Objective This study aims to develop high-performing Machine Learning and Deep Learning models in predicting hospital length of stay (LOS) while enhancing interpretability. We compare performance and interpretability of models trained only on structured tabular data with models trained only on unstructured clinical text data, and on mixed data. Methods The structured data was used to train fourteen classical Machine Learning models including advanced ensemble trees, neural networks and k-nearest neighbors. The unstructured data was used to fine-tune a pre-trained Bio Clinical BERT Transformer Deep Learning model. The structured and unstructured data were then merged into a tabular dataset after vectorization of the clinical text and a dimensional reduction through Latent Dirichlet Allocation. The study used the free and publicly available Medical Information Mart for Intensive Care (MIMIC) III database, on the open AutoML Library AutoGluon. Performance is evaluated with respect to two types of random classifiers, used as baselines. Results The best model from structured data demonstrates high performance (ROC AUC = 0.944, PRC AUC = 0.655) with limited interpretability, where the most important predictors of prolonged LOS are the level of blood urea nitrogen and of platelets. The Transformer model displays a good but lower performance (ROC AUC = 0.842, PRC AUC = 0.375) with a richer array of interpretability by providing more specific in-hospital factors including procedures, conditions, and medical history. The best model trained on mixed data satisfies both a high level of performance (ROC AUC = 0.963, PRC AUC = 0.746) and a much larger scope in interpretability including pathologies of the intestine, the colon, and the blood; infectious diseases, respiratory problems, procedures involving sedation and intubation, and vascular surgery. Conclusions Our results outperform most of the state-of-the-art models in LOS prediction both in terms of performance and of interpretability. Data fusion between structured and unstructured text data may significantly improve performance and interpretability.
IntroductionThere is a lack of quantitative evidence on the role of food innovations-new food ingredients and processing techniques-in the nutrition transition.ObjectiveDocument the distribution of food innovations across 67 high-income (HIC) and middle-income (MIC) countries between 1970 and 2010, and its association with the nutritional composition of food supply.MethodsWe used all available data on food patents, as compiled by the European Patent Office, to measure food innovations. We considered innovations directly received by countries from inventors seeking protection in their territories, and those embedded in processed food imports. Food and Agricultural Organization data were used to estimate the associations between international diffusion of food innovations and trends in total food supply and its macronutrient composition, after adjusting for confounding trends in demand-side factors. We identified the role of trade by simulating the changes in average diet due to innovations embedded in food imports.ResultsTrends in food innovations were positively and significantly associated with changes in daily per capita calorie supply available for human consumption in MIC between 1990 and 2010 (elasticity of 0.027, 95% CI 0.019 to 0.036). Food innovations were positively correlated with the share of animal and free fats in total food supply (elasticities of 0.044, 95% CI 0.030 to 0.058 for MIC between 1970 and 1989 and 0.023, 95% CI 0.003 to 0.043 for HIC between 1990 and 2010). Food innovations were associated with substitutions from complex carbohydrates towards sugars in total food supply for MIC after 1990 (elasticities of -0.037, 95% CI -0.045 to -0.029 for complex carbs, 0.082, 95% CI 0.066 to 0.098 for sugars). For these countries, the trade channel capturing access to innovations through imports of processed food played a key role.ConclusionPolicy-makers should consider the impacts of the international diffusion of food innovations in assessing the costs and benefits of international trade regulations.
Many people obtain job information from friends and acquaintances. However, one factor influencing labor-market outcomes that is ignored in the literature is the presence of overlapping friendship circles in social networks. We find that overlapping friendship networks produce correlated information flows, resulting in an increased probability of two events: either receiving redundant job offers or receiving no job offers at all. Consequently, people with common contact networks exhibit worse employment prospects even if they have the same number of information providers and compete with the same number of people for vacancies. In quantitative terms, the impact of overlapping friendship circles rivals that of the number of direct contacts and contacts' contacts. This implies that the results in Calvo-Armengol (2004) only apply for networks where people's friends are neither connected nor have common contacts. Because overlapping friendship circles are a crucial aspect of strong relationships, our findings uncover an alternative mechanism behind "The Strength of Weak Ties"(Granovetter, 1973): their ability to maintain independence in job information flows. We further show that people with common job contacts earn lower incomes on average. However, conditional on being employed, their expected wage is higher because they can take advantage of the multiple job offers received by selecting the one with the highest pay.
The paper examines the question of non-anonymous Growth Incidence Curves (na-GIC) from a Bayesian inferential point of view. Building on the notion of conditional quantiles of Barnett (1976. “The Ordering of Multivariate Data.” Journal of the Royal Statistical Society: Series A 139: 318–55), we show that removing the anonymity axiom leads to a complex and shaky curve that has to be smoothed, using a non-parametric approach. We opted for a Bayesian approach using Bernstein polynomials which provides confidence intervals, tests and a simple way to compare two na-GICs. The methodology is applied to examine wage dynamics in a US university with a particular attention devoted to unbundling and anti-discrimination policies. Our findings are the detection of wage scale compression for higher quantiles for all academics and an apparent pro-female wage increase compared to males. But this pro-female policy works only for academics and not for the para-academics categories created by the unbundling policy.
The objective of this paper is to emphasize the differences between a call and a warrant as well as the different valuation methods of warrants which have been introduced in the financial literature. For the sake of simplicity and applicability, we only consider a debt-free equity-financed firm. More recently a formal distinction between structural and reduced form pricing models has been introduced. This distinction is important whether one wishes to price a new warrant issue or outstanding warrants. If we are interested in pricing a new issue of warrants, e.g. in the context of a management incentive package, one has to rely on a structural model. However most of practitioners use the simple Black-Scholes formula. In this context, we analyze the accuracy of the approximation of the “true” price of a warrant by the Black-Scholes formula. We show that in the current low interest rate environment, the quality of the approximation deteriorates and the sensitivity of this approximation to the volatility estimate increases.
We estimate the causal effect of losing a father in the U.S. Civil War on children’s long-run socioeconomic outcomes. Linking military records from the 2.2 million Union Army soldiers with the 1860 U.S. population census, we track soldiers’ sons into the 1880 and 1900 census. Sons of soldiers who died had lower occupational income scores and were less likely to work in a high- or semi-skilled job as opposed to being low-skilled or farmers. These effects persisted at least until the 1900 census. Our results are robust to instrumenting paternal death with the mortality rate of the father’s regiment, which we argue was driven by military strategy that did not take into account the social origins of soldiers. Pre-war family wealth is a strong mitigating factor: there is no effect of losing a father in the top quartile of the wealth distribution.
In this paper, we develop an overlapping generations model with endogenous fertility and calibrate it to the Swedish historical data in order to estimate the economic cost of the 1918–19 influenza pandemic. The model identifies survivors from younger cohorts as main benefactors of the windfall bequests following the influenza mortality shock. We also show that the general equilibrium effects of the pandemic reveal themselves over the wage channel rather than the interest rate, fertility or labor supply channels. Finally, we demonstrate that the influenza mortality shock becomes persistent, driving the aggregate variables to lower steady states which costs the economy 1.819% of the output loss over the next century.
This paper describes an empiric study of aggregation and deliberation—used during citizens’ workshops—for the elicitation of collective preferences over 20 different ecosystem services (ESs) delivered by the Palavas coastal lagoons located on the shore of the Mediterranean Sea close to Montpellier (S. France). The impact of deliberation is apprehended by comparing the collectives preferences constructed with and without deliberation. The same aggregation rules were used before and after deliberation. We compared two different aggregation methods, i.e. Rapid Ecosystem Services Participatory Appraisal (RESPA) and Majority Judgement (MJ). RESPA had been specifically tested for ESs, while MJ evaluates the merit of each item, an ES in our case, in a predefined ordinal scale of judgment. The impact of deliberation was strongest for the RESPA method. This new information acquired from application of social choice theory is particularly useful for ecological economics studying ES, and more practically for the development of deliberative approaches for public policies.
We study repeated zero-sum games where one of the players pays a certain cost each time he changes his action. We derive the properties of the value and optimal strategies as a function of the ratio between the switching costs and the stage payoffs. In particular, the strategies exhibit a robustness property and typically do not change with a small perturbation of this ratio. Our analysis extends partially to the case where the players are limited to simpler strategies that are history independent―namely, static strategies. In this case, we also characterize the (minimax) value and the strategies for obtaining it.