IntroductionThe Department of Veterans Affairs (VA) is the largest provider of HIV care in the United States. Changes in healthcare delivery became necessary with the COVID‐19 pandemic. We compared HIV healthcare delivery during the first year of the COVID‐19 pandemic to a prior similar calendar period.MethodsWe included 27,674 people with HIV (PWH) enrolled in the Veterans Aging Cohort Study prior to 1 March 2019, with ≥1 healthcare encounter from 1 March 2019 to 29 February 2020 (2019) and/or 1 March 2020 to 28 February 2021 (2020). We counted monthly general medicine/infectious disease (GM/ID) clinic visits and HIV‐1 RNA viral load (VL) tests. We determined the percentage with ≥1 clinic visit (in‐person vs. telephone/video [virtual]) and ≥1 VL test (detectable vs. suppressed) for 2019 and 2020. Using pharmacy records, we summarized antiretroviral (ARV) medication refill length (<90 vs. ≥90 days) and monthly ARV coverage.ResultsMost patients had ≥1 GM/ID visit in 2019 (96%) and 2020 (95%). For 2019, 27% of visits were virtual compared to 64% in 2020. In 2019, 82% had VL measured compared to 74% in 2020. Of those with VL measured, 92% and 91% had suppressed VL in 2019 and 2020. ARV refills for ≥90 days increased from 39% in 2019 to 51% in 2020. ARV coverage was similar for all months of 2019 and 2020 ranging from 76% to 80% except for March 2019 (72%). Women were less likely than men to be on ARVs or to have a VL test in both years.ConclusionsDuring the COVID‐19 pandemic, the VA increased the use of virtual visits and longer ARV refills, while maintaining a high percentage of patients with suppressed VL among those with VL measured. Despite decreased in‐person services during the pandemic, access to ARVs was not disrupted. More follow‐up time is needed to determine whether overall health was impacted by the use of differentiated service delivery and to evaluate whether a long‐term shift to increased virtual healthcare could be beneficial, particularly for PWH in rural areas or with transportation barriers. Programmes to increase ARV use and VL testing for women are needed.

Deep learning models have been the subject of study from various perspectives, for example, their training process, interpretation, generalization error, robustness to adversarial attacks, etc. A trained model is defined by its decision boundaries, and therefore, many of the studies about deep learning models speculate about the decision boundaries, and sometimes make simplifying assumptions about them. So far, finding exact points on the decision boundaries of trained deep models has been considered an intractable problem. Here, we compute exact points on the decision boundaries of these models and provide mathematical tools to investigate the surfaces that define the decision boundaries. Through numerical results, we confirm that some of the speculations about the decision boundaries are accurate, some of the computational methods can be improved, and some of the simplifying assumptions may be unreliable, for models with nonlinear activation functions. We advocate for verification of simplifying assumptions and approximation methods, wherever they are used. Finally, we demonstrate that the computational practices used for finding adversarial examples can be improved and computing the closest point on the decision boundary reveals the weakest vulnerability of a model against adversarial attack.Preprint. Under review.

No abstract