This study was approved by Austin Hospital’s Human Research Ethics Committee with a waiver of informed consent (LNR/17/Austin/49).
For the study population, we used univariate analysis to compare patients’ characteristics according to survival, pulmonary oedema resolution and interventions. We used the χ2 test to test for equal proportions, the Student t
test to compare normally distributed data, and the Mann–Whitney U
test as appropriate. We report results as number (percentage), mean (standard deviation) or median (interquartile range [IQR], 25th to 75th percentile), respectively. We assessed resolution of pulmonary oedema and use of CRRT using the log-rank test, and we present the results as Kaplan–Meier survival curves. Independent predictors of mortality and resolution of pulmonary oedema were determined using multivariable logistic regression and Cox proportional hazards models. The models were constructed using stepwise selection and backward elimination techniques considering potential confounders such as age, chronic comorbidities and APACHE III score, with each of the results presented as odds ratio (95% confidence interval). For statistical tests, a two-sided P
value of less than 0.05 was taken to indicate statistical significance. Statistical analysis was performed using IBM SPSS Statistics for Windows, version 25.0 (IBM Corp) and Stata Statistical Software, release 16 (StataCorp).
Out of 2001 ICU patients, 264 were identified as having radiologically diagnosed pulmonary oedema. Of them, 238 were included in the primary analysis; 17 patients were excluded owing to missing data or duplications, and nine were excluded because of an alternative clinical diagnosis (Figure 1
Demographics and baseline characteristics of patients with and without RCDPO are shown in Table 1
. Patients with RCDPO were more severely ill. They also had more comorbidities and were more commonly admitted with acute cardiovascular disease. They were more likely to receive invasive MV and CRRT; had longer duration of ICU and hospital stay; were more likely to die in hospital; and, if discharged alive, they were more likely to be admitted to a chronic care facility.
Invasive MV was used in 175 patients with RCDPO (73.5%). Those who received invasive MV were more likely to also receive CRRT and were treated with CRRT for longer. They received less non-invasive ventilation, received a lower median daily dose of frusemide and achieved a less negative median daily fluid balance. Although patients given invasive MV cleared their RCDPO more often, they stayed in the ICU longer and had a significantly higher ICU mortality rate (Table 2
CRRT was used in 69 patients with RCDPO (29.0%). These patients were more severely ill, more likely to receive invasive MV and had longer duration of mechanical ventilation. They were also more likely to have chronic liver disease and to receive a larger total dose of frusemide. In addition, they had longer ICU and hospital length of stay and greater ICU and hospital mortality (Figure 2
; Supporting Information
, Table S1).
Diuretics were administered in 214 patients (89.9%). The most commonly used diuretic was frusemide. The online Supporting Information
(Table S2) shows a summary of the characteristics of patients receiving diuretics versus those who did not receive diuretics. The patients who did not receive diuretics all received CRRT and, did not achieve a negative fluid balance; they also were more likely to have chronic liver disease, and had significantly higher in-hospital mortality compared with those who did receive diuretics.
Pulmonary oedema resolution and mortality
Radiological resolution of pulmonary oedema occurred in 177 patients (74.4%). Patients who achieved RCDPO resolution were more likely to be male and more likely to receive invasive MV. However, they also had a less negative median daily fluid balance. Such patients had higher rates of ICU and hospital survival (Figure 3
), but longer length of stay in ICU and hospital (Table 3
Multivariable analysis for the prediction of radiological pulmonary oedema resolution showed that invasive MV was independently associated with a 2.5-fold increased odds of resolution (Supporting Information
, Table S3). Moreover, receiving CRRT and having a higher APACHE III score were associated with a decreased hazard ratio for resolution of pulmonary oedema (Supporting Information
, Table S4).
In total, 46 RCDPO patients died (19.3%). Non-survivors were older, more acutely ill, more likely to have chronic liver disease, more likely to receive invasive MV, more likely to receive CRRT, more likely to receive a smaller dose of frusemide, less likely to achieve a substantial negative fluid balance, and less likely to achieve radiological resolution (Supporting Information
, Table S5).
Multivariable logistic regression analysis for baseline predictors of mortality across all patients adjusted for RCDPO, age, chronic liver disease, chronic renal failure, chronic respiratory disease and chronic cardiovascular disease showed that RCDPO, chronic liver disease and age were independently associated with increased risk of death (Supporting Information
, Table S6). RCDPO was independently associated with a 59% increase in the risk of death.
Multivariable analysis for the prediction of mortality among RCDPO patients adjusted for CRRT, MV, APACHE III score and age showed that only age and CRRT were independently associated with mortality in this group. CRRT was associated with a 3.9-fold increased risk of death, while age was associated with an increased risk of death of 4.8% per year of age (Supporting Information
, Table S7).
A Cox regression proportional hazards model for mortality adjusted for age, MV, CRRT, APACHE III score and chronic liver disease (Supporting Information
, Table S8) confirmed that each 1-year increase in age was independently associated with an increased risk of death of 4.0%.