Seasonal variations of arrhythmias and their impact on mortality in cancer patients with health disparities: A propensity score adjusted machine learning analysis of over 100 million hospitalizations across 3 years

The National Inpatient Sample (NIS) is the largest publicly available U.S. all-payer inpatient healthcare administrative dataset spanning approximately 4,500 hospitals in 50 states, and it was the data source for this study. The dataset includes demographic, comorbidity, procedural, complication, mortality, length of stay, total cost, and hospital characteristics for each hospitalization. The 2016, 2017, and 2018 NIS datasets were selected for this study as they are among the latest available datasets and the first to use International Classification of Disease-10 (ICD-10) coding and as such better reflect current clinical trends in diagnoses, treatments, and outcomes compared to prior years. Study inclusion criteria included all NIS hospitalizations for adults aged 18 years or older during the above index time periods. Per the US Department of Health and Human Services (DHHS) and National Bureau of Economic Research, no review by an Institutional Review Board (IRB) is required for the NIS under the HIPPA Privacy Rule since the NIS is a limited data set (in which 16 direct identified specified by the Privacy Rule have been removed).^{[5, 6]} This study used de-identified data and was conducted according to the ethical principles in the Declaration of Helsinki.

To conduct a more comprehensive analysis more broadly and practically applicable within current healthcare systems, the primary analysis consisted of AI-driven Computational Ethics and policy analysis (AiCE) according to its first empirical (clinical then economic) step then the second ethical-policy step.^[7–9] The first empirical step featured a nationally representative retrospective longitudinal multicenter cohort analysis of inpatient mortality and total cost among all hospitalized adults including by active cancer, arrhythmia, and season (by High Influenza Activity Season [HIA] defined by the US Centers for Disease Control as December to February).^[10] It additionally utilized ML-PSr and deep learning (DL) artificial neural network to assess inpatient mortality and cost. A cost analysis was then conducted using the above clinical results. This empirical step was followed by the final ethical-policy step in which the above AI-augmented empirical results informed a pluralistic-based global bioethical analysis to optimize equitable care for the above patient populations.

Descriptive statistics were performed for the full sample to define the arrhythmia among all adult hospitalizations. Bivariable analysis was then performed according to active cancer (yes/no) and HIA (yes/no) across the full 2016-2018 duration and within each year separately (2016, 2017, and 2018). For continuous variables, independent sample t-tests were performed to compare means, and Wilcoxon rank sum tests were performed for medians. For categorical variables, Pearson Chi-square tests or Fisher exact tests were performed to compare proportions as applicable.

Demographics included age, sex, race, income, insurance, urban density, and region. Comorbidities were selected for analysis (and identified in the dataset by their ICD-10 codes) based on their clinical and/or statistical significance identified in prior published studies and current clinical practice. They included cancer status (active, prior, and metastatic), hypertension, diabetes, coronary artery disease, congestive heart failure, chronic obstructive pulmonary disease, cirrhosis, and chronic kidney disease. The 26 primary malignancies investigated included the brain and nervous system, head or neck, thyroid, breast, lung, esophagus, stomach, pancreas, liver or bile system, rectum or anus, colon, peritoneum, bone or connective tissue system, hematological malignancies (including Hodgkin lymphoma, non-Hodgkin lymphoma, leukemia, and multiple myeloma), melanoma, non-melanoma skin, uterus, cervix, ovarian, prostate, testes, bladder, and renal.

The primary outcome was inpatient mortality (yes/no), and the secondary outcomes were arrhythmia (yes/no; defined as non-sinus rhythm and non-eucardia) and total hospitalization cost (in U.S. dollars [$]). The regression statistical analysis, machine learning analysis, and model optimization methods have been explained in our previously published study.^[11]

Regression analysis featured the particular technique within BAM-PSr performed for the above NIS dataset. ^{[10, 12–14]} This technique integrates ML-PSr, in which the traditional statistical methodology of causal inference-based propensity score analysis is augmented (a) by machine learning capable of handling higher dimensional, more complex, and faster data streams, and then translates its results as informative priors for (b) Bayesian regression. Regression was conducted on the above outcomes and stratified by active cancer and HIA (and additionally included sub-group analysis by primary malignancy among patients with active cancer). The propensity score (modified as a disease risk score) for the likelihood of presenting during HIA was first created (utilizing the same above variables used in the final regression model given the double propensity score adjustment method), a balance was confirmed among blocks, and then the propensity score was included in the final regression models as an adjusted variable.^{[15, 16]}

Mortality disparity analysis was conducted using ML-PSr to assess possible significant disparities independent of clinical confounding.

Cost analysis was conducted by calculating the significant and independent mortality disparities (if any) from ML-PSr multiplied by the 2022 inflation adjusted statistical value of a human life as utilized by the US federal government for health and public policy. ^[17]

The second or ethical-policy step within AiCE was then conducted by integrating the above quantitative analyses with ethical analysis using the pluralistic global bioethical framework of the Personalist Social Contract (PSC).^[18–20] The PSC is a novel integration of modern ethics (principally utilitarianism-informed Rawlsian social contract of political liberalism, bounded by Kantian deontology and informed by feminist, Marxist, deconstructionist, and ecological ethics) and classical ethics (principally Thomistic-Aristotelian virtue ethics, articulated by William Carlo’s esse/essence revision of Norris Clarke’s Strong Thomistic Personalism, a derivative formulation of Thomism as a development of classical Aristotelianism metaphysics).^[21–27] It uniquely articulates the philosophical foundation and framework of the United Nation’s 1948 Universal Declaration of Human Rights, founded on the primary metaphysical principle of human dignity and resultant rights and duties, which has since united the world’s diverse belief systems and 193 nations in what has become the dominant modern ethical framework and foundation of international law.

An academic physician-data scientist, biostatistician, and ethicist (DJM) confirmed that the final analytic models were sufficiently supported by the existing literature and related theories. Mean values are reported with standard deviation (SDs). Fully adjusted regression results were reported with 95% confidence intervals (CIs) with statistical significance set at a 2-tailed p-value of <30.05. Statistical analysis was performed with STATA 17.0 MP edition (STATACorp, College Station, TX, USA), and ML and DL analyses were performed with Java 9 (Oracle, Redwood Chores, CA, USA).

Among 90,869,382 adult hospitalizations, 16,795,379 (18.48%) presented with arrhythmia of whom 3,214,914 (19.14%) were during HIA. Among patients with arrhythmia during HIA, there was a longitudinal increase among those with active cancer from 1.95% (2016) to 7.41% (2017) to 7.54% (2018).

Arrhythmia was more prevalent among adults during HIA compared to non-HIA in 2016 (18.32% versus 17.43%), in 2017 (16.52% versus 15.70%), and 2018 (17.41% versus 16.44%) (all p<0.001) (Table 1; Figure 1). Arrhythmia was less likely for those with active cancer versus those without in 2016 during HIA (19.02% versus 26.36%) and non-HIA (19.44% versus 27.32%), more likely in 2017 during HIA (19.65% versus 16.32%) and non-HIA (19.50% versus 15.44%), and more likely in 2018 during HIA (20.92% versus 17.17%) and non-HIA (20.36% versus 16.17%) (all p<0.001). From 2016-2018, the primary malignancies with the highest prevalence of arrhythmias during HIA were lung (19.60%), leukemia (11.49%), NHL (8.24%), prostate (8.15%), and MM (6.21%) (p<0.001).

Figure 1

In ML-PSr, HIA did not significantly increase the odds of arrhythmia for patients with active cancer (OR 1.01, 95%CI 0.99-1.03, p=0.37), but it did for those without it (OR 1.01, 95%CI 1.01-1.02, p=0.003) (Table 1). Among patients with active cancer, arrhythmia increased mortality odds during HIA (OR 1.19, 95%CI 1.12-1.27, p<0.001) greater than non-arrhythmia during non-HIA (OR 1.17, 95%CI 1.13-1.22, p<0.001). Arrhythmia-related total hospitalization cost was not significantly different during HIA ($2,312.96, p=0.003) and non-HIA seasons ($3,474.68, p<0.001).

Among patients with arrhythmia and active cancer during HIA, the mortality rate increased the most in different primary malignancies in different years as follows: GI in 2016 (OR 1.15, 95%CI 1.01-1.32, p=0.039), leukemia in 2017 (OR 1.31, 95%CI 1.10-1.54, p=0.002), GI in 2018 (OR 1.14, 95%CI 1.01-1.29, p=0.029), and renal in 2018 (OR 1.54, 95%CI 1.06-2.23, p=0.025).

In ML-PSr, among patients with active cancer and arrhythmia, the only significant mortality disparities across all 3 years were for African Americans who had significantly greater mortality rates than Caucasians (OR 1.13, 95%CI 1.03-1.23, p=0.013), independent of the above socio-economic and clinical confounders otherwise. Based on the related population averaged (predictive margins) increased mortality likelihood multiplied by the number of African Americans and Caucasians with cancer and arrhythmia and the 2022 inflation adjusted statistical value of a human life, this translates into an excess annual mortality of 406 African Americans and $5.16 billion among adult patients with cancer and arrhythmia.

The above health and economic results then informed the final or focused computational ethical-policy analysis step of AiCE. The primary material object of this ethical analysis was arrhythmia, the primary context was inpatient healthcare delivery to patients with active cancer and arrhythmia, and the primary formal object or ethical analytic framework is the PSC. Applied to this concrete ethical situation, the formal PSC argument is as follows. (Premise 1) Arrhythmia in active cancer diagnosed inpatient carries a high mortality risk, which may be significantly increased during HIA. (Premise 2) Among this patient population, there appears to be significant disparities for African Americans who are more likely to die inpatient than their Caucasian counterparts, even when adjusting for clinical severity and other such relevant factors. (Premise 3) Life and equal societal protection are fundamental individual and state rights logically derivative from the human person’s dignity and are politically enshrined across the United Nations, multiple other international institutions, and the majority of nations’ constitutions and legal statutes. (Premise 4) Respect for dignity at the individual level requires respecting the person’s rights to goods (beginning with the primary good of life) necessary for the person to develop through just and stable commitment to the common good and thus the community in reciprocal care for the individual. (Premise 5) Respect for dignity at the communal level requires respecting other cultures as the communal manifestations of their constitutive individuals seeking through justice the common good (as the objective good of the community, entailing the objective good of individual flourishing, and subjectively experienced as the ultimate individual good of self-actualization through justice). (Premise 6) Racial disparities in hospitalization outcomes for patients with arrhythmia and cancer can produce disproportionate morbidity and even related mortality in those clinical and social sub-communities, resulting in a disproportionate threat to the preservation of those persons and related cultures, leading to the global human community’s impoverishment with the loss or diminishment of those individuals and cultures. (Premise 7) The reduction of such disparities may result in over hundreds of additional patients living and $5 billion saved annually. (Premise 8) Disparities in the effective and equitable chronic and acute treatment of patients with arrhythmia and cancer undermines respect for the rights of patients and respect for their cultures, which is critical to the wellbeing of societies that encompass all peoples and cultures. (Premise 9) (Conclusion) Therefore, clinical, economic, and ethical justifications support greater healthcare policy and healthcare system investment reducing disparities in the health and cost burden of arrhythmia, particularly in those with active cancer to ensure equitable, value-based healthcare for all peoples regardless of socio-demographics.