Infectious diseases and Control
Presenter: Kylie Ainslie
When: Tuesday, July 12, 2016 Time: 11:00 AM - 12:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Estimation of the effectiveness of influenza vaccination from household studies
Influenza vaccination is recommended every season due to changes in influenza virus types, subtypes, and phenotypes from one season to the next. The variation in the influenza virus requires the production of a new vaccine each season, thus vaccine effectiveness (VE) must be estimated each year. VE is defined as one minus the risk ratio (RR), where risk is defined as the probability of getting infected throughout the influenza season. Since 2010, influenza vaccination is recommended for all individuals over the age of 6 months in the U.S., making randomized clinical trials (RCT) unethical. Observational studies are being increasingly used to estimate influenza VE. Recent studies have followed households rather than separate individuals to determine VE against influenza transmission from the household compared to VE against transmission from the community. We present a likelihood approach to estimate vaccine-related protection against transmission of influenza from the household and from the community. We use symptomatic influenza, defined as infection with the influenza virus that develops into an acute respiratory infection (ARI), as our outcome of interest and allow for vaccination to occur at any time within the study period. Previous methods require vaccination to occur prior to the study period. However, because infection source cannot actually be observed, we present an estimation strategy using data in which infection source is unknown to estimate source-specific VE (i.e., VE against household transmission and VE against community transmission) by estimating source-specific daily transmission probabilities. Estimates will be validated via stochastic simulations of an influenza season. Advantages of our approach include: (a) estimation of source specific VE, (b) accommodation of vaccination during the season, and (c) lack of reliance on the assumptions required by the proportional hazards model.
Infectious diseases and Control
Presenter: Ahmadou Alioum
When: Tuesday, July 12, 2016 Time: 11:00 AM - 12:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Estimating HIV incidence from HIV diagnoses surveillance data: a penalized likelihood approach
The availability of accurate estimates of HIV incidence in recent periods is important for tracking the epidemic as well as for evaluating the impact of prevention programs. A virological monitoring of new HIV diagnoses with quantification of two biomarkers of infection was implemented in France since 2003. Assuming that the dynamics of the markers from the time of infection is known from other sources, the use of information on behaviors towards HIV testing before HIV diagnosis, clinical stage and markers values at the time of HIV diagnosis can help to improve estimation of the infection time for newly HIV diagnosed individuals. We propose a penalized likelihood approach to estimate smooth HIV incidence from HIV diagnoses surveillance data. The choice of the smoothing parameter is achieved using an approximated cross-validation criterion. We investigate the performance of the proposed approach through a simulation study. The method is illustrated to estimate HIV incidence among men who have sex with men in France.
Infectious diseases and Control
Presenter: Francesco Brizzi
When: Tuesday, July 12, 2016 Time: 11:00 AM - 12:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Age-Specific Back-calculation to Estimate HIV Incidence
Bayesian multi-state back-calculation is an important tool for monitoring the HIV epidemic amongst men who have sex with men (MSM) in England and Wales [1], with latest estimates suggesting non-decreasing, although highly uncertain, incidence in the most recent years. Targeted public health interventions require, however, the identification of population sub-groups at increased risk of infection. In particular, whether incidence is distributed equally, or concentrated within, specific age-groups. Here we propose a development of the back-calculation model in [1], also extending existing age-specific back-calculation approaches [2], to allow characterisation of the infected population through various disease stages stratified by time and age. Epidemic evolution is described by a population level multi-state model combining three distinct processes: infection incidence, disease progression, diagnosis. Both time and age specific incidence and time-dependent diagnosis probabilities can be estimated, using data on new HIV and AIDS diagnoses over time together with information on CD4 cell count around diagnosis. Modelling and estimating the latent (age and time-specific) HIV incidence is challenging. We adopt a semi-parametric formulation using bivariate splines to discover that resulting estimates depend on the type of splines employed (thin plate or tensor product splines) and on the choice of smoothing parameters. A further challenge arises from the non-standard form of the multi-state back-calculation likelihood, which cannot be expressed as a GLM (as in [2]). We explore estimation using both penalized likelihood [2] and Bayesian [1] approaches. A relation exists between the two approaches, as the penalty term can be interpreted as a prior for the model parameter. In the likelihood framework, cross validation, routinely used to estimate the smoothing parameter, cannot be efficiently implemented in our non-GLM setup. So in this situation, we resort to a complex approximation of the AIC [3] to estimate the smoothing parameters. The Bayesian approach allows for a more natural estimation of the smoothing parameters and incorporation of model uncertainty. Both approaches are assessed in a simulation study, before applying them to data for MSM in England and Wales. References: [1]Birrell, PJ, Lancet Ifect. Dis. 13(4), 313-8(2013) [2]Marschner, IC, Stat. in Med. 17.9(1998):1017-1031 [3]Wood, SN, arXiv 1511.03864
Infectious diseases and Control
Presenter: Hiroshi Nishiura
When: Tuesday, July 12, 2016 Time: 11:00 AM - 12:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Determining the end of an epidemic with human-to-human transmission
Declaring the end of an epidemic would enable allocation of human resources to healthcare facilities to return to normal and would help restore international travel to the country. The present study was aimed to objectively calculate the probability of observing additional cases at a given time. To clearly define the end of the outbreak, we excluded reintroduction of imported cases and infections resulting from a zoonotic reservoir. We defined the end of the outbreak as the end of continued chains of transmission. The probability of observing additional cases was derived by using the serial interval; that is, the time from illness onset in the primary case-patient to illness onset in a secondary case-patient, and the transmissibility of the disease. The end of outbreak could be declared if that probability is <5%, a threshold value. Considering that the use of the incubation period distribution would be fully supported only when the exact times of infection were known for exposed potential contacts, an objective decision of the end of an outbreak should explicitly rest on the risk of observing at least 1 more case on or after a specified date, as was shown here.
Infectious diseases and Control
Presenter: Jan Van den Broel
When: Tuesday, July 12, 2016 Time: 11:00 AM - 12:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
A non-homogeneous proportional-odds model for outbreak data
The early stages of a disease outbreak might be dominated by an infectious disease process, yet soon the spread becomes more complex due to at least three other factors. First, the contact process can change during the outbreak due to different behavior of the individuals in the population, or due to measures taken. Control measures are taken to reduce the number of contacts but also to limit the reproduction ability of infected individuals. Second, it can be difficult to determine the population at risk at the start of the outbreak. Moreover, during the outbreak the population at risk can change due to measures taken. Finally, measurements during an outbreak are often limited to whether an individual was infected in the past, disregarding times of entering or leaving a compartment. To address these factors a non-homogeneous birth model is proposed as an approximation to this complicated data generating process. This leads to a negative binomial model for the number of new cases in a discrete time interval. This model depends on the non-homogeneous hazard (reproductive power), by which events can occur. It is illustrated that a log link for the mean leads to a proportional odds model for the reproductive probability. The over-dispersion parameter for the negative binomial model is the cumulative number of previous cases. Since this number increases with time, the model will approximate a Poisson model when time goes on. This model captures important aspects of the data generating process. The birth process can give a good description of the early stage of the outbreak. Secondly, it does not need a definition of the population at risk because reproduction is modeled.Third, the ability to reproduce is characterized in the model by the reproductive power (hazard). This probability can change over time, which is crucial because it can be influenced by other dynamics besides the infectious disease outbreak. Moreover, a variation in susceptibility of individuals can show an infection rate that is declining over time because highly susceptible individuals tend to be infected earlier. Also, the probability of observing new cases depends on the cumulative number of previous cases . As a last aspect, the model depends on the prevalence and not on the exact infection times.The proposed model is demonstrated using data from the H5N1 outbreak among poultry in Thailand.
