Le Bao

Going from bad to worse: A stochastic model of transitions in deficit accumulation, in relation to mortality

As people age, they accumulate deficits. The more deficits they accumulate, the greater their vulnerability, which can be expressed as the probability to accumulate even more deficits, or to die. The probability of death is known to be exponentially related to the number of deficits. Using data from elderly (aged 65 + years) participants in the Canadian Study of Health and Aging (n = 9008), we investigated the relationship between the number of deficits and the change in the number of deficits over two successive 5 year intervals. We show that the probabilities of changes in the number of deficits, in relation to baseline, are well fitted (R(2) > 0.98) by a simple distribution, with two parameters. The model suggests a maximum to deficit accumulation, and illustrates no level of deficit accumulation at which there is a net gain in fitness. Age-related deficit accumulation is highly characteristic, and can be modeled as a stochastic process with readily interpretable parameters.

Modelling HIV epidemics in the antiretroviral era: the UNAIDS Estimation and Projection package 2009

Objective The UNAIDS Estimation and Projection Package (EPP) is a tool for country-level estimation and short-term projection of HIV/AIDS epidemics based on fitting observed HIV surveillance data on prevalence. This paper describes the adaptations made in EPP 2009, the latest version of this tool, as new issues have arisen in the global response, in particular the global expansion of antiretroviral therapy (ART). Results By December 2008 over 4 million people globally were receiving ART, substantially improving their survival. EPP 2009 required modifications to correctly adjust for the effects of ART on incidence and the resulting increases in HIV prevalence in populations with high ART coverage. Because changing incidence is a better indicator of program impact, the 2009 series of UNAIDS tools also focuses on calculating incidence alongside prevalence. Other changes made in EPP 2009 include: an improved procedure, incremental mixture importance sampling, for efficiently generating more accurate uncertainty estimates; provisions to vary the urban/rural population ratios in generalised epidemics over time; introduction of a modified epidemic model that accommodates behaviour change in low incidence settings; and improved procedures for calibrating models. This paper describes these changes in detail, and discusses anticipated future changes in the next version of EPP.

Estimating and projecting trends in HIV/AIDS generalized epidemics using incremental mixture importance sampling.

The Joint United Nations Programme on HIV/AIDS (UNAIDS) has decided to use Bayesian melding as the basis for its probabilistic projections of HIV prevalence in countries with generalized epidemics. This combines a mechanistic epidemiological model, prevalence data, and expert opinion. Initially, the posterior distribution was approximated by sampling-importance-resampling, which is simple to implement, easy to interpret, transparent to users, and gave acceptable results for most countries. For some countries, however, this is not computationally efficient because the posterior distribution tends to be concentrated around nonlinear ridges and can also be multimodal. We propose instead incremental mixture importance sampling (IMIS), which iteratively builds up a better importance sampling function. This retains the simplicity and transparency of sampling importance resampling, but is much more efficient computationally. It also leads to a simple estimator of the integrated likelihood that is the basis for Bayesian model comparison and model averaging. In simulation experiments and on real data, it outperformed both sampling importance resampling and three publicly available generic Markov chain Monte Carlo algorithms for this kind of problem.

Bias Correction and Bayesian Model Averaging for Ensemble Forecasts of Surface Wind Direction

Wind direction is an angular variable, as opposed to weather quantities such as temperature, quantitative precipitation, or wind speed, which are linear variables. Consequently, traditional model output statistics and ensemble postprocessing methods become ineffective, or do not apply at all. This paper proposes an effective bias correction technique for wind direction forecasts from numerical weather prediction models, which is based on a state-of-the-art circular‐circular regression approach. To calibrate forecast ensembles, a Bayesian model averaging scheme for directional variables is introduced, where the component distributions are von Mises densities centered at the individually bias-corrected ensemble member forecasts. These techniques are applied to 48-h forecasts of surface wind direction over the Pacific Northwest, using the University of Washington mesoscale ensemble, where they yield consistent improvements in forecast performance.

A cross-national study of transitions in deficit counts in two birth cohorts: implications for modeling ageing.

Generally, health does not improve with age, and many physical and physiological functions are known to decline. These changes do not occur uniformly, however; for many reasons, some people experience significant improvement in their health over non-trivial time intervals. Earlier, we showed that 5-year transitions in health status in elderly people (age 65+ years) can be modeled as a stochastic process, using a modified Poisson distribution with four readily interpretable parameters. The original description was based on follow-up of a single cross-sectional study, thus mixing age and cohort effects. Here, we again used a multistate Markov chain to model 5-year deficit accumulation in relation to frailty in both a Swedish birth cohort (aged 70 years at inception) and, from the original cross-sectional study, a Canadian birth cohort, aged 69-71. In both datasets, we found again that a modified Poisson describes the transition in health status with high precision. The parameters of the model though different, are close to each other, even though the cohorts are from different countries, were assembled 20 years apart, and counted different deficits. The model suggests that all health transitions, including health improvement, worsening, and death, can be summarized in a unified stochastic model with a few interpretable parameters.

Improvements in prevalence trend fitting and incidence estimation in EPP 2013.

Objective:Describe modifications to the latest version of the Joint United Nations Programme on AIDS (UNAIDS) Estimation and Projection Package component of Spectrum (EPP 2013) to improve prevalence fitting and incidence trend estimation in national epidemics and global estimates of HIV burden. Methods:Key changes made under the guidance of the UNAIDS Reference Group on Estimates, Modelling and Projections include: availability of a range of incidence calculation models and guidance for selecting a model; a shift to reporting the Bayesian median instead of the maximum likelihood estimate; procedures for comparison and validation against reported HIV and AIDS data; incorporation of national surveys as an integral part of the fitting and calibration procedure, allowing survey trends to inform the fit; improved antenatal clinic calibration procedures in countries without surveys; adjustment of national antiretroviral therapy reports used in the fitting to include only those aged 15–49 years; better estimates of mortality among people who inject drugs; and enhancements to speed fitting. Results:The revised models in EPP 2013 allow closer fits to observed prevalence trend data and reflect improving understanding of HIV epidemics and associated data. Conclusion:Spectrum and EPP continue to adapt to make better use of the existing data sources, incorporate new sources of information in their fitting and validation procedures, and correct for quantifiable biases in inputs as they are identified and understood. These adaptations provide countries with better calibrated estimates of incidence and prevalence, which increase epidemic understanding and provide a solid base for program and policy planning.

Modelling national HIV/AIDS epidemics: revised approach in the UNAIDS Estimation and Projection Package 2011

Objective United Nations Programme on HIV/AIDS reports regularly on estimated levels and trends in HIV/AIDS epidemics, which are evaluated using an epidemiological model within the Estimation and Projection Package (EPP). The relatively simple four-parameter model of HIV incidence used in EPP through the previous round of estimates has encountered challenges when attempting to fit certain data series on prevalence over time, particularly in settings with long running epidemics where prevalence has increased recently. To address this, the most recent version of the modelling package (EPP 2011) includes a more flexible epidemiological model that allows HIV infection risk to vary over time. This paper describes the technical details of this flexible approach to modelling HIV transmission dynamics within EPP 2011. Methodology For the flexible modelling approach, the force of infection parameter, r, is allowed to vary over time through a random walk formulation, and an informative prior distribution is used to improve short-term projections beyond the last year of data. Model parameters are estimated using a Bayesian estimation approach in which models are fit to HIV seroprevalence data from surveillance sites. Results This flexible model can yield better estimates of HIV prevalence over time in situations where the classic EPP model has difficulties, such as in Uganda, where prevalence is no longer falling. Based on formal out-of-sample projection tests, the flexible modelling approach also improves predictions and CIs for extrapolations beyond the last observed data point. Conclusions We recommend use of a flexible modelling approach where data are sufficient (eg, where at least 5 years of observations are available), and particularly where an epidemic is beyond its peak.

Methods for selecting fixed-effect models for heterogeneous codon evolution, with comments on their application to gene and genome data

Models of codon evolution have proven useful for investigating the strength and direction of natural selection. In some cases, a priori biological knowledge has been used successfully to model heterogeneous evolutionary dynamics among codon sites. These are called fixed-effect models, and they require that all codon sites are assigned to one of several partitions which are permitted to have independent parameters for selection pressure, evolutionary rate, transition to transversion ratio or codon frequencies. For single gene analysis, partitions might be defined according to protein tertiary structure, and for multiple gene analysis partitions might be defined according to a genes functional category. Given a set of related fixed-effect models, the task of selecting the model that best fits the data is not trivial. In this study, we implement a set of fixed-effect codon models which allow for different levels of heterogeneity among partitions in the substitution process. We describe strategies for selecting among these models by a backward elimination procedure, Akaike information criterion (AIC) or a corrected Akaike information criterion (AICc). We evaluate the performance of these model selection methods via a simulation study, and make several recommendations for real data analysis. Our simulation study indicates that the backward elimination procedure can provide a reliable method for model selection in this setting. We also demonstrate the utility of these models by application to a single-gene dataset partitioned according to tertiary structure (abalone sperm lysin), and a multi-gene dataset partitioned according to the functional category of the gene (flagellar-related proteins of Listeria). Fixed-effect models have advantages and disadvantages. Fixed-effect models are desirable when data partitions are known to exhibit significant heterogeneity or when a statistical test of such heterogeneity is desired. They have the disadvantage of requiring a priori knowledge for partitioning sites. We recommend: (i) selection of models by using backward elimination rather than AIC or AICc, (ii) use a stringent cut-off, e.g., p = 0.0001, and (iii) conduct sensitivity analysis of results. With thoughtful application, fixed-effect codon models should provide a useful tool for large scale multi-gene analyses.BackgroundModels of codon evolution have proven useful for investigating the strength and direction of natural selection. In some cases, a priori biological knowledge has been used successfully to model heterogeneous evolutionary dynamics among codon sites. These are called fixed-effect models, and they require that all codon sites are assigned to one of several partitions which are permitted to have independent parameters for selection pressure, evolutionary rate, transition to transversion ratio or codon frequencies. For single gene analysis, partitions might be defined according to protein tertiary structure, and for multiple gene analysis partitions might be defined according to a genes functional category. Given a set of related fixed-effect models, the task of selecting the model that best fits the data is not trivial.ResultsIn this study, we implement a set of fixed-effect codon models which allow for different levels of heterogeneity among partitions in the substitution process. We describe strategies for selecting among these models by a backward elimination procedure, Akaike information criterion (AIC) or a corrected Akaike information criterion (AICc). We evaluate the performance of these model selection methods via a simulation study, and make several recommendations for real data analysis. Our simulation study indicates that the backward elimination procedure can provide a reliable method for model selection in this setting. We also demonstrate the utility of these models by application to a single-gene dataset partitioned according to tertiary structure (abalone sperm lysin), and a multi-gene dataset partitioned according to the functional category of the gene (flagellar-related proteins of Listeria).ConclusionFixed-effect models have advantages and disadvantages. Fixed-effect models are desirable when data partitions are known to exhibit significant heterogeneity or when a statistical test of such heterogeneity is desired. They have the disadvantage of requiring a priori knowledge for partitioning sites. We recommend: (i) selection of models by using backward elimination rather than AIC or AICc, (ii) use a stringent cut-off, e.g., p = 0.0001, and (iii) conduct sensitivity analysis of results. With thoughtful application, fixed-effect codon models should provide a useful tool for large scale multi-gene analyses.

A new infectious disease model for estimating and projecting HIV/AIDS epidemics

Objectives As the global HIV pandemic enters its fourth decade, countries have collected longer time series of surveillance data, and the AIDS-specific mortality has been substantially reduced by the increasing availability of antiretroviral treatment. A refined model with a greater flexibility to fit longer time series of surveillance data is desired. Methods In this article, we present a new epidemiological model that allows the HIV infection rate, r(t), to change over years. The annual change of infection rate is modelled by a linear combination of three key factors: the past prevalence, the past infection rate and a stabilisation condition. We focus on fitting the antenatal clinic (ANC) data and household surveys which are the most commonly available data source for generalised epidemics defined by the overall prevalence being above 1%. A hierarchical model is used to account for the repeated measurement within a clinic. A Bayesian approach is used for the parameter estimation. Results We evaluate the performance of the newly proposed model on the ANC data collected from urban and rural areas of 31 countries with generalised epidemics in sub-Sahara Africa. The three factors in the proposed model all have significant contributions to the reconstruction of r(t) trends. It improves the prevalence fit over the classic Estimation and Projection Package model and provides more realistic projections when the classic model encounters problems. Conclusions The proposed model better captures the main pattern of the HIV/AIDS dynamic. It also retains the simplicity of the classic model with a few interpretable parameters that are easy to interpret and estimate.

A stochastic infection rate model for estimating and projecting national HIV prevalence rates.

Background Every 2 years, the Joint United Nations Programme on HIV/AIDS (UNAIDS) produces probabilistic estimates and projections of HIV prevalence rates for countries with generalised HIV/AIDS epidemics. To do this they use a simple epidemiological model and data from antenatal clinics and household surveys. The estimates are made using the Bayesian melding method, implemented by the incremental mixture importance sampling technique. This methodology is referred to as the ‘estimation and projection package (EPP) model’. This has worked well for estimating and projecting prevalence in most countries. However, there has recently been an ‘uptick’ in prevalence in Uganda after a long sustained decline, which the EPP model does not predict. Methods To address this problem, a modification of the EPP model, called the ‘r stochastic model’ is proposed, in which the infection rate is allowed to vary randomly in time and is applied to the entire non-infected population. Results The resulting method yielded similar estimates of past prevalence to the EPP model for four countries and also similar median (‘best’) projections, but produced prediction intervals whose widths increased over time and that allowed for the possibility of an uptick after a decline. This seems more realistic given the recent Ugandan experience.