Potential and limitations of machine meta-learning (ensemble) methods for predicting COVID-19 mortality in a large inhospital Brazilian dataset.

The majority of early prediction scores and methods to predict COVID-19 mortality are bound by methodological flaws and technological limitations (e.g., the use of a single prediction model). Our aim is to provide a thorough comparative study that tackles those methodological issues, considering multiple techniques to build mortality prediction models, including modern machine learning (neural) algorithms and traditional statistical techniques, as well as meta-learning (ensemble) approaches. This study used a dataset from a multicenter cohort of 10,897 adult Brazilian COVID-19 patients, admitted from March/2020 to November/2021, including patients [median age 60 (interquartile range 48-71), 46% women]. We also proposed new original population-based meta-features that have not been devised in the literature. Stacking has shown to achieve the best results reported in the literature for the death prediction task, improving over previous state-of-the-art by more than 46% in Recall for predicting death, with AUROC 0.826 and MacroF1 of 65.4%. The newly proposed meta-features were highly discriminative of death, but fell short in producing large improvements in final prediction performance, demonstrating that we are possibly on the limits of the prediction capabilities that can be achieved with the current set of ML techniques and (meta-)features. Finally, we investigated how the trained models perform on different hospitals, showing that there are indeed large differences in classifier performance between different hospitals, further making the case that errors are produced by factors that cannot be modeled with the current predictors.

Synchronous Teleconsultation and Monitoring Service Targeting COVID-19: Leveraging Insights for Postpandemic Health Care.

BACKGROUND: Although a great number of teleconsultation services have been developed during the COVID-19 pandemic, studies assessing usability and health care provider satisfaction are still incipient. OBJECTIVE: This study aimed to describe the development, implementation, and expansion of a synchronous teleconsultation service targeting patients with symptoms of COVID-19 in Brazil, as well as to assess its usability and health care professionals' satisfaction. METHODS: This mixed methods study was developed in 5 phases: (1) the identification of components, technical and functional requirements, and system architecture; (2) system and user interface development and validation; (3) pilot-testing in the city of Divinópolis; (4) expansion in the cities of Divinópolis, Teófilo Otoni, and Belo Horizonte for Universidade Federal de Minas Gerais faculty and students; and (5) usability and satisfaction assessment, using Likert-scale and open-ended questions. RESULTS: During pilot development, problems contacting users were solved by introducing standardized SMS text messages, which were sent to users to obtain their feedback and keep track of them. Until April 2022, the expanded system served 31,966 patients in 146,158 teleconsultations. Teleconsultations were initiated through chatbot in 27.7% (40,486/146,158) of cases. Teleconsultation efficiency per city was 93.7% (13,317/14,212) in Teófilo Otoni, 92.4% (11,747/12,713) in Divinópolis, and 98.8% (4981/5041) in Belo Horizonte (university campus), thus avoiding in-person assistance for a great majority of patients. In total, 50 (83%) out of 60 health care professionals assessed the system's usability as satisfactory, despite a few system instability problems. CONCLUSIONS: The system provided updated information about COVID-19 and enabled remote care for thousands of patients, which evidenced the critical role of telemedicine in expanding emergency services capacity during the pandemic. The dynamic nature of the current pandemic required fast planning, implementation, development, and updates in the system. Usability and satisfaction assessment was key to identifying areas for improvement. The experience reported here is expected to inform telemedicine strategies to be implemented in a postpandemic scenario.

Implementation of a text messaging intervention to patients on warfarin therapy in Brazilian primary care units: a quasi-experimental study.

BACKGROUND: Warfarin remains the most affordable oral anticoagulant in many countries. However, it may have serious side effects, and the success of the therapy depends on the patient's understanding of the medication and their adherence to treatment. The use of short messages services (SMS) is a strategy that can be used to educate patients, but there are no studies evaluating this intervention in patients taking warfarin. Therefore, we aimed to develop, implement, and assess the feasibility of an intervention using SMS to primary care patients taking warfarin in a medium-sized Brazilian city. METHODS: A bank of 79 SMS was drafted and validated by an expert panel. During 6 months, three times a week, patients received messages about anticoagulation with warfarin. At baseline and after 3 months, we assessed their knowledge and adherence with validated instruments. At the end of the follow-up, participants answered a satisfaction questionnaire. Subsequently, a scale-up phase was conducted, with another round of the intervention including 82 participants (29 from the first phase and 53 newly recruited). Seven months after the end of the scale-up, we asked the patients for their insights about the long-term effects of this program. All patients signed informed consent. The study was approved by the Research and Ethics committee of the Universidade Federal de Minas Gerais. RESULTS: In the pilot, 33 (89.2%) patients completed the follow-up. Among the participants who answered the satisfaction questionnaire (n = 29), 86.2% considered that the intervention motivated a healthy lifestyle and improved their understanding of warfarin therapy. All patients were willing to continue receiving the messages. Adherence measured by the Measure of Adherence to Treatment (MAT) test was high in the pre-intervention assessment and remained high (96.7% vs. 93.3%; p = 1.0000). The proportion of patients who achieved > 75% correct answers on the Oral Anticoagulation Knowledge (OAK) test increased from 6.5% to 25.6, p = 0.0703. In the scale-up, 23 patients answered the long-term assessment questionnaire. The main long-term knowledge reported was dietary information. Nine patients received the messages but did not remember their content. CONCLUSION: The intervention was well-accepted and had a positive impact on patient's knowledge about oral anticoagulation therapy. The scale-up assessment reinforced the need to constantly monitor digital interventions.