P value
Age, years	51.33 [13.09]	54.79 [8.85]	49.96 [14.28]	0.18
Male	50 (73.5)	15 (75.0)	35 (72.9)	>0.99
BMI, kg/m2	22.27 [3.40]	22.60 [3.44]	22.14 [3.41]	0.62
Diabetes mellitus	28 (42.4)	6 (31.6)	22 (46.8)	0.39
Hypertension	33 (50.0)	10 (52.6)	23 (48.9)	>0.99
IHD	32 (48.5)	7 (36.8)	25 (53.2)	0.35
INTERMACS				0.41
1	11 (44.0)	4 (40.0)	7 (46.7)
2	7 (28.0)	2 (20.0)	5 (33.3)
3	6 (24.0)	4 (40.0)	2 (13.3)
4	1 (4.0)	0 (0.0)	1 (6.7)
NT-proBNP_pre, pg/mL	8,801.65 [10,795.43]	5,277.34 [5,121.59]	10,113.02 [12,044.82]	0.13
eGFR_pre, mL/min/1.73 m²	92.95 [200.33]	150.34 [359.74]	68.17 [40.13]	0.14
Creatinine_pre, μmol/L	161.59 [227.64]	166.94 [289.84]	159.18 [197.74]	0.91
LVEF_pre, %	29.38 [11.97]	34.54 [15.11]	27.46 [10.17]	0.07
LVEDD_pre, mm	62.73 [10.66]	62.90 [15.39]	62.67 [8.69]	0.95
Surgical duration, min	428.12 [82.02]	423.28 [67.11]	429.94 [87.53]	0.77
CPB intra, min	170.69 [54.90]	138.28 [29.83]	183.11 [57.42]	0.003**
Blood transfusion	41 (61.2)	13 (68.4)	28 (58.3)	0.63
Respiratory dysfunction	34 (50.0)	1 (5.0)	33 (68.8)	<0.001***
Cadio-vascular dysfunction	48 (70.6)	8 (40.0)	40 (83.3)	0.001**
Renal dysfunction	25 (36.8)	1 (5.0)	24 (50.0)	0.001**
Hepatic dysfunction	41 (60.3)	8 (40.0)	33 (68.8)	0.053
Neuro dysfunction	3 (4.4)	0 (0.0)	3 (6.2)	0.62
Coagulative dysfunction	8 (11.8)	1 (5.0)	7 (14.6)	0.48
MOD_score	2.34 [1.18]	0.95 [0.22]	2.92 [0.90]	<0.001***
ICU stay, days	7.92 [6.61]	6.82 [2.58]	8.34 [7.60]	0.43
Hospital stay, days	37.72 [17.27]	41.16 [14.11]	36.35 [18.32]	0.31
Inpatient mortality	6 (8.8)	0 (0.0)	6 (12.5)	0.24
Mortality 6 months	20 (29.4)	3 (15.0)	17 (35.4)	0.16
Mortality 1-year	21 (30.9)	3 (15.0)	18 (37.5)	0.12
Mortality 2-year	26 (38.2)	4 (20.0)	22 (45.8)	0.09

*, P<0.05; **, P<0.01; ***, P<0.001. CPB, cardiopulmonary bypass; eGFR, estimated glomerular filtration rate; ICU, intensive care unit; IHD, ischemic heart disease; INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LVEDD, left ventricular end-diastolic diameter; LVEF, left ventricular ejection fraction; MOD, multiple organ dysfunction; MODS, multi-organ dysfunction syndrome; NT-proBNP, N-terminal pro-B-type natriuretic peptide.

LCA-MODS phenotypes

LCA identifies unobserved subgroups from binary organ dysfunction indicators and returns probabilistic class membership, AIC/BIC fit, and item-response profiles.

Model selection

To determine the optimal number of latent classes for the study, an LCA model selection process was performed using the Akaike Information Criterion (AIC) and BIC.

As shown in Figure 3, both the AIC and BIC values were at their lowest when the number of latent classes (k) was set to 2. As the number of classes increased beyond two, the IC values for both criteria rose steadily, indicating decreasing model parsimony and fit. Therefore, the 2-class model was selected as the optimal statistical fit for classifying the patient phenotypes in this cohort.

Figure 3.

LCA model selection—AIC and BIC by number of classes. AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; IC, information criterion;LCA, latent class analysis.

Following the selection of the optimal 2-class model, the distribution of surgical modalities, HTx and LVAD, was analysed for each latent class Table 4.

Table 4. LCA class distribution by surgical group.

LCA class	HTX	LVAD
1	23	6
2	26	13

HTX, heart transplant; LCA, latent class analysis; LVAD, left ventricular assist device.

Class 1 consisted of 29 patients, of whom 23 underwent HTx, and 6 received an LVAD. Class 2 represented a larger group of 39 patients, comprising 26 HTx procedures and 13 LVAD implantations. While both classes were predominantly composed of HTx patients, class 2 showed a relatively higher utilisation of LVADs compared to class 1.

Figure 4 shows that, across both classes, cardiovascular, coagulatory, and neurological dysfunction remained low-probability events, confirming that renal, hepatic, and respiratory systems are the primary drivers of phenotype differentiation in this cohort. The two-class LCA identified distinct phenotypes: a high severity/multisystem classes with 100% renal dysfunction and high probabilities of hepatic (78%) and respiratory (72%) failure, and a low severity/selective class with overall lower organ dysfunction, most prominent in hepatic (47%) and low renal (<20%) and respiratory (37%) involvement. Cardiovascular, coagulatory, and neurological dysfunctions were rare in both classes, indicating that renal, hepatic, and respiratory dysfunction primarily drive class differentiation.

Figure 4.

Conditional probability of each organ dysfunction per LCA class. LCA, latent class analysis.

Kaplan-Meier survival analysis by LCA class

Figure 5 below illustrates longitudinal survival probabilities for the two LCA phenotypes. The analysis reveals a highly significant difference in survival trajectories between the two groups over the 2-year follow-up period (P=0.005).

Figure 5.

Kaplan-Meier survival curves by LCA phenotype class. LCA, latent class analysis.

• Class 1 (high severity): this cohort (n=29) experienced a sharp initial decline in survival probability to approximately 79.3% within the first six months. This survival level remained stable through the remainder of the study until the 2-year mark, where the final probability dropped to approximately 44.8%. At the end of the follow-up, 23 patients remained at risk.

• Class 2 (low severity): this cohort (n=39) demonstrated superior clinical stability, maintaining a 100% survival probability through the first 1.5 years of the study. A decline was observed only at the final 2-year endpoint, with the survival probability reaching approximately 74.4%. All 39 patients remained at risk until the final study interval.

Cox proportional hazards—MODS

In the multivariable Cox regression model, MODS involving ≥2 organ systems were significantly associated with mortality; patients with MODS had approximately 5.5-fold higher hazard of mortality than those without MODS (HR =5.47, 95% CI: 1.07–27.9; P=0.04) (Figure 6).

Figure 6.

Forest plot—Cox Model hazard ratios for 2-year mortality. HTX, heart transplantation; INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LVAD, left ventricular assist device; MODS, multi-organ dysfunction syndrome.

In contrast, LVAD recipients did not have a significantly different hazard of mortality compared with HTx recipients after adjustment for MODS, age, sex, and INTERMACS profile (HR =1.32, 95% CI: 0.22–8.11; P=0.80). Neither age (HR =1.00, 95% CI: 0.96–1.05; P>0.90) nor male sex (HR =0.98, 95% CI: 0.16–6.19; P>0.90) was significantly associated with mortality.

Similarly, the INTERMACS profile was not significantly associated with mortality; none of its components reached statistical significance (linear: HR =1.16, P=0.90; quadratic: HR =1.17, P=0.80; cubic: HR =1.51, P=0.60) (Table 5).

Table 5. Cox model for 2-year mortality with MODS.

Variable	HR [95% CI]	P value
LVAD vs. HTX	1.32 [0.22–8.11]	0.80
Sex
Female	1
Male	0.98 [0.16–6.19]	>0.90
Age (per year)	1.00 [0.96–1.05]	>0.90
MODS (≥2 organs)	5.47 [1.07–27.9]	0.04
INTERMACS profile
INTERMACS. Linear	1.16 [0.12–11.4]	0.90
INTERMACS. Quadratic	1.17 [0.28–4.94]	0.80
INTERMACS. Cubic	1.51 [0.34–6.73]	0.60

CI, confidence interval; HR, hazard ratio; HTX, heart transplantation; INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LVAD, left ventricular assist device; MODS, multi-organ dysfunction syndrome.

In the multivariable Cox regression model incorporating LCA class, MODS, treatment group, INTERMACS profile, age, and sex, none of the covariates were significantly associated with 2-year mortality (all P>0.05) (Figure 7).

Figure 7.

Cox model—LCA class as predictor of 2-year mortality. INTERMACS, Interagency Registry for Mechanically Assisted Circulatory Support; LCA, latent class analysis; MODS, multi-organ dysfunction syndrome.

Discussion

In this retrospective experience of patients undergoing advanced heart failure therapies, postoperative MODS emerged as a frequent and clinically important complication. MODS was observed in both HTx and LVAD recipients, reflecting the complex multisystem response that often follows advanced cardiac surgery. The cardiovascular, hepatic, and respiratory systems were most commonly involved, and dysfunction in the cardiovascular and respiratory systems tended to cluster together, suggesting interrelated injury mechanisms. Patients who developed MODS had longer exposure to CPB and were more likely to experience dysfunction across multiple organ systems, particularly respiratory, cardiovascular, and renal. After accounting for surgical modality and baseline patient characteristics, including age, sex, and INTERMACS profile, MODS remained independently associated with poorer long-term survival. This reinforces the idea that MODS itself, rather than the specific procedure, is a key driver of outcome. Importantly, the LCA suggests that MODS is not a single condition. We identified a high-severity, multi-system phenotype defined by widespread renal involvement together with a high likelihood of hepatic and respiratory dysfunction. This phenotype carried a substantially worse prognosis compared with a lower-severity phenotype with more limited organ involvement. The discoveries highlight that prognosis after HTx and LVAD is shaped not only by whether MODS occurs, but by the specific pattern and severity of organ dysfunction. Taken together, this study supports moving beyond a binary definition of MODS toward a phenotype-based approach that may better capture risk and informed targeted management in this high-risk population. The high frequency of MODS in our cohort aligns with its well-recognised role as a key driver of morbidity and mortality in critically ill patients. Prior studies have described how an initial physiologic insult can trigger systemic inflammation, endothelial injury, microcirculatory disruption, and mitochondrial dysfunction, leading to progressive organ injury (12,13). These pathways are especially pertinent after major cardiac surgery, where CPB, ischemia-reperfusion, hemodynamic instability, transfusion, and inflammatory activation can converge to produce postoperative organ failure. Our data extend this framework to patients undergoing contemporary advanced heart failure therapies, a group that carries a particularly high perioperative physiologic burden. The clinical relevance of postoperative organ dysfunction is also supported by studies using serial SOFA assessments. In cardiac surgery cohorts, higher SOFA scores in the early postoperative period, especially at 48 hours and peak SOFA, have been linked to worse 12- and 24-month survival (14). These observations suggest that the degree of organ dysfunction after surgery provides prognostic information beyond preoperative risk alone. In a large non-cardiac surgical population with brief ICU stay, deterioration in SOFA during the postoperative ICU course predicted higher 3-year mortality and a greater incidence of acute kidney injury (AKI) (15). Building on this, our study demonstrates that in patients undergoing HTx or LVAD implantation, MODS itself was associated with mortality even when accounting for surgical type and key preoperative risk factors. Taken together, these findings support evaluating postoperative organ dysfunction as a dynamic process, rather than treating it only as a binary postoperative complication. The association between prolonged CPB duration and MODS observed in our cohort is both biologically and clinically plausible. Recent literature has identified longer CPB duration as a significant risk factor for AKI after HTx (16). Similarly, longer CPB duration, lower mean arterial pressure, and greater cumulative fluid balance have each been associated with postoperative AKI (17,18). Although these studies focused primarily on renal dysfunction rather than MODS as a composite outcome, the consistent association between prolonged CPB and postoperative organ injury supports our findings. Mechanistically, prolonged CPB may contribute to organ dysfunction through several pathways, including systemic inflammatory activation, hemodilution, altered oxygen delivery, microcirculatory impairment, and ischemia-reperfusion injury. Despite these plausible biological associations, our study is observational. Therefore, the observed association between CPB duration and MODS should not be interpreted as evidence that reducing CPB duration will necessarily prevent MODS. Instead, CPB duration should be viewed as a potentially perioperative factor that warrants further prospective investigation. Respiratory dysfunction was another major component of postoperative organ injury in our study. Prior research in cardiac surgery has associated postoperative respiratory failure with several perioperative factors, including prolonged CPB times, cardiogenic shock, prior cardiac surgery, elevated body mass index (BMI), and increased intraoperative transfusion requirements. Collectively, these factors indicate that postoperative pulmonary complications may result from the combined effects of more complex surgical procedures, hemodynamic instability, and heightened perioperative physiologic stress (19). Acute hypoxemic respiratory failure carries high mortality in contemporary ICUs, with risk rising as the PaO₂/FiO₂ ratio declines (20). Cardiovascular and respiratory dysfunction co-occurred most strongly. This likely reflects underlying cardiopulmonary interdependence, in which reduced cardiac output, fluid accumulation, impaired ventilation, and abnormal gas exchange amplify one another. The clustering may therefore represent a shared pathophysiological pathway rather than independent organ injuries (21). Renal dysfunction was a defining feature of postoperative MODS, most pronounced in high-severity, multi-system phenotype. The MELD-XI score, which integrates bilirubin and creatinine, has been widely used to assess combined hepatic and renal dysfunction in HTx and LVAD patients. Emerging data suggest that HTx recipients often achieve sustained MELD-XI improvement, whereas LVAD patients tend to show only transient gains, likely reflecting different postoperative trajectories and device-related complications (6). Our findings extend this observation; renal dysfunction was not merely more common among those with adverse outcomes but was universal in the high-severity, multi-system LCA phenotype. This suggests renal impairment may serve as a robust marker of overall systemic injury severity. Moreover, the MODS mortality association persisted, indicating that the extent and combination of organ dysfunction carry independent prognostic value beyond the specific advanced therapy received. The prognostic importance of postoperative complications is also evident in present LVAD trials. In the MOMENTUM 3 cohort, post-implant complications and adverse events predicted subsequent mortality, underscoring that outcomes after LVAD are shaped by both pre-implant risk and the postoperative course (22). Our data extended this observation by framing those events within broader patterns of multi-organ dysfunction. Rather than treating renal, hepatic, respiratory, and cardiovascular complications as separate events, the LCA results suggested they cluster as coordinated patterns of systemic injury with distinct weight. A second contribution of this study is the identification of distinct MODS phenotypes. Traditional MODS definitions treat the syndrome as a single entity, despite marked variability in which organs are involved and the severity of dysfunction. In clinical care, LCA has been applied to identify clinically distinct subphenotypes in sepsis with differing mortality and potentially different treatment response (23). Such work has shown the limits of syndrome-level labels that can mask biologically and clinically relevant heterogeneity. The high-severity/multi-system phenotype, defined by near-universal renal dysfunction and frequent hepatic and respiratory involvement, experienced substantially worse survival than the lower-severity/selective phenotype. These LCA-derived groups should be viewed as clinical subphenotypes rather than biological endotypes, because our investigation relied on clinical laboratory data and did not include molecular or mechanistic biomarkers. The ability to distinguish patients with different prognoses in MODS highlights the value of phenotype-based postoperative risk stratification. In our study, MODS remained associated with mortality regardless of surgical treatment, indicating that postoperative organ dysfunction offers prognostic insight beyond therapy type or baseline INTERMACS status. These findings should not be seen as proving equivalence between HTx and LVAD. Limitations include small sample size, single-center settings, and potential selection bias. Rather, our results suggest that postoperative physiologic response may be a key determinant of outcome in both treatment cohorts. Clinical implications of these findings center on improving postoperative monitoring, targeting modifiable factors, and refining risk stratifications after HTx and LVAD. The high prevalence of MODS suggests that serial assessment of cardiovascular, renal, hepatic, and respiratory function may better capture postoperative risk than isolated measurements. Prolonged CPB time was associated with the development of MODS. Although causality cannot be determined from these observational studies, minimising unnecessary CPB duration through optimised surgical and perfusion strategies represents a testable intervention. Phenotype-based risk stratification could further inform management. Patients with high-severity/multi-system phenotype, marked by renal dysfunction with hepatic or respiratory involvement had the poorest survival. For this group, strategies such as intensified hemodynamic monitoring, renal-protective interventions, careful fluid management, and earlier escalation of support warrant consideration, though these approaches remain hypotheses for prospective evaluation. Because MODS predicted mortality independent of surgical modality and baseline risk, incorporating early postoperative organ trajectories into risk assessment could improve identification of patients at highest risk for adverse outcomes. Despite these findings, our study has limitations. It is a single-center analysis with a modest sample size and an imbalanced control group, which may affect generalisability and statistical power. For instance, the small patient sample and the use of orthogonal polynomial coding for the INTERMACS profile, which consumes three degrees of freedom, limit statistical power. This limitation helps explain the loss of significance in models that included the LCA class. Trajectory modeling depends on complete longitudinal data, and we used an available-case approach for missing data. We did not exclude patients for missing values, but this could introduce selection bias. We also did not evaluate advanced biomarkers such as cytokines, angiogenic factors, or metabolomics that might further refine these phenotypes. To our knowledge, this is among the first studies to link dynamic MODS trajectories with cardiohepatic physiology following cardiac surgery, utilizing daily clinical data and robust statistical methods. These results provide a basis for validation in larger multicentre registries and for future adaptive trials testing interventions tailored to specific phenotypes.

Conclusions

MODS frequently occurs after HTx and LVAD implantation. Through latent class analyses, we identified two distinct postoperative MODS phenotypes: a high-severity multisystem phenotype linked to poor 2-year survival, and a lower-severity selective phenotype associated with a more favorable prognosis. Importantly, the overall burden and pattern of organ failure, rather than the specific surgical intervention, emerged as the primary determinant of mortality after adjustment for baseline severity. These findings support adopting a phenotype-based framework for postoperative MODS. Early identification of organ dysfunction phenotype may enhance risk stratification and enable more targeted management strategies. Further validation of this trajectory-based approach in multicenter cohorts is needed to confirm its clinical utility.

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, and approved by the Ethics committee of Huazhong University of Science and Technology of Tongji Medical College, Tongji Hospital. Due to the retrospective nature of the study, there was a waiver of informed consent by the Ethics Committee.

Data Sharing Statement

Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1983/dss

jtd-18-10-1100-dss.pdf ^{(104.2KB, pdf)}

DOI: 10.21037/jtd-2026-1983