The adequate management of healthcare resources is essential to provide optimal patient care, especially under high stress/strain conditions or limited resources. Benchmarking is helpful to evaluate the performance and quality of care within these systems and provide targets for improvement. This is especially important for the intensive care units (ICUs), which deal with complex cases, high costs and provides relevant insights for treating severe diseases. Under usual conditions, assessing performance in intensive care is complex since metrics must account for the patient s case-mix and the unit s organizational settings. When high strain or stress conditions arise, the resource use increases, and the unit performs in unusual conditions. One of these settings is the COVID-19 pandemic, which has overwhelmed healthcare systems worldwide since December 2019, notably intensive care resources. This thesis aims to evaluate the use of resources and performance of healthcare systems under the perspectives of before and during the COVID-19 pandemic. Using data from Brazilian hospitals, we developed six individual studies aiming to perform ICU benchmarking in a pre-pandemic period and understand the use of ICU resources and outcomes during the progression of the pandemic. We managed each work as data science projects following the Data Science Life Cycle, under the Design Science research framework, and used different statistical approaches to analyze data. Our main results show that before the pandemic, the assessment of quality-of-care metrics and active surveillance of infections were associated with efficient ICU units. During the pandemic period, the use of resources and outcomes varied temporally and regionally in Brazil. North and Northeast, regions with more vulnerable healthcare systems, showed poor outcomes and lower availability of ICU resources than South and Southeast regions. The impact on the Brazilian healthcare system was intensified in a second pandemic surge, showing increasing use of respiratory resources and mortality. Finally, when evaluating the mortality evolution in a network of private hospitals that underwent preparedness and presented large availability of resources, the overall mortality was low and decreased over time. Noninvasive respiratory support was independently associated with a reduction in mortality.