Accuracy, Validation, and Usability of an Intelligent Respiratory Pattern Monitoring System in Patients at Risk of Acute Respiratory Failure

Monitoring respiratory patterns is essential in the management of respiratory diseases, yet it often still relies on subjective and visual assessments. Health technologies based on artificial intelligence (AI) can enhance clinical decision-making by providing more objective and accurate analyses. Given the high prevalence of acute and chronic respiratory diseases, implementing devices capable of detecting variables such as flow, volume, and time has become a priority for enabling more effective diagnosis and therapeutic planning. This study aims to evaluate the accuracy, validity, and usability of an intelligent system for monitoring respiratory patterns in patients at risk of acute respiratory failure.

This is a prospective cohort study to be conducted in the emergency departments of Hospital Otávio de Freitas and Urgent Care Units (UPAs), involving volunteers of both sexes, aged 18 years or older, breathing spontaneously, and under suspicion of acute respiratory failure. Daily screening will be performed, collecting sociodemographic, blood gas, laboratory, and additional clinical data. When indicated, pulmonary function tests, respiratory muscle strength assessments, and diaphragmatic ultrasonography will be performed. Respiratory patterns will be recorded using the Respiratory Diagnostic Assistant (RDA), with data collected directly at the patient's bedside, preferably in a seated position or, if not feasible, in the supine position with the head of the bed elevated to 30°. The device will be used with appropriate protective filters and a face mask properly fitted to the patient, preceded by a clinical evaluation that includes peripheral oxygen saturation, respiratory rate, and signs of respiratory distress. The protocol comprises three minutes of spontaneous basal breathing to record time, volume, and flow variables.

Simultaneously with the RDA assessment, respiratory parameters will also be measured using conventional methods-manual or multiparameter monitor respiratory rate, arterial blood gas analysis (when clinically indicated), pulse oximetry, and spirometry-serving as reference standards for diagnostic accuracy analysis. The collected data will be analyzed using correlation coefficients, agreement tests, and ROC curves to assess the sensitivity, specificity, and overall performance of the RDA algorithm. In addition to accuracy, clinical usability of the device will be evaluated using the System Usability Scale (SUS) questionnaire, assessing interface clarity, ease of mask fitting, examination duration, data interpretation, and clinical applicability. The mean SUS score will be used as an indicator of acceptance, with values ≥68 considered satisfactory.

All clinical and technical data will be securely stored on an encrypted server with access restricted to the research team, in compliance with the Declaration of Helsinki, Brazilian regulations, and the General Data Protection Law (LGPD). Participation will be voluntary, requiring the signing of an informed consent form (ICF) by patients or, when applicable, their legal representatives. Data will be stored in Microsoft Excel 2016 (Microsoft®, USA) and analyzed using SPSS Statistics v.22.0. Descriptive variables will be presented as means and standard deviations or as medians and interquartile ranges, depending on their distribution, assessed using the Kolmogorov-Smirnov test.

The analysis will be guided by three main hypotheses: (1) Accuracy - to assess whether the intelligent monitoring system provides superior performance compared to conventional methods in detecting respiratory pattern alterations, using performance metrics such as accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the ROC curve (AUC), with comparisons made using McNemar's test for paired binary data and AUC comparisons using the z-test; (2) Validation - to verify the system's precision and reliability using the Intraclass Correlation Coefficient (ICC) and Bland-Altman analysis, as well as Cohen's Kappa index for categorical variables, with ICC values above 0.75 indicating satisfactory validation; (3) Usability - to assess system acceptance based on SUS scores, complemented by analysis of average training time and operational error rates, using the Student's t-test or Mann-Whitney test depending on data distribution.