Derivation and Validation of an Automated Search Strategy to Retrospectively Identify Acute Respiratory Distress Patients Per Berlin Definition

Abstract

Purpose: Acute respiratory distress syndrome (ARDS) is common in critically ill patients and linked with serious consequences. A manual chart review for ARDS diagnosis could be laborious and time-consuming. We developed an automated search strategy to retrospectively identify ARDS patients using the Berlin definition to allow for timely and accurate ARDS detection.Methods: The automated search strategy was created through sequential steps, with keywords applied to an institutional electronic medical records (EMRs) database. We included all adult patients admitted to the intensive care unit (ICU) at the Mayo Clinic (Rochester, MN) from January 1, 2009 to December 31, 2017. We selected 100 patients at random to be divided into two derivation cohorts and identified 50 patients at random for the validation cohort. The sensitivity and specificity of the automated search strategy were compared with a manual medical record review (gold standard) for data extraction of ARDS patients per Berlin definition.Results: On the first derivation cohort, the automated search strategy achieved a sensitivity of 91.3%, specificity of 100%, positive predictive value (PPV) of 100%, and negative predictive value (NPV) of 93.1%. On the second derivation cohort, it reached the sensitivity of 90.9%, specificity of 100%, PPV of 100%, and NPV of 93.3%. The strategy performance in the validation cohort had a sensitivity of 94.4%, specificity of 96.9%, PPV of 94.4%, and NPV of 96.9%.Conclusions: This automated search strategy for ARDS with the Berlin definition is reliable and accurate, and can serve as an efficient alternative to time-consuming manual data review.

Clinical Characteristics and Predictors of Complications and Mortality in Hospitalized Octogenarian Patients with COVID-19: An Ambispective Study

“…8,23,24 Despite being high, the incidence of ARDS in hospitalized patients is likely underestimated in most studies, as physicians do not always document this diagnosis in the medical records. 29 After ARDS, the most common complications were altered mental status and renal failure, with one-fth of the hospitalized elderly patients developing acute kidney injury. Other cohort studies assessing hospitalized elderly patients have reported acute kidney injury, which is likely related to tubular injury caused by local and systemic in ammation and immune systems, aggravated by hemodynamic instability.…”

Open-source machine learning pipeline automatically flags instances of acute respiratory distress syndrome from electronic health records

“…In contrast, our study considers any adult patient placed on mechanical ventilation, which requires evaluating all Berlin Definition components. Finally, Song and Li developed a fully rules-based tool that automates the entire Berlin Definition, both achieving identical high performance 13,14 . However, their models were constructed within a single hospital, which may not be reproducible across different health systems.…”

“…In addition, a ML algorithm to risk stratify patients for ARDS using structured clinical data derived from the EHR was shown to have good discriminative performance 12 . Regarding automating the entire ARDS diagnostic algorithm, two studies 13,14 have recently reported implementation of keyword search (i.e. rule-based approach) in the EHR with validation conducted for 100 intensive care unit (ICU) admissions from a single time period and from a single institution.…”

Uncertainty-Aware Convolutional Neural Network for Identifying Bilateral Opacities on Chest X-rays: A Tool to Aid Diagnosis of Acute Respiratory Distress Syndrome

“…Hospital billing codes are insufficient to create such a dataset due to clinical under-recognition [ 14 ]. Automated methods to filter patient encounters using constraints on clinical features from the Electronic Medical Record (EMR) like PaO 2 /FiO 2 ratio, or positive end-expiratory pressure (PEEP) defined by the Berlin criteria are being approached [ 17 ], but are limited by non-standard documentation across EMR [ 18 ], and the heterogeneity of clinical manifestations of ARDS [ 19 ]. Thus, clinician adjudication of retrospective data is needed, which involves meticulous inspection of patient history using the EMR.…”