Designing Accountable Health Care Algorithms: Lessons from Covid-19 Contact Tracing

AI THEME ISSUE: How can health care organizations ensure that there is accountability of algorithms for accuracy, bias, and the wide range of unintended consequences when deployed in real-world settings? A machine-learning system for Covid-19 contact tracing serves as a model to scope out, develop, interrogate, and assess an algorithmic solution that produces improvements in care, mitigates risk, and enables evaluation by many stakeholders.

The Influence of the COVID-19 Pandemic on Author Gender and Manuscript Acceptance Rates Among Pulmonary and Critical Care Journals

Rationale: The COVID-19 pandemic has profoundly disrupted academic endeavors worldwide, disproportionately influencing female physicians and scientists, and potentially negatively impacting the quantity and quality of research unrelated to COVID-19. We sought to evaluate whether the COVID-19 pandemic was associated with a) changes in manuscript submission and acceptance rates among pulmonary/critical care medicine journals, and b) gender-based disparities in these rates. Methods: We used a validated database of author gender to analyze first, senior, and corresponding authorship of all manuscripts submitted to four pulmonary/critical care journals based in the United States (US) between January 1, 2018 and December 31, 2020. We constructed interrupted time series regression models to evaluate whether the proportion of female first and senior authors of non-COVID-19 original research manuscripts changed coincident with the COVID- 19 pandemic. Next, we performed multivariable logistic regressions to evaluate the association of author gender with acceptance of original research manuscript after adjusting for subject category, author world region, and journal. We then conducted sensitivity analyses including all non-original research manuscripts. Results: Among 8,332 original research submissions, women comprised 39.9% and 28.3% of first and senior authors, respectively. We found no change in the proportion of female first or senior authors of non-COVID-19 or COVID-19 submitted research manuscripts during the COVID-19 era. Although female first authorship was not associated with manuscript acceptance, female senior authorship was associated with decreased acceptance of non-COVID research manuscripts (adjusted odds ratio [aOR] 0.84, 95% confidence interval [CI] 0.71-0.99). Non-COVID-19 manuscripts submitted during the COVID-19 era had reduced odds of acceptance, regardless of author gender (first author analysis: aOR 0.46 [95% CI 0.36-0.59];senior author analysis: aOR 0.46 [95% CI 0.37-0.57]). Conclusions: Women comprise a minority of first and senior authors among research manuscripts submitted to US-based pulmonary and critical care journals, but this was not influenced by the pandemic. Female senior authorship of non-COVID-19 research manuscripts was associated with reduced odds of acceptance. However, non-COVID manuscripts were nearly 50% less likely to be accepted during the COVID-19 era, independent of author gender. These results provide important insights into the influence of the pandemic on gender disparities in academic medicine and on the publication of high-quality research focused on topics unrelated to COVID-19.

Leveraging administrative data for bias audits: Assessing disparate coverage with mobility data for COVID-19 Policy

Anonymized smartphone-based mobility data has been widely adopted in devising and evaluating COVID-19 response strategies such as the targeting of public health resources. Yet little attention has been paid to measurement validity and demographic bias, due in part to the lack of documentation about which users are represented as well as the challenge of obtaining ground truth data on unique visits and demographics. We illustrate how linking large-scale administrative data can enable auditing mobility data for bias in the absence of demographic information and ground truth labels. More precisely, we show that linking voter roll data - -containing individual-level voter turnout for specific voting locations along with race and age - -can facilitate the construction of rigorous bias and reliability tests. Using data from North Carolina's 2018 general election, these tests illuminate a sampling bias that is particularly noteworthy in the pandemic context: older and non-white voters are less likely to be captured by mobility data. We show that allocating public health resources based on such mobility data could disproportionately harm high-risk elderly and minority groups. © 2021 Owner/Author.