UNASSIGNED: Little is known about stage 1 and 2 pressure injuries that are health care-acquired. We report incidence rates of health care-acquired stage 1 and stage 2 pressure injuries, and, estimate the excess length of stay using four competing analytic methods. We discuss the merits of the different approaches.
UNASSIGNED: We calculated monthly incidence rates for stage 1 and 2 health care-acquired pressure injuries occurring in a large Singapore acute care hospital. To estimate excess stay, we conducted unadjusted comparisons with a control cohort, performed linear regression and then generalized linear regression with a gamma distribution. Finally, we fitted a simple state-based model. The design for the cost attribution work was a retrospective matched cohort study.
UNASSIGNED: Incidence rates in 2016 were 0.553% (95% confidence interval [CI] 0.55, 0.557) and 0.469% (95% CI 0.466, 0.472) in 2017. For data censored at 60 days\' maximum stay, the unadjusted comparisons showed the highest excess stay at 17.68 (16.43-18.93) days and multi-state models showed the lowest at 1.22 (0.19, 2.23) days.
UNASSIGNED: Poor-quality methods for attribution of excess length of stay to pressure injury generate inflated estimates that could mislead decision makers. The findings from the multi-state model, which is an appropriate method, are plausible and illustrate the likely bed-days saved from lowering the risk of these events. Stage 1 and 2 pressure injuries are common and increase costs by prolonging the length of stay. There will be economic value investing in prevention. Using biased estimates of excess length of stay will overstate the potential value of prevention.

Post hoc attribution methods can provide insights into the learned patterns from deep neural networks (DNNs) trained on high-throughput functional genomics data. However, in practice, their resultant attribution maps can be challenging to interpret due to spurious importance scores for seemingly arbitrary nucleotides. Here, we identify a previously overlooked attribution noise source that arises from how DNNs handle one-hot encoded DNA. We demonstrate this noise is pervasive across various genomic DNNs and introduce a statistical correction that effectively reduces it, leading to more reliable attribution maps. Our approach represents a promising step towards gaining meaningful insights from DNNs in regulatory genomics.