Don't let your analysis go to seed: on the impact of random seed on machine learning-based causal inference.

Machine learning techniques for causal effect estimation can enhance the reliability of epidemiologic analyses, reducing their dependence on correct model specifications. However, the stochastic nature of many machine learning algorithms implies that the results derived from such approaches may be influenced by the random seed that is set prior to model fitting. In this work, we highlight the substantial influence of random seeds on a popular approach for machine learning-based causal effect estimation, namely doubly robust estimators. We illustrate that varying seeds can yield divergent scientific interpretations of doubly robust estimates produced from the same dataset. We propose techniques for stabilizing results across random seeds and, through an extensive simulation study, demonstrate that these techniques effectively neutralize seed-related variability without compromising the statistical efficiency of the estimators. Based on these findings, we offer practical guidelines to minimize the influence of random seeds in real-world applications, and we encourage researchers to explore variability due to random seed when implementing any method that involves random steps.

Paid Sick Leave Among Working Cancer Survivors and Its Associations With Use of Preventive Services in the United States.

BACKGROUND: We sought to examine the lack of paid sick leave among working cancer survivors by sociodemographic/socioeconomic and employment characteristics and its association with preventive services use in the United States. METHODS: Working cancer survivors (ages 18-64 years; n=7,995; weighted n=3.43 million) were identified using 2001-2018 National Health Interview Survey data. Adjusted prevalence of lack of paid sick leave by sociodemographic and socioeconomic characteristics, as well as job sector, working hours, and employer size, were generated using multivariable logistic regression models. Separate analyses examined the associations of lack of paid sick leave with use of various preventive services. RESULTS: Of all working cancer survivors, 36.4% lacked paid sick leave (n=2,925; weighted n=1.25 million), especially those working in food/agriculture/construction/personal services occupations or industries (ranging from 54.9% to 88.5%). In adjusted analyses, working cancer survivors with lower household income (<200% of the federal poverty level, 48.7%), without a high school degree (43.3%), without health insurance coverage (70.6%), and who were self-employed (89.5%), were part-time workers (68.2%), or worked in small businesses (<50 employees, 48.8%) were most likely to lack paid sick leave. Lack of paid sick leave was associated with lower use of influenza vaccine (ages 18-39 years, 21.3% vs 33.3%; ages 40-49 years, 25.8% vs 38.3%; ages 50-64 years, 46.3% vs 52.4%; P<.001 for all), cholesterol screening (ages 18-39 years, 43.1% vs 62.5%; P<.05), and blood pressure check (ages 18-39 years, 43.1% vs 62.5%; P<.05) compared with survivors having paid sick leave. CONCLUSIONS: In the United States, more than one-third of all working cancer survivors and more than half of survivors working for small employers and in certain occupations/industries lack paid sick leave. Survivors with lower household income or educational attainment are particularly vulnerable. Moreover, lack of paid sick leave is associated with lower use of some recommended preventive services, suggesting that ensuring working cancer survivors have access to paid sick leave may be an important mechanism for reducing health disparities.

Association of Medical Financial Hardship and Mortality Among Cancer Survivors in the United States.

BACKGROUND: Cancer survivors frequently experience medical financial hardship in the United States. Little is known, however, about long-term health consequences. This study examines the associations of financial hardship and mortality in a large nationally representative sample of cancer survivors. METHODS: We identified cancer survivors aged 18-64 years (n = 14 917) and 65-79 years (n = 10 391) from the 1997-2014 National Health Interview Survey and its linked mortality files with vital status through December 31, 2015. Medical financial hardship was measured as problems affording care or delaying or forgoing any care because of cost in the past 12 months. Risk of mortality was estimated with separate weighted Cox proportional hazards models by age group with age as the timescale, controlling for the effects of sociodemographic characteristics. Health insurance coverage was added sequentially to multivariable models. RESULTS: Among cancer survivors aged 18-64 years and 65-79 years, 29.6% and 11.0%, respectively, reported financial hardship in the past 12 months. Survivors with hardship had higher adjusted mortality risk than their counterparts in both age groups: 18-64 years (hazard ratio [HR] = 1.17, 95% confidence interval [CI] = 1.04 to 1.30) and 65-79 years (HR = 1.14, 95% CI = 1.02 to 1.28). Further adjustment for health insurance reduced the magnitude of association of hardship and mortality among survivors aged 18-64 years (HR = 1.09, 95% CI = 0.97 to 1.24). Adjustment for supplemental Medicare coverage had little effect among survivors aged 65-79 years (HR = 1.15, 95% CI = 1.02 to 1.29). CONCLUSION: Medical financial hardship was associated with mortality risk among cancer survivors in the United States.

Financial Hardship, Healthcare Utilization, and Health Among U.S. Cancer Survivors.

INTRODUCTION: This study examined associations of both medical and nonmedical financial hardships with healthcare utilization and self-rated health among cancer survivors. METHODS: The National Health Interview Survey (2013-2017) was used to identify cancer survivors (aged 18-64 years: n=4,939; aged ≥65 years: n=6,972). A total of 4 levels of medical financial hardship intensities were created with measures from material, psychological, and behavioral domains. A total of 5 levels of nonmedical financial hardship intensities were created with measures in food insecurity and worry about other economic needs (e.g., housing expenses). Generalized ordinal logistic regression examined associations between medical and nonmedical financial hardship intensities and emergency department visits, use of preventive services and cancer screenings, and self-rated health. All analyses were performed in 2019. RESULTS: In adjusted analyses, cancer survivors with higher medical financial hardship intensity (Level 4 vs Level 1; aged 18-64 years: 42% vs 26.2%, p<0.001; aged ≥65 years: 37.6% vs 24.3%, p=0.001) and higher nonmedical financial hardship intensity (Level 5 vs Level 1; aged 18-64 years: 37.2% vs 27.9%, p=0.011) had more emergency department visits. Moreover, cancer survivors with higher medical financial hardship intensity had lower influenza vaccine (Level 4 vs Level 1; aged 18-64 years: 45.6% vs 52.5%, p=0.036; aged ≥65 years: 64.6% vs 75.6%, p=0.008) and lower breast cancer screening levels (Level 4 vs Level 1; 46.8% vs 61.2%, p=0.001). Similar patterns were found between higher financial hardship intensities and worse self-rated health. CONCLUSIONS: Higher medical and nonmedical financial hardships are independently associated with more emergency department visits, lower receipt of some preventive services, and worse self-rated health in cancer survivors. With growing healthcare costs, unmet medical and nonmedical financial needs may worsen health disparities among cancer survivors.