Model predicts acute respiratory distress syndrome risk in patients with sepsis

Key takeaways:

• Thirteen factors were identified as potential predictors of acute respiratory distress syndrome.
• With these factors, the model had positive predictive and discriminatory ability in sepsis patients.

A model featuring 13 potential predictors of acute respiratory distress syndrome was found to have “good predictive ability” in patients with sepsis, according to study results published in BMC Pulmonary Medicine.

“The developed predictive model may also be a potential tool to guide clinicians in predicting the risk of ARDS in septic patients in the ICU, which help take early interventions to prevent ARDS progression in sepsis patients admitted to ICU and improved clinical outcomes,” Chi Xu, of the emergency department at the Affiliated Wuxi People’s Hospital of Nanjing Medical University, and colleagues wrote.

In a retrospective cohort study, Xu and colleagues analyzed 16,523 ICU patients with sepsis to see if they could create and validate a nomogram that could predict the risk for ARDS in this patient population. According to researchers, a nomogram is a tool that can predict the likelihood of a clinical event in patients.

Of the total cohort, 11,566 patients (mean age, 65.45 years; 60.17% men) belonged to the training set, which was where the nomogram was established, and 4,957 belonged to the testing set, which was where the model was validated.

In the training cohort of patients, researchers discovered risk factors for ARDS through univariate and multivariate logistic regression analysis, and these prognostic factors then went into the nomogram.

To test the model’s predictive ability, researchers evaluated the receiver operating characteristic (ROC) and calibration curves.

Of the total cohort, 2,422 (20.66%) patients experienced ARDS.

Researchers observed 13 possible predictors of ARDS in multivariate logistic regression analysis:

• BMI between 18.5 kg/m² and less than 25 kg/m² (RR = 0.732; 95% CI, 0.542-0.987) and BMI between 25 kg/m² and less than 30 kg/m² (RR = 0.748; 95% CI, 0.555-1.008);
• respiratory rate (RR = 1.049; 95% CI, 1.041-1.058);
• urine output (RR = 1; 95% CI, 1-1);
• partial pressure of carbon dioxide (RR = 1.017; 95% CI, 1.013-1.021);
• blood urea nitrogen (RR = 1.005; 95% CI, 1.002-1.007);
• vasopressin (RR = 1.711; 95% CI, 1.491-1.964);
• continuous renal replacement therapy (RR = 4.87; 95% CI, 4.054-5.851);
• chronic pulmonary disease (RR = 1.353; 95% CI, 1.208-1.514);
• malignant cancer (RR = 1.278; 95% CI, 1.097-1.488);
• liver disease (RR = 1.72; 95% CI, 1.505-1.967);
• septic shock (RR = 1.268; 95% CI, 1.044-1.541);
• pancreatitis (RR = 2.273; 95% CI, 1.693-3.053); and
• ventilation status of invasive vent (RR = 5.617; 95% CI, 3.122-10.107), noninvasive vent (RR = 7.387; 95% CI, 4.825-11.308) and trach (RR = 2.906; 95% CI, 1.914-4.412).

In terms of the predictive performance of the created model, researchers found that the training set had an area under the curve (AUC) of 0.811 (95% CI, 0.802-0.82), and the testing set had a similar AUC of 0.812 (95% CI, 0.798-0.826).

When comparing these results with the AUC of the sequential organ failure assessment (0.539; 95% CI, 0.518-0.559) and the simplified acute physiology score II (0.609; 95% CI, 0.589-0.629) in the testing cohort, researchers found that their created model had “favorable discriminatory ability” in predicting ARDS.

Additionally, within the testing set, the prediction model had 70.5% accuracy, 79.8% sensitivity and 68.2% specificity, according to researchers.

Lastly, between the predicted risk for ARDS and the observed risk, researchers noted good concordance based on calibration curves.

“Further prospective studies are warranted to validate the effectiveness and applicability of this prediction model,” Xu and colleagues wrote.