Mortality prediction for patients with acute respiratory distress syndrome based on machine learning: a population-based study

Abstract

Background: Traditional scoring systems for patients' outcome prediction in intensive care units such as Oxygenation Saturation Index (OSI) and Oxygenation Index (OI) may not reliably predict the clinical prognosis of patients with acute respiratory distress syndrome (ARDS). Thus, none of them have been widely accepted for mortality prediction in ARDS. This study aimed to develop and validate a mortality prediction method for patients with ARDS based on machine learning using the Medical Information Mart for Intensive Care (MIMIC-III) and Telehealth Intensive Care Unit (eICU) Collaborative Research Database (eICU-CRD) databases.Methods: Patients with ARDS were selected based on the Berlin definition in MIMIC-III and eICU-CRD databases. The APPS score (using age, PaO 2 /FiO 2 , and plateau pressure), Simplified Acute Physiology Score II (SAPS-II), Sepsis-related Organ Failure Assessment (SOFA), OSI, and OI were calculated. With MIMIC-III data, a mortality prediction model was built based on the random forest (RF) algorithm, and the performance was compared to those of existing scoring systems based on logistic regression. The performance of the proposed RF method was also validated with the combined MIMIC-III and eICU-CRD data. The performance of mortality prediction was evaluated by using the area under the receiver operating characteristics curve (AUROC) and performing calibration using the Hosmer-Lemeshow test.Results: With the MIMIC-III dataset (308 patients, for comparisons with the existing scoring systems), the RF model predicted the in-hospital mortality, 30-day mortality, and 1-year mortality with an AUROC of 0.891, 0.883, and 0.892, respectively, which were significantly higher than those of the SAPS-II, APPS, OSI, and OI (all P<0.001). In the multi-source validation (the combined dataset of 2,235 patients in MIMIC-III and 331 patients in eICU-CRD), the RF model achieved an AUROC of 0.905 and 0.736 for predicting inhospital mortality for the MIMIC-III and eICU-CRD datasets, respectively. The calibration plots suggested good fits for our RF model and these scoring systems for predicting mortality. The platelet count and lactate level were the strongest predictive variables for predicting in-hospital mortality.Conclusions: Compared to the existing scoring systems, machine learning significantly improved performance for predicting ARDS mortality. Validation with multi-source datasets showed a relatively robust generalisation ability of our prediction model.

A propensity score‐matching analysis of angiotensin‐converting enzyme inhibitor and angiotensin receptor blocker exposure on in‐hospital mortality in patients with acute respiratory failure

“…In the present study, patients with ARDS were included in the sensitivity analysis to improve credibility of our conclusion. Due to the lack of ICD‐9 code for ARDS, we screened for patients with ARDS according to a previous study, 18 which resulted in the identification of a higher number of patients with ARDS than total number of patients with ARF. In short, the general conclusion of patients with ARDS was similar with what we found in the ARF cohort.…”

“…Berlin definition was more widely used than ICD-9 code in these studies. 18,[41][42][43][44][45] Unfortunately, the large number of missing information for oxygen concentration (itemid = "50816") or oxygen flow (itemid = "50815") in the MIMICIII database, 362,668/490,629 (73.92%) of oxygen concentration and 478,321/490,629 (97.49%) of oxygen flow (Table S15), make it difficult to calculate the PaO 2 /FiO 2 ratio which is the core factor of Berlin definition. We recognized that using the nearest FiO 2 to replace the missing values might be a good alternative.…”

“…However, these codes representing pulmonary insufficiency and respiratory failure are not in agreement with the definition of ARDS. Berlin definition was more widely used than ICD‐9 code in these studies 18,41–45 . Unfortunately, the large number of missing information for oxygen concentration (itemid = “50816”) or oxygen flow (itemid = “50815”) in the MIMICIII database, 362,668/490,629 (73.92%) of oxygen concentration and 478,321/490,629 (97.49%) of oxygen flow (Table ), make it difficult to calculate the PaO 2 /FiO 2 ratio which is the core factor of Berlin definition.…”

“…Sensitivity analyses were constructed to demonstrate the robustness of our results by changing the research subject from ARF to ARDS and the database from MIMIC III to MIMIC IV. Our screening standards for patients with ARDS in the MIMIC III database were as previously reported 18 . The same inclusion and exclusion criteria as MIMIC III database were used in the MIMIC IV database.…”

“…Notably, no significant differences in AKI ratio and ventilation percentage during hospitalization were found in the matched cohort (p = 0.121 and p = 0.175, respectively). Meanwhile, patients with ACEI/ARB exposure had lower serum F I G U R E 1 Flowchart of patient selection MIMIC-III Data Hospitalized patients with acute respiratory failure (ICD-9: 51881) between 2001 to 2012, n=7,497 First admission for acute respiratory failure, n=5,389 Adult patients with acute respiratory failure, n=5,335 -2,108 -54 Keep only the first admission for patients with acute respiratory failure (n=2,108) Non-ACEI and ARB users, n=4,972 Exclude Patient age<18 years (n=4) No laboratory data (n=50) Non-ACEI and ARB users, n=1,401 ACEI or ARB users, n=363 ACEI or ARB users, n=363 PSM method MIMIC-III = Medical Information Mart for Intensive Care-III ICD-9 = International classification of diseases-9 ACEI = Angiotensin converting enzyme inhibitors ARB = Angiotensin receptor blockers PSM = Propensity score matchingcreatinine and BUN levels in the matched cohort (1.2 (0.9, 1.8) mg/ dl vs. 1.3 (0.9, 2.2) mg/dl, p = 0.003 and 26(18,44) mg/dl vs. 31(20, 50) mg/dl, p < 0.001), respectively, which did not occur in the original cohort.…”