Explainable Artificial Intelligence to Identify Dosimetric Predictors of Toxicity in Patients with Locally Advanced Non-Small Cell Lung Cancer: A Secondary Analysis of RTOG 0617.

PURPOSE: Dosimetric predictors of toxicity in patients treated with definitive chemoradiation for locally advanced non-small cell lung cancer are often identified through trial and error. This study used machine learning (ML) and explainable artificial intelligence to empirically characterize dosimetric predictors of toxicity in patients treated as part of a prospective clinical trial. METHODS AND MATERIALS: A secondary analysis of the Radiation Therapy Oncology Group (RTOG) 0617 trial was performed. Multiple ML models were trained to predict grade ≥3 pulmonary, cardiac, and esophageal toxicities using clinical and dosimetric features. Model performance was evaluated using the area under the curve (AUC). The best performing model for each toxicity was explained using the Shapley Additive Explanation (SHAP) framework; SHAP values were used to identify relevant dosimetric thresholds and were converted to odds ratios (ORs) with confidence intervals (CIs) generated using bootstrapping to obtain quantitative measures of risk. Thresholds were validated using logistic regression. RESULTS: The best-performing models for pulmonary, cardiac, and esophageal toxicities, outperforming logistic regression, were extreme gradient boosting (AUC, 0.739), random forest (AUC, 0.706), and naive Bayes (AUC, 0.721), respectively. For pulmonary toxicity, thresholds of a mean dose >18 Gy (OR, 2.467; 95% CI, 1.049-5.800; P = .038) and lung volume receiving ≥20 Gy (V20) > 37% (OR, 2.722; 95% CI, 1.034-7.163; P = .043) were identified. For esophageal toxicity, thresholds of a mean dose >34 Gy (OR, 4.006; 95% CI, 2.183-7.354; P < .001) and V20 > 37% (OR, 3.725; 95% CI, 1.308-10.603; P = .014) were identified. No significant thresholds were identified for cardiac toxicity. CONCLUSIONS: In this data set, ML approaches validated known dosimetric thresholds and outperformed logistic regression at predicting toxicity. Furthermore, using explainable artificial intelligence, clinically useful dosimetric thresholds might be identified and subsequently externally validated.

Condition-Dependent and Dynamic Impacts of Indoor Masking Policies for Coronavirus Disease 2019 Mitigation: A Nationwide, Interrupted Time-Series Analysis.

BACKGROUND: The effectiveness and sustainability of masking policies as a pandemic control measure remain uncertain. Our aim was to evaluate different masking policy types on severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) incidence and to identify factors and conditions impacting effectiveness. METHODS: Nationwide, retrospective cohort study of US counties from 4/4/2020-28/6/2021. Policy impacts were estimated using interrupted time-series models with the masking policy change date (eg, recommended-to-required, no-recommendation-to-recommended, no-recommendation-to-required) modeled as the interruption. The primary outcome was change in SARS-CoV-2 incidence rate during the 12 weeks after the policy change; results were stratified by coronavirus disease 2019 (COVID-19) risk level. A secondary analysis was completed using adult vaccine availability as the policy change. RESULTS: In total, N = 2954 counties were included (2304 recommended-to-required, 535 no-recommendation-to-recommended, 115 no-recommendation-to-required). Overall, indoor mask mandates were associated with 1.96 fewer cases/100 000/week (cumulative reduction of 23.52/100 000 residents during the 12 weeks after policy change). Reductions were driven by communities with critical and extreme COVID-19 risk, where masking mandated policies were associated with an absolute reduction of 5 to 13.2 cases/100 000 residents/week (cumulative reduction of 60 to 158 cases/100 000 residents over 12 weeks). Impacts in low- and moderate-risk counties were minimal (<1 case/100 000 residents/week). After vaccine availability, mask mandates were not associated with significant reductions at any risk level. CONCLUSIONS: Masking policy had the greatest impact when COVID-19 risk was high and vaccine availability was low. When transmission risk decreases or vaccine availability increases, the impact was not significant regardless of mask policy type. Although often modeled as having a static impact, masking policy effectiveness may be dynamic and condition dependent.

Lack of Association Between Pandemic School Mode Policy and Pediatric Body Mass Index Percentile Over Time.

During the 2020-2021 academic year, schools across the country were closed for prolonged periods. Prior research suggests that children tend to gain more weight during times of extended school closures, such as summer vacation; however, little is known about the impact of school learning mode on changes. Thus, the aim of this study was to measure the association between school mode (in-person, hybrid, remote) and pediatric body mass index (BMI) percentile increases over time. In this longitudinal, statewide retrospective cohort study in Massachusetts, we found that BMI percentile increased in elementary and middle school students in all learning modes, and that increases slowed but did not reverse following the statewide reopening. Body mass percentile increases were highest in elementary school aged children. Hispanic ethnicity and receipt of Medicaid insurance were also associated with increases. Additional research is needed to identify strategies to combat pediatric body mass percentile increases and to address disparities.

The Impact of the Different Stages of COVID-19, Time of the Week and Exercise Frequency on Mental Distress in Men and Women.

The COVID-19 pandemic produced life disturbances and loss of routine which affected diet and sleep quality as well as physical exercise frequency. Interestingly, mental distress was higher even in those who exercised. The purpose of this study was to assess exercise frequency in relation to different levels of mental distress severity in men and women while accounting for working days and weekends. A de-identified secondary data set was analyzed. Regression analyses produced models of the different stages of COVID-19 in relation to physical exercise frequency and mental distress levels. Margin analysis generated predictive models that could be used prophylactically to customize physical exercise frequencies in men and women to reduce their risk of mental distress during future pandemics. Mental distress during the lockdown and after ease of restrictions was associated with different physical exercise frequencies, with a noticeable difference between men and women. During a pandemic, sedentary men are more likely to be mentally distressed during working days. Nevertheless, moderately active, but not very active women, may be less distressed during pandemic weekends. These findings may provide a framework to optimize mental health during different stages of a pandemic by customizing physical exercise frequencies based on gender and time of the week.

A decision analytic approach for social distancing policies during early stages of COVID-19 pandemic.

The COVID-19 pandemic has become a crucial public health problem in the world that disrupted the lives of millions in many countries including the United States. In this study, we present a decision analytic approach which is an efficient tool to assess the effectiveness of early social distancing measures in communities with different population characteristics. First, we empirically estimate the reproduction numbers for two different states. Then, we develop an age-structured compartmental simulation model for the disease spread to demonstrate the variation in the observed outbreak. Finally, we analyze the computational results and show that early trigger social distancing strategies result in smaller death tolls; however, there are relatively larger second waves. Conversely, late trigger social distancing strategies result in higher initial death tolls but relatively smaller second waves. This study shows that decision analytic tools can help policy makers simulate different social distancing scenarios at the early stages of a global outbreak. Policy makers should expect multiple waves of cases as a result of the social distancing policies implemented when there are no vaccines available for mass immunization and appropriate antiviral treatments.

The impact of school opening model on SARS-CoV-2 community incidence and mortality.

The role that traditional and hybrid in-person schooling modes contribute to the community incidence of SARS-CoV-2 infections relative to fully remote schooling is unknown. We conducted an event study using a retrospective nationwide cohort evaluating the effect of school mode on SARS-CoV-2 cases during the 12 weeks after school opening (July-September 2020, before the Delta variant was predominant), stratified by US Census region. After controlling for case rate trends before school start, state-level mitigation measures and community activity level, SARS-CoV-2 incidence rates were not statistically different in counties with in-person learning versus remote school modes in most regions of the United States. In the South, there was a significant and sustained increase in cases per week among counties that opened in a hybrid or traditional mode versus remote, with weekly effects ranging from 9.8 (95% confidence interval (CI) = 2.7-16.1) to 21.3 (95% CI = 9.9-32.7) additional cases per 100,000 persons, driven by increasing cases among 0-9 year olds and adults. Schools can reopen for in-person learning without substantially increasing community case rates of SARS-CoV-2; however, the impacts are variable. Additional studies are needed to elucidate the underlying reasons for the observed regional differences more fully.

The Impact of School Opening Model on SARS-CoV-2 Community Incidence and Mortality: A Nationwide Cohort Study.

The role that in-person schooling contributes to community incidence of SARS-CoV-2 infections and deaths remains unknown. We conducted an event study evaluating the effect of in-person school on SARS-CoV-2 cases and deaths per 100,000 persons during the 12-weeks following school opening, stratified by US Census region. There was no impact of in-person school opening and COVID-19 deaths. In most regions, COVID-19 incidence rates were not statistically different in counties with in-person versus remote school modes. However, in the South, there was a significant and sustained increase in cases per week among counties that opened for in-person learning versus remote learning, with weekly effects ranging from 7.8 (95% CI: 1.2-14.5) to 18.9 (95% CI: 7.9-29.9) additional cases per 100,000, driven by increases among 0-9 year olds and adults.

Inter-urban mobility via cellular position tracking in the southeast Songliao Basin, Northeast China.

Position tracking using cellular phones can provide fine-grained traveling data between and within cities on hourly and daily scales, giving us a feasible way to explore human mobility. However, such fine-grained data are traditionally owned by private companies and is extremely rare to be publicly available even for one city. Here, we present, to the best of our knowledge, the largest inter-city movement dataset using cellular phone logs. Specifically, our data set captures 3-million cellular devices and includes 70 million movements. These movements are measured at hourly intervals and span a week-long duration. Our measurements are from the southeast Sangliao Basin, Northeast China, which span three cities and one country with a collective population of 8 million people. The dynamic, weighted and directed mobility network of inter-urban divisions is released in simple formats, as well as divisions' GPS coordinates to motivate studies of human interactions within and between cities.

Optimal multi-source forecasting of seasonal influenza.

Forecasting the emergence and spread of influenza viruses is an important public health challenge. Timely and accurate estimates of influenza prevalence, particularly of severe cases requiring hospitalization, can improve control measures to reduce transmission and mortality. Here, we extend a previously published machine learning method for influenza forecasting to integrate multiple diverse data sources, including traditional surveillance data, electronic health records, internet search traffic, and social media activity. Our hierarchical framework uses multi-linear regression to combine forecasts from multiple data sources and greedy optimization with forward selection to sequentially choose the most predictive combinations of data sources. We show that the systematic integration of complementary data sources can substantially improve forecast accuracy over single data sources. When forecasting the Center for Disease Control and Prevention (CDC) influenza-like-illness reports (ILINet) from week 48 through week 20, the optimal combination of predictors includes public health surveillance data and commercially available electronic medical records, but neither search engine nor social media data.