Local and Distributed Machine Learning for Inter-hospital Data Utilization: An Application for TAVI Outcome Prediction

Background: Machine learning models have been developed for numerous medical prognostic purposes. These models are commonly developed using data from single centers or regional registries. Including data from multiple centers improves robustness and accuracy of prognostic models. However, data sharing between multiple centers is complex, mainly because of regulations and patient privacy issues.Objective: We aim to overcome data sharing impediments by using distributed ML and local learning followed by model integration. We applied these techniques to develop 1-year TAVI mortality estimation models with data from two centers without sharing any data.Methods: A distributed ML technique and local learning followed by model integration was used to develop models to predict 1-year mortality after TAVI. We included two populations with 1,160 (Center A) and 631 (Center B) patients. Five traditional ML algorithms were implemented. The results were compared to models created individually on each center.Results: The combined learning techniques outperformed the mono-center models. For center A, the combined local XGBoost achieved an AUC of 0.67 (compared to a mono-center AUC of 0.65) and, for center B, a distributed neural network achieved an AUC of 0.68 (compared to a mono-center AUC of 0.64).Conclusion: This study shows that distributed ML and combined local models techniques, can overcome data sharing limitations and result in more accurate models for TAVI mortality estimation. We have shown improved prognostic accuracy for both centers and can also be used as an alternative to overcome the problem of limited amounts of data when creating prognostic models.

Date of introduction and epidemiologic patterns of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in Mogadishu, Somalia: estimates from transmission modelling of satellite-based excess mortality data in 2020

Background: In countries with weak surveillance systems, confirmed coronavirus disease 2019 (COVID-19) deaths are likely to underestimate the pandemic’s death toll. Many countries also have incomplete vital registration systems, hampering excess mortality estimation. Here, we fitted a dynamic transmission model to satellite imagery data of cemeteries in Mogadishu, Somalia during 2020 to estimate the date of introduction and other epidemiologic parameters of the early spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in this low-income, crisis-affected setting. Methods: We performed Markov chain Monte Carlo (MCMC) fitting with an age-structured compartmental COVID-19 model to provide median estimates and credible intervals for the date of introduction, the basic reproduction number (R0) and the effect of non-pharmaceutical interventions (NPIs) up to August 2020. Results: Under the assumption that excess deaths in Mogadishu March-August 2020 were attributable to SARS-CoV-2 infections, we arrived at median estimates of November-December 2019 for the date of introduction and low R0 estimates (1.4-1.7) reflecting the slow and early rise and long plateau of excess deaths. The date of introduction, the amount of external seeding, the infection fatality rate (IFR) and the effectiveness of NPIs are correlated parameters and not separately identifiable in a narrow range from deaths data. Nevertheless, to obtain introduction dates no earlier than November 2019 a higher population-wide IFR (≥0.7%) had to be assumed than obtained by applying age-specific IFRs from high-income countries to Somalia’s age structure. Conclusions: Model fitting of excess mortality data across a range of plausible values of the IFR and the amount of external seeding suggests an early SARS-CoV-2 introduction event may have occurred in Somalia in November-December 2019. Transmissibility in the first epidemic wave was estimated to be lower than in European settings. Alternatively, there was another, unidentified source of sustained excess mortality in Mogadishu from March to August 2020.

Follow-up after emergency laparotomy suggests high 1- and 5-year mortality with risk stratified by ASA

Introduction We investigated all-cause mortality following emergency laparotomy at 1 and 5 years. We aimed to establish a basis from which to advise patients and relatives on long-term mortality. Methods Local data from a historical audit of emergency laparotomies from 2010 to 2012 were combined with National Emergency Laparotomy Audit (NELA) data from 2017 to 2020. Covariates collected included deprivation status, preoperative blood work, baseline renal function, age, American Society of Anesthesiologists (ASA) grade, operative time, anaesthetic time and gender. Associations between covariates and survival were determined using multivariate logistic regression and Kaplan–Meier analysis. We used patients undergoing laparoscopic cholecystectomy between 2015 and 2020 as controls. Results ASA grade was the best discriminator of long-term outcome following laparotomy (n=894) but was not a predictor of survival following cholecystectomy (n=1,834), with mortality being significantly greater in the laparotomy group. Following cholecystectomy, 95% confidence intervals for survival at 5 years were 98–99%. Following laparotomy these intervals were: ASA grade 1, 79–96%; ASA grade 2, 69–82%; ASA grade 3, 44–58%; ASA grade 4, 33–48%; and ASA grade 5, 4–51%. The majority of deaths (%) occurred after 30 days. Conclusions Emergency laparotomy is associated with a significantly increased risk of death in the following 5 years. The risk is strongly correlated to ASA. Thirty-day mortality estimation is not a good basis on which to advise patients and carers on long-term outcomes. ASA score can be used to predict long-term outcomes and to guide patient counsel.

Insights Into Excess Mortality During the First Months of the COVID-19 Pandemic From a Rural, Demographic Surveillance Site in Bangladesh

Background: Coronavirus disease 2019 (COVID-19) has spread globally, and the government of each affected country is publishing the number of deaths every day. This official figure is an underestimate as it excludes anybody who did not die in a hospital, who did not test positive, who had a false result, or those who recovered on their own without a test.Objective: This study aimed to measure the community level excess mortality using health and demographic surveillance in a rural area of Bangladesh.Method: The study was conducted in Matlab, in a rural area of Bangladesh, with a Health and Demographic Surveillance System (HDSS) covering a population of 239,030 individuals living in 54,823 households in 142 villages. We examined the mortality in January-April from 2015 to 2020 and compared the mortality in 2020 with the historical trend of 2015–2019. Between 2015 and 2020, we followed 276,868 people until migration or death, whichever occurred first. We analyzed mortality using crude mortality rate ratio (MRR) and adjusted MRR (aMRR) from a Cox proportional hazard model. Mortality was analyzed according to age, sex, and period.Results: During follow-up, 3,197 people died. The mortality rate per 1,000 person-years increased from 10 in 2019 to 12 in 2020. Excess mortality was observed among the elderly population (aged 65 years and above). The elderly mortality rate per 1,000 person-years increased from 80 in 2019 to 110 in 2020, and the aMRR was 1.40 (95% CI: 1.19–1.64). Although an increasing tendency in mortality was observed between 2015 and 2019, it was statistically insignificant.Conclusions: The study reported a 28% increase in excess deaths among the elderly population during the first months of the pandemic. This all-cause mortality estimation at the community level will urge policymakers, public health professionals, and researchers to further investigate the causes of death and the underlying reasons for excess deaths in the older age-group.

Robust and Timely Estimation of Excess Mortality: Statistical Analysis of Cremation Data During the COVID-19 Pandemic (Preprint)

BACKGROUND Early estimates of excess mortality are crucial for understanding the impact of COVID-19. However, there is a lag of several months in the reporting of vital statistics mortality data for many jurisdictions. In Ontario, a Canadian province, certification by a coroner is required before cremation can occur, creating timely mortality data that encompasses the majority of deaths within the province. OBJECTIVE Our objectives were to (1) validate the ability of cremation data in permitting real-time estimation of excess all-cause mortality, interim of vital statistics data, and (2) describe the patterns of excess mortality. METHODS Cremation records from January 2020 until April 2021 were compared to the historical records from 2017-2019, grouped according to week, age, sex, and COVID-19 status. Cremation data were compared to Ontario’s provisional vital statistics mortality data released by Statistics Canada. The 2020 and 2021 records were then compared to previous years to determine whether there was excess mortality and if so, which age groups had the greatest number of excess deaths during the COVID Pandemic, and whether deaths attributed to COVID-19 account for the entirety of the excess mortality. RESULTS Between 2017-2019, cremations were performed for 67.4% (95% CI: 67.3–67.5%) of deaths; the proportion of cremated deaths remained stable throughout 2020, establishing that the COVID-19 pandemic did not significantly alter cremation practices, even within age and sex categories. During the first wave (from April to June 2020), cremation records detected a 16.9% increase (95% CI: 14.6–19.3%) in mortality. The accuracy of this excess mortality estimation was later confirmed by vital statistics data. CONCLUSIONS The stability in the percent of Ontarians cremated and the completion of cremation data several months before vital statistics data, enables accurate estimation of all-causes mortality in near real-time with cremation data. These findings demonstrate the utility of cremation data to provide timely mortality information during public health emergencies.

Challenges of Implementing Occupational Health Services (OHS) in Malaysia -Perspective of Occupational Health Doctors

Mortality estimation due to work-related illness has reached up to 2.4 million each year. The current coverage of occupational health services (OHS) in Malaysia is still low. Occupational health doctors (OHDs) are one of the essential personnel to ensure proper execution of OHS. This study was conducted to explore the experiences and views of OHDs on the challenges in implementing OHS in Malaysia. Four focus group discussions were conducted with OHDs (N = 23) from four different states in Malaysia in 2016. Another five OHDs participated in in-depth interviews to implement the identified codes or themes. The discussions were recorded and transcribed verbatim. NVivo version 11.0 was used to facilitate data analysis. The data were analysed following the thematic analysis guidelines. Three themes were identified from the discussions: difficulties in diagnosing occupational diseases and poisoning; poor practices, attitudes, and commitment by both workers and employers; and non-compliance with laws and regulations related to the industries. The common challenges discussed by the participants were the lack of knowledge and skills among OHDs, and the shortage of standard procedures, leading to difficulties to screen occupational diseases. The poor cooperation and behaviour from the industries also hindered OHDs when performing their services. This study suggests better training and provision of standard tools or guideline to assist OHDs in making occupational disease diagnoses, increasing OHS awareness among the industries, and enacting OHS as part of the laws and regulations with adequate enforcement.

Cherish your children: socio-economic and demographic characteristics associated with child mortality

Background The United Nations Interagency Group for Child Mortality Estimation (UNIGME) indicates that child mortality is the death rate of children between age zero to five. The importance of this area of research is high where worldwide a number of studies have been led on infant and child mortality, despite limited research discoveries with regards to Sri Lanka. The aim of this study is to investigate the socio-economic and demographic characteristics associated with child mortality in Sri Lanka. Methods Using the context of Sri Lanka as a case study, this study carried out based on data gathered from the micro level national survey. Using the logit regression model through the step-wise technique, the study investigate the socio-economic and demographic characteristics associated with child mortality in Sri Lanka. Results According to the generated results, place of residence province-wise, household head’s education level and source of drinking water have negative effect (lower risk) on child mortality in Sri Lanka. Exceptionally, the Western province has the highest negative effect on child mortality which demonstrates it as the least harmful region in Sri Lanka in child endurance. Household heads who owns private entities and Sri Lankan Moors has a positive effect on child mortality as well. Conclusion This study is helpful to address the population health of local arena and results can be supportive to the government and policymakers to gain an overview of physical health status of the country and able to uplift their policies based on the new findings.

Date of introduction and epidemiologic patterns of SARS-CoV-2 in Mogadishu, Somalia: estimates from transmission modelling of 2020 excess mortality data

Introduction In countries with weak surveillance systems confirmed COVID-19 deaths are likely to underestimate the death toll of the pandemic. Many countries also have incomplete vital registration systems, hampering excess mortality estimation. Here, we fitted a dynamic transmission model to satellite imagery data on burial patterns in Mogadishu, Somalia during 2020 to estimate the date of introduction, transmissibility and other epidemiologic characteristics of SARS-CoV-2 in this low-income, crisis-affected setting. Methods We performed Markov chain Monte Carlo (MCMC) fitting with an age-structured compartmental COVID-19 model to provide median estimates and credible intervals for the date of introduction, the basic reproduction number (R0) and the effect of non-pharmaceutical interventions in Mogadishu up to September 2020. Results Under the assumption that excess deaths in Mogadishu February-September 2020 were directly attributable to SARS-CoV-2 infection we arrived at median estimates of October-November 2019 for the date of introduction and low R0 estimates (1.3-1.5) stemming from the early and slow rise of excess deaths. The effect of control measures on transmissibility appeared small. Conclusion Subject to study assumptions, a very early SARS-CoV-2 introduction event may have occurred in Somalia. Estimated transmissibility in the first epidemic wave was lower than observed in European settings.

Social disadvantage, economic inequality, and life expectancy in nine Indian states

An extensive literature documents the contributions of discrimination and social exclusion to health disparities. This study investigates life expectancy differentials along lines of caste, religion, and indigenous identity in India, home to some of the largest populations of marginalized social groups in the world. Using a large, high-quality survey that measured mortality, social group, and economic status, we are the first to estimate and decompose life expectancy differences between higher-caste Hindus and three of India's most disadvantaged social groups: Adivasis, Dalits, and Muslims. Relative to higher-caste Hindus, Adivasi life expectancy is more than four years lower, Dalit life expectancy is more than three years lower, and Muslim life expectancy is about one year lower. Economic status explains less than half of these gaps. The differences between the life expectancy of higher-caste Hindus and the life expectancies of Adivasis and Dalits are comparable to the Black-White gap in the US in absolute magnitude. The differences are larger in relative terms because overall life expectancy in India is lower. Our findings extend the literature on fundamental causes of global health disparities. Methodologically, we contribute to the literature on mortality estimation and demographic decomposition using survey data from low- and middle-income contexts.

Correlation of Caesarean Section Rates to Maternal and Neonatal Mortality in the Eastern Mediterranean Region; A Population-Based Ecological Study

Purpose: This study aims to explore the relationship between population-level caesarean section rates (CSRs) with maternal and neonatal mortality rates (MMR, NMR) in the Eastern Mediterranean Region (EMR). Design: A populationbased ecological study was performed with data obtained from the World Health Organization, Global Health Observatory database, 2015, United Nations Inter-agency Group for Child Mortality Estimation (UN-IGME) and the United Nations Maternal Mortality Estimation Inter-Agency Group (UN-MMEIG) 2015). Mean ± standard deviation (SD), range, median and Interquartile range (IQR) were used to describe the quantitative data. We performed multivariate logistic regression analysis to explore the effect (a) of (a) Antenatal clinic visits (ANC %), (b), Skilled Birth Attendance (SBA) rate (% of deliveries attended by SBA), (c) Total Health Expenditure (THE) per capita and (d) Female Literacy Rate (FLR%) on the studied relationship. Spline linear regression was used to find the most predictive variable for MMR, and the NMR. Statistical significance was accepted at P<0.05. Results: The mean CSR was 21.20±13.38, (1.8-52). The CSR of <10% was linked with the highest NMR and MMR, 33.0 (24.0-39.0) and 390.5(329.5-648.0) respectively. The most predictable variables for NMR and MMR were SBA % [B=-0.875; p< .001; R2=0.766 and adjusted R2=0.754] and FLR (F=15-24) [B=0.877; P<0.001; R2=0.77 and adjusted R2=0.758] respectively. Conclusions: We found a statistically significant inverse relationship between CSRs and maternal and neonatal mortality in MSs with <10% of CSR. The improved mortality rates in MSs with >15% of CSR were significantly linked with better socioeconomic and healthcare variables than higher CSRs. Keywords: Caesarean Section, Maternal and Neonatal Mortality, Eastern Mediterranean Region.