Real-time parameter estimation of Zika outbreaks using model averaging.

Early prediction of the final size of any epidemic and in particular for Zika disease outbreaks can be useful for health authorities in order to plan the response to the outbreak. The Richards model is often been used to estimate epidemiological parameters for arboviral diseases based on the reported cumulative cases in single- and multi-wave outbreaks. However, other non-linear models can also fit the data as well. Typically, one follows the so called post selection estimation procedure, i.e., selects the best fitting model out of the set of candidate models and ignores the model uncertainty in both estimation and inference since these procedures are based on a single model. In this paper we focus on the estimation of the final size and the turning point of the epidemic and conduct a real-time prediction for the final size of the outbreak using several non-linear models in which these parameters are estimated via model averaging. The proposed method is applied to Zika outbreak data in four cities from Colombia, during the outbreak ocurred in 2015-2016.

Determinants of between-hospital variations in outcomes for patients admitted with COPD exacerbations: findings from a nationwide clinical audit (AUDIPOC) in Spain.

BACKGROUND: Previous studies have demonstrated significant variability in the processes of care and outcomes of chronic obstructive pulmonary disease (COPD) exacerbations. The AUDIPOC is a Spanish nationwide clinical audit that identified large between-hospital variations in care and clinical outcomes. Here, we test the hypothesis that these variations can be attributed to either patient characteristics, hospital characteristics and/or the so-called hospital-clustering effect, which indicates that patients with similar characteristics may experience different processes of care and outcomes depending on the hospital to which they are admitted. METHODS: A clinical audit of 5178 COPD patients consecutively admitted to 129 Spanish public hospitals was performed, with a 90-day follow-up. Multilevel regression analysis was conducted to model the probability of patients experiencing adverse outcomes. For each outcome, an empty model (with no independent variables) was fitted to assess the clustering effect, followed by a model adjusted for the patient- and hospital-level covariables. The hospital-clustering effect was estimated using the intracluster correlation coefficient (ICC); the cluster heterogeneity was estimated with the median odds ratio (MOR), and the coefficients of predictors were estimated with the odds ratio (OR). RESULTS: In the empty models, the ICC (MOR) for inpatient mortality and the follow-up mortality and readmission were 0.10 (1.80), 0.08 (1.65) and 0.01 (1.24), respectively. In the adjusted models, the variables that most represented the patients' clinical conditions and interventions were identified as outcome predictors and further reduced the hospital variations. By contrast, the resource factors were primarily unrelated with outcomes. CONCLUSIONS: This study demonstrates a noteworthy reduction in the observed crude between-hospital variation in outcomes after accounting for the hospital-cluster effect and the variables representing patient's clinical conditions. This emphasises the predictor importance of the patients' clinical conditions and interventions, and understates the impacts of hospital resources and organisational factors.

Bayesian hierarchical Poisson models with a hidden Markov structure for the detection of influenza epidemic outbreaks.

Considerable effort has been devoted to the development of statistical algorithms for the automated monitoring of influenza surveillance data. In this article, we introduce a framework of models for the early detection of the onset of an influenza epidemic which is applicable to different kinds of surveillance data. In particular, the process of the observed cases is modelled via a Bayesian Hierarchical Poisson model in which the intensity parameter is a function of the incidence rate. The key point is to consider this incidence rate as a normal distribution in which both parameters (mean and variance) are modelled differently, depending on whether the system is in an epidemic or non-epidemic phase. To do so, we propose a hidden Markov model in which the transition between both phases is modelled as a function of the epidemic state of the previous week. Different options for modelling the rates are described, including the option of modelling the mean at each phase as autoregressive processes of order 0, 1 or 2. Bayesian inference is carried out to provide the probability of being in an epidemic state at any given moment. The methodology is applied to various influenza data sets. The results indicate that our methods outperform previous approaches in terms of sensitivity, specificity and timeliness.

Spatial moving average risk smoothing.

This paper introduces spatial moving average risk smoothing (SMARS) as a new way of carrying out disease mapping. This proposal applies the moving average ideas of time series theory to the spatial domain, making use of a spatial moving average process of unknown order to define dependence on the risk of a disease occurring. Correlation of the risks for different locations will be a function of m values (m being unknown), providing a rich class of correlation functions that may be reproduced by SMARS. Moreover, the distance (in terms of neighborhoods) that should be covered for two units to be found to make the correlation of their risks 0 is a quantity to be fitted by the model. This way, we reproduce patterns that range from spatially independent to long-range spatially dependent. We will also show a theoretical study of the correlation structure induced by SMARS, illustrating the wide variety of correlation functions that this proposal is able to reproduce. We will also present three applications of SMARS to both simulated and real datasets. These applications will show SMARS to be a competitive disease mapping model when compared with alternative proposals that have already appeared in the literature. Finally, the application of SMARS to the study of mortality for 21 causes of death in the Comunitat Valenciana will allow us to identify some qualitative differences in the patterns of those diseases.

Bayesian methods in cost-effectiveness studies: objectivity, computation and other relevant aspects.

In a probabilistic sensitivity analysis (PSA) of a cost-effectiveness (CE) study, the unknown parameters are considered as random variables. A crucial question is what probabilistic distribution is suitable for synthesizing the available information (mainly data from clinical trials) about these parameters. In this context, the important role of Bayesian methodology has been recognized, where the parameters are of a random nature. We explore, in the context of CE analyses, how formal objective Bayesian methods can be implemented. We fully illustrate the methodology using two CE problems that frequently appear in the CE literature. The results are compared with those obtained with other popular approaches to PSA. We find that the discrepancies can be quite marked, specially when the number of patients enrolled in the simulated cohort under study is large. Finally, we describe in detail the numerical methods that need to be used to obtain the results.

An autoregressive approach to spatio-temporal disease mapping.

Disease mapping has been a very active research field during recent years. Nevertheless, time trends in risks have been ignored in most of these studies, yet they can provide information with a very high epidemiological value. Lately, several spatio-temporal models have been proposed, either based on a parametric description of time trends, on independent risk estimates for every period, or on the definition of the joint covariance matrix for all the periods as a Kronecker product of matrices. The following paper offers an autoregressive approach to spatio-temporal disease mapping by fusing ideas from autoregressive time series in order to link information in time and by spatial modelling to link information in space. Our proposal can be easily implemented in Bayesian simulation software packages, for example WinBUGS. As a result, risk estimates are obtained for every region related to those in their neighbours and to those in the same region in adjacent periods.

Spatial analysis of bovine spongiform encephalopathy in Galicia, Spain (2000-2005).

In Spain, the first bovine spongiform encephalopathy (BSE) case was detected in 2000 in a cow born in the Galicia region (Northwestern Spain). From then and until October 2005, 590 cases were detected, 223 of them in Galicia. In 1994, meat and bone meal (MBM) was banned on ruminant feed and, in 1996, an EU decision mandating an overall change in MBM processing was implemented. This decision was gradually applied in the territory and not enforced before July 1998. The objective of this study was to explore clustering of BSE cases and estimate the standard incidence ratio (SIR) of BSE in Galicia. Our study was based on the BSE cases detected during the surveillance period 2000-2005 in the Galicia region. These cases were divided, based on birth date, into two periods: animals born from 1994 to July 1998, and those born after July 1998. We tested the role of cross-contamination on the geographical SIR distribution for both periods. Hierarchical Bayesian models were used to model the overdispersion and lack of independence of the SIR estimates. The geographical distribution of the standard incidence ratio of BSE between both periods was different. In the second period, the SIR was reduced in some areas. The reduction in these areas could be attributable to the changes in the processing of MBM. We did not find any statistical link between the poultry population and the standard incidence ratio, but pig population had a positive effect.

[Point processes as a tool for analyzing possible sources of contamination]. / Procesos puntuales como herramienta para el análisis de posibles fuentes de contaminación.

Point pattern analysis pattern comprises a series of techniques that enables the distribution of a series of events occurring in the vicinity of a particular region of a map to be studied. In epidemiology, this problem arises when a potential source of environmental contamination, possibly leading to cases of a specific disease, is investigated.The present study provides a brief description of point pattern analysis. The approach is illustrated through determination of the environmental source and study of the areas of greatest risk of incidence of an outbreak of legionella pneumonia that occurred between the middle of September and beginning of October in the city of Alcoi in Alicante (Spain).Point pattern analysis was able to confirm the environmental source of the outbreak and identify the areas of the city at greatest risk. This provided the justification for an exhaustive inspection of the installations generating aerosols after which, to date, the epidemics ceased.

Procesos puntuales como herramienta para el análisis de posibles fuentes de contaminación / Point processes as a tool for analyzing possible sources of contamination

El análisis de un patrón puntual engloba una serie de técnicas que permiten estudiar la distribución de un conjunto de eventos ocurridos sobre una región del plano. Este problema surge en epidemiología cuando se investiga una potencial fuente de contaminación ambiental alrededor de la cual se sospecha que surgen casos de una determinada enfermedad. En el presente trabajo, se explica brevemente en qué consiste el análisis de un patrón puntual y se ilustra con una aplicación a la determinación del origen medioambiental y al estudio de las zonas de mayor riesgo de incidencia en un brote de neumonía por Legionella ocurrido entre mediados de septiembre y principios de octubre en la ciudad de Alcoi (Alicante). El estudio permitió confirmar el origen medioambiental del brote y señalar las zonas de la ciudad con mayor riesgo, convirtiéndose en el argumento básico para llevar a cabo una exhaustiva inspección de las instalaciones generadoras de aerosoles, tras la cual, hasta la fecha, cesaron los brotes epidémicos (AU)

Point pattern analysis pattern comprises a series of techniques that enables the distribution of a series of events occurring in the vicinity of a particular region of a map to be studied. In epidemiology, this problem arises when a potential source of environmental contamination, possibly leading to cases of a specific disease, is investigated. The present study provides a brief description of point pattern analysis. The approach is illustrated through determination of the environmental source and study of the areas of greatest risk of incidence of an outbreak of legionella pneumonia that occurred between the middle of September and beginning of October in the city of Alcoi in Alicante (Spain). Point pattern analysis was able to confirm the environmental source of the outbreak and identify the areas of the city at greatest risk. This provided the justification for an exhaustive inspection of the installations generating aerosols after which, to date, the epidemics ceased (AU)