Latent class analysis of imaging and clinical respiratory parameters from patients with COVID-19-related ARDS identifies recruitment subphenotypes

Background Patients with COVID-19-related acute respiratory distress syndrome (ARDS) require respiratory support with invasive mechanical ventilation and show varying responses to recruitment manoeuvres. In patients with ARDS not related to COVID-19, two pulmonary subphenotypes that differed in recruitability were identified using latent class analysis (LCA) of imaging and clinical respiratory parameters. We aimed to evaluate if similar subphenotypes are present in patients with COVID-19-related ARDS. Methods This is the retrospective analysis of mechanically ventilated patients with COVID-19-related ARDS who underwent CT scans at positive end-expiratory pressure of 10 cmH2O and after a recruitment manoeuvre at 20 cmH2O. LCA was applied to quantitative CT-derived parameters, clinical respiratory parameters, blood gas analysis and routine laboratory values before recruitment to identify subphenotypes. Results 99 patients were included. Using 12 variables, a two-class LCA model was identified as best fitting. Subphenotype 2 (recruitable) was characterized by a lower PaO2/FiO2, lower normally aerated lung volume and lower compliance as opposed to a higher non-aerated lung mass and higher mechanical power when compared to subphenotype 1 (non-recruitable). Patients with subphenotype 2 had more decrease in non-aerated lung mass in response to a standardized recruitment manoeuvre (p = 0.024) and were mechanically ventilated longer until successful extubation (adjusted SHR 0.46, 95% CI 0.23–0.91, p = 0.026), while no difference in survival was found (p = 0.814). Conclusions A recruitable and non-recruitable subphenotype were identified in patients with COVID-19-related ARDS. These findings are in line with previous studies in non-COVID-19-related ARDS and suggest that a combination of imaging and clinical respiratory parameters could facilitate the identification of recruitable lungs before the manoeuvre. Supplementary Information The online version contains supplementary material available at 10.1186/s13054-022-04251-2.

1. Formulas used 2. Figure S1: correlation plots 3. Stepwise description of discarded variables due to correlation 4. Missing data 5. Figure S2a: density plots of imputed variables. Figure S2b: strip plots of imputed variables 6. Table S1: main outcomes and transitions between complete case and imputation models 7. Figure S3a: profile plot of all recruitable subphenotypes 8. Figure S3b: profile plot of all non-recruitable subphenotypes 9. Figure S4: alluvial plot of patient flow in complete case and imputation models 10. Table S2: changes per lung region 11. Figure S5a: changes in in end-expiratory lung volumes before and after recruitment 12. Figure S5b: changes in lung weight before and after recruitment 13. Figure S6a: volumes in different aeration regions before and after recruitment 14. Figure S6b: weight in different aeration regions before and after recruitment 15. Table S3: LASSO regression results 16. Table S4: GLM results of nested variable models. 17. Table S5: AUROCs for variable subsets Figure S7: ROC curves for variable subsets 18. Table S6a: Fine and Gray regression results of subphenotype membership and duration of MV. Table S6b: Cox regression results of subphenotype membership and survival 19. Figure S8: Kaplan-Meier plot of survival. 20. Table S7. Goodness-of-fit tests. Figure S9: Shoenfeld plots for covariates used in survival analysis. 21. Figure S10: Cumulative incidence plot using only complete case analyses 22. Figure S11: Kaplan-Meier using only complete cases 23. Table S8a: Fine and Gray regression results of subphenotype membership and duration of MV using complete cases only. In which N is the number of slices, t is the slice thickness, d is the distance between slices and M i is the lung mass in the ith slice.

× 100%
In which M is as lung weight in grams and RM indicating recruitment manoeuvre.

Data collection
Clinical respiratory parameters: last values before transport to CT-scan (estimated maximum 30 minutes prior) Blood gas results: values used in decision making for the recruitment manoeuvre (estimated maximum 3 hours prior) Laboratory results: morning round (maximum 24 hours prior) Figure S1. Correlation plots of all variables used in the latent class analysis (left) compared to all available variables before selecting (right). The size of the dots represents the p-value of the correlation (bigger dot equals lower p-value) and the colour represents the correlation coefficient (legend included). Spearman rank-order correlation coefficient was used for calculation. The stars plotted correspond with significance levels; * > 0.05, ** > 0.01, *** > 0.001.

Missing data
As LCA requires complete datasets, missing data was imputed by fully conditional specification with predictive mean matching under the missing at random assumption 1,2 . Imputation was limited to the variables used in the LCA, other variables and outcomes were not imputed. Prediction of imputed variables was based only on variables used in the LCA. Five imputation models with a maximum of 50 iterations each were created. The quality of the imputation models was visually assessed using stripplots and density plots for each imputated variable 2,3 . The effects of imputation on latent class assignment were tested by calculating class transitions between complete case and imputation models. Outcomes and class characteristics were compared among complete case and imputation models.
Supplementary materials -Page 5 Figure S2a. Density plots of variables in complete case dataset compared to imputated datasets. Complete case is plotted in black, colored lines are different imputation sets. Only imputated variables are included. Figure S2b. Stripplots of imputated variables only. Complete case data is displayed with grey dots, imputated datapoints are colored.
Supplementary materials -Page 6 Supplementary materials -Page 7 Figure S3a: profile plot of all recruitable subphenotypes in complete case (CC) and imputation models (I1 -I5). Variables used in the LCA are displayed on the X-axis and their corresponding standardized mean difference (SMD) on the Y-axis.

Figure S3b
: Profile plot of all non-recruitable subphenotypes in complete case (CC) and imputation models (I1 -I5). Variables used in the LCA are displayed on the X-axis and their corresponding standardized mean difference (SMD) on the Y-axis. Figure S4. Alluvial plot of complete case (left) and imputation models (subsequent 5). Patient flow is coloured according to class membership in complete case analysis.  Figure S5a. Changes in end-expiratory lung volumes before and after the recruitment manoeuvre. Data is stratified per lung aeration category and per subphenotype. Lung regions were classified according to Hounsfield Units (HU) with normally aerated ranging from −900 to −501 HU, poorly aerated from −500 to −101 HU, non-aerated from −100 to 100 HU and hyper-inflated from -1000 to -901 HU. Figure S5b. Changes in lung weight before and after the recruitment manoeuvre. Data is stratified per lung aeration category and per subphenotype. Lung regions were classified according to Hounsfield Units (HU) with normally aerated ranging from −900 to −501 HU, poorly aerated from −500 to −101 HU, nonaerated from −100 to 100 HU and hyper-inflated from -1000 to -901 HU. Figure S6a. End-expiratory lung volume (in mL) of different aeration regions displayed before and after the recruitment manoeuvre. Corresponding patient data points are connected by a line. Data is stratified and coloured by subphenotype. The p-values before and after recruitment (bottom p-values) compare relative amounts (mL / total mL) by the recruitment manoeuvre and are derived by Wilcoxon signed-rank tests. The p-values between subphenotypes (upper p-values) compare the changes in relative amounts by the recruitment manoeuvre (change in mL / total mL) between subphenotypes and are derived by Mann-Whitney U tests (table S2). Figure S6b. Mass (in grams) of different aeration regions displayed before and after the recruitment manoeuvre. Corresponding patient data points are connected by a line. Data is stratified and coloured by subphenotype. The p-values before and after recruitment (bottom p-values) compare relative amounts (grams / total grams) by the recruitment manoeuvre and are derived by Wilcoxon signed-rank tests. The p-values between subphenotypes (upper p-values) compare the changes in relative amounts by the recruitment manoeuvre (change in grams / total grams) between subphenotypes and are derived by Mann-Whitney U tests (table S2).

Supplementary materials -Page 15
Subphenotype identification using variable subsets and standard ICU severity scores. Table S3. LASSO coefficient estimates, with the lambda parameter tuned in order to arrive at 4 variables.

Prediction model considering all variables
Intercept 0.5655465   Table S5. Areas under the receiver operating curves (AUROCs) for the variable subsets and standard severity scores, in terms of subphenotype prediction. 95% confidence intervals are derived by means of bootstrapping the AUROCs.