Cardiovascular Subphenotypes in Acute Respiratory Distress Syndrome* : Critical Care Medicine

Journal Logo

Feature Articles

Cardiovascular Subphenotypes in Acute Respiratory Distress Syndrome*

Chotalia, Minesh BMBCh1,2; Ali, Muzzammil MRCP2; Alderman, Joseph E. MBChB1,2; Bansal, Sukh MBChB1,2; Patel, Jaimin M. PhD1,2; Bangash, Mansoor N. PhD1,2; Parekh, Dhruv PhD1,2

Author Information
Critical Care Medicine 51(4):p 460-470, April 2023. | DOI: 10.1097/CCM.0000000000005751



Question: Using latent class analysis of echocardiographic parameters, what are the characteristics of derived cardiovascular subphenotypes in acute respiratory distress syndrome (ARDS)?

Findings: In this retrospective cohort study, latent class analysis identified four cardiovascular subphenotypes that were described as states of preserved cardiovascular function, right ventricular (RV) dilation with preserved cardiac output, RV dilation with impaired cardiac output and high cardiac output.

Meaning: Clustering methods identified cardiovascular subphenotypes that better describe circulatory failure mechanisms and more closely align with mortality in ARDS than current definitions of RV dysfunction.

Acute respiratory distress syndrome (ARDS) is common in mechanically ventilated ICU patients and is associated with a high attributable mortality (1). Shock is a sign of poor prognosis in ARDS (2) with echocardiography studies implicating right ventricular dysfunction (RVD) as the main contributor to cardiovascular dysfunction in the condition (3–8).

However, defining RVD remains problematic, as it is a heterogeneous syndrome with no consensual echocardiography definition available (9). Traditional cutoff values of right ventricular (RV) size or systolic function have been used in a binary fashion by clinicians to define RVD yet are not validated in ARDS populations, and subsequently, these definitions have inconsistently been associated with mortality (3–8). A classification system that is simple and defines consistently prevalent subgroups, which strongly associated with mortality is needed to provide prognostic enrichment for trials of RV protective strategies. Furthermore, although RVD is a common focus of ARDS circulatory dysfunction, ventricular interdependence suggests that an all-encompassing classification system would not solely focus on RV pathology alone.

A method of characterizing circulatory failure phenotypes in ARDS that is nonbinary, unbiased, and multifaceted is required to better identify pathophysiological mechanisms of shock that may benefit from heterogeneous therapy (10). Techniques clustering physiologic and blood parameters have been utilized in this endeavor and have consistently described two patient subgroups: termed hyperinflammatory and hypoinflammatory ARDS that potentially have different responses to therapies (11–13). In COVID-ARDS, we employed latent class analysis (LCA) of cardiovascular parameters and identified three subphenotypes of increasingly severe RVD and mortality risk (14). To the best of our knowledge, a similar study has not been conducted in patients with non-COVID ARDS, and it is therefore unclear whether any of these or other subphenotypes are found in non-COVID ARDS.

The objective of this study was to perform LCA of transthoracic echocardiography (TTE)/clinical parameters to characterize circulatory failure mechanisms in non-COVID ARDS. Secondary aims were to delineate which hemodynamic parameters were most crucial to subphenotype derivation and to compare the association with mortality between LCA-derived subphenotypes and current definitions of RVD in the literature. The over-arching goal is to describe cardiovascular subphenotype(s) in ARDS that are prevalent, have pathophysiological rationale, and are associated with mortality and thus may benefit from targeted treatment.


This retrospective cohort study was approved by the hospital’s Institutional Review Board (RRK-7414) and by the Research Ethics Committee of the National Health Service Health Research Authority (“Latent Class Analysis in ICU patients with ARDS; integrated research application system identification 301564 on July 15, 2021) and procedures were followed in accordance with institutional ethical standards of the responsible committee on human experimentation and with the Helsinki Declaration of 1975.

All patients that received TTE within 7 days of meeting Berlin criteria for ARDS (15) at the Queen Elizabeth Hospital, Birmingham ICU, between April 4, 2016, and December 31, 2021, were included. Three-hundred sixty-six patients with ARDS from a previously published sepsis cohort were included (16). Patients were excluded if they had a positive nasal/throat/sputum polymerase chain reaction test for severe acute respiratory syndrome coronavirus 2, were receiving mechanical circulatory support, had preexisting left ventricular (LV)/RV dilation or systolic dysfunction on TTE or did not meet the Berlin ARDS criteria, or received TTE more than 7 days after meeting criteria. The primary outcome measure was the optimum number of LCA-derived cardiovascular subphenotypes present in a non-COVID ARDS cohort. Secondary outcomes included comparing clinical and outcome measures between identified subphenotypes, assessing their association with mortality after multivariable logistic regression analysis and measuring the diagnostic accuracy of three-variable models in identifying each latent subphenotype. Chest radiograph opacification (17), ARDS severity (15), dead space fraction (18), dynamic compliance (19), and vasopressor dose (20) were calculated as described previously and clinical data from the day of TTE was extracted from electronic health records and Intensive Care National Audit and Research Centre Case Mix Programme by analysts. The mean value of continuous variables with multiple measurements on the day of TTE was recorded.

Transthoracic Echocardiography

A British Society of Echocardiography level 2 TTE protocol was performed by echocardiographers/clinicians with advanced TTE accreditation after being requested by the treating clinician. RV fractional area change (RVFAC) less than 35% or tricuspid annular plane systolic excursion (TAPSE) less than 17 mm defined RV systolic dysfunction (RVSD) and RV:LV end-diastolic area (RV:LVEDA) greater than 0.6 defined RV dilation. Acute cor pulmonale (ACP) was defined as RV dilation with septal dyskinesia (3). LV outflow tract velocity time integral derived stroke volume, RV:LVEDA and RVFAC were measured offline in accordance with guidelines by two independent observers accredited in echocardiography and blinded to clinical data.

Statistical Analysis

LCA was performed as previously described (14) using Latent Gold (v 6.0; Statistical Innovations, Arlington, MA) and is detailed in the supplementary appendix ( Briefly, all TTE and hemodynamic parameters with less than 10% missing data were included. Any one of two variables that demonstrated collinearity after Pearson correlation testing or local dependence after bivariate residual testing were excluded, with sensitivity analyses performed with inclusion of the excluded variable. Bayesian information criteria (BIC), Vuong-Lo-Mendell-Rubin (VLMR) p value test, entropy, class size, and clinical expertise regarding the pathophysiological rationale of identified classes were used to evaluate best-fit of models ranging from one to five classes.

Analyses other than LCA were carried out using SPPS (IBM, Armonk, NY) and GraphPad Prism (v9.1; GraphPad, San Diego, CA). Categorical variables are presented as n (%) and compared between classes using a chi-square test. Continuous variables were tested for normality and presented as median (interquartile range [IQR]) and compared between subphenotypes using a Kruskal-Wallis test. A p value of less than 0.05 was considered statistically significant and all tests were two-sided. A univariate analysis was performed on all collected clinical variables and those with a p value of less than 0.05 were included in multivariable binary logistic regression analysis. If parameters were related, only one of them was included in the model. Each LCA subphenotype and RVD definition cohort was included in turn. Missing data were imputed and a p value of greater than 0.05 on Hosmer-Lemeshow test indicated goodness of fit. Last, we determined the three most important variables for each class on the basis of the greatest difference in the mean standardized values when compared with other classes. The diagnostic performance of the three-variable model in identifying latent cardiovascular (CV) subphenotypes was evaluated using area under the receiver operating characteristic curve of a multivariate logistic regression analysis with the dependent variable being cluster allocation and the three variables as independent variables and by calculated sensitivity specificity, positive, and negative predictive values for the three-variable combinations above or below optimal thresholds.


Four-thousand one-hundred sixteen ICU patients received TTE within the study time frame, of whom 801 met criteria for inclusion (Fig. 1). Table 1 and s-Table 1 ( outlines patient demographics. Patients had a median age of 62 years (IQR, 50–72 yr), with TTE performed on a median day 4 (2–7) after ARDS diagnosis. Most patients were intubated (86%; n = 691) and receiving vasopressor agents (61%; n = 492) at the time of TTE. The 90-day mortality rate for the entire cohort was 40% (n = 319).

TABLE 1. - Comparison of Demographics and Clinical Variables in Cardiovascular Subphenotypes
Parameter All, n = 801 Class 1, n = 347 Class 2, n = 188 Class 3, n = 102 Class 4, n = 164 p
Age, yr 62 (50–72) 64 (52–73) 59 (45–70) 63 (53–76) 60 (47–71) 0.012
Sex, % male 508 (63.4) 219 (63.1) 120 (63.8) 68 (66.7) 101 (61.5) 0.866
Day of transthoracic echocardiography 4 (2–7) 3 (2–6) 5 (3–7) 3 (2–5) 3 (2–7) 0.0040
ARDS severity 0.298
 Mild 401 (50.1) 184 (53.0) 100 (53.2) 44 (43.1) 73 (44.5)
 Moderate 378 (47.2) 153 (44.1) 83 (44.2) 54 (52.9) 88 (53.7)
 Severe 22 (2.7) 10 (2.9) 5 (2.7) 4 (3.9) 3 (1.8)
Sequential Organ Failure Assessment score 7 (4–11) 6 (4–9) 6 (4–10) 10 (7–12) 9 (7–12) < 0.0001
Tracheal intubation 691 (86.3) 285 (82.1) 166 (88.3) 89 (87.3) 151 (92.1) 0.015
Risk factor for ARDS 0.0021
 Pneumonia 420 (52.4) 195 (56.2) 103 (54.8) 62 (60.8) 60 (36.6)
 Nonrespiratory sepsis 208 (26.0) 72 (20.8) 49 (26.1) 24 (23.5) 63 (38.4)
 Chest trauma 101 (12.6) 44 (12.7) 22 (11.7) 8 (7.8) 27 (16.5)
 Aspiration 54 (6.7) 27 (7.8) 11 (5.9) 7 (6.9) 9 (5.5)
 Other 18 (2.3) 9 (2.6) 3 (1.6) 1 (1.0) 5 (3.1)
Septic shock, n (%) 175 (21.9) 50 (14.4) 26 (13.8) 34 (33.3) 65 (39.6) < 0.0001
 Pao 2:Fio 2 ratio 203 (156–248) 210 (158–255) 210 (150–240) 188 (143–248) 210 (165–248) 0.122
 Paco 2, kPa 6.0 (5.2–7.0) 5.7 (5.1–6.7) 6.1 (5.2–7.1) 6.4 (5.5–7.9) 5.9 (5.0–7.0) 0.0009
 pH 7.37 (7.32–7.42) 7.39 (7.33–7.44) 7.36 (7.32–7.42) 7.34 (7.28–7.39) 7.36 (7.30–7.40) 0.0012
 Mean tidal volume, mL/kg/predicted body weight 5.9 (5.0–6.8) 5.7 (4.7–6.6) 6.0 (5.3–6.7) 5.9 (5.0–6.8) 6.0 (5.2–7.1) 0.118
 Chest radiograph opacification (0–16) 6 (4–8) 5 (4–8) 6 (4–8) 8 (6–9) 5 (3–7) 0.036
 Deadspace fraction (n = 691) 0.63 (0.55–0.70) 0.60 (0.53–0.69) 0.64 (0.53–0.71) 0.67 (0.60–0.74) 0.63 (0.56–0.68) 0.0067
 Dynamic compliance, mL/cm H2O (n = 691) 30 (25–36) 31 (26–36) 28 (23–34) 30 (24–34) 32 (26–37) 0.044
 Peak airway pressure, cm H2O (n = 691) 24 (21–27) 23 (20–26) 25 (21–29) 25 (22–27) 24 (20–27) 0.020
 Positive end-expiratory pressure, cm H2O (n = 691) 7 (5–8) 7 (5–8) 7 (5–9) 7 (5–8) 7 (5–9) 0.454
 Mean value across entire day, mL/kg/hr 0.68 (0.28–1.1) 0.74 (0.40–1.1) 0.75 (0.30–1.2) 0.39 (0.13–0.60) 0.62 (0.23–1.0) < 0.0001
Management, n (%)
 Prone ventilation 39 (4.9) 4 (1.2) 10 (5.3) 15 (14.7) 10 (6.1) < 0.0001
 Neuromuscular blockade 188 (23.5) 54 (15.6) 49 (26.1) 44 (43.1) 41 (25.0) < 0.0001
 Renal replacement therapy 210 (26.2) 57 (16.4) 46 (24.5) 43 (42.2) 64 (39.0) < 0.0001
 90-d mortality 319 (39.8) 67 (19.4) 75 (39.7) 80 (78.4) 97 (59.1) < 0.0001
ARDS = acute respiratory distress syndrome.
Values are number (proportion) or median (interquartile range).

Figure 1.:
Flowchart for inclusion of patients in the study. ARDS = acute respiratory distress syndrome, LV = left ventricular, RV = right ventricular, SARS-CoV-2 = severe acute respiratory syndrome coronavirus 2, TTE = transthoracic echocardiography.

After exclusion of TTE/hemodynamic parameters due to missing values, collinearity, and local dependence (supplementary appendix,, the following nine variables were included in the LCA model: RV:LVEDA, RVFAC, TAPSE, inferior vena cava diameter, LV end-diastolic area index, cardiac index (CI), heart rate, central venous pressure (CVP), and vasopressor dose category. Table 2 outlines the fit statistics for the five LCA models. A four-class model was chosen based on having the lowest BIC, significant VLMR test p value compared with class 3 and the strongest clinical and pathophysiological rationale compared with other classes. Subphenotypes were well separated with median posterior probabilities of class assignment greater than 90% for all classes (supplementary appendix, and were stable after exclusion of patients undergoing spontaneous ventilation, with chronic respiratory conditions, ischemic heart disease, or liver disease (supplementary appendix, LCA of patients that received a second TTE (n = 44) demonstrated that the majority (86.4%) remained in the same class.

TABLE 2. - Fit Statistics for One to Five Class Models of Latent Class Analysis
Cluster Likelihood Ratio Bayesian Information Criteria Akaike Information Criteria Maximum Bivariate Residual VLMR p Adjusted VLMR p Entropy
1 –9,664.3 19,448.9 19,364.6 105.3 1.00
2 –9,282.7 18,806.2 18,637.5 54.3 < 0.0001 < 0.0001 0.76
3 –9,059.8 18,480.6 18,227.6 17.4 < 0.0001 < 0.0001 0.72
4 –8,950.0 18,381.3 18,043.9 10.6 < 0.0001 < 0.0001 0.72
5 –8,905.5 18,412.7 17,990.9 7.0 0.0002 0.0003 0.72
VLMR = Vuong-Lo-Mendell-Rubin test.

Forty-three percent of patients (n = 347) were allocated to class 1, 24% (n = 188) to class 2, 13% (n = 102) to class 3, and 21% (n = 164) to class 4. Comparison of TTE and hemodynamic parameters between classes is outlined in Table 3, s-Table 1 (, and Figure 2. Class 1 was predominantly formed of patients with normal RV and LV size and function. Class 2 was characterized by a high prevalence of mild RV dilation (RV:LVEDA 0.6–0.8 in 61.2%) and relative preservation of RV systolic function (in 72.3%) and CI (98.9%). Class 3 was distinguished by markedly increased RV size with concomitant RV systolic impairment in 93.1%. The dilation and impairment was most commonly severe (in 35.3% and 33.3%, respectively) and accompanied by a lower CI. In contrast, most patients in class 4 had normal RV size (81.1%) and systolic function (84.8%) but a high CI (in 82.3%), hyperdynamic LV ejection fraction (in 60.4%), and low systemic vascular resistance index.

TABLE 3. - Comparison of Echocardiographic and Hemodynamic Variables in Cardiovascular Subphenotypes
Parameter All, n = 801 Class 1, n = 347 Class 2, n = 188 Class 3, n = 102 Class 4, n = 164 p
Right ventricle
 Right ventricular:left ventricular  end-diastolic area 0.58 (0.51–0.70) 0.53 (0.47–0.57) 0.74 (0.65–0.82) 0.93 (0.77–1.07) 0.54 (0.48–0.59) < 0.0001
 Right ventricular fractional area change, % 40.7 (32.5–48.2) 41.6 (35.6–47.9) 41.6 (33.8–48.1) 22.4 (17.0–27.5) 46.2 (38.5–52.8) < 0.0001
 Tricuspid annular plane systolic  excursion, mm (n = 746) 21 (19–24) 21 (19–23) 22 (20–26) 16 (12–20) 22 (20–27) < 0.0001
 Dilation 335 (41.8) 33 (9.5) 171 (91.0) 100 (98.0) 31 (18.9) < 0.0001
 Systolic impairment 247 (30.8) 75 (21.6) 52 (27.7) 95 (93.1) 25 (15.2) < 0.0001
 Acute cor pulmonale 93 (11.6) 5 (1.5) 35 (15.4) 47 (46.1) 6 (3.7) < 0.0001
Probability pulmonary hypertension < 0.0001
 Low 291 (36.3) 134 (38.6) 66 (35.1) 21 (20.6) 70 (42.7)
 Intermediate 115 (14.4) 48 (13.8) 21 (11.2) 23 (22.5) 23 (14.0)
 High 80 (10.0) 14 (4.0) 19 (10.1) 36 (35.3) 11 (6.7)
 Unable to determine 315 (39.3) 151 (43.5) 82 (43.6) 22 (21.6) 60 (36.6)
Maximum tricuspid regurgitation  velocity, m/s (n = 486) 259 (220–257) 245 (219–280) 261 (216–310) 290 (257–324) 248 (209–290) < 0.0001
Left ventricle
 Ejection fraction < 0.0001
  Normal (55–70%) 498 (62.2) 244 (70.3) 135 (71.8) 59 (57.8) 60 (36.6)
  Depressed (< 55%) 122 (15.2) 73 (21.0) 16 (8.5) 28 (27.5) 5 (3.1)
  Hyperdynamic (> 70%) 181 (22.6) 30 (8.7) 37 (19.7) 15 (14.7) 99 (60.4)
 Stroke volume index, mL/m2 (n = 792) 40.2 (31.7–49.4) 37.4 (31.7–43.9) 44.7 (35.4–52.1) 25.4 (19.6–31.4) 51.7 (43.2–62.4) < 0.0001
 Cardiac index, L/min/m2 (n = 792) 3.3 (2.6–4.3) 2.9 (2.5–3.5) 3.7 (3.2–4.4) 2.0 (1.6–2.5) 5.2 (4.4–6.3) < 0.0001
 Systemic vascular resistance index,  dynes cm–5 s–1 (n = 792) 1,600 (1,070–2,020) 1,880 (1,530–2,270) 1,420 (1,050–1,830) 2,190 (1,570–2,700) 880 (680–1,070) < 0.0001
 IVC diameter, cm (n = 773) 1.8 (1.5–2.2) 1.8 (1.6–2.0) 2.0 (1.5–2.2) 2.2 (2.0–2.5) 1.5 (1.3–1.8) < 0.0001
 Collapsibility IVC, % (n = 773) 24.5 (0–50.0) 28.2 (0–47.9) 20.0 (0–50.0) 0 (0–12.0) 45.8 (4.9–73.1) < 0.0001
 Mean arterial pressure, mm Hg 73 (65–86) 75 (69–89) 72 (64–83) 67 (59–72) 67 (61–75) < 0.0001
 Vasoactive agent, n (%) 492 (61.4) 190 (54.8) 99 (52.7) 85 (83.3) 118 (72.0) < 0.0001
 Vasopressor dose, µg/kg/min 0.08 (0–0.26) 0.05 (0–0.17) 0.06 (0–0.19) 0.16 (0.07–0.45) 0.23 (0–0.52) < 0.0001
 Heart rate, beats/min 85 (74–100) 80 (70–90) 83 (73–95) 88 (72–105) 100 (90–112) < 0.0001
 Central venous pressure, mm Hg (n = 768) 9 (7–12) 9 (5–11) 10 (8–12) 12 (7–15) 9 (7–13) < 0.0001
IVC = inferior vena cava.
Values are number (proportion) or median (interquartile range).

Figure 2.:
Profile plot of continuous class defining variables in the latent class analysis model. CI = cardiac index, CVP = central venous pressure, HR = heart rate, IVC = inferior vena cava diameter, LVEDAi = left ventricular end-diastolic area index, RV:LVEDA = right ventricular end-diastolic area, RVFAC = right ventricular fractional area change, TAPSE = tricuspid annular plane systolic excursion.

When comparing clinical variables between subphenotypes, classes 3 and 4 had higher Sequential Organ Failure Assessment (SOFA) scores, vasopressor requirements, and renal replacement therapy (RRT) prevalence. Class 3, had the most deranged respiratory function, with a higher Paco2, deadspace fraction, and chest radiograph opacification. Class 4 had an increased WBC count, temperature, international normalized ratio (INR), bilirubinemia, and lower platelet counts suggesting either coagulative or hepatic dysfunction (although liver enzyme levels were similar among groups). Classes 3 and 4 were associated with a significantly higher 90-day mortality rate (78.4% and 59.1%, respectively) compared with classes 1 (19.4%) and 2 (39.7%; p < 0.0001 overall). Subphenotype proportion and associated TTE characteristics/90-d mortality were nonsignificantly different comparing patients with early (< 72 hr) versus late (> 72 hr) TTE.

One clinical and five easily measured TTE variables produced different three-variable models that could identify the latent cardiovascular system subphenotypes with high sensitivity and specificity (Table 4). Following univariate analysis (s-Table 2,, the following clinical variables were included in a multivariable model: age, Pao2/Fio2 ratio, temperature, lactate, platelet count, bilirubin, alkaline phosphatase, INR, Charlson Comorbidity Index, performance status, need for intubation, and RRT. Following logistic regression analysis, classes 3 (odds ratio [OR], 6.9 [4.0–11.8]) and 4 (OR, 2.5 [1.6–3.7]) were independently associated with mortality, while class 1 associated with survival (OR, 0.20 [0.14–0.29]; s-Table 3, Most standard RVD definitions were also independently associated with mortality in our cohort (s-Table 3,, with RV dilation and impairment (RVD&I) having the highest OR for mortality (OR, 6.0 [3.9–9.3]). However, these standard TTE-derived RVD definitions produced four groups with essentially two mortality levels (70% and 30–37%), while every other definition of RVD also produced dichotomous mortality groups with the non-RVD groups sharing similar mortality rates of approximately 35% (data not shown).

TABLE 4. - Predictive Accuracy of Three-Variable Models in Identifying Latent Cardiovascular Subphenotypes
Cluster Three-Variable Model With Cutoff Values Sensitivity Specificity Positive Predictive Value Negative Predictive Value
Class 1 RV:LV end-diastolic area < 0.65 89.6% (311/347) 91.2% (414/454) 88.6% (311/351) 92.0% (414/450)
Cardiac index < 4
Heart rate < 110
Class 2 RV:LV end-diastolic area > 0.6 81.4% (153/188) 87.9% (539/613) 67.4% (153/227) 93.9% (539/574)
RV fractional area change > 0.25
Tricuspid annular plane systolic excursion > 16
Class 3 RV:LV end-diastolic area > 0.6 90.2% (92/102) 95.0% (664/699) 73.0% (92/126) 98.5% (664/674)
RV fractional area change < 0.35
Cardiac index < 4
Class 4 Cardiac index > 4 80.5% (132/164) 90.7% (578/637) 69.1% (132/191) 94.8% (578/610)
Heart rate > 70
Inferior vena cava diameter < 2.2
LV = left ventricular, RV = right ventricular.


In this large study of critically ill patients with ARDS, LCA of cardiovascular parameters identified four subphenotypes with distinct characteristics and clinical outcomes. Importantly, the first three classes appear to match those identified in hemodynamic clustering analyses in COVID-ARDS (14). Class 1 was associated with normal cardiovascular function and survival, and we suggest be labeled as “preserved RV.” The second class was described by mild RV dilation with preserved systolic function and cardiac output and was termed “RV dilation with preserved function.” Class 3 was characterized by RV dilation with systolic impairment and a low cardiac output, strongly associated with mortality and was labeled “RV failure.” Finally, a fourth class, not present in COVID-ARDS, but appearing to match a high mortality group present in a recent sepsis echocardiography study (16), was associated with a high cardiac output state and mortality and was termed “hyperdynamic.” These LCA-derived subphenotypes more closely aligned with circulatory failure mechanisms and outcomes than current RVD definitions, underscoring their prognostic utility. Furthermore, it was possible to derive these LCA classes with high sensitivity and specificity using only a few commonly recorded cardiovascular parameters, highlighting their potential for clinical use.

Subphenotype Pathophysiology

The “RV dilation with preserved function” and “RV failure” subphenotypes resemble previously outlined RV pathophysiology (20) and have also been identified in COVID-19 ARDS (14). The “RV dilation with preserved function” class may represent a state of adapted RV dilation in response to raised pulmonary afterload, due to thrombotic, inflammatory, pulmonary, or ventilatory burden—indeed markers of pulmonary pathology were worse in this class than the preserved RV class. RV dilation, importantly via the Frank-Starling mechanism, preserves systolic function to maintain LV filling and cardiac output (21). In this sense, this class was best derived by three variables that delineate RV dilation but relatively preserved RV systolic function. The class was not independently associated with mortality, perhaps by maintaining cardiac output, preserving end-organ perfusion. In this regard, the preserved urine output and SOFA score despite several surrogates of systemic congestion being increased is notable. Nonetheless, a propensity for decompensation from this adapted state of RV dilation likely exists and may be reflected in the higher 90-day mortality rate for the subphenotype when compared with class 1 (Preserved RV).

The “RV failure” class likely represents this decompensation, whereby progressive dilation precipitates systolic impairment and low cardiac output, perhaps by stretching sarcomeres above their optimal interactive capacity (22). Decreased intrinsic contractility as a result of septic cardiomyopathy could also be a cause (23), with the class demonstrating a higher prevalence of concomitant septic shock. Notably, CVP was also markedly elevated in this class denoting that systemic venous congestion, as well as low cardiac output, may contribute to reduced organ perfusion pressure, resulting in organ failure that is denoted by the marked rise in SOFA score in this subphenotype. Multiple organ dysfunction is likely to precipitate to death, with a striking seven-fold increased odds of mortality for this cohort. This class was described by RV dilation, systolic impairment and low cardiac output with high sensitivity/specificity. The “hyperdynamic” class was characterized by a heightened inflammatory state (evident through raised WBC, temperature) that may precipitate vasodilation, tachycardia, and increased stroke volume index. This “hyperdynamic circulation” has been commonly described in sepsis (16), and of note, nonrespiratory sepsis was the predominant risk factor for ARDS in this cohort. This may be why the subphenotype was not identified in COVID-ARDS, in which a homogenous respiratory insult was the precipitating cause.

Interestingly, Pao2/Fio2 ratio was not significantly different between CV subphenotypes with differing degrees of RV function. However, other factors that increase RV afterload (e.g., hypercarbia and acidosis [22]) or worsen cardiac contractility (e.g., septic shock by precipitating septic cardiomyopathy [23]) were significantly different.

Clustering of blood and physiologic variables in multiple ARDS cohorts have consistently demonstrated two subphenotypes with different responses to treatments, termed hypoinflammatory and hyperinflammatory ARDS (11–13). Although we identify four CV subphenotypes, their degree of overlap warrants prospective study and reverse translation, as the hyperinflammatory subgroup (characterized by tachycardia, shock, renal dysfunction, and mortality) resembles the third and fourth CV subphenotypes that may be separated as a result of the organ-specific information included in their analysis. Nonetheless, clustering analyses that incorporate inflammatory, respiratory and cardiovascular variables may be desirable in order to better characterize the latent structures underlying ARDS pathology.

Broader Implications

In defining RVD in ARDS, the “RV failure” LCA class and “RVD&I” definition more closely aligned with mortality than ACP, RV dilation with CVP greater than 8 and RVSD, despite the latter three being the more frequently utilized definitions in the literature (3–5). While RVD&I may be simpler to diagnose at the bedside, the third LCA class incorporates global cardiovascular function. The two classifications substantially overlap, as the three-variable model that best derived the third LCA class only differed from RVD&I by incorporating CI. However, RV-centric definitions of circulatory failure in ARDS may be too simplistic and neglect the hyperdynamic LCA subphenotype, which closely aligned with mortality despite having mostly normal RV function.

As described, circulatory failure mechanisms in ARDS are varied, yet the downstream consequence of shock is managed uniformly. In the hyperdynamic LCA subphenotype, efforts to increase systemic vascular resistance using selective vasopressor agents (e.g., vasopressin) make physiologic sense, while in patients with RVD and impaired cardiac output, inotropic agents, pulmonary vasodilators, or mechanical circulatory support to increase RV forward flow may more coherently address the pathophysiology. Whether early intervention with prone ventilation or negative fluid balance strategies can mitigate transition from a state of compensated RV dilation to RV failure also warrants study.

Strengths and Limitations

It is unclear whether the CV subphenotypes derived are mechanistic determinants of mortality or merely signal extracardiac disease severity (e.g., inflammatory, thrombotic, ventilatory or comorbid burden) and, as such, whether stratifying therapies in relation to subphenotypes will lead to heterogeneous treatment response and/or improve outcomes. Nonetheless, this does not detract from their utility in prognostication/circulatory failure characterization. Patients with preexisting ventricular dysfunction were excluded to determine the effects of acute disease on normal hearts. Although patients with cardiorespiratory comorbidities were included, sensitivity analyses demonstrate that they had no significant effect on subphenotype derivation. Although the timing of echocardiography was not standardized, the prevalence, characteristics, and mortality of subphenotypes was no different if TTE was performed early or late within the first 7 days. Subphenotype stability across ICU admission requires prospective study with serial echocardiography as RV injury may be an evolving process (24), although in the small percentage of patients that received a second TTE, the majority remained in the same subphenotype. It is also unknown whether subphenotype derivation was influenced by the selection bias introduced by retrospective analysis, single-center cohort of patients or variables of CV function employed, with more sensitive markers (e.g., diastolic dysfunction, longitudinal strain, RV:pulmonary artery coupling or 3D ejection fraction) not measured. However, the derived subphenotypes were stable across numerous sensitivity analyses in which different CV variables and demographic cohorts were included/excluded and in derivation/validation cohorts in the largest TTE dataset in ARDS patients to date. Furthermore, three of the four subphenotypes were described in clustering analyses performed in separate COVID-ARDS datasets (14), demonstrating their reproducibility and external validity. Regarding model selection, while statistically, the grounds for the four-class model may be stronger (lowest BIC value, VLMR p value strongly positive compared with three classes), the weak clinical/pathophysiological rational for choosing 2, 3, or 5 classes (supplementary appendix, pushed selection to the four-class model as is noted can (and indeed should) happen by Sinha et al (25). The classes and especially their three-variable models still require prospective validation at serial time points in multicenter ARDS cohorts to describe their true prevalence. The three-variable models, although highly sensitive/specific in subphenotype derivation, may demonstrate overlap between Subphenotypes.


This is the first description of an unbiased, data-driven, nonbinary, and multimodal approach to characterizing circulatory failure mechanisms in ARDS and identifies four subphenotypes with clear separation, stability across sensitivity analyses, pathophysiological rationale, and association with mortality that have potential clinical utility as they can be reliably derived using only a few commonly measured TTE/clinical variables.


1. Bellani G, Laffey JG, Pham T, et al.: Epidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries. JAMA. 2016; 23:788–800
2. Vieillard-Baron A, Girou E, Valente E, et al.: Predictors of mortality in acute respiratory distress syndrome. Focus on the role of right heart catheterization. Am J Respir Crit Care Med. 2000; 161:1597–1601
3. Chotalia M, Ali M, Alderman J, et al.: Right ventricular dysfunction and its association with mortality in coronavirus disease 2019 acute respiratory distress syndrome. Crit Care Med. 2021; 49:1757–1768
4. Mekontso Dessap A, Boissier F, Charron C, et al.: Acute cor pulmonale during protective ventilation for acute respiratory distress syndrome: Prevalence, predictors, and clinical impact. Intensive Care Med. 2016; 42:862–870
5. Boissier F, Katsahian S, Razazi K, et al.: Prevalence and prognosis of cor pulmonale during protective ventilation for acute respiratory distress syndrome. Intensive Care Med. 2013; 39:1725–1733
6. Vieillard-Baron A, Prigent A, Repesse X, et al.: Right ventricular failure in septic shock: Characterization, incidence and impact on fluid responsiveness. Crit Care. 2020; 24:630
7. Sato R, Dugar S, Cheungpasitporn W, et al.: The impact of right ventricular injury on the mortality in patients with acute respiratory distress syndrome: A systematic review and meta-analysis. Crit Care. 2021; 25:172
8. Vallabhajosyula S, Kumar M, Pandompatam G, et al.: Prognostic impact of isolated right ventricular dysfunction in sepsis and septic shock: An 8-year historical cohort study. Ann Intensive Care. 2017; 7:94
9. Lahm T, Douglas I, Archer SL, et al.: Assessment of right ventricular function in the research setting: Knowledge gaps and pathways forward. An official American Thoracic Society Research Statement. Am J Respir Crit Care Med. 2018; 98:e15–e43
10. Shankar-Hari M, McAuley DF: Acute respiratory distress syndrome phenotypes and identifying treatable traits. The dawn of personalized medicine for ARDS. Am J Respir Crit Care Med. 2017; 195:280–281
11. Famous KR, Delucchi K, Ware LB, et al.; ARDS Network: Acute respiratory distress syndrome subphenotypes respond differently to randomized fluid management strategy. Am J Respir Crit Care Med. 2017; 195:331–338
12. Calfee CS, Delucchi KL, Sinha P, et al.; Irish Critical Care Trials Group: Acute respiratory distress syndrome subphenotypes and differential response to simvastatin: Secondary analysis of a randomised controlled trial. Lancet Respir Med. 2018; 6:691–698
13. Sinha P, Delucchi KL, McAuley DF, et al.: Development and validation of parsimonious algorithms to classify acute respiratory distress syndrome phenotypes: A secondary analysis of randomised controlled trials. Lancet Respir Med. 2020; 8:247–257
14. Chotalia M, Ali M, Alderman JE, et al.: Cardiovascular subphenotypes in patients with COVID-19 pneumonitis whose lungs are mechanically ventilated: A single-centre retrospective observational study. Anaesthesia. 2022; 77:763–771
15. Ferguson ND, Fan E, Camporota L, et al.: The Berlin definition of ARDS: An expanded rationale, justification, and supplementary material. Intensive Care Med. 2012; 38:1573–1582
16. Chotalia M, Ali M, Hebballi R, et al.: Hyperdynamic left ventricular ejection fraction in ICU patients with sepsis. Crit Care Med. 2021; 50:770–779
17. Mason SE, Dieffenbach PB, Englert JA, et al.: Semi-quantitative visual assessment of chest radiography is associated with clinical outcomes in critically ill patients. Respir Res. 2019; 20:218
18. Beitler JR, Thompson BT, Matthay MA, et al.: Estimating dead-space fraction for secondary analyses of acute respiratory distress syndrome clinical trials. Crit Care Med. 2015; 43:1026–1035
19. Stahl CA, Möller K, Schumann S, et al.: Dynamic versus static respiratory mechanics in acute lung injury and acute respiratory distress syndrome. Crit Care Med. 2006; 34:2090–2098
20. Chotalia M, Matthews T, Arunkumar S, et al.: A time-sensitive analysis of the prognostic utility of vasopressor dose in septic shock. Anaesthesia. 2021; 76:1358–1366
21. Holubarsch C, Ruf T, Goldstein DJ, et al.: Existence of the Frank-Starling mechanism in the failing human heart. Investigations on the organ, tissue, and sarcomere levels. Circulation. 1996; 94:683–689
22. Mebazaa A, Karpati P, Renaud E, et al.: Acute right ventricular failure--from pathophysiology to new treatments. Intensive Care Med. 2004; 30:185–196
23. Vieillard-Baron A, Naeile R, Haddad F, et al.: Diagnostic workup, etiologies and management of acute right ventricle failure. Intensive Care Med. 2018; 44:774–790
24. Evrard B, Goudelin M, Giraudeau B, et al.: Right ventricular failure is strongly associated with mortality in patients with moderate-to-severe COVID-19-related ARDS and appears related to respiratory worsening. Intensive Care Med. 2022; 12:1–3
25. Sinha P, Calfee CS, Delucchi KL: Practitioner’s guide to latent class analysis: Methodological considerations and common pitfalls. Crit Care Med. 2021; 49:e63–e79

acute respiratory distress syndrome; latent class analysis; right ventricular dysfunction; right ventricular failure; subphenotypes; transthoracic echocardiography

Supplemental Digital Content

Copyright © 2023 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.